SESSIONS AND TALKS DAY BY DAY - EURO...When the slower transportation mode is less-polluting, higher inventory levels need to be kept. Both the choice for a clean but slow transportation

SESSIONS AND TALKS DAY BY DAY

Wednesday, 9:00-10:15

WA-01: Opening and Plenary: Alexander Rinnooy Kan (see page 26) (Rotterdam Hall) . . . . . . 49


WB-03: Sustainable Development (Leeuwen Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50WB-04: Auctions and Game Theory (Leeuwen Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51WB-05: Managerial Accounting (Goudriaan Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52WB-06: Neural Networks and Fuzzy (Goudriaan Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53WB-07: Logistics scheduling (Antwerp Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54WB-08: Scheduling and Assignments (Penn Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55WB-09: ATO systems and SCM (Diamond Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57WB-10: Perishable Inventory Management (Diamond Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58WB-11: TSP and Vehicle Routing (Penn Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60WB-12: Routing I (Van Oldenbarnevelt Room ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61WB-13: Rolling Stock Rescheduling (Mees Auditorium) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62WB-14: Decision support systems (Tokyo Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64WB-15: Networks I (New York Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65WB-16: GOR Master Awards (Oscar Auditorium) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66WB-17: Statistical methods (New Orleans Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67


WC-03: Electricity Markets (Leeuwen Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70WC-04: Decision Making (Leeuwen Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71WC-05: Kidney exchange (Goudriaan Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73WC-06: Decision Support for Market Analysis and Risk Management (Goudriaan Room 2) . . 74WC-07: Maritime Shipping (Antwerp Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76WC-08: Resource allocation in practice (Penn Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77WC-09: Maintenance Planning I (Diamond Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78WC-10: Inventory I (Diamond Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .80WC-11: Supply Chain Optimization I (Penn Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81WC-12: Supply Chain Contracts (Van Oldenbarnevelt Room ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83WC-13: Optimization Methods for Vehicle Routing (Mees Auditorium) . . . . . . . . . . . . . . . . . . . . 84WC-14: Decision Analysis (Tokyo Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86WC-15: Networks II (New York Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87WC-16: GOR PhD Awards (Oscar Auditorium) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

45


WC-17: Queueing systems (New Orleans Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90


WD-01: Semi-plenary speaker: Miguel Anjos (Rotterdam Hall) . . . . . . . . . . . . . . . . . . . . . . . . . . . 92WD-09: Semi-plenary speaker: Fred Espen Benth (Diamond Room 1) . . . . . . . . . . . . . . . . . . . . . . 92WD-13: Semi-plenary speaker: Horst Tempelmeier (Mees Auditorium) . . . . . . . . . . . . . . . . . . . . 92WD-16: Semi-plenary speaker: Henk Zijm (Oscar Auditorium) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93


WE-03: Enhanced Energy Network Models (Leeuwen Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . 94WE-04: Economic Modeling (Leeuwen Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95WE-05: Facility Location in Healthcare Management Systems (Goudriaan Room 1) . . . . . . . . . 96WE-06: Text Mining and Clustering (Goudriaan Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98WE-07: Hinterland Container Logistics (Antwerp Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100WE-08: Scheduling (Penn Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101WE-09: Meta-Heuristics and Multi-criteria Programming (Diamond Room 1) . . . . . . . . . . . . . . 102WE-10: Inventory II (Diamond Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104WE-11: Supply Chain Optimization II (Penn Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105WE-12: Production and Distribution Planning (Van Oldenbarnevelt Room ) . . . . . . . . . . . . . . . . . 107WE-13: Railway Networks, Routing and Scheduling (Mees Auditorium) . . . . . . . . . . . . . . . . . . 108WE-14: Multi-objective optimization (Tokyo Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109WE-15: Graphs and Networks (New York Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111WE-16: Young Participant with Most Academic/Practical Impact Awards (Oscar Auditorium) 112WE-17: Resource allocation (New Orleans Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

Thursday, 9:00-10:30

TA-03: Energy Storages: Location and Operation (Leeuwen Room 1) . . . . . . . . . . . . . . . . . . . . . 115TA-04: Option Pricing and Market Models (Leeuwen Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . 117TA-05: Ambulance planning (Goudriaan Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118TA-06: SAMS-solvers (Goudriaan Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119TA-07: Scheduling Applications I (Antwerp Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120TA-08: Novel problems and methodological advances in response to needs of practitioners

(Penn Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122TA-09: Planning & control in the process industry (Diamond Room 1) . . . . . . . . . . . . . . . . . . . . 123TA-10: Coordination and Demand Fulfillment (Diamond Room 2) . . . . . . . . . . . . . . . . . . . . . . . . 125TA-11: Robust Supply Chain Management (Penn Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126TA-12: Crowdedness, Evacuation and Inland Shipping (Van Oldenbarnevelt Room ) . . . . . . . . . 128TA-13: Integer programming (Mees Auditorium) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129TA-14: Vehicle Routing with Time Windows (Tokyo Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

46


TA-15: Robust communication networks (New York Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132TA-17: Nonlinear Optimization I (New Orleans Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

Thursday, 11:00-12:30

TB-02: Capacity Control (Veder Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136TB-03: Environmental Management (Leeuwen Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137TB-04: Financial Modeling, Optimization and Risk (Leeuwen Room 2) . . . . . . . . . . . . . . . . . . . 138TB-05: Healthcare modelling and optimization (Goudriaan Room 1) . . . . . . . . . . . . . . . . . . . . . . 139TB-06: SAMS-modelling (Goudriaan Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140TB-07: Scheduling Applications II (Antwerp Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142TB-08: Job Shop Scheduling (Penn Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143TB-09: Maintenance Planning II (Diamond Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144TB-10: Demand Modeling (Diamond Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146TB-11: Supply Chain Optimization III (Penn Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147TB-12: Facility Location and Logistics (Van Oldenbarnevelt Room ) . . . . . . . . . . . . . . . . . . . . . . . 148TB-13: Decomposition methods (Mees Auditorium) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149TB-14: Airline Applications (Tokyo Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150TB-15: Branch-and-Price (New York Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151TB-17: Conic Optimization (New Orleans Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

Thursday, 13:00-13:45

TC-01: Roundtable Discussion (see page 28/29) (Rotterdam Hall) . . . . . . . . . . . . . . . . . . . . . . . . 154

Thursday, 14:00-14:45

TD-01: Semi-plenary speaker: GOR Science Prize (Rotterdam Hall) . . . . . . . . . . . . . . . . . . . . . . 155TD-09: Semi-plenary speaker: Kalyan Talluri (Diamond Room 1) . . . . . . . . . . . . . . . . . . . . . . . . 155TD-10: Semi-plenary speaker: Andrea Lodi (Diamond Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . 155TD-12: Semi-plenary speaker: Leo Kroon (Van Oldenbarnevelt Room ) . . . . . . . . . . . . . . . . . . . . . 156TD-13: Semi-plenary speaker: Dick den Hertog (Mees Auditorium) . . . . . . . . . . . . . . . . . . . . . . 156

Thursday, 15:15-16:45

TE-02: Dynamic Pricing (Veder Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157TE-03: Modeling Gas Transportation: Yesterday, Today, and Tomorrow (Leeuwen Room 1) . 158TE-04: Topics in Covered Bond Modeling (Leeuwen Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . 159TE-05: Home healthcare routing; ambulance planning (Goudriaan Room 1) . . . . . . . . . . . . . . . . 161

47


TE-06: Public Transport and Bike Sharing (Goudriaan Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . 162TE-07: Scheduling Applications III (Antwerp Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163TE-08: Resource Constrained Scheduling (Penn Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164TE-09: Production and Inventory control (Diamond Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166TE-10: Closed-loop supply chains (Diamond Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167TE-11: Real Life Applications in Supply Chain Management (Penn Room 2) . . . . . . . . . . . . . . 168TE-12: Warehousing (Van Oldenbarnevelt Room ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170TE-13: Combinatorial Optimization (Mees Auditorium) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171TE-14: Vehicle Routing Applications and Network Design (Tokyo Room) . . . . . . . . . . . . . . . . 173TE-15: Traveling Salesman Problem and Branch-and-Bound (New York Room) . . . . . . . . . . . 174TE-17: Nonlinear Optimization II (New Orleans Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

Friday, 9:00-10:30

FA-02: Revenue Management in Non-Service Industries (Veder Room) . . . . . . . . . . . . . . . . . . . 177FA-03: Simulation and System Dynamics (Leeuwen Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178FA-04: Portfolio Optimization (Leeuwen Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180FA-05: Healthcare planning and scheduling (Goudriaan Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . 181FA-06: SAMS-decision-support (Goudriaan Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182FA-07: Container Terminal Operations (Antwerp Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184FA-08: Staff Scheduling (Penn Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186FA-09: Queuing and Scheduling (Diamond Room 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187FA-10: Perishability in Supply Chain Management (Diamond Room 2) . . . . . . . . . . . . . . . . . . . . 189FA-11: Supply Chain Design (Penn Room 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191FA-12: Routing II (Van Oldenbarnevelt Room ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191FA-13: Railway Optimization (Mees Auditorium) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193FA-14: Distribution, Pickup&Delivery, Dial-a-Ride (Tokyo Room) . . . . . . . . . . . . . . . . . . . . . . . 194FA-15: Mixed Integer Programming (New York Room) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196FA-17: Nonlinear Optimization and Applications (New Orleans Room) . . . . . . . . . . . . . . . . . . . . 197

Friday, 11:00-11:45

FB-01: Semi-plenary speaker: Rommert Dekker (Rotterdam Hall) . . . . . . . . . . . . . . . . . . . . . . . . 199FB-09: Semi-plenary speaker: Stein W. Wallace (Diamond Room 1) . . . . . . . . . . . . . . . . . . . . . . 199FB-12: Semi-plenary speaker: Nelly Litvak (Van Oldenbarnevelt Room ) . . . . . . . . . . . . . . . . . . . 199FB-13: Semi-plenary speaker: Erik Demeulemeester (Mees Auditorium) . . . . . . . . . . . . . . . . . . 200

Friday, 12:00-13:00

FC-01: Closing and Plenary: Steve Sashihara (see page 27) (Rotterdam Hall) . . . . . . . . . . . . . . 200

48


� WA-01Wednesday, 9:00-10:15Rotterdam Hall

Opening and Plenary: Alexander Rinnooy Kan (see page 26)Stream: Plenary SessionsPlenary sessionChair: Albert Wagelmans, Econometric Institute, Erasmus University Rotterdam, POBox 1738, 3000DR , Rotterdam, Netherlands, [email protected]

WB-03


� WB-03Wednesday, 10:45-12:15Leeuwen Room 1

Sustainable DevelopmentStream: Energy and EnvironmentInvited sessionChair: Laura Elisabeth Hombach, School of Business and Economics, Chair of OperationsManagement, RWTH Aachen, 52056, Aachen, Germany, [email protected]

1 - Transportation Costs and Carbon Emissions in a Vendor Managed Inventory Situ-ationMarcel Turkensteen, CORAL, Economics, Aarhus University, Fuglesangs Alle 4, 8210, AarhusV, Denmark, [email protected], Christian Larsen

Recently, there has been large focus on carbon emissions and fuel consumptions from road transport.When the slower transportation mode is less-polluting, higher inventory levels need to be kept. Both thechoice for a clean but slow transportation mode and consolidation of shipments requires higher inventorylevels because of longer intervals between deliveries. In this paper, we consider a vendor who decides onthe degree to which deliveries to geographically dispersed retailers should be consolidated. The vendorhas the option to ship to several retailers during a joint delivery; such a group of retailers is called a zone.Such a consolidation of a shipment has the advantage that frequent joint deliveries can be made, but has thedisadvantage that each delivery tour is long and overall emissions can be high. On the other hand, deliveriesto small groups of retailers are less frequent, but have shorter delivery tours and may have lower carbonemissions. For different values of the transportation costs per kilometer, we determine a minimum costsolution for the vendor using a Joint Replenishment Problem approach. Carbon emissions are computedfrom both the daily driven distance and the ton kilometers using an advanced engine emission model. Wefind that, for products with frequent demand, an increase in transportation costs leads to lower fuel usageand carbon emissions, as it is not costly to increase inventory levels at the retailer. However, for productswith infrequent demand, this is so costly that the optimal strategies and hence, carbon emissions, remainmore or less unchanged. The policy implications are that measures such as road tariffs and greener vehiclesmay have a different effect on vendors of fast moving and slow moving products.

2 - Simulation based evaluation of strategies for the market diffusion of alternativedrivetrain technologiesLaura Elisabeth Hombach, School of Business and Economics, Chair of OperationsManagement, RWTH Aachen, 52056, Aachen, Germany, [email protected],Grit Walther

In 2012 the transportation sector was the second largest contributor to worldwide CO2 emissions, androad transportation was with 75% the leading emitter within this sector. Additionally, depletion of finiteresources like fossil oil by road transportation is an important topic. To deal with these external effects oftransportation, alternative drivetrain technologies like battery electric, fuel cell or gas vehicles are intro-duced into the marked. However, the market diffusion depends on customer acceptance and manufacturerstrategies, which both are uncertain and can profoundly be influenced by political instruments. Further-more, the economic and ecological impact of manufacturers’ strategies and political instruments is notknown so far. Thus, the aim of this contribution is to evaluate manufacturers’ strategies and politicalinstruments for market introduction and diffusion of new drivetrain technologies with regard to their eco-nomic and ecological impact. In order to do so, we consider the interaction of political decision makers,manufacturers and customers using a hybrid agent-based and system dynamics model.

50

WB-04

3 - Measurement of Risk for Wind Energy Projects: A Critical Analysis of Full LoadHoursAndré Koukal, Leibniz Universität Hannover, Institut für Wirtschaftsinformatik, Germany,[email protected], Stefan Lange, Michael H. Breitner

In scientific literature, profitability analyses of on- and offshore wind energy projects as well as assessmentsof general conditions for such projects usually make use of full load hours to determine the annuallyproduced energy. Furthermore, it also serves for the calculation of the project value and other financial keyfigures. This procedure leads to accurate results, if only the expected value of each parameter is taken intoaccount. However, it is difficult to choose an adequate type of distribution and to set suitable distributionparameters for the aggregate key figure of full load hours when risks of a project are considered. This isthe case probability distributions are applied for the parameters. In this paper, a different approach usingthe more basic parameter of the average wind speed and a Weibull distribution in combination with thetechnical availability and other discounts is provided. It aims at estimating the annual energy productionby simultaneously taking uncertainties into account. This approach is integrated into a discounted cash-flow (DCF) model on which a Monte Carlo simulation is applied. Finally, a case study for a fictitiousoffshore wind park in the German North Sea is conducted. It is shown that the application of the presentedapproach leads to more precise distributions of the outcomes than the standard analysis with full load hours.

� WB-04Wednesday, 10:45-12:15Leeuwen Room 2

Auctions and Game TheoryStream: Game Theory and Experimental EconomicsInvited sessionChair: Martin Bichler, TU München, Germany, [email protected]

1 - Pricing and Coordination in Ascending Combinatorial AuctionsZhen Hao, Informatics, Technical University of Munich, I 18, Boltzmannstr 3, 85748, Munich,Germany, [email protected], Martin Bichler

Combinatorial auctions have received a lot of attention recently due to their relevance for industrial pro-curement and logistics. Bayesian models of ascending combinatorial auctions focus on situations withincomplete information and known bundles where bidders know with certainty which bundles other bid-ders are interested in. One of the main strategic challenges in most applications is that bidders do notknow which bundles are of interest to other bidders. In such situations, the coordination problem is central.We propose a combinatorial auction format where we leverage information on losing coalitions which iscollected by the auctioneer. In addition to ask prices, which constitute lower bounds on new bids, theauctioneer provides coalitional winning levels. The latter inform bidders about bundle bids which com-plement the interests of others and enable promising coalitions. The properties of the new auction formatare analyzed theoretically and in lab experiments. One of the main results is that using the new auctionformat the communication in terms of number of rounds and bids between auctioneer and bidders is muchreduced without hurting efficiency or revenue of the auction.

2 - Mechanism design with value-maximizing biddersSalman Fadaei, TU München, Germany, [email protected], Martin Bichler

We discuss auction design for bidders which maximize valuations for particular allocations or overall. Wewill refer to such bidders as value-maximizing bidders or short value bidders. Examples are markets forTV ads sold by a TV station, where bidders are media agencies who are given a budget by clients andtheir goal is to win the most preferred allocation. The valuations are decided a priori and devoted to aparticular campaign and the bidder’s task is to maximize the value of the TV slots. Similar assumptionshave been met for Internet ad auctions. The environment is different from mechanism design without

51

WB-05

money, because bidders actually submit bids expressing their cardinal value. It is also different from quasi-linear mechanism design as bidders maximize value, not payoff. We analyze mechanism design goals suchas Pareto efficiency and social welfare maximization and discuss when strategy-proof mechanisms arepossible. Although, strategy-proof and social welfare maximizing mechanisms are impossible for generalvaluations, we show that simple first-price multi-unit auctions satisfy strategy-proofness in the large. Inaddition, we will show some environments, where strategy-proof approximation mechanisms are available.

3 - Is the solidarity value close to the equal split value?Tadeusz Radzik, Institute of Mathematics and Computer Science, Wroclaw University ofTechnology, Wybrzeze Wyspianskiego 27, Wroclaw, Poland, [email protected]

An inspiration for this paper was a short remark (about two TU-values for cooperative games) found in thearticle of Hart and Mas-Colell (Bargaining and value. Econometrica, 64 (1996)), which can be summarizedto the following: "Under a minor boudedness condition, for large seize n of the ground coalition N, thesolidarity value (introduced by Nowak and Radzik in (1994)) is close to the equal split value (awardingeach player by the same payoff v(N)/n in a game v). However Hart and Mas-Colell did not explain whatthey understood under "a minor boudedness condition" and "one value is close to other value". Just this iswidely discusseed in our paper.

We study the asymptotic behavior of the solidarity value and the equal split value in different classes ofcooperative games, and their mutual "convergence". In particular, we discuss three natural definitions ofasymptotic equivalence of values for TU-games, and identify for each of them the classes of games forwhich the solidarity value and the equal split value are asymptotically equivalent. Besides, a computerillustration of the obtained results is given.

4 - Spectrum Auction Design - Simple Auctions For Complex SalesMartin Bichler, TU München, Germany, [email protected], Jacob Goeree, Stefan Mayer, PashShabalin

Recently, a number of countries have introduced the combinatorial clock auction (CCA) to sell spectrum.However, the number of possible packages grows exponentially with the number of licenses, which addscomplexity to the auction. We analyze the impact of two main design choices that spectrum auctiondesigners face: "compact" bid languages versus "fully expressive" bid languages and "pay-as-bid" paymentrules versus "bidder-optimal core-selecting" payment rules. We consider these design choices both forascending and sealed-bid formats. We find that simplicity of the bid language has a substantial positiveimpact on the auction’s efficiency and that simplicity of the pricing rule has as a significant impact on theauction’s revenue.

� WB-05Wednesday, 10:45-12:15Goudriaan Room 1

Managerial AccountingStream: Managerial AccountingInvited sessionChair: Katja Schimmelpfeng, Lehrstuhl für Beschaffung und Produktion, Universität Hohenheim,Schloss Osthof-Nord, 70599, Stuttgart, Germany, [email protected]

1 - Commercial efforts of dealers to reach the product efficiency frontier and intensi-fication of competition in the Spanish auto marketEduardo Gonzalez, Business Administration, University of Oviedo, Av Cristo s(n, 33071,Oviedo, Asturias, Spain, [email protected], Juan Ventura, Ana Cárcaba

52

WB-06

This paper investigates the relationship between the dynamics of price discounts at the dealership leveland to product efficiency in the Spanish auto market. Using Data Envelopment Analysis (DEA), productefficiency scores are estimated for 2092 different vehicles commercialized during 2010, using an innovativedatabase that accounts for more than 75 technical attributes of each model. By alternating the official andthe discounted prices within the DEA specification, we are able to propose a measure of product efficiencyimprovement in the marketplace. This measure is further decomposed into two indexes that measure the"commercial effort’ of the dealer and the "intensification of competition’ that results from the discountingpractices of the set of all dealers. Finally, we explore the relevance of some drivers of dealership discounts,with the focus on product efficiency. As expected, the results confirm a strong inverse relationship betweenproduct efficiency and dealer discount. Also as expected, discounts are larger for generalist brands, agedmodels and gasoline engines. In contrast, car size is found to be positively related to the magnitude ofdiscounts, which was an unexpected result, although it has also been reported in previous research aboutthe European auto market.

2 - The effect of depreciation rules on residual income: The case of overlapping ca-pacity investmentsChristian Lohmann, Bergische Universität Wuppertal, Juniorprofessur für Controlling,Gaußstraße 20, D-42119, Wuppertal, Germany, [email protected] examine a setting where the owner of a company delegates the authority to make overlapping capacityinvestments to an impatient manager. If the manager’s internal interest rate exceeds the owner’s cost ofcapital, a discrepancy between the owner’s marginal cost of capacity and the manager’s perceived marginalcost arises and induces capacity underinvestment by the manager. The underinvestment incentive of themanager can be prevented by using the performance measure residual income in conjunction with particulardepreciation rules, e.g. the relative practical capacity (RPC) depreciation rule. First, we examine theeffect direction of a deviation from the RPC depreciation rule on the average historical cost of capacitywhich is a component of the performance measure residual income. Second, we analyze the magnitudeof the distortion of average historical cost of capacity if the straight-line depreciation rule and the annuitydepreciation rule are used. Thereby, we characterize the bias in the average historical cost of capacitybetween these depreciation rules. Additionally, we investigate the effect of partial direct expensing on theaverage historical cost of capacity. Finally, we analyze the development of the average historical cost ofcapacity during the initiation phase of a depreciation rule which is more or less accelerated than the RPCdepreciation rule.

3 - Judgmental Forecasting and Production Decisions: Incentives and NegotiationsClint Pennings, Rotterdam School of Management, Erasmus University, Netherlands,[email protected] forecasting is central to the forecasting process at many companies, but is subject to manybiases. An experiment was conducted in which respondents forecast demand and provide input to a meet-ing to examine these biases. Psychological traits and attitudes are examined to analyze differences inunintentional biases; different departmental roles and incentives (asymmetrical cost functions) are used toinfluence intentional biases. Insights gained are based on various measures of operational performanceand forecast accuracy, and learning is explicitly taken into account. Another experiment was conducted toexamine negotiation behaviors.

� WB-06Wednesday, 10:45-12:15Goudriaan Room 2

Neural Networks and FuzzyStream: Information Systems, Neural Nets and Fuzzy SystemsInvited sessionChair: Hans Georg Zimmermann, Corporate Technology CT RTC BAM, Siemens AG,Otto-Hahn-Ring 6, 81730, München, Germany, [email protected]

53

WB-07

1 - New Applications of Neural Networks in IndustryRalph Grothmann, Corporate Technology CT RTC BAM LSY, Siemens AG, Otto-Hahn-Ring 6,81730, München, Germany, [email protected], Hans Georg Zimmermann,Christoph TietzWe give an overview of recent developments in the field of neural networks for time series forecating anddiagnosis. The focus will be on time-delay recurrent neural networks. We also deal with the application ofthe neural network models in the energy and industry sector of Siemens.

2 - Advances in Forecasting with Recurrent Neural NetworksHans Georg Zimmermann, Corporate Technology CT RTC BAM, Siemens AG, Otto-Hahn-Ring6, 81730, München, Germany, [email protected], Ralph Grothmann,Christoph TietzRecurrent neural networks (RNNs) are typically considered as relatively simple architectures, which comealong with complicated learning algorithms. This talk has a different view: We start from the fact thatRNNs can model any high dimensional, nonlinear dynamical system. Rather than focusing on learningalgorithms, we concentrate on the design of network architectures. Unfolding in time is a well-knownexample of this modeling philosophy. Here a temporal algorithm is transferred into an architectural frame-work such that the learning can be performed by an extension of standard error backpropagation.

We will also deal with the application of RNNs in recent projects carried out at Siemens Corporate Tech-nology.

3 - Evaluating Employer Attractiveness: development of a Fuzzy Rule-based SystemJessica Naundorf, Department of Economics & Management, Otto-von-Guericke-UniversityMagdeburg, Chair of Strategic Management and Organisation, P.O. Box 41 20, 39 016,Magdeburg, Germany, [email protected] is a war going on: Companies compete to attract the most talented employees as ensuring a sustainedsupply with human resources that match company needs is a precondition for competitiveness. This "warfor talent’ causes the necessity for companies to be viewed as highly attractive employer by potential em-ployees. Therefore, employer branding and the development of a positive employer image play a vital rolein intending to influence job choice decisions of applicants. One way of creating and transporting a spe-cific employer image is to take part in so-called employer awards. In this paper, we understand employerawards as a competition between a limited number of employers that proactively take part in the contest,aiming at creating and/or shaping a positive employer image. The degree of attractiveness of a companyas an employer is thereby subjectively assessed by third parties. Employer awards need to comply withrequirements of ranking criteria and ranking procedure to be an adequate signal of employer attractiveness.The aim of this paper is to develop a transparent and traceable, multi-criteria rule-based system to deter-mine employer attractiveness by making use of the concept of fuzzy logic. Thereby both, criteria valuesand weighted criteria hierarchy can be mirrored best by referring to fuzzy-based thinking. We develop anevaluation system that is based on fuzzy linguistic variables and fuzzy rules. The evaluation system enablesus to measure and display employer attractiveness in a traceable and objective way. Our rule-based systemprovides an opportunity for business practice to better deal with decision making complexity. Furthermore,it can serve as groundwork for future quantitative-empirical research on this topic.

� WB-07Wednesday, 10:45-12:15Antwerp Room

Logistics schedulingStream: Scheduling and Project ManagementInvited sessionChair: Nils Boysen, Lehrstuhl für ABWL/ Operations Management, Friedrich-Schiller-UniversitätJena, Carl-Zeiß-Str. 3, 07743, Jena, Germany, [email protected]

54

WB-08

1 - Layout planning of sortation conveyors in parcel distribution centersStefan Fedtke, Friedrich-Schiller-University Jena, 07743, Jena, Germany,[email protected]

Fully automated sortation processes play a crucial role in modern distribution networks of the parcel serviceindustry. In this context, this study systematically investigates different design alternatives of closed-loop tilt tray sortation conveyors. In particular, the number of loading stations for feeding parcels into aconveyor system and its relation to the subordinate decision problem of assigning inbound and outbounddestinations to docking doors, is shown to considerably influence the throughput of a hub terminal. Anovel destination assignment problem is formalized, appropriate solution procedures are presented and, bymeans of simulation, the resulting throughput is quantified, so that an experienced terminal manager canweigh the operational gains against the investment cost of different conveyor layouts.

2 - Complexity results for some basic yard crane scheduling problemsKonrad Stephan, Friedrich-Schiller-University Jena, 07743, Jena, Germany,[email protected]

Rail-road transshipment yards constitute essential nodes in intermodal transportation networks. In thesefacilities, huge gantry cranes are the bottleneck resource when transshipping containers between freighttrains and trucks. To speed-up container handling crane schedules with minimal completion time forexecuting all container moves are of special interest. By differentiating two different metrics for cranemovements (Manhattan and Maximum metric), different transshipping directions (from rail to road, viceversa and both directions), and different policies for determining the trucks’ parking positions, this talkprovides complexity results for some basic crane scheduling problems with a single gantry crane.

3 - The berth allocation problem with mobile quay wallsNils Boysen, Lehrstuhl für ABWL/ Operations Management, Friedrich-Schiller-Universität Jena,Carl-Zeiß-Str. 3, 07743, Jena, Germany, [email protected], Dirk Briskorn, Simon Emde

The berth allocation problem (BAP), which defines a processing interval and a berth at the quay wall foreach ship to be (un-)loaded, is an essential decision problem for efficiently operating a container port.In this talk we integrate mobile quay walls into the BAP. Mobile quay walls are huge propelled floatingplatforms, which encase ships moored at the immobile quay and provide additional quay cranes for accel-erating container processing. Furthermore, additional ships can be processed at the seaside of the platform,so that scarce berthing space at a terminal is enlarged. We formalize the BAP with mobile quay walls andprovide suitable solution procedures.

� WB-08Wednesday, 10:45-12:15Penn Room 1

Scheduling and AssignmentsStream: Scheduling and Project ManagementInvited sessionChair: Dirk Briskorn, University of Siegen, Germany, [email protected]

1 - Approximation Algorithms for Multi-Dimensional Vector Assignment ProblemsFrits Spieksma, Operations Research and Business Statistics, Katholieke Universiteit Leuven,Naamsestraat 69, B-3000, Leuven, Belgium, [email protected], TrivikramDokka, Yves Crama

55

WB-08

We consider multi-dimensional vector assignment (MVA) problems which form a special class of ax-ial multi-dimensional assignment problems. In a vector assignment problem, one is given the followingingredients: m disjoint sets, each of which contains the same number n of p-dimensional vectors withnonnegative integral components; a component-wise operation on sets of vectors; and a cost function de-fined on vectors. The problem is to partition these m sets of vectors into n m-tuples so that no two vectorsfrom the same set are in the same m-tuple and so that the total cost of the m-tuples is minimized. Themain motivation comes from a yield optimization problem in semi-conductor manufacturing. We considertwo classes of polynomial-time heuristics for MVA: sequential heuristics (among which the heaviest-firstheuristics) and the hub heuristics, and we study their performance ratio. In particular, we show that whenthe cost function is monotone and submodular, sequential heuristics, as well as hub heuristics, have finiteperformance ratio for every fixed m. Moreover, we establish a better ratio for heaviest-first heuristics whenthe cost function is additive. We provide examples to illustrate tightness of our analysis. Furthermore, weshow that the MVA problem is APX-hard even for the case m = 3 and binary vectors. Finally, we show thatthe problem can be solved in polynomial time in the special case of binary vectors with fixed dimension p.

2 - Battery scheduling under various objectivesThijs van der Klauw, EEMCS, University of Twente, Netherlands, [email protected],Johann Hurink, Gerhard Post, Gerard Smit

We consider the problem of scheduling an energy storage device to match a given energy (production)profile to a predefined target profile. This problem has applications in e.g. smart grids where we can useenergy storage devices to control the output of otherwise uncontrollable energy sources such as wind andPV. The problem shows some similarities with warehouse scheduling problems. Depending on the type ofenergy storage device the use conditions which have a negative effect on the aging of the device differ. Thisleads to different objectives for the scheduling of the storage device. We present the resulting problems asMIPs and show the boundary between polynomial solvable and NP-hard problems. We present results forthe problem of minimizing the number of charging cycles or minimizing the total energy flow through thebattery. We also investigate the possibility of using multiple batteries simultaneously.

3 - An efficient algorithm for scheduling groups of jobs with individual release datesAlexander Lieder, Chair of production management, University of Mannheim, Schloss,Schneckenhof Ost, 68131, Mannheim, Germany, [email protected], Dirk Briskorn, RaikStolletz

We present a new optimization algorithm that schedules groups of identical jobs on identical parallel ma-chines. The jobs can have release dates, limited processing time windows and sequence-dependent setuptimes. Possible objectives include minimal make-span, total (weighted) delay, and maximum delay ofa job. The developed algorithm is a dynamic program which exploits dominance criteria to reduce thecombinatorial complexity of the problem. The development of this approach is motivated by the "aircraftscheduling problem" that is to minimize the total delay costs of runway operations at an airport while en-suring that the necessary separation time between each pair of operations on the same runway is met. Ournumerical study demonstrates that, for a small number of different groups, problem instances with up to100 jobs can be solved to optimality within seconds, instead of hours which standard MIP formulationsmay take.

4 - Meeting Project Deadlines under Uncertainty ?How (and When) to EncourageHelp?Fabian Sting, Technology & Innovation Management Department, Erasmus University,Rotterdam School of Management, Netherlands, [email protected], Pascale Crama, Yaozhong Wu

Despite sophisticated project management methods, companies struggle with costly project delays. Whilecooperative behavior has been identified to be a critical factor for on-time project completion, it has notbeen explicitly embedded into project management (PM) systems. Inspired by an innovative real-lifenew product development management practice, we model a PM system that incorporates and shapesproject managers’ cooperative behavior. Help is at the core of this system in which project managers maymutually ask for and provide help, while top management formally monitors the exchange of help. Wefind that the company should take a nuanced approach to the provision of help. In projects with a verylow cost of effort, help is allowed to occur informally, that is, without top management’s monitoring. For

56

WB-09

higher costs, the company benefits from shaping the compensation structure to monitor and reward help.Against all intuition, the company may prefer to avoid cooperative behavior in projects for which earlycompletion benefits taper off quickly with decreasing earliness. Finally, it is striking that a natural orderamong projects emerges, even with symmetrical projects: cooperative behavior may lead the company tochoose an equilibrium in which projects receive a different level of effort. This asymmetry also explainsthe persisting value of help even under correlated task uncertainties across projects.

� WB-09Wednesday, 10:45-12:15Diamond Room 1

ATO systems and SCMStream: Production and Operations ManagementInvited sessionChair: Willem van Jaarsveld, Econometrie, Erasmus universiteit rotterdam, Burgemeester Oudlaan 50,Room H07-11, NL 3000DR, Rotterdam, Netherlands, [email protected]

1 - Demand Fulfillment in an Assemble-to-Order (ATO) production systemSebastian Geier, Chair for Sustainable Operations and Logistics, University of Augsburg,Universitätsstraße 16, 86159, Augsburg, Deutschland, Germany,[email protected], Bernhard Fleischmann

We consider a computer manufacturer who assembles customized final products from various components.Customer orders specify the product configuration, the quantity and a desired delivery date. They arriveduring the day and have to be confirmed instantly. The "online order promising’ (OP) process must an-nounce a "first promised delivery date’ to the customer. Demand fulfillment in this ATO case is still littleinvestigated and differs remarkably from the more popular Make-to-Stock (MTS) case: Bottlenecks are theassembly capacity and the stocks of components, which are "available to promise’ (ATP). The componentshave a n:m-relationship to the orders. During order fulfillment time, which is typically longer than in theMTS case, various unforeseen events may happen, such as faulty material supply, machine breakdownsor arrival of urgent new orders. Therefore, an important task of the demand fulfillment, besides OP, is"Demand Supply Matching’ (DSM), i.e. deciding on the assembly date of orders and eventually changingthe delivery date of promised orders ("repromising’). We present a new concept for demand fulfillment inthe ATO case which consists of online OP for single orders arriving during the day and DSM once a day,linked in a rolling-horizon scheme. The DSM is based on a MIP model which simultaneously determinesassembly and delivery dates for all promised orders. Constraints consider the assembly capacity, the stockand planned supplies of components, as well as the interdiction of split deliveries. Objectives are to mini-mize delivery delay regarding the desired or first promised date and the number of repromisings. We reporton a case study with real data of a computer manufacturer with 20,000 orders per day and 2,000 differentcomponents.

2 - Pre-shipment Financing: Credit Capacities and Supply Chain ConsequencesAnne Lange, Department of Business Policy and Logsitics, University of Colgone,Albertus-Magnus-Platz, 50923, Köln, Germany, [email protected], Fehmi Tanrisever,Matthew Reindorp

We study a supply chain where a wealthy retailer buys from a debt-constrained supplier who cannot in-ternally finance his entire production operations. The retailer commits to a minimum purchase quantity tofacilitate pre-shipment financing, which enables the supplier to extend his debt capacity and thereby alsohis production level. Thus, our work fits with the recently developing literature stream on the operations-finance-interface. Moreover, while pre- and post-shipment financing solutions have attracted increasedattention in the industry during the recent economic crisis, this paper is the first contribution to analyze thepotential of pre-shipment financing jointly considering operational and financial constraints.

57

WB-10

Key parameters of our model are the retailer’s creditworthiness, which determines the extent to which thecommitment will relax the supplier’s financial constraint, the supplier’s independent financial resources,and the demand characteristics of the product market. Using a newsvendor model of operations, we analyzethe strategic interaction of the two players as a Stackelberg game. In equilibrium, we illustrate that theretailer’s profits are increasing in her own creditworthiness and in her supplier’s debt capacity. Hence, shehas an interest to collaborate with a financially strong supplier. In contrast, we find that the supplier maybe at a disadvantage when doing business with a highly creditworthy retailer, since the latter may reduceher purchase order commitment and extract profit from the supplier. We highlight numerically that thesupplier’s debt capacity and the retailer’s creditworthiness are generally substitutes for the retailer but arecomplements for the supplier in medium value ranges.

3 - Optimization of industrial-scale assemble-to-order systemsWillem van Jaarsveld, Econometrie, Erasmus universiteit rotterdam, Burgemeester Oudlaan 50,Room H07-11, NL 3000DR, Rotterdam, Netherlands, [email protected], AlanScheller-Wolf

We provide insights to improve inventory control in industrial-sized Assemble-To-Order (ATO) systems.By developing a novel stochastic programming (SP) formulation, we develop an algorithm that has unpar-alleled efficiency and scalability. Applying our algorithm on several numerical examples, we generate newinsights with respect to the control and optimization of industrial-sized ATO systems.

We consider a continuous time model, seeking base-stock levels for components that minimize the sum ofholding costs and product-specific backorder costs. Our initial focus is on first-come first-serve (FCFS)allocation of components to products; for this setting our algorithm quickly computes solutions that areprovably within one percent of the optimal base-stock/FCFS policy. We then turn to two related questions:How do common heuristics used in practice compare to our performance, and how costly is the FCFSassumption.

For the first question, we investigate the effectiveness of ignoring simultaneous stockouts (ISS), a heuristicthat has been used by companies such as IBM and Dell to optimize inventories. We show that ISS perfor-mance, when compared to the optimal FCFS base-stock policy, improves as the average newsvendor (NV)fractiles increase. In addition, lead time demand correlations have an adverse impact on ISS performance.For the second question, we adapt the SP formulation of Dogru, Reiman and Wang (2010), yielding anefficiently computable upper bound on the benefit of optimal allocation over FCFS. In summary, we findthat for many large ATO systems, FCFS performs surprisingly well, and that its performance improveswith decreasing NV fractile asymmetry among products and, again, with increasing average NV fractiles.

� WB-10Wednesday, 10:45-12:15Diamond Room 2

Perishable Inventory ManagementStream: Production and Operations ManagementInvited sessionChair: Stefan Minner, TUM School of Management, Technische Universität München, Arcisstrasse 21,80333, Munich, Germany, [email protected]

1 - Production planning of a perishable food product with non-stationary demand ina rolling planning horizonKarin G.J. Pauls-Worm, Operations Research and Logistics, Wageningen UR, Hollandseweg 1,6706KN, Wageningen, Netherlands, [email protected], René Haijema, Eligius M.T. Hendrix,Jack van der Vorst

58

WB-10

Planning the production of a perishable food product can be a challenging task, as products have a shortshelf life, while demand is highly uncertain (erratic) and service levels have to be met. For food safetyreasons, equipment has to be cleaned thoroughly when switching between different products. Possibly,machines have to be adjusted and settings have to be changed. Given the significant costs involved, eco-nomic reasons trigger long production runs, but the perishable nature of food products require short runsto reduce product waste.

In previous work we have shown an MILP model that approximately solves the production-inventorydecision for a fixed planning horizon (of 12 periods) for a product with a fixed maximum shelf life of 3periods. Given initial inventory levels the MILP model finds an order policy that aims at a non-stock-outprobability of at least 0.95. The model corrects for the average waste, but the actual waste depends on theactual inventory levels. We apply this model in a rolling planning horizon. We execute a simulation studybased on a case of a cheese producer. Both the MILP model and the simulation results are presented fordifferent parameter settings. The results indicate that the approach is well suited for practical use, as thesolution is close to optimal and the solution process is fast using an existing solver (like Gurobi or Cplex).

2 - Assortment planning and inventory management under stock-out based substitu-tion for perishable products.Rob Broekmeulen, OPAC, TU Eindhoven, P.O. Box 516, Pav. E10, 5600 MB, Eindhoven, -,Netherlands, [email protected], Karel van Donselaar

In this research we study a combined assortment planning and inventory optimization problem for per-ishable products in a retail environment which is characterized by stock-out based demand substitution.While many papers have been published on the effects of demand substitution for non-perishable products,this research addresses these effects for perishable products specifically. We develop a fast heuristic whichenables the joint determination of optimal assortments and optimal reorder levels. We use this heuristicto run a full factorial experiment with 46,656 different perishable inventory systems in which we vary 10system parameters in a systematic way. We evaluated the impact of demand substitution on expected profit,outdating, assortment size and product availability, but we also show for which system parameters substi-tution has the biggest impact and we explain why. Due to demand substitution the profit in our numericalstudy increases on average 7.3%, the amount of outdating decreases 40.7%, the assortment size decreases12.8% and the product availability increases 3.1%. The impact of demand substitution on profit is largefor product subcategories with a large assortment size, low average total demand for the assortment, andsmall product lifetimes. The increase in profit due to demand substitution only depends in a minor wayon the outdating costs and lost sales costs. The amount of outdating can be reduced most from demandsubstitution in situations with assortments consisting of many slow moving products having small productlifetimes, especially if the outdating costs are relatively low.

3 - New heuristic policies for managing perishable inventoriesStefan Minner, TUM School of Management, Technische Universität München, Arcisstrasse 21,80333, Munich, Germany, [email protected]

The optimal replenishment policy for perishable items is complex and therefore can typically only beobtained by stochastic dynamic programming. Therefore, several simple policies have been proposed inthe literature. The base-stock and the constant order policy are the two most well-known ones. Based on theobservation that each one can outperform the other and realizing the reasons for performance deterioration,we propose a new combined policy that applies a constant order size for a certain range of inventorystates and modified base-stock policies otherwise. A numerical study illustrates the improvements of thiscombined policy compared to the two pure policies.

59

WB-11

� WB-11Wednesday, 10:45-12:15Penn Room 2

TSP and Vehicle RoutingStream: Traffic and TransportationInvited sessionChair: Achim Koberstein, Business Administration, Goethe-University of Frankfurt, Grueneburgplatz1, 60323, Frankfurt am Main, Germany, [email protected]

1 - Core-based Cost Allocation in Cooperative Traveling Salesmen Problem withRolling HorizonIgor Kozeletskyi, Mercator School of Management, University of Duisburg-Essen, Lotharstr 65,47057, Duisburg, Germany, [email protected], Alf Kimms

This paper presents a cost allocation scheme for the horizontal cooperation that is implemented a prioriand provides expected costs for the coalition members. The cost allocation is determined using the coreconcept. To compute the value of characteristic function over the whole planning horizon the TSP withrelease dates combined with simulation is used. The developed core computation algorithm, based onmathematical programming techniques, provides a core element or, in case of an empty core, a least coreelement. To decrease the computational effort of core computation a row generation procedure is imple-mented. The developed computation study tests the computational performance of the solution procedureand compares obtained results with the results of a previous study, where the Shapley value was used.

2 - The Driver Assignment Vehicle Routing ProblemRemy Spliet, Econometric Institute, Erasmus University Rotterdam, Burgemeester Oudlaan 50,3000DR, Rotterdam, Netherlands, [email protected], Rommert Dekker

We introduce the problem of assigning customers to drivers before the quantity to be delivered to thesecustomers is known. We consider a set of demand scenarios, and for each scenario a delivery schedule hasto be made which minimizes the transportation costs while satisfying the vehicle capacity constraints. Fur-thermore, the delivery schedules per scenario should be such that at least a given fraction of the customersthat are assigned to a driver is actually visited by that driver, where the fraction is provided by the decisionmaker. The driver assignment vehicle routing problem, DAVRP, is to assign customers to drivers such thatthe expected transportation costs over all scenarios are minimized. To solve this problem, we propose acluster first-route second heuristic. We use it to find good solutions to the DAVRP for instances with upto 100 customers and instances with up to 100 scenarios. In our computational experiments we study thecosts of adhering to the driver assignments. We compare the costs of always having a customer visited bythe same driver, with the costs of relaxing this requirement entirely. Such an analysis aids a policy makerin determining whether it is worthwhile to require customers to be visited by the same driver. Furthermore,using two variants of the cluster first-route second algorithm, we study the increase in transportation costsof only constructing new routes with customers that cannot be visited by their assigned drivers, instead oftrying to assign them to different drivers.

3 - When Vehicle Routing needs Project Scheduling . . .Jörn Schönberger, University of Bremen, 28359, Bremen, Germany, [email protected]

Vehicle routing and scheduling problems (VRSP) form an important subset of Resource ConstrainedProject Scheduling Problems (RCPSP). In VRSP time-related constraints like time windows or prece-dence among loading and unloading operations assigned to a certain vehicle are investigated but temporaldependencies among operations assigned to different vehicles do not occur. Compared to the RCPSP, thescheduling of operations is less challenging since each route-specific scheduling can be made individually.We introduce the VRSP with Coordinated Operation Starting Times (VRSP-COST). Two types of requestsmust be served. A regular request is assigned to exactly one vehicle but a flexible request is assigned totwo vehicles. A customer site associated with a flexible request is visited by two vehicles. It is necessaryto coordinate the operations of these two vehicles at the customer locations, i.e. time constraints between

60

WB-12

starting times of the (un)loading operations must be kept. In a typical applications it is necessary to sched-ule the arrival times of all vehicles at a customer node within a period whose length does not exceed agiven threshold (e.g. two vehicles must be unloaded in short succession in order to avoid costly cleaning orrefitting of the handling equipment). We investigate the situation of the VRSP-COST in which the secondvehicle serving a flexible request must arrive at the associated customer location within the implicit timewindow implied by the vehicle arriving firstly at the customer. We integrate basic RCPSP-techniques intotypical vehicle routing algorithms in order to address the operation starting time coordination for flexiblerequests. Results from comprehensive computational experiments are presented.

4 - A Heuristic Solution Approach for the Hybrid Electric Vehicle Traveling SalesmanProblemChristian Doppstadt, Logistics and Supply Chain Management, Goethe University Frankfurt,Grueneburgplatz 1, 60323, Frankfurt, Germany, [email protected], AchimKoberstein, Daniele VigoThe usage of Hybrid Electric Vehicles (HEVs) is one currently ongoing endeavor to reduce exhaust gasemission, especially, within urban areas. Although, hybrid electric passenger cars are on the rise, the devel-opment of small transporters and trucks is far behind. The reason for this is basically that the economicalbreak-even point has not been reached yet and ecological motives are second-ranked. The Hybrid ElectricVehicle Traveling Salesman Problem (HEVTSP) is an extension of the well-known Traveling SalesmanProblem (TSP) and considers a HEV for the delivery tours. This allows us to examine economical as-pects for the usage of hybrid electric transporters and determine the point when they become profitable.Moreover, additional questions can be answered, e.g., whether a plug-in hybrid is reasonable for a specifictour profile or not. For the HEVTSP we assume four different modes of operation: mere combustion,mere electric, charging while driving with combustion engine and combined combustion electric as a boostmode. The modes differ in costs and required travel time for each arc and a maximal tour duration has tobe met. The assumption of four different modes of operation makes the problem very difficult to solve,as the number of possible solutions for the already NP-hard TSP is increased by a factor of four times thenumber of cities to be visited. To be able to handle practical problem sizes, we develop a heuristic solutionmethod based on Variable Neighborhood Search (VNS) and having the ability to deal with an exchange ofthe different modes of operation. Our solution approach is tested on benchmark instances, generated basedon actual data of delivery tours from a large Courier Express Parcel (CEP) company.

� WB-12Wednesday, 10:45-12:15Van Oldenbarnevelt Room

Routing IStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Leendert Kok, Ortec Software Development, ORTEC, P.O. Box 490, 2800AL, Gouda,Netherlands, [email protected]

1 - Column generation based heuristic for three dimensional vehicle routing problemSatyaveer S. Chauhan, John Molson School of Business, Concordia University, MB 11-317,1455 de Maissoneuve, Blvd. west, H3G 1M8, Montreal, Quebec, Canada,[email protected], Batoul Mahvash, Anjali AwasthiIn this work we present a vehicle loading problem that consists of packing a given set of items into aminimum number of vehicles for delivery. A heuristic approach based on Dantzig-Wolfe decompositionis developed and tested on the widely available data set. A greedy search procedure is developed to speedup the column generation approach. A problem specific branching technique is used to generate integersolutions in a reasonable time. Numerical experimentation is done to compare the performance of thedeveloped approach with the available results in the literature.

61

WB-13

2 - Vehicle routing with redistributing equipmentChagiet Bloemendal, Software Development, ORTEC, Biesbosch 101, 1181HZ , Amstelveen,Netherlands, [email protected], Leendert Kok, Joaquim Gromicho

We consider a VRP motivated by practice. A company does quality control on the milk produced byfarmers. For this quality control milk samples need to be collected. These milk-samples are obtained byspecial equipment. This equipment needs to be delivered at the farmer before the samples can be collected.Generally the equipment is needed at the farmer for multiple days. Each vehicle departs from the homelocation of its driver, resulting in a heterogeneous vehicle fleet. Furthermore there are multiple depots,where the vehicles as well as the equipment for a specific farmer are linked to a single depot. In additionto the collection of milk samples, there are some other tasks like delivering sperm that are not linked to adepot. As the equipment is expensive, the total amount of equipment is limited. For efficient use of themeasuring equipment, it is transported directly between farmers as much as possible. Since this may notalways be possible or efficient, equipment can be transported from and to the depot. However, the capacityat the depot is limited. This restricts the sum of the start capacity of all routes. Though the time restrictionscan encompass multiple days, we propose to solve the VRP per day and take the time restrictions intoaccount by distinguishing between tasks that have to be planned on the current day and tasks that can beplanned on the current day but are also allowed on a succeeding day. We propose to use a combination ofrecursive parallel cheapest insertion and local search to solve this VRP. The reason for the recursion is thatit can occur that the delivery of equipment cannot be planned initially because there is not enough stock atthe depot, but once a pickup is planned in the route, the delivery can be planned after this pickup.

3 - Routing in practice: current major challengesLeendert Kok, Ortec Software Development, ORTEC, P.O. Box 490, 2800AL, Gouda,Netherlands, [email protected]

Automated vehicle route planning is a major challenge in practice. On the one hand, vehicle route planningmethods should become faster to cope with growing problem sizes. On the other hand, there is a strongwish for more elaborate vehicle routing models to cope with more real-life restrictions. Restrictions thatrequire synchronization between vehicle routes, such as a limited loading capacity at a depot, are partic-ularly challenging, for at least two reasons. First, an effective way to speed up vehicle routing methodsis to use estimations for changes to the vehicle route plan, before evaluating them exactly. For example,when evaluating a move of a customer, the local change in distance may give a proper estimation of thequality of that move with respect to other moves. However, when synchronization constraints are present,a change in one route may lead to substantial changes in other routes. Therefore, coming up with strongestimations is in general difficult when synchronization constraints are present. Second, this propagationeffect of synchronization constraints may make exact evaluations of changes to the plan very expensive. Inthe worst case, a small change leads to a re-evaluation of the entire solution. In addition to synchronizationconstraints, some other constraints have a big impact on solution methods for vehicle routing problems.For example, companies that transport animal feed face contamination constraints, such that certain ordersmay not succeed directly in the same vehicle. However, placing an extra order in between may be feasible.This makes the sequence in which a solution is built up very important. In this talk, I will give an overviewof the most important routing challenges we currently face in practice.

� WB-13Wednesday, 10:45-12:15Mees Auditorium

Rolling Stock ReschedulingStream: Traffic and TransportationInvited sessionChair: Thijs Dewilde, Centre for Industrial Management/Traffic & Infrastructure, KU Leuven,University of Leuven, Celestijnenlaan 300A, bus 2422, B-3001, Leuven, Belgium,[email protected]

62

WB-13

1 - Maintenance in the Rolling Stock Rescheduling ModelJoris Wagenaar, Rotterdam School of Management, Erasmus University Rotterdam, P.O. Box1738, 3000 DR, Rotterdam, Netherlands, [email protected], Leo Kroon

In this paper we propose a formulation to take maintenance into account for disruption management mod-els for passenger railway transportation. This formulation is tested on real life instances from NetherlandsRailways. Unfortunately, disruptions disturb the daily operations of Netherlands Railways almost everyday. During a disruption part(s) of the railway infrastructure can no longer be used, so the planned sched-ules become infeasible. To uphold as much of the passenger service as possible, a new rolling stockschedule should be created as soon as possible. A selected amount of the allocated rolling stock unitsare scheduled to have a maintenance check somewhere during the operations, at one of the preselectedmaintenance locations. Note that rolling stock units are no longer allowed to be employed for passengertransportation after a threshold amount of time without a maintenance check. So, it is of utmost importanceto take maintenance into account while rescheduling, or else the train units scheduled for maintenance aremost likely not in time for their inspection and are possibly no longer allowed to be used during the oper-ations. Currently, maintenance is ignored during the first steps of the rescheduling phase and the rollingstock schedule is in a later stadium slightly adjusted to take maintenance into account. This procedure isfar from optimal. That is why we propose an integrated formulation to take maintenance right from thestart of the disruption into account.

2 - Ultra-short term railway rolling stock reschedulingGabor Maroti, Logistics, Innovation and Information, VU University Amsterdam, De Boelelaan1105, 1105 HV, Amsterdam, Netherlands, [email protected]

In this talk we consider the problem of rescheduling the rolling stock of a passenger railway operatorwith an extremely short planning horizon. This problem arises when the train operator is faced witha severe weather condition for the next day. As a preventive measure, the operator decides to run analternative, slightly reduced timetable. The adjusted timetable requires large-scale changes to the rollingstock schedules.

The objective of the ultra-short term rolling stock rescheduling is twofold. On one hand, the availablecapacity is to be distributed is such a way that it meets the passenger seat demand as well as possible.On the other hand, one tries to minimise the consequences of the adjusted rolling stock schedules on theshunting processes.

The main challenge of the studied problem is its time-pressure. There is less than a day available forcreating the adjusted schedules and for implementing them in practice.

We propose a sequence of optimisation models to solve the ultra-short term rolling stock reschedulingproblem of Netherlands Railways (NS). These models proved their value in practice in the winter of 2013by helping NS to deal with the cold and snowy weather.

3 - Re-optimization of Rolling Stock RotationsThomas Schlechte, Optimization, Zuse-Institute-Berlin, Takustrasse 7, 14195, Berlin, Berlin,Germany, [email protected], Markus Reuther, Ralf Borndörfer

The Rolling Stock Rotation Planning Problem is to schedule rail vehicles in order to cover timetabled tripsby a cost optimal set of vehicle rotations. The problem integrates several facets of railway optimization, i.e.,vehicle composition, maintenance constraints, and regularity aspects. In industrial applications existingschedules often have to be re-optimized to integrate timetable changes or construction sites. We present anintegrated modeling and algorithmic approach for this task as well as computational results for industrialproblem instances of DB Fernverkehr AG.

4 - Adapting the train schedule due to planned infrastructure unavailability in com-plex railway stationsThijs Dewilde, Centre for Industrial Management/Traffic & Infrastructure, KU Leuven,University of Leuven, Celestijnenlaan 300A, bus 2422, B-3001, Leuven, Belgium,[email protected], Peter Sels, Dirk Cattrysse, Pieter Vansteenwegen

63

WB-14

In this paper, we consider the influence of planned infrastructure works like maintenance actions that blockcertain parts of the infrastructure. In case a certain part of a single track is blocked, busses may be usedto fulfill the service since otherwise rides have to be canceled which is very inconvenient for travelers.In situations where alternative routes exist, we can consider rerouting actions but often this is not enoughto solve the problem for all trains. In this paper, we delay or reschedule some trains and reroute otherswhile minimizing the number of cancelations (primary objective) and maintaining the level of robustnessas much as possible (secondary objective).

In previous work [Dewilde et al. (2013) Robust Railway Station Planning: an Interaction between Routing,Timetabling and Platforming, in: IAROR RailCopenhagen, pp 1-20.], we developed an algorithm to im-prove the robustness of a train schedule in the surroundings of large and complex railway stations. Usingthat algorithm, we succeeded in improving the robustness of a real-life schedule restricted to the neigh-borhood of a large and complex station with —according to our simulation model- more than 8% whichresulted in a decrease in the amount of knock-on delay with 33%. For the purpose of the current research,we reworked this algorithm without losing its efficiency. As a consequence, one can quickly evaluate theimpact of planned infrastructure changes and update the schedule accordingly. The first computationalresults indicate that cancelations can be avoided for an instance in which 14 of the 84 considered trains areharmed. This goes at a robustness cost of 0.7% and a 5% increase in the amount of knock-on delay.

� WB-14Wednesday, 10:45-12:15Tokyo Room

Decision support systemsStream: Decision Analysis and Multiple Criteria Decision MakingInvited sessionChair: Alexander Scherrer, Optimization, Fraunhofer Institute for Industrial Mathematics (ITWM),Germany, [email protected]

1 - Measuring criticality of elements based on multi-criteria decision makingDennis Stindt, Department of management sciences, University of Augsburg, Universitätsstr. 16,86159, Augsburg, Afghanistan, [email protected]

Corporations and economies are increasingly dependent on a growing number of elements used as rawmaterials. A shortage in availability of certain elements would significantly impact those systems. Mid-and long-term availability, including the relevant factors that influence availability, is described throughthe concept of criticality. This concept is widely discussed from academic as well as practitioners’ stand-point due to the according risks that emanate from potential shortages. Nevertheless, criticality and thepatterns of criticality are not adequately defined within this relatively young field of research. State-of-the-art studies name different biophysical, social and economic factors that influence criticality, which arequalitative and quantitative in nature. These factors have to be aggregated in order to derive a judgmentabout the criticality. The logic of measurement as well as aggregation of factors is mostly non-transparentor over-simplified. Concluding, a structured methodology that accounts for the specific characteristics andprovides transparency of the decision-making process is missing. Therefore, the article presents a method-ology for the assessment of criticality based on multi-criteria decision making (e.g. AHP). The decisionproblem itself is characterized by multiplicity of factors ranging from production rates to political stability.In addition, the decision process should be flexible and able to reflect uncertainties, limitations of knowl-edge as well as the decision makers’ preferences. Several methods of multi-criteria decision making arediscussed and assessed regarding their appropriateness for the stated issue. The article concludes with ademonstration of the proposed method focusing on a certain element.

64

WB-15

2 - A decision support system for advanced breast cancer treatment planningAlexander Scherrer, Optimization, Fraunhofer Institute for Industrial Mathematics (ITWM),Germany, [email protected], Patrick Ruediger, Andreas Dinges,Karl-Heinz Kuefer, Hans Hagen, Ilka Schwidde, Sherko Kuemmel

Breast cancer is the most common and mortal carcinosis among women. The clinician plans the nexttreatment steps for a patient case based on its previous course and current status according to standardtherapy protocols and similar former cases. This process requires the time-consuming search for suitabletreatment options and a thorough decision making on these options in order to come up with the best pos-sible treatment for the individual patient. These tasks form quite a challenge in clinical routine with itsenormous and always present stress, implying urgent need for a decision support system, which facilitatesthe time-efficient conception of high-quality breast cancer therapy. This talk presents the joint work ofmathematicians and clinicians on the way towards this ultimate goal. The major aspects in the character-ization of patient cases (patient age, tumor type, . . . ) and the indications and quality measures (10 yeardeath rates, . . . ) of treatment options yield the different-type criteria (ordinal, nominal, . . . ) underlying tothe decision problem. The clinical understanding of similarity between cases and applicability of treatmentoptions transposes into pseudo-metrics on the criteria and their suitable scalarization up to measures forthe overall relevance of a treatment option or former case for the current one. The algorithmic level withflexible and efficient search routines initialized with the current patient case builds upon this mathematicalmodel. The operational procedures of clinical routine are converted into a decision support functionality,which features the conduction of suitable search runs, the intuitive visualization of search results and theirfurther processing by the clinician in terms of treatment planning and case documentation.

� WB-15Wednesday, 10:45-12:15New York Room

Networks IStream: Discrete and Combinatorial Optimization, Graphs and NetworksInvited sessionChair: Marco Lübbecke, Operations Research, RWTH Aachen University, Kackertstraße 7, 52072,Aachen, Germany, [email protected]

1 - Single-Commodity Network Design with Uncertain DemandsFrauke Liers, Department Mathematik, FAU Erlangen-Nuremberg, Cauerstrasse 1, 91058,Erlangen, Germany, [email protected], Valentina Cacchiani, Michael Juenger,Andrea Lodi, Daniel Schmidt

We study a single-commodity robust network design model with dynamic routing. The model can beinterpreted as an intermediate step between deterministic minimum cost flows and multi-commodity net-work loading. It has the benefit of allowing for a capacity-based integer-programming formulation whoseinequalities can be separated in polynomial time. We present polyhedral insights as well as a branch-and-cut algorithm that is based on the theoretical results. Evaluating the approach on random and realisticinstances, we show the effectiveness of our method.

2 - Improving flow solutions by path decompositionJan Peter Ohst, Mathematics, Universität Koblenz, Germany, [email protected], StefanRuzika

Network flows have become a common tool for many OR models such as evacuation planning. Techniquesto solve those problems are very well studied and optimal flows can be calculated efficiently. However torealize the calculated flow in reality the flow has to be decomposed into paths which then are assigned tothe units of flow. While it is straight forward to find an arbitrary decomposition it might not be unique andthe choice of the decomposition can have a tremendous effect on the properties of the single paths.

65

WB-16

For example consider an evacuation scenario where the overall exposure to a toxic substance is minimizedand an optimal flow is computed. For one decomposition of the optimal flow, every person might beexposed to nearly the same level of toxicity, while in another, some persons are exposed to an unacceptablyhigh degree. Note that although those two solutions are equivalent in terms of the total exposure to thetoxin, they are obviously very different in quality.

This example illustrates the necessity as well as the potential use of a very rarely addressed type of op-timization problem, which is optimizing over the set of all path decompositions of a given flow. In thistalk we focus on the problem of finding a decomposition minimizing the maximum costs of any path inthe decomposition. The problem has been shown to be NP-hard but has not been investigated in detail sofar. An integer program is formulated for this problem and an approximation algorithm based on a greedystrategy is developed and analyzed.

3 - A contraction-expansion algorithm for the capacitated min-cost network flowproblemMarco Lübbecke, Operations Research, RWTH Aachen University, Kackertstraße 7, 52072,Aachen, Germany, [email protected], Jacques Desrosiers, Jean-BertrandGauthier

The recently proposed "improved primal simplex" (IPS) algorithm only makes non-degenerate pivots tosolve linear programs (LPs). It therefore needs a finite number of iterations on integral data network flowproblems. We show that IPS is strongly polynomial for assignment and shortest path problems. Moreover,with a slight modification, it becomes strongly polynomial for the capacitated min-cost network flow prob-lem. The resulting algorithm is similar to the minimum-mean cycle canceling algorithm of Goldberg andTarjan (1989).

4 - Vertex coloring with bounds and distances constraints for minmax channel as-signment problemRosiane deFreitas, Institute of Computing, UFAM / UFRJ, Brazil, [email protected],Bruno Cardoso Dias, Nelson Maculan

This paper discusses about some coloring problems in graphs, based on the classical NP-hard problem ofvertex coloring with bounded lists of colors and distance constraints. Such characteristics are found in thechannel assignment in wireless mobile networks, which is explored in this work. An integer programmingformulation is presented, as well as exact and heuristic approaches. We have obtained better results thanthose available for CALM and Philadelfia benchmark instances.

� WB-16Wednesday, 10:45-12:15Oscar Auditorium

GOR Master AwardsStream: GOR AwardsInvited sessionChair: Kathrin Klamroth, Department of Mathematics and Informatics, University of Wuppertal,Gaussstr. 20, 42097, Wuppertal, Germany, [email protected]

1 - New Lower Bounds for the Three-Dimensional Strip Packing ProblemKirsten Hoffmann, Lehrstuhl für BWL, insb. Industrielles Management, Technische UniversitätDresden, Helmholtzstraße 10, 01062, Dresden, Germany, [email protected]

66

WB-16

In production and transportation, growing scarcity of resources and competition pressure will force com-panies into efficient management of raw material or storage area. Such optimisation problems can bemodelled as bin or strip packing problems. In this thesis, we study the three-dimensional strip packingproblem (SPP3) which involves packing a set of non-rotatable boxes into a three-dimensional strip (con-tainer) of fixed length and width but unconstrained height. The goal is to pack all of the boxes orthogonaloriented and without overlapping into the container, minimising its resulting height. The problem is NP-hard in the strong sense and, therefore, lower bounds are used to limit the optimal solution and to estimatethe performance of some heuristic solutions. We present new lower bounds based on geometric consider-ations and derived from different relaxations of the mathematical formulation of the SPP3. Furthermore,we show dominance relations between different bounds and limit the worst case performance ratio of somebounds.

2 - Two-stage robust combinatorial optimization with priced scenariosRoman Rischke, Technische Universität Berlin, 10623, Berlin, Germany,[email protected] applications of combinatorial optimization often require decision making under data uncertainty.Uncertainty in optimization problems is usually represented by a set of possible scenarios, where a scenariois a particular realization of the uncertain input parameters. Two-stage robust combinatorial optimizationis a methodology for handling combinatorial optimization problems with uncertain input. The goal of two-stage robust optimization problems is to construct a solution in a two-stage manner that is feasible in allscenarios, i.e., robust against uncertainty, and we want to minimize the worst-case cost of the constructedsolution. This brings up a central question: Which scenarios do we take into account? If we consider allpossible scenarios the robust solution is often too conservative and too expensive. Restricting the set ofpossible scenarios is a common approach, but this usually depends on subjective decision criteria like thewillingness to take risks or the expectation on the future. We propose an alternative concept. Instead ofrestricting the set of scenarios we price all scenarios, which affects the objective function. This leads to newtwo-stage robust combinatorial optimization problems. We study complexity and devise approximationalgorithms for such problems. We show that the complexity significantly depends on the pricing method.

3 - Time-dependent dynamic location and relocation of ambulancesLara Wiesche, Fac. of Management and Economics, Ruhr-University Bochum,Universitätsstraße 150, 44801, Bochum, Germany, [email protected] high quality medical services and ensuring accessibility to these services for the public is a keytask for health care systems. Given a limited budget the available resources, e. g. ambulances, have to beused economically in order to ensure a high quality coverage. Emergency vehicles have to be positionedand repositioned such that emergencies can be reached within a legal time frame. Especially the useof flexible ambulance locations, e. g. hospitals or voluntary fire departments, can be useful to reach asuitable coverage. Empirical studies have shown temporal and spatial variations of emergency demand aswell as variations of travel times during a day. The numbers of emergency calls with respect to the 24hours differ significantly between night and day and show peaks especially during rush hours. Existingmodels do not consider time-dependency of important model parameters as demand and travel times forEMS-vehicles. A mixed-integer linear program is formulated using time periods in order to model time-dependent demand and time-dependent travel times explicitly for reducing the uncertainty of unavailableambulances. It is shown on large empirical data records that the presented dynamic model outperformsexisting static models with respect to coverage and utilization of resources.

� WB-17Wednesday, 10:45-12:15New Orleans Room

Statistical methodsStream: Applied Probability and Stochastic Programming, ForecastingInvited sessionChair: Theo Berger, Empirical Economics and Applied Statistics, University of Bremen,

67

WB-17

Wilhelm-Herbst-Str. 5, Bremer Forum, Raum 0.16, 28359 , Bermen, Germany,[email protected]

1 - Galton VariationsIngo Althoefer, Mathematics and Computer Science, FSU Jena, Ernst-Abbe-Platz 2, 07743,Jena, Germany, [email protected], Petar Hofmann

The Galton box (also called bean machine) was invented and described by Francis Galton to demonstratethe binomial distribution. We did many experiments and found that in real world only few Galton boxeslead to good approximations of the binomial distribution.

Several variants of the classical box are presented, for instance boxes with different propagation rules forthe balls and 3-d Galton boxes.

Together with biologist J. Wöstemeyer, we sent non-flying Drosophila Melanogaster in a Galton box. Theflies can crawl forth and BACK in the Galton "labyrinth". Modelling their bidirectional movements inan abstract way gives a Galton box with the following properties: a particle starts in the top cell and isabsorbed in all bottom cells. Each normal inner cell has four neighbors: two below and two above. Cellsat the left and right borders have only three neighbors: two below and one above. The cell for the next stepis chosen by equal distribution; the particle makes independent steps. Main findings: Average time untilabsorption is about 0.6 * k * k, when depth k of the box is large. Outer bottom cells get much much higherprobability to be hit than in Binomial distribution. A slight paradox: Particles ending in outer bottom havein the average made more steps than those ending in central cells.

The "Galton Game" for two players was invented by I. Althöfer in 2012. He was inspired by the classicgame "Connect Four", but with three layers of nails above the seven columns. The winning condition is tomake four own balls in a row. The game is described online at http://www.althofer.de/galton-game.html .Monte Carlo agents play the Galton Game well.

2 - Adjustable Robust Parameter Design with Unknown DistributionsIhsan Yanikoglu, Tilburg University, Afghanistan, [email protected]

This article presents a novel combination of robust optimization developed in mathematical programming,and robust parameter design developed in statistical quality control. Robust parameter design uses meta-models estimated from experiments with both controllable and environmental inputs (factors). These ex-periments may be performed with either real or simulated systems; we focus on simulation experiments.For the environmental inputs, classic robust parameter design assumes known means and covariances, andsometimes even a known distribution. We, however, develop a robust optimization approach that uses onlyexperimental data, so it does not need these classic assumptions. Moreover, we develop ‘adjustable’ robustparameter design which adjusts the values of some or all of the controllable factors after observing thevalues of some or all of the environmental inputs. We also propose a new decision rule that is suitable foradjustable integer decision variables. We illustrate our novel method through several numerical examples,which demonstrate its effectiveness.

3 - On the Impact of Copulas on VaR Forecasts: A Simulation StudyTheo Berger, Empirical Economics and Applied Statistics, University of Bremen,Wilhelm-Herbst-Str. 5, Bremer Forum, Raum 0.16, 28359 , Bermen, Germany,[email protected]

Forecasting portfolio risk requires both, estimation of marginal return distributions for individual assets anddependence structure of returns as well. Due to the fact, that the marginal return distribution represents themain impact factor on portfolio volatility (Fantazzini (2009), Aloui et al. (2011)), the impact of dependencymodeling which is required for instance in the field of Credit Pricing, Portfolio Sensitivity Analysis orCorrelation Trading is rarely investigated that far.

Triggered by the introduction of copulas to financial time series (Embrechts et al. (2002)), flexible de-pendency approaches are recently applied as an alternative to the classical Markowitz Approach (1952).However, an ex-post separation between the isolated impact of misspecified margins and misspecified de-pendency on portfolio risk forecasts represents a staggering task in modern portfolio analysis.

In addition to Fantazzini (2009), who addresses the theoretical impact of misspecified margins on the es-timation of copula parameters, we explicitly focus on the impact of decoupled dependency modeling in

68

WB-17

the context of risk measurement. We do so, by setting up an extensive simulation analysis which enablesus to analyze competing copula approaches (Clayton, Frank, Gauss, Gumbel and t copula) under the as-sumption that the "true" marginal distribution is known. By simulating return series with different realisticdependency schemes accounting for time varying dependency as well as tail dependence, we show that thechoice of copula becomes crucial for VaR, especially in volatile dependency schemes. Albeit the Gausscopula approach does neither account for time variance nor for tail dependence, it represents a solid toolthroughout all investigated dependency schemes.

69

WC-03


� WC-03Wednesday, 12:45-14:15Leeuwen Room 1

Electricity MarketsStream: Energy and EnvironmentInvited sessionChair: Stephan Raths, High Voltage Technology, RWTH Aachen University, Schinkelstr. 2, 52056,Aachen, NRW, Germany, [email protected]

1 - Multi Level European Electricity Market Simulation using Network Flow Algorithmand Lagrangian RelaxationStephan Raths, High Voltage Technology, RWTH Aachen University, Schinkelstr. 2, 52056,Aachen, NRW, Germany, [email protected]

Due to a growing share of renewables in the European power system and other political decisions thereis a need for electricity network expansion. As grid utilization is based on market results, a realisticpower market simulation is crucial. The proposed approach determines the most cost efficient powerplant schedules for all European countries, taking into account the cross-border electricity exchange. Thesimulation is split into three stages: In the first stage the international electricity exchange between theEuropean countries is calculated by means of a linear network flow model, where each country representsa network node. The NTC (Net Transfer Capacities) are used as capacities for the network’s edges. The firststage considers time coupling ramping constraints and the maximum power output for thermal generators.Hydro storage is also integrated into the LP. Based on the first stage results the countries can be separatedfrom each other by summing up the national power demand and import/export in order to calculate thenational residual power demand to be covered by the national generators. Thus the second stage focuseson each national power system. Langrangian Relaxation in combination with MILP is applied in order tosolve a highly accurate UC and Economic Dispatch for each European country considering also spinningreserve requirements. In a third stage the integer decisions for all generators (obtained from stage two) arefixed and - based on the Unit Commitment — again an integrated European network flow (from stage one)is solved in order to consider the integer decisions within the energy exchange calculation. Final results arehourly schedules for each generator and price signals taken from the Lagrange Multipliers in the secondstage.

2 - A Stochastic-Global Optimization Approach to Electricity Market ModelMiray Hanım Yıldırım, Institute of Applied Mathematics, Middle East Technical University;Department of Industrial Engineering, Çankaya University; European Commission, JointResearch Centre, Institute for Energy and Transport, Institute of Applied Mathematics MiddleEast Technical University, 06531, Ankara, Turkey, [email protected], Gerhard-WilhelmWeber

Since many countries defined their pioneering policy as sustainable development, electricity market be-came vital in the whole energy system. While reforms and revolutions are established by policy makers,restrictions on greenhouse gas emissions, uncertainties in electricity demand and supply are formulated andexplored by scientists. The main source of randomness in demand and supply is the competitive marketstructure. In this study, we present a model that analyzes the restrictions and uncertainties in a competitiveelectricity market. In particular, a stochastic-global optimization framework incorporating numerical solu-tion methods is proposed. Therefore, we explore initial conditions of this global optimization problem byusing numerical approximation techniques. The model is implemented for a sample electricity network.The presentation ends with a conclusion and an outlook on future investigations and applications.

70

WC-04

3 - Forecating the distribution of hourly electricity spot pricesChristian Pape, Chair for Management Science and Energy Economics, UniversitätDuisburg-Essen, Universitätsstraße 12, 45117, Essen, NRW, Germany,[email protected], Christoph Weber, Oliver Woll

Short-term forecasting of hourly electricity prices is gaining importance because many market participantsin the energy industry are exposed to the risk of spot price variations. For the evaluation of optimal bid-ding and dispatch strategies or for the optimal use of power plants it is not only relevant to know theexpected price. Especially for risk management issues it is more important to assess possible price vari-ations and to get reliable interval price forecasts, e.g. to make right decisions about dynamic strategiesor to find the value of flexibility and optionality. Therefore, forecasting electricity spot prices in practicehas to meet three important requirements. (1) Capture the main characteristics of electricity spot pricedynamics to deliver useful and reliable forecasting results. (2) Deliver not only point forecasts but alsomultivariate price distributions for valuation and bidding purposes. (3) Establish a parameter estimatingand price simulation procedure that keeps time and effort small. This work presents an applicable mod-elling procedure to describe the dynamics of hourly electricity prices. A stepwise combination of severalmathematical operations is used to reach adequate distribution characteristics. One major element of thestochastic modelling procedure is to identify principal components of hourly electricity prices by usingan eigenvalue-eigenvector de-composition. Moreover, non-normality of residuals is accounted for by per-forming normal quantile transformation. Another crucial step within the modeling procedure is to specifythe stochastic process for the identified principal components.

� WC-04Wednesday, 12:45-14:15Leeuwen Room 2

Decision MakingStream: Game Theory and Experimental EconomicsInvited sessionChair: Erik Kropat, Department of Computer Science, Universität der Bundeswehr München,Werner-Heisenberg-Weg 39, 85577, Neubiberg, Germany, [email protected]

1 - A Multi-Objective Online Terrain Coverage ApproachMichael Preuß, OR, UniBw München, Werner-Heisenberg-Weg 39, 85577, Neubiberg,Germany, [email protected]

This paper introduces a new multi-objective optimization approach in the field of terrain coverage. Withthe help of the multi-objective online terrain coverage model, a decentralized autonomous swarm is able tocover an unknown environment. This innovative terrain coverage model has a high impact on autonomousvehicle applications because it enables to consider conflicting objective functions during the coverageprocess. That important improvement opens up new possibilities for real world applications. The designmethodology is based on combining an auction-based algorithm with a multiple ant colony optimizationroute planning algorithm. Experimental analyses are performed on the presented online terrain coveragemodel which includes for one thing the multi-objective route optimization and also for another thing asingle-objective route optimization. The analysis shows that a multi-objective approach can reduce thetotal exploration time and therefore the repeated coverage. Especially difficult obstacle structures can betreated by the presented multi-objective online terrain coverage model in a robust way.

2 - A Human Resource Model for Performance Optimization to Gain Competitive Ad-vantagesJoachim Block, Institut für Theoretische Informatik, Mathematik und Operations Research,Universität der Bundeswehr München, Werner-Heisenberg-Weg 39, 85577, Neubiberg,Germany, [email protected], Stefan Wolfgang Pickl

71

WC-04

Strategic management has to guarantee the existence of an organization on the long run. In an ever fasterchanging environment with growing globalization and rapid technology cycles, performance is a criticalsuccess factor. The core questions for strategic management are: how to counter depreciation of compe-tences and how to increase total performance in a sustainable fashion? Human resources are a cornerstonefor competitive advantages. This is especially true in the ongoing transformation process from blue towhite collar work and the entrance of generation Y into the workplace. Therefore, implementing efficientand effective human resource policies is one important aspect of strategic management. In contrast to poli-cies which aim to influence staff directly and on an individual level, designing a system which makes useof adaptive behavior of employees can result in a much more promising and sustainable human resourcepolicy. Such a policy will positively affect labor productivity, innovation, and overall performance of anorganization. We present a model of the human resource system of an organization from a holistic perspec-tive, where staff is regarded as a complex adaptive system. Our model takes into account factors whichare assumed to influence human behavior, like career prospects, work climate or work-life balance, andtherefore affect performance of the employees. Main intention is to apply specific OR techniques like sys-tem dynamics or agent based modeling and simulation upon the model in order to identify effective humanresource policies in a next step. This will enable managers to optimize organization’s total performance.First numerical results will be presented.

3 - Learning in Highly Polarized ConflictsSigifredo Laengle, Department of Management Control, University of Chile, Diagonal Paraguay257, Of. 1302A, 833015, Santiago, Santiago de Chile, Chile, [email protected], GinoLoyola

Negotiations are often conducted in highly polarized environments, which are also uncertain and dynamic.However, the intense rivalry involved in these conflicts does not always prevent an agreement from be-ing reached. A recently proposed static model sets out the conditions under which either an agreement isachieved or negotiations break down in this environment (Laengle and Loyola, Optim Lett, 2012). Never-theless, important aspects related to partial mutual knowledge of players in a dynamic context are not yetbeen studied. To fill this gap, we develop an extension of the static game to modelling highly polarizedconflicts in an uncertain, asymmetric and dynamic environment. In this extension both parties bargain mul-tiple negotiation rounds under uncertain threats that are materialised only if an agreement is not reached.If a negotiation breakdown occurs, each party learns about these threats from the outcome observed inthe previous round. This paper presents the most important results, and a short discussion about possibleapplications. In particular, we provide the conditions that characterise different paths for negotiations heldunder polarized environments, which matches the observed evolution of many of these conflicts in the realworld.

4 - Dynamical Supply Networks for Crisis and Disaster Relief - Networks Resilienceand Decision Support in Uncertain EnvironmentsErik Kropat, Department of Computer Science, Universität der Bundeswehr München,Werner-Heisenberg-Weg 39, 85577, Neubiberg, Germany, [email protected], SiljaMeyer-Nieberg

Recent natural disasters affected many parts of the world and resulted in an extensive loss of life anddisruption of infrastructure. The randomness of impacts and the urgency of response efforts require arapid decision making in an often uncertain and complex environment. In particular, the organization andcontrolling of efficient humanitarian supply chains are challenging the operational analyst from both thetheoretical and practical perspective. A far-sighted and comprehensive emergency planning can allevi-ate the effects of sudden-onset disasters and facilitate the efficient delivery of required commodities andhumanitarian aid to the victims.

Methods from computational networks and agent-based modelling supported by sophisticated data farmingexperiments allow a detailed analysis of network performance measures and an evaluation of the vulnera-bility of infrastructure and supply networks. These approaches can be used for relief planning as well asfor a simulation of continuous aid work threatened by severe disruptions. In particular, we investigate howthe allocation of transportation resources and the distribution of materials affect the networks efficiency. Inaddition, we study in which way physical disruptions caused for example by aftershocks or perturbationsof the logistics operations itself are affecting the system as a whole.

72

WC-05

We demonstrate the effectiveness of this approach for the example of an earthquake scenario in a coastalregion. We analyze the operability of a complex supply chain as well as the networks resilience.

� WC-05Wednesday, 12:45-14:15Goudriaan Room 1

Kidney exchangeStream: Health Care ManagementInvited sessionChair: Kristiaan Glorie, Econometric Institute, Erasmus University Rotterdam, Postbus 1738, 3000DR, Rotterdam, Netherlands, [email protected]

1 - Kidney exchange in sparse and dynamic graphsKristiaan Glorie, Econometric Institute, Erasmus University Rotterdam, Postbus 1738, 3000 DR,Rotterdam, Netherlands, [email protected], Albert Wagelmans, J.j. van de Klundert

Kidney exchange has recently emerged as a way to alleviate the worldwide shortage of kidney donorsby facilitating the exchange of kidneys between incompatible patient-donor pairs. Typically, the goal ofkidney exchange is to find an allocation of donors to patients that is optimal with respect to multiplehierarchically ordered criteria; the usual primary criterion being to maximize the number of transplants.Mathematically, this corresponds to finding a maximum weight k-cycle cover in a directed graph. Weaddress several new insights into the optimization of kidney exchange when the underlying graph is sparse,such as in the exchange pools that now begin to arise in practice, and/or dynamic. Our results extend to ageneral class of barter exchange markets.

We first show why for static sparse graphs k should be larger than for non-sparse graphs in order to max-imize the benefits of exchange, and why, in part because of that, the mixed integer programming formu-lation for the kidney exchange optimization problem with one variable per cycle is difficult to solve usingthe branch-and-price approach with depth-first pricing previously suggested in the literature. We then pro-pose how the pricing problem can be solved in polynomial time for a general class of hierarchical criteria,making the approach tractable for sparse graphs. Second, we focus on the dynamic optimization of kidneyexchanges. Our approach and its effects are demonstrated using two realistic simulators. The first is asimulator using empirical data from the Dutch kidney exchange program, which is the oldest nationallycoordinated program. The second is a modified version of a simulator often used in the literature and isbased on US population data.

2 - New Optimization Tools for Kidney Exchange ProgramsAna Viana, INESC TEC/ISEP, Campus da FEUP, Rua Dr. Roberto Frias, 4220-465, Porto,Portugal, [email protected], Filipe Alvelos, Miguel Constantino, Xenia Klimentova, JoaoPedro Pedroso, Abdur Rais, Nicolau Santos, Paolo Tubertini

This presentation updates current work and optimization tools being developed under a project funded bythe Portuguese Foundation for Science and Technology (FCT). The main goals of the project are to developcomprehensive mathematical models and optimization tools for "optimally matching’ the donor kidneyswith the kidney recipients under a Kidney Exchange Program. The matching optimization problem hasbeen modeled as both compact and non-compact mixed integer programs. Two different compact modelshave been tested on randomly-generated problems using a commercially-available solution methodologywith promising results. Models combining column generation and metaheuristics are also being tested.Given uncertainties in the last-minute availability of the kidney donors, maximization of the expected valueof the size of kidney exchange has been studied. A simulation tool will be used to capture and analyzevarious dynamic aspects associated with the optimization problem. Moreover, it will also help to assessbenefits and drawbacks of different market policies. The tools developed in this project directly support

73

WC-06

the Portuguese Institute for Blood and Transplantation Services (IPST) in the national kidney exchangeprogram.

Acknowledgment: This work is financed by the ERDF - European Regional Development Fund - throughthe COMPETE Programme (operational programme for competitiveness) and by National Funds throughthe FCT - Fundação para a Ciência e a Tecnologia (Portuguese Foundation for Science and Technology) -within the project KEP: New mechanisms for enhancing the kidney transplantation process, PTDC/EGE-GES/110940/2009.

3 - Paired and Altruistic Kidney Donation in the UK: A Column Generation ApproachDavid Manlove, School of Computing Science, University of Glasgow, Sir Alwyn WilliamsBuilding, G12 8QQ, Glasgow, United Kingdom, [email protected], GreggOMalley

The National Living Donor Kidney Sharing Schemes (NLDKSS) is a matching scheme run by the NationalHealth Service Blood and Transplant and used to find optimal sets of kidney exchanges in the UK. It is runevery quarter and has been in existence since April 2007.

The optimality criteria comprise five hierarchical objectives, including maximising the number of trans-plants, and subject to this, maximising the total weight of the transplants using a system for scoring poten-tial transplants. An optimal solution is found by solving a series of integer programming models that buildon those described by Roth et al (2007). Such a solution may contain both pairwise and 3-way exchangesand additionally allows altruistic donors to be involved in domino paired donation (DPD) chains involvingone couple. Software developed by the second author has been used to find an optimal solution since July2008, so far identifying 393 potential transplants, 201 of which have proceeded.

In this talk we describe how the column generation approach introduced by Abraham et al (2007) wasadapted to the UK optimality criteria in order to cope with potential increases in the numbers of patient-donor pairs and altruistic donors, and the maximum lengths of cycles and DPD chains. We show that oursoftware can be used to maximise the number of transplants when using larger datasets (e.g., with 1000patient-donor pairs). Most instances can be solved to optimality for the first four criteria, and to within veryclose to optimality for the fifth. However we describe some problems experienced in proving optimalityduring the branching phase with certain objectives.

� WC-06Wednesday, 12:45-14:15Goudriaan Room 2

Decision Support for Market Analysis and Risk ManagementStream: Information Systems, Neural Nets and Fuzzy SystemsInvited sessionChair: Ralph Grothmann, Corporate Technology CT RTC BAM LSY, Siemens AG, Otto-Hahn-Ring 6,81730, München, Germany, [email protected]

1 - Structural Econometric Analysis and Its Application on the Dutch Flower AuctionsYixin Lu, Rotterdam School of Management, Erasmus University, Netherlands, [email protected],Alok Gupta, Wolfgang Ketter, Eric van Heck

We apply the Structural Econometric Analysis to model the Dutch Flower Auctions (DFA). The DFA aremulti-unit sequential Dutch auctions. They account for more than 60% of the global flower trade and thusplay a vital role in maintaining the Netherlands’ leadership in the flower industry. They serve as efficientcenters for the exchange of flowers between suppliers and buyers.

Using a large real-world dataset, we estimate the structural model for the DFA using a non-parametricmethod. The structural properties are then used to optimize the key auction parameters under differentmarket conditions.

74

WC-06

Note that almost all the existing research on sequential auctions restricts attention to the sale of a singleindivisible unit per round. We, on the other hand, deal with a more general setting where bidders canacquire multiple units in each round. Such multi-unit sale in each transaction makes it difficult to predictthe (residual) supply and demand in the upcoming rounds and hence introduce extra complexities in themodeling process. Therefore, our research adds new insights to the auction literature. From the managerialperspective, our research provides valuable implications to auctioneers in their choices of key auctionparameters in the DFA.

2 - A simulation-optimization approach for information security risk managementElmar Kiesling, Information & Software Engineering Group, Vienna University of Technology,Favoritenstrasse 9-11/188, 1040, Vienna, Austria, [email protected], AndreasEkelhart, Bernhard Grill, Christine Strauss, Christian Stummer

Most enterprises have implemented various measures to protect their critical information systems fromsecurity threats. Such measures have been relatively effective in coping with random, opportunistic attacks(e.g., worms and viruses not directed at a particular target). Attacks by motivated threat agents, however,are much more difficult to deal with because these adversaries differ in resources, capabilities, as well aspoints of access, and they exploit multiple attack vectors to achieve their particular goals — frequentlyin unforeseen ways. We introduce a simulation-based optimization approach that addresses the problemof assessing and improving the security of complex information systems, particularly against the lattertype of attacks. To this end, we have designed and implemented a framework that integrates conceptualmodeling of security knowledge, behavioral modeling of threat agents, discrete event simulation of attacks,and genetic algorithms to identify efficient portfolios of security measures. Based on a general model ofsecurity knowledge and a model of the information system to be protected, our approach simulates a largenumber of attacks on different system configurations and records various outcome metrics. Based on thesemetrics, we optimize the system with respect to multiple cost and benefit objectives (e.g., minimize cost,maximize detection of attacks, minimize expected impact on availability, integrity, and confidentialityof data etc.). Finally, we provide interactive decision support for the selection of a proposed efficientportfolio of security measures to implement. We describe a prototypical implementation of our approachand illustrate its applicability by means of an exemplary application scenario.

3 - Interactive Decision Support within Corporate Environmental Management Infor-mation SystemsFabian Renatus, Chair of Production and Logistics, Georg-August-Universität Göttingen, Platzder Göttinger Sieben 3, 37073, Göttingen, Germany, [email protected],Jutta Geldermann

Corporate Environmental Management Information Systems (CEMIS) enable companies to provide the re-quired environmental related information to various stakeholders on the basis of material and energy flows.However, the currently available CEMIS usually offer only an operational perspective on the business pro-cesses. They focus solely on the current enquiry period and lack a strategic point of view. Modern CEMIShave to address this need for a strategic forecast which means they are also required to offer decision sup-port tools. Multi-criteria decision aiding (MCDA) methods provide an appropriate way to analyze possiblebusiness actions. This raises one question though: Which method is suitable to be implemented in such amodern CEMIS decision support tool? Methods as PROMETHEE, ELECTRE, AHP and MAUT/MAVTare well established among researchers, but can be intimidating to decision makers in companies. Thus, itis an important requirement to offer a method which is easy to understand and concurrently trustworthy.With the absence of an analyst, the applied method should offer some kind of assistance that helps thedecision maker to familiarize himself with the decision problem and his preference towards the availablealternatives. Interactive methods try to achieve this through an ongoing iterative process where input andoutput alternate. Furthermore, this interactive procedure is far more likely to generate an analysis thatis consistent with decision makers’ preferences. The other important requirement, besides the trustwor-thiness, is that the applied method should be easy to understand even for people with no experience inmulti-criteria decision making.

75

WC-07

� WC-07Wednesday, 12:45-14:15Antwerp Room

Maritime ShippingStream: Maritime LogisticsInvited sessionChair: Rutger de Mare, Oil, Gas, and Chemicals, ORTEC, Houtsingel 5, PO Box 75, 2700 AB ,Zoetermeer, Netherlands, [email protected]

1 - Designing robust liner shipping schedules: Optimizing recovery actions andbuffer timesJudith Mulder, Erasmus University Rotterdam, Netherlands, [email protected], RommertDekker

Liner shipping networks consist of fixed routes and time schedules that are published beforehand. However,ships can encounter delays both when they are sailing between ports and when they are berthing in a port.When ships arrive delayed, the berthing schedule of a port has to be adapted, which results in additionalcosts for the liner shipping company. Furthermore, a delayed ship will have a higher probability of arrivingdelayed in the next ports to visit. The incurred delay can increase during a round tour, because new delaysare encountered or because delayed ships have to wait until the berths are free before they can enter the port.Therefore, it is important for liner shipping companies to develop robust shipping routes. Ships can reducetheir delay by taking recovery actions against certain costs. Examples of recovery actions are: increasingthe sailing speed or port handling capacity. Furthermore, buffer time can be incorporated in the route tocapture the delay. The goal of the ship delay recovery problem is to determine a recovery policy and buffertime allocation that minimizes the costs associated with delays and recovery actions for a given route. Theship delay recovery problem with fixed buffer times can be formulated using a Markov decision process.The states of the Markov process denote the position of the ship and the amount of delay encountered bythe ship with respect to the original schedule. In each state, it has to be decided which recovery action isperformed. The buffer times can also be modeled as actions in a Markov decision process. However, thedifference with the recovery actions is that only one buffer time has to be chosen for each ship position.Therefore, integer variables are needed to add buffer time decisions to the model.

2 - Routing and Scheduling in Tramp Shipping - Integrating Bunker OptimizationCharlotte Vilhelmsen, Management Engineering, Technical University of Denmark,Produktionstorvet, Building 426, room 033C, 2800, Kgs. Lyngby, Denmark,[email protected], Richard Lusby, Jesper Larsen

A tramp ship operator typically has some contracted cargoes that must be carried and seeks to maximizeprofit by carrying optional cargoes. Hence, tramp ships operate much like taxies following available car-goes and not according to a fixed route network and itinerary as liner ships. Marine fuel is referred to asbunker fuel or simply bunker and bunker costs constitute a significant part of the daily operating costs.There can be great variations in bunker prices across bunker ports so it is important to carefully planbunkering for each ship. As ships operate 24 hours a day, they must refuel during operations. Therefore,route and schedule decisions affect the options for bunkering. Current practice is, however, to separatethe two planning problems by first constructing fleet schedules and then plan bunkering for these fixedschedules. In this paper we explore the effects of integrating bunker planning in the routing and schedulingphase and present a mixed integer programming formulation for the integrated problem of optimally rout-ing, scheduling and bunkering a tramp fleet. Aside from the integration of bunker, this model also extendsstandard tramp formulations by using load dependent costs, speed and bunker consumption. We devise asolution method based on column generation with a dynamic programming algorithm to generate columns.The method is heuristic mainly due to a discretization of the continuous bunker purchase variables. Weshow that the integrated planning approach can increase profits and that the decision of which cargoes tocarry and on which ships is affected by the bunker integration and by changes in the bunker prices.

76

WC-08

3 - A Branch-and-Price Approach for a Ship Routing Problem with Multiple Productsand Inventory ConstraintsRutger de Mare, Oil, Gas, and Chemicals, ORTEC, Houtsingel 5, PO Box 75, 2700 AB ,Zoetermeer, Netherlands, [email protected], Remy Spliet, Dennis Huisman

In the oil industry, different oil components are blended in a refinery to fuel products. These productsare transported to different harbors by ship. Due to the limited storage capacity at the harbors and theundesirability of a stock-out, inventory levels at the harbors have to be taken into account during theconstruction of the ship routes. In this paper, we give a detailed description of this problem, which we callthe ship routing problem with multiple products and inventory constraints. Furthermore, we formulate thisproblem as a generalized set-covering problem, and we present a Branch-and-Price algorithm to solve it.The pricing problems have a very complex nature. We discuss a dynamic programming algorithm to solvethem to optimality.

� WC-08Wednesday, 12:45-14:15Penn Room 1

Resource allocation in practiceStream: Scheduling and Project ManagementInvited sessionChair: Norbert Trautmann, Department of Business Administration, University of Bern, Ordinariat fürQuantitative Methoden der BWL, Schützenmattstrasse 14, 3012, Bern, BE, Switzerland,[email protected]

1 - An MBLP model for scheduling assessment centersAdrian Zimmermann, Department of Business Administration, University of Bern,Schuetzenmattstrasse 14, 3012, Bern, Switzerland, [email protected], NorbertTrautmann

Firms aim at employing qualified and motivated people and at assigning these people to the most suitablejobs. Human resources managers often conduct assessment centers before making such personnel deci-sions. By means of an assessment center, the potential and skills of a firm’s current employees and newjob applicants (referred to as candidates) can be assessed more objectively than with job interviews. In anassessment center, the participating candidates complete a series of exercises, during each of which theyare observed and assessed by one or two assessors, e.g., managers or trained psychologists.

Planning an assessment center represents a challenging optimization problem. This problem consists offinding a schedule, i.e., determining the start times of all exercises and other activities and of assigning therequired number of assessors to specific exercises, i.e., to the candidates performing these exercises, suchthat the duration of the assessment center is minimized. For each candidate, a subset of assessors mustbe chosen which can then be assigned to the exercises of that candidate. This subset should not containassessors who have a no-go relation with the candidate; such no-go relations arise if a candidate and anassessor know each other. Some exercises are designed as role-plays and therefore additionally requireactors. Other activities such as lunch breaks have to be scheduled for each candidate within a given timewindow on each day of the assessment center.

In this paper, we formulate this problem as a mixed-binary linear program. We report on computationalresults for four real-life examples. Even though finding optimal solutions is computationally intractable,the MBLP provides good feasible solutions within a reasonable computational time.

77

WC-09

2 - Short and long-term scheduling in an operating roomFrancisco Ballestin, Matematicas para la Economia, Universidad de Valencia, Avenida de losNaranjos, s/n, 46022, Valencia, Spain, [email protected], M. Angeles Pérez, M.PilarLino, Sacramento Quintanilla, Vicente Valls

A health care specialist unit in a hospital has a list of patients awaiting surgery. There are several piecesof information associated with each patient: 1) the priority of the operation, indicating its urgency, 2) adue date that establishes a soft limit to operate on a patient, 3) the subspecialty to which the patient’sillness corresponds and 4) the duration of the operation. In addition, there are several operating roomsavailable for the specialist unit each week. There are only few hours available for some subspecialties(called critical), whereas others subspecialties can be done on many days. Some types of operations arevery long, and are therefore critical if they belong to critical specialties because very few operations can bedone in a week. This paper studies how to choose the operations to do in the short-term (two weeks) and inthe long-term (several months). The operations have to be planned in their corresponding subspecialties sothat urgent patients are scheduled first, non-urgent patients are scheduled before their due dates if possible,and the efficiency of the operating rooms is maximised. To plan operations in the short-term the currentpatient list is considered. To plan operations in the long-term we simulate operations to come. Short-termplanification selects the specific operations that are actually going to be scheduled. Long-term planificationhelps the decision maker to make strategic decisions: a) correct number of operating rooms with the currentnumber of patients, or if there is an increase/decrease in patients, b) correct number of operating roomsneeded per week for each subspecialty, c) policy to follow when scheduling long operations.

3 - Resource allocation for a make-to-order production process in the printing indus-tryNorbert Trautmann, Department of Business Administration, University of Bern, Ordinariat fürQuantitative Methoden der BWL, Schützenmattstrasse 14, 3012, Bern, BE, Switzerland,[email protected], Philipp Baumann, Salome Forrer

Offset printing is a common method to produce large amounts of printed matter. We consider a real-world offset printing process that is used to imprint customer-specific designs on napkin pouches. Theprinting technology used yields a number of specific constraints. The planning problem consists of allo-cating designs to printing-plate slots such that the given customer demand for each design is fulfilled, alltechnological and organizational constraints are met and the total overproduction and setup costs are mini-mized. We formulate this planning problem as a mixed-binary linear program, and we develop a multi-passmatching-based savings heuristic. We report computational results for a set of problem instances devisedfrom real-world data.

� WC-09Wednesday, 12:45-14:15Diamond Room 1

Maintenance Planning IStream: Production and Operations ManagementInvited sessionChair: Geert-Jan van Houtum, Fac. of Technology Management, Eindhoven University of Technology,P.O. Box 513, 5600 MB Eindhoven, Netherlands, [email protected]

1 - Optimal maintenance policies for a safety-critical system and its deteriorating sen-sorChiel van Oosterom, Industrial Engineering & Innovation Sciences, Eindhoven University ofTechnology, PO Box 513, 5600MB, Eindhoven, Netherlands, [email protected], LisaMaillart, Jeffrey Kharoufeh, Hao Peng

78

WC-09

We consider the integrated problem of optimally maintaining an imperfect, deteriorating sensor and thesafety-critical binary state system it monitors. Costless observations on the state of the system becomeless informative as the sensor ages. By conducting a costly site visit, the state of the system can be iden-tified perfectly, after which the system is maintained if it is identified to be in the out-of-control state. Inaddition, a site visit provides the opportunity to replace the sensor. We formulate the problem of adap-tively scheduling site visits and sensor replacements using a partially observable Markov decision process(POMDP) model. The objective is to minimize total expected discounted cost due to system operation, sitevisits, system replacement, and sensor replacement. We characterize the structure of the optimal policyand provide numerical examples.

2 - The optimal line replaceable unit definition problemJorge Eduardo Parada Puig, Design, Production and Management, Chair of MaintenanceEngineering, University of Twente, Drienerlolaan 5, Postbus 217, 7500AE, Enschede,Overijssel, Netherlands, [email protected], Rob Basten, Leo van Dongen

Defective capital assets may be quickly restored to their operational condition by replacing the item that hasfailed. The item that is replaced is called the line replaceable unit (LRU), and the so-called LRU-definitionproblem is the problem of deciding on which item to replace upon each type of failure. When a replacementaction is required in the field, one option is to replace the failed item itself, but another option is to replacea parent assembly that holds the failed item. Items that are located deep in the indenture structure aretypically smaller and have lower holding costs. On the other hand, replacing these items can require moremaintenance resources, and longer time to identify and isolate failures. The reverse effects occur whenreplacing assemblies higher in the indenture structure. Therefore, defining what to replace is fundamentalto the trade-off between maintenance costs and availability of the capital asset. We consider a fleet ofassets (e.g., trains) so that we can improve the probability of having a sufficient number of assets availableby acquiring additional assets. The objective of the LRU-definition problem is then to minimize the totalcosts of item replacement and the investment in additional assets, given a constraint on the availability ofthose assets. We propose an integer linear programming formulation to solve the problem. We illustratethe model and the potential cost savings that it can lead to using a case in passenger service rolling stock.

3 - A comparison of age-based maintenance and condition-based maintenanceBram de Jonge, Operations, University of Groningen, P.O. Box 800, 9700 AV GRONINGEN,9700 AV, Groningen, Netherlands, [email protected], Ruud Teunter, Warse Klingenberg, TiedoTinga

We compare condition-based maintenance with age-based maintenance for situations where maintenancehas to be planned in advance. Advances in sensor technology have enabled companies to use the conditionof components when planning maintenance actions. Condition-based maintenance has the advantage thatmaintenance actions can be performed just before the component fails. However, the moment of failureis uncertain implying that this benefit can only be fully utilized if maintenance does not have to plannedin advance. The more traditional age-based maintenance strategy does not suffer from this disadvantage.Maintenance will be performed when the component reaches a specific age and can therefore easily beplanned. Disadvantages of age-based maintenance are that remaining useful life is wasted if the componentis not about to fail when maintenance is performed and that failure might occur before maintenance takesplace.

The aim of this study is to determine to what extent condition-based maintenance is beneficial if main-tenance has to be planned in advance, if the objective is to minimize the total costs of maintenance andbreakdowns. Various cases will be considered with different degradation processes, required times be-tween planning and performing maintenance, and cost structures. The results indicate in which situationsit is advantageous to invest in condition monitoring.

79

WC-10

� WC-10Wednesday, 12:45-14:15Diamond Room 2

Inventory IStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Karl Inderfurth, Faculty of Economics and Management, Otto-von-Guericke University ofMagdeburg, Universitaetsplatz 2, 39106, Magdeburg, Germany, [email protected]

1 - A new heuristic for an inventory system with random yield and positive lead timeWanrong Ju, Econometric Institute, Erasmus University Rotterdam, Burgmeester Oudlaan 50,H8-23, 3062PA, Rotterdam, Netherlands, [email protected], Adriana F. Gabor, Jan-Keesvan Ommeren

We study an infinite-horizon periodic-review inventory model with an unreliable supplier. The supplier hasa lead time and random yield, which is common in practice. The part of order received is assumed to be bi-nomially distributed. Positive lead time is considered which hasn’t been done extensively in literature. Westudy a simple order-up-to type of heuristic and proved that the above mentioned model can be simplifiedto a newsvendor problem with a modified demand. We found the steady-state distribution of the modifieddemand without any assumption on demand distribution. The performance of the heuristic is verified bycomparing with the optimal policy derived from dynamic programming in several scenarios.

2 - Stability of re-entrant flow linesCarlos Bispo, Electrical and Computer Engineering, Instituto Superior Técnico, Av. RoviscoPais, 1049-001 Lisboa, Lisboa, Portugal, [email protected]

The paper addresses the issue of stability for re-entrant flow lines, producing multiple products, withcapacitaded machines, random demand and random yield. The policies of interest are multi-echelon basestock policies, combined with a set of static and dynamic management rules of the available capacity.

We introduce three classes of capacitated echelon base stock policies: P1, the pure multi-echelon basestock; P2, like P1 with a possibly finite upper bound on the admission of raw materials; and P3, like P2with a possibly finite upper bound on the utilization of intermediate inventories.

The order of business is: establishing conditions for the stability of the shortfall echelon process whendemands are stationary and ergodic; examining the regenerative structure of the shorfall process whendemands are given by an i.i.d. sequence. The regenerative properties are valuable in establishing the con-vergence of costs and also simulation estimators, which enables the utilization of Infinitesimal PerturbationAnalysis to optimize the policy parameters. We use a coupling argument for shortfalls while establishingthe stability conditions, which will, by itself, render the Harris ergodicity of the shortfall process.

We show that the stability condition suffices to ensure that the shortfall process possesses the regenerativestructure of a Harris ergodic markov chain. Under a stronger condition, we establish that the vector ofshortfalls returns to the origin infinitely often, with probability one.

We show that the necessary stability condition is also sufficient for any sort of re-entrant system, in thepresence of random yield, provided the control policy is in P2.

3 - Optimal and Approximate Policies for Inventory Systems with Multiple Supply Op-tionsMahesh Srinivasan, Management, The University of Akron, 259 South Broadway, 365 CBA,44325, Akron, OH, United States, [email protected], Douglas Thomas

We consider a single item periodic-review system with lost sales having iid demand and lead times. Thesupply options are characterized by a faster source (mode) of supply which is usually more expensiveand a slower source (mode) of supply which is comparatively more economical. Based on our interactionwith sourcing professionals in the industry, this decision making problem is very relevant and one which

80

WC-11

managers face routinely in their day to day working. Moreover, due to the increasing energy prices, thereis a shift towards using more economical, but at the same time less reliable modes of transportation forregular shipments and an option to use the expedited mode for emergency shipments. Hence, firms arefaced with greater instances of this problem scenario in practice. Using dynamic programming basedoptimization and simulation; we investigate the optimal and near-optimal inventory policies under suchscenarios and compute the performance of these policies under various demand lead-time distributions.We provide insights to the cost threshold limit ? below which the fast supply option is preferred. Throughnumerical experiments, we show that for a vast majority of problem instances, not considering the use ofthe fast supply option can result in sub-optimal policies and significant additional costs as high as 21-23percent, even when the cost of using the fast supply option is ten times more than the cost of using the slowsupply option. Even when the cost of using the fast supply option is substantially higher, (twenty timesmore expensive), a significant portion of the orders are placed using this option, especially when stock-outcosts are high.

4 - Exact and heuristic linear-inflation policies for an inventory model with randomyield and arbitrary lead timesKarl Inderfurth, Faculty of Economics and Management, Otto-von-Guericke University ofMagdeburg, Universitaetsplatz 2, 39106, Magdeburg, Germany,[email protected], Gudrun Kiesmüller

We investigate a periodic inventory system for a single item with stochastic demand and random yield.Since the optimal policy for such a system is complicated we study the class of stationary linear-inflationpolicies where orders are only placed if the inventory position exceeds a critical stock level and where theorder quantity is controlled by a yield inflation factor. We consider two different models for the uncertainsupply: binomial and stochastically proportional yield. Linear-inflation policies for random yield modelshave already been addressed in the literature but contributions devoted to positive and constant lead timesor asymmetric demand distributions are rare. In this paper we propose two novel approaches to deriveoptimal and near-optimal numerical values for the critical stock level, minimizing the average holdingand backorder cost for a given inflation factor. First, we present a Markov chain approach, which isexact in case of negligible lead time. Second, we provide a steady-state analysis to derive approximateclosed-form expressions for the optimal critical stock level. We conduct an extensive numerical study, alsoincluding instances with symmetric and asymmetric demand and yield distributions, to test the performanceof our approaches. The numerical experiments reveal an excellent performance of both approaches. Sinceour derived formulas are easily implementable and highly accurate they are very valuable for practicalapplication.

� WC-11Wednesday, 12:45-14:15Penn Room 2

Supply Chain Optimization IStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Marcus Brandenburg, Chair of Supply Chain Management, University of Kassel, UntereKönigsstr. 71, D-34117, Kassel, Germany, [email protected]

1 - Solving Supply Chain Management Problems with the SCIP Optimization SuiteMatthias Miltenberger, Optimization, Zuse Institute Berlin, Zuse Institute Berlin, Takustr. 7,14195, Berlin, Germany, [email protected], Gerald Gamrath, Weninger Dieter

Supply chain management problems are an important class of real world instances and cover a wide rangeof applications. Additionally, they can be almost arbitrarily large and complex, depending on scale and

81

WC-11

level of detail of the supply chain. We present current research on instances that constantly test the bound-aries of our computational possibilities. To solve these problems we use the SCIP Optimization Suite, oneof the fastest MIP solvers available in source code.

Strong presolving techniques are necessary to reduce the problem to manageable dimensions and to providethe practitioner with a good solution after a reasonable amount of time. We will demonstrate some newpresolving implementations and their effects on the instances.

Another area of research is the solution of linear programming relaxations which are an essential part ofthe MIP solving process. Considerable progress was made improving the performance of our LP solverSoPlex, especially regarding very large and sparse problems, as they are provided by our industry partner.

2 - Territory Design for Arc RoutingAlexander Butsch, Institute of Operations Research (IOR), Karlsruhe Institute of Technology(KIT), Englerstr. 11, Building 11.40, 2nd floor, 76128, Karlsruhe, Baden-Wuerttemberg,Germany, [email protected], Jörg Kalcsics, Gilbert Laporte, Stefan Nickel

Territory design is the problem of grouping small geographic areas, called basic areas, into larger clusters,called territories, subject to a number of relevant planning criteria. Three important criteria are balance,compactness and contiguity. Balance describes the requirement for territories to have approximately thesame size with respect to a quantitative measure. A territory is said to be geographically compact if itis more or less square or round-shaped and undistorted. Furthermore, territories should be continuous,respectively connected. We consider the territory design problem for arc routing, which arises e.g. in thedesign of territories for waste collection, winter services or leaflet delivery. The basic areas correspond tostreets and each street must be serviced exactly once. We are given the required service time as well as the(unproductive) deadheading time along each street. Besides the criteria mentioned above, we minimize thetotal deadheading time. We present a heuristic approach that combines a geometric construction procedureand a two-stage iterative improvement approach, using tabu search and adaptive randomized neighborhoodsearch. Since the neighborhood of a solution can be quite large, we apply different strategies which restrictthe search to subsets of the neighborhood. Each strategy focuses on an improvement with respect to oneplanning criterion. For each strategy we have developed a subroutine based on tabu search to explore thecorresponding neighborhood. The subroutines are randomly selected according to their past performanceand to user-defined weights. Finally, we present results of computational experiments conducted on real-world street data.

3 - Supply chain design for new product introduction by mixed-integer linear pro-grammingMarcus Brandenburg, Chair of Supply Chain Management, University of Kassel, UntereKönigsstr. 71, D-34117, Kassel, Germany, [email protected]

The paper focuses on a case example of a globally operating manufacturer that must link its productionplants to a supplier and three demand regions for a new consumer good [1]. Due to short product lifecycles (PLC), this decision must be made before the product is launched in the market [2]. Pipeline fill-ing requirements that decouple market demands from available capacity and region-specific launch plansincrease the problem complexity. Due to these characteristics, such SC design decisions call for reflectingcash flows and the time value of money [3]. To support value-based decision-making, quantitative modelsshould include discounted cash flow (DCF) or economic value added (EVA) concepts [4]. In contrast to [1],who employ discrete-event simulation to assess value impacts of SC design options, a MILP model is for-mulated to optimize DCF over the whole PLC. This model complements other approaches for EVA-basedsales & operations planning [5]. — [1] Brandenburg M, Schilling R (2012): Value impacts of dynamicsand uncertainty in tactical supply chain design for new product introduction. In Wirtschaftsinformatik,Entscheidungstheorie&—praxis, Geiger M et al. (eds.), Aachen:23-46 [2] Higuchi T, Troutt M (2004):Dynamic simulation of the supply chain for a short life cycle product — lessons from the Tamagotchi case.Comput Oper Res 31(7): 1097-1114 [3] Brandenburg, M. (2013): Quantitative Models for Value-BasedSupply Chain Management. Heidelberg [4] Hahn, G., Kuhn, H. (2012): Designing decision support sys-tems for value-based management: A survey and an architecture. Decis Support Syst 53(3): 591-598 [5]Hahn, G., Kuhn, H. (2012): Value-based performance and risk management in supply chains: A robustoptimization approach. Int J Prod Econ 139:135-144

82

WC-12

� WC-12Wednesday, 12:45-14:15Van Oldenbarnevelt Room

Supply Chain ContractsStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Gerd J. Hahn, Assistant Professorship for Supply Chain Management, University of Mannheim,Schloss, Schneckenhof Ost, 68131, Mannheim, Germany, [email protected]

1 - Capacity reservation contracts for risk-neutral and risk-sensitive suppliersMehdi Sharifyazdi, Department of Strategy and Logistics, BI Norwegian Business School,Nydalsveien 37, 0442, Oslo, Norway, [email protected], Hoda Davarzani

This paper formulates a specific type of capacity reservation contract between a supplier and multiplebuyers. The supplier offers price discounts to the buyers, in case they reserve a minimum level of capacity,to make them prefer her over the other sources of capacity in the market and to maximize her capacityutilization. It is assumed that the customers can always buy as much capacity as they need from the othersources. The supplier wants to determine the optimal incentives and maximum capacity offered to eachcustomer, while customers decide on how much capacity to reserve given the supplier’s offer. Buyers’demand is stochastic and has a distribution which is known to them as well as to the supplier. Furthermore,in order to study the behaviour of the suppliers who are either risk-neutral or risk-sensitive, two differentobjective functions are formulated respectively: one that aims to maximize the supplier’s revenue andanother one which tries to maximize the chance of a minimum level of capacity utilization. To take careof the complexity of the model, an evolutionary numerical method is developed to find good solutions in areasonable time.

2 - Managing After-sales Service Contracts with Lifetime Buy PlanningDong Li, Rotterdam School of Management, Burgemeester Oudlaan 50, 3062 PA , Rotterdam,Netherlands, [email protected], Nishant Mishra

We look at contracting between a customer and a supplier, in a durable goods supply chain where equip-ment downtime impacts both the customer’s and the supplier’s revenues, and lifetime buy planning isconsidered to manage component obsolescence issues. For such a supply chain, we study how the choiceof contract can impact the customer’s initial purchase decision regarding the number of products, and cor-respondingly, the overall supply chain’s profit and efficiency. We find that PBC can result in higher productavailability, but only when penalty is incurred on lost revenue. However, penalty on lost revenue leads tolower profit and efficiency for both players and for the supply chain. The W+T works well with low inven-tory and long warranty, when replenishment cost is low. With low obsolescence cost, longer warranty canresult in higher profit, despite spare parts inventory backorder. Furthermore, we find that the above insightsare robust to the choice of the revenue functions and when we include risk-aversion of the customer andthe supplier.

3 - Optimal design of buyback contract in the magazine industryKei Takahashi, The Institute of Statisitical Mathematics, 10-3, Midori-cho, 190-8562,Tachikawa-shi, Tokyo, Japan, [email protected], Ko Nagai, Takahiro Ohno

This study analyzes a responsible sales system (RSS) in the magazine industry ensures Pareto improvementfrom other contracts, spot procurement and resale price maintenance (RPM), between publishers and bookstores via theoretical models. There are various supply chain contracts in the world. In Japan, RPMhas been widely adapted to the magazine industry. RPM is one of the buyback contacts, which coercespublishers to buyback unsold magazines at wholesale from book stores. In 2008, the RSS could be adoptedto the magazine industry because of high ratio of returned unsold magazines via RPM. Under the RSS,publishers need to buyback unsold magazines at lower price than wholesales. However, it is not obviousthat the RSS ensures Pareto improvement from other contract. There are optimal buyback price and amountfor publishers and desirable ones for book stores.

83

WC-13

We employ a real option game approach in order to analyze supply chain contracts. In this analysis,an optimal design of buyback contract, buyback price and amount, is emerged as a result of a subgameperfect equilibrium in a leader-follower game between a publisher and a book store. Under this approach,a Geometric Brownian Motion is adopted to demand of magazines. Many studies in supply chain contractsassume that demand follows independent and identical normal distribution in each period. Obviously, thischaracter cannot sketch demand for magazines, which has non-negativity and high serial correlation.

We conduct sensitivity analyses with actual parameter sets via numerical calculation. As a result, wefind that initial demand for magazines and volatility (degree of stochastic demand) influence particularlyoptimal designs of buyback contract for the publisher and desirable ones for the book store.

4 - Value-based Supply Chain Management: A newsvendor perspectiveGerd J. Hahn, Assistant Professorship for Supply Chain Management, University of Mannheim,Schloss, Schneckenhof Ost, 68131, Mannheim, Germany, [email protected], MoritzFleischmann

Increasing shareholder value is commonly considered the paramount business goal and thus approaches forvalue-based management are widely discussed in theory and practice. Value-based management uses top-level performance metrics such as Economic Value Added (EVA) to determine whether a business createsvalue by earning profits that exceed total costs of invested capital. Value driver trees are a common methodto drill down the top-level metric into distinct performance levers. Since investment and financing decisionsare inextricably interlinked when capital markets are imperfect (e.g. tax-deductible debt financing), anintegrated approach for both perspectives is required.

In this presentation, we use EVA as a prevalent concept of value-based management and derive optimalinvestment and financing decisions for the newsvendor that are in accordance with common valuationtheory (e.g. CAPM). Corporate taxes and bankruptcy are explicitly considered in the approach and areconsistently reflected in the calculation of the weighted average cost of capital (WACC). The EVA-basedapproach is compared against a standard profit-based newsvendor both analytically and numerically tohighlight differences in the order quantities.

Having analyzed the problem for the newsvendor firm, we extend the scope towards the supply chainconsidering a capital-constrained retailer who uses working capital financing from the supplier besidesshort-term borrowing. Different means of working capital financing such as trade credit and reverse fac-toring are analyzed. Further supply chain contracts are investigated to analyze different risk allocationsbetween the supplier and the retailer as well as the implications on the respective WACCs and channelcoordination.

� WC-13Wednesday, 12:45-14:15Mees Auditorium

Optimization Methods for Vehicle RoutingStream: Traffic and TransportationInvited sessionChair: Stefan Irnich, Chair of Logistics Management, Gutenberg School of Management andEconomics, Johannes Gutenberg University Mainz, Jakob-Welder-Weg 9, 55128, Mainz, Germany,[email protected]

1 - Dynamic Programming Formulations for the Minimum Tour Duration ProblemChristian Tilk, Chair of Logistics Management, Johannes Gutenberg University Mainz,Germany, [email protected], Stefan Irnich

84

WC-13

In this presentation, we consider a variant of the traveling salesman problem with time windows (TSPTW),called minimum tour duration problem (MTDP), where the objective is the minimization of the tour dura-tion. We present a new effective dynamic programming (DP) approach to solve the MTDP. It is motivatedby the results recently presented by Baldacci et al. (2011), who solve the TSPTW with a DP-based al-gorithm. When solving TSPTW, two independent resources are propagated along partial paths, one forcosts and one for earliest arrival times. For dealing with tour duration minimization, we define consistentresource extension functions meaning that dominance is straightforward and the forward DP and any ofits relaxations provide bounds for the backward DP, and vice versa. This is a non-trivial task because inthe MTDP at least three resources are needed, and two resources depend on each other in a non-additiveand non-linear way. To obtain lower bounds, we present a new relaxation for the MTDP with only oneresource, which is attractive due to its low computational complexity. This and other relaxations can becombined with the ng-tour relaxation and the ngL-tour relaxation. To improve the lower bounds, we usetwo methods: First, we adapt a penalty method. Second, we generate the neighborhoods for the ng-tourand ngL-tour relaxations dynamically. To the best of our knowledge, we present the first exact algorithmfor the MTDP.

2 - New solution approaches to integrate the vehicle-type scheduling and timetablegenerationMonize Sâmara Visentini, Management Science, Universidade Federal do Rio Grande do Sul,ESTRADA RS 509, 7260, 97110620, santa maria, Rio Grande do Sul, Brazil,[email protected], Denis Borenstein, Olinto Araújo, Pablo Guedes

We propose three new approaches to solve the vehicle-type scheduling problem (VTSP), which comprisesthe integration of the vehicle scheduling problem and the vehicle type required for each trip, keeping acertain level of service for each trip such as the degree of passengers’ comfort, seat availability, and otheroperational features. Two of them are mixed integer linear programming (MILP) models aiming at opti-mizing the VTSP and the timetabling generation problem based on a time-space network (TSN). The thirdone is a new methodology for inserting time windows to the proposed problem, based on small changesin the TSN structure, with easy computational implementation and optimal solution at low computationrun-times. By including small changes to the timetable and/or including the time windows for timetablingtrips, we introduced some flexibility levels in the departure times of trips, resulting in operational advan-tages for the service provider. Since we use a very short time window interval, the current timetable isonly slightly modified, minimally changing the passenger routines. The developed approaches were testedusing real instances in a Brazilian city, where the timetable presents irregular headways, in which the busesdeparture time do not have a fixed interval. All three approaches have resulted in relevant savings in thedaily operations of the public transportation service, reducing the required number of scheduled vehiclesto carry out the historic demand.

3 - Static vs. Dynamic Policies for Vehicle Routing Problems with Backhauling andDynamic Customer DemandsSubrata Mitra, Operations Management, IIM Calcutta, Diamond Harbour Road, Joka, 700104,Kolkata, West Bengal, India, [email protected]

Dynamic vehicle routing problems (VRP) without backhauling have attracted more attention than dynamicvehicle routing problems with backhauling (VRPB) in the relevant literature. Dynamic VRPB is morecomplex than dynamic VRP without backhauling, and since VRP without backhauling is a special case ofVRPB, models and algorithms for dynamic VRPB can easily be adapted for dynamic VRP. In this paper,we compare static vs. dynamic policies for solving dynamic VRPB with dynamically occurring customerdelivery and pickup demands. We develop MILP formulations and heuristic algorithms of worst-case-complexity of O(n2) where n is the number of customer locations, in line with the Nearest Neighbour(NN) routing policy, for small-to-medium-sized problems under static and dynamic policies. Althoughdynamic policies are always at least as good as static policies, we observe from numerical experimentationsthat static policies perform relatively well for small-sized problems and low degrees of dynamism (dod),defined as the ratio of the number of dynamic customer requests to the total number of customer requests.On the other hand, dynamic policies are expected to perform significantly better than static policies forlarge-sized problems, high dod and early availabilities of dynamic customer delivery and pickup demandinformation. We conclude the paper by providing directions for future research on dynamic VRPB.

85

WC-14

� WC-14Wednesday, 12:45-14:15Tokyo Room

Decision AnalysisStream: Decision Analysis and Multiple Criteria Decision MakingInvited sessionChair: Andreas Kleine, Faculty of Economics, University of Hagen, D-58084 , Hagen, Germany,[email protected]

1 - On the Aggregation of the Benchmark Technologies in the Global Malmquist Pro-ductivity IndexMohsen Afsharian, Institute of Management Control and Business Accounting, BraunschweigUniversity of Technology, Fallersleber-Tor-Wall 23 38100 Braunschweig, Germany, 38100 ,Braunschweig, Germany, [email protected], Heinz Ahn

This paper deals with a special case of the non-homogeneity problem related to the determination of theglobal benchmark technology when measuring productivity changes over time. The authors propose a newMalmquist-type index, the overall Malmquist Index. Rather than assuming all observations in the globalproduction possibility set being homogeneous with respect to the underlying technology, this index onlyassumes contemporaneous observations — i.e. observations belonging to the same period — to be ho-mogeneous, permitting the observations to be considered non-homogeneous over time. The benchmarktechnology is obtained by aggregation of the technologies corresponding to the contemporaneous observa-tions. No further assumptions are necessary for the aggregation of these technologies, e.g. concerning theconvexity of the global benchmark technology. On this basis, mixed integer linear programming problemsare developed for the computation of the corresponding distance functions. In terms of productivity mea-surement, the proposed index possesses the circularity property, generates a single measure of productivitychange and does not need to be recomputed completely when a new time period is incorporated, i.e. itmaintains previous productivity calculations. Similar to traditional indices, it is decomposed into variouscomponents such as efficiency change, scale efficiency change, and best practice change. The suggestedindex will be illustrated by means of a real-world example from banking. In particular, in order to demon-strate its advantages, it will be compared to the global Malmquist index introduced into the literature byPastor and Lovell (2005).

2 - Predicting student retention by comparing histograms of bootstrapping forCharnes-Cooper transformation-linear programming discriminant analysisJaan Ubi, Department of Informatics, Tallinn University of Technology, Akadeemia tee 15A,12618, Tallinn, Harjumaa, Estonia, [email protected], Evald Ubi, Innar Liiv, Leo Vohandu

The goal of the paper is to predict student retention by using linear discriminant analysis with bootstrap-ping. The result (93%) provides accuracy superior to the bootstrapping of a comparative method, as wellas to the non-bootstrapping variations. In order to perform discriminant analysis, we linearize a fractionalprogramming method by using Charnes-Cooper transformation and apply linear programming, while thecomparative approach uses deviation variables to tackle a similar multiple criteria optimization problem.We train the discriminatory hyperplane family and apply it to the testing set — thus arriving at a set ofhistograms. We analyze the histograms by using the simple mean — best for prediction — and a five-foldKolmogorov-Smirnov test — best used for resources allocation, in order to act on the final results. Finalresults are the outcome of applying the hyperplane family on freshman data.

3 - Sensitivity Analysis of BCC Efficiency in DEA with Application to European HealthServicesAndreas Kleine, Faculty of Economics, University of Hagen, D-58084 , Hagen, Germany,[email protected], Andreas Dellnitz, Wilhelm Rödder

86

WC-15

The CCR model by Charnes, Cooper and Rhodes (1978) on the one hand and the BCC model by Banker,Charnes and Cooper (1984) on the other hand are the most common used approaches of data envelopmentanalysis (DEA). If we measure the efficiency of decision making units (DMUs) by the BCC model thetechnology is characterized by variable returns to scale. The inputs and outputs of a DMU are scalable bytwo parameters such that the BCC (in)efficiency score is unchanged. We call this adaptation a bicentricscaling (BS). We introduce a linear program to calculate the BS stability region of all DMUs, efficientor inefficient. Moreover we determine the scale efficiency within in the stability region. The approach isillustrated by a numerical example of the European health service. We demonstrate the stability region forvarious states and illustrate the consequence on scale efficiency. It is shown how states can improve scaleefficiency without losing BCC efficiency.

� WC-15Wednesday, 12:45-14:15New York Room

Networks IIStream: Discrete and Combinatorial Optimization, Graphs and NetworksInvited sessionChair: Diptesh Ghosh, Production and Quantitative Methods, Indian Institute of Management,Ahmedabad, Vastrapur, 380015, Ahmedabad, Gujarat, India, [email protected]

1 - Identifying defective valves in a blowout preventer valve networkDiptesh Ghosh, Production and Quantitative Methods, Indian Institute of Management,Ahmedabad, Vastrapur, 380015, Ahmedabad, Gujarat, India, [email protected]

Blowouts are financially damaging for drilling companies and are ecological hazards. Hence blowout pre-vention equipment is critical infrastructure for drilling companies. Drilling companies attempt to preventblowouts through blowout preventer (BOP) valve networks. BOP valves are high pressure valves that canbe closed rapidly. In a BOP valve network, BOP valves are arranged in a primarily series-parallel configu-ration which can contain any unwanted influx from a well-bore and divert it though manifolds away fromthe rig to be released in a safe and controlled manner. Once the influx is contained, the valves are returnedto their original configuration so that normal drilling operations can continue. The advantage of a series-parallel structure is that it allows, if any of the valves leak, fail or wear out, to continue to deal with theinflux in a controlled fashion while the leaking or worn out valves are repaired or replaced. Typical BOPvalve networks have between 10 and 50 BOP valves. Since BOP valves are critical pieces of equipment,they need to be checked regularly. However, since these valves are often physically inaccessible, they arechecked in batches called test sets. In this paper we present an exact method to check the functional statusof all BOP valves in a network using a minimum number of test sets. We also present a heuristic methodto identify malfunctioning valves if they exist. We illustrate both methods using a real world example.

2 - The intersection of all maximum independent sets and the intersection of all criti-cal sets of a unicyclic graph are equalEugen Mandrescu, Computer Science, Holon Institute of Technology, Golomb Str. 52, Holon,Israel, [email protected], Vadim Levit

A set of vertices in a graph is (a) independent if its vertices are pairwise non-adjacent, and (b) critical ifthe difference between its size and the size of its neighborhood is maximum. The independence number isthe size of a largest independent set. A graph where the sum of its independence and matching numbersequals the cardinality of its vertex set is a König-Egerváry graph. By core, corona, ker of a graph we meanthe intersection of all its maximum independent sets, the union of all its maximum independent sets, andthe intersection of all its critical sets, respectively. While finding a maximum independent set is an NP-hard problem, the critical independent set problem is tractable, which leads to a practically efficient wayof approximating the independence number (C. Q. Zhang, Finding critical independent sets and criticalvertex subsets are polynomial problems, SIAM J. Discrete Math. 3 (1990) 431-438). A connected graph

87

WC-16

is unicyclic if it has a unique cycle. It is known that: (a) ker is a subset of core in every graph (Levit andMandrescu, Vertices belonging to all critical independent sets of a graph, SIAM J. Discrete Math. 26 (2012)399-403); (b) these two sets coincide in bipartite graphs (Levit and Mandrescu, Critical sets in bipartitegraphs, Annals of Combinatorics (2013) (accepted)); (c) in a König-Egerváry graph, the sum of sizes ofits core and corona is twice its independence number (Levit and Mandrescu, A set and collection lemma,(2011) arXiv:1101.4564). In this work we show that for a unicyclic non-König-Egerváry graph, ker andcore coincide, while the sum of sizes of its core and corona is greater by one than twice its independencenumber.

3 - An Optimal Placement of a Liaison with Short Communication Lengths betweenTwo Members of the Same Level in an Organization Structure of a Complete K-aryTreeKiyoshi Sawada, Department of Policy Studies, University of Marketing and DistributionSciences, 3-1, Gakuen-nishi-machi, Nishi-ku, 651-2188, Kobe, Japan,[email protected]

Liaisons which have roles of coordinating different sections are placed as a means to become effective incommunication of information in an organization. We proposed some models of placing a liaison whichforms relations to members in the same level of a pyramid organization structure which is a completeK-ary tree of height H. When a liaison node which gets adjacent to nodes with the same depth is placed,an optimal depth is obtained by maximizing the total shortening distance which is the sum of shorteninglengths of shortest paths between every pair of all nodes in the complete K-ary tree. These models areexpressed as all edges have the same length. However, we should consider that edges between the liaisonand the other members are shorter than those between members except the liaison in the organization. Thisstudy proposes a model of placing a liaison which forms relations to two members in the same level ofa pyramid organization structure when lengths between the liaison and the other members are less thanthose between members except the liaison in the organization such that the communication of informationbetween every member in the organization becomes the most efficient. For a model of adding a node ofliaison which gets adjacent to two nodes with the same depth N in a complete K-ary tree of height H wherethe lengths of edges between the liaison and the other members are L which is less than 1 while thoseof edges between members except the liaison are 1, we have formulated the total shortening distance andhave obtained an optimal pair of nodes with the optimal depth N* which maximizes the total shorteningdistance.

� WC-16Wednesday, 12:45-14:15Oscar Auditorium

GOR PhD AwardsStream: GOR AwardsInvited sessionChair: Alf Kimms, Mercator School of Management, University of Duisburg-Essen, Lotharstr. 65, LB125, 47057, Duisburg, Germany, [email protected]

1 - Competition for Resources: The Equilibrium Existence Problem in CongestionGamesMax Klimm, Institut für Mathematik, Technische Universität Berlin, Sekr. MA 5-1, Straße des17. Juni 136, 10623, Berlin, Germany, [email protected]

Weighted congestion games are an important class of noncooperative games that constitute and elegantmodel of the resource usage by selfish users. Unfortunately, they need not possess a pure Nash equilibrium,in general. We give a complete characterization of the maximal sets of cost functions that one can allowon the resources in order to guarantee the existence of a pure Nash equilibrium. This characterization has

88

WC-16

several applications in the modeling and simulation of traffic in road and data networks, habitat choicesby animal populations, and - perhaps surprisingly - in the design of integrated circuits. We give furthercharacterizations of the existence of pure Nash equilibria for several variants of congestion games that serveas a more fine tuned model of the aforementioned applications, including games with resource-dependentdemands, games with variable demands, and games with bottleneck objectives. For the latter class ofgames we also characterize the combinatorial properties of the strategy spaces for which an equilibriumcan be computed efficiently.

2 - Operational Planning Problems in Intermodal Freight TransportationJenny Nossack, Institute of Information Systems, University of Siegen, Hölderlinstraße 3,57068, Siegen, North Rhine-Westphalia, Germany, [email protected]

In the context of intermodal freight transportation, various optimization problems arise. This disserta-tion makes several contributions to operational planning problems that are encountered in the pre- andend-haulage and at transshipment yards of intermodal transport. For each optimization problem, we statethe relevant literature and present several mathematical formulations. Moreover, we develop solution ap-proaches that are specially designed for the problems that arise in the context of intermodal transport, butcan be applied to other areas as well. We address, for example, the problem of scheduling the containermovements between customers and terminals under hard time window constraints and the simultaneousreposition of empty containers. We model this problem as a full-truckload pickup-and-delivery problemwith time windows and develop a two-stage heuristic solution approach. A simplification of this prob-lem can be modeled as an asymmetric TSP on a bipartite graph and as an integrated model that capturesthe problem’s assignment and routing structure. We solve the integrated model by the classical and thegeneralized Benders decomposition. The performance of our solution methods is evaluated on severalcomputational studies. For each addressed problem, we were able to demonstrate the successfulness of ourapproaches by outperforming algorithms known from the literature.

3 - Application of operational research methods for ergonomic design of workingplaces at assembly linesAlena Otto, University of Siegen, 57068, Siegen, Germany, [email protected]

Mitigation of ergonomic risks, or risks for health of workers, is a high-priority objective in manufacturing.We draw attention, that consideration of ergonomic risks can be integrated into the planning routines. Weemphasize the necessity to develop Operations Research models and effective solution methods to assistindustrial engineers during such a planning process. On the example of a mid-term planning problem ofassembly line balancing and a short-term planning problem of job rotation scheduling, we illustrate theeffectiveness of the proposed approach. We show that it is possible to reduce and redistribute ergonomicrisks at low costs with the help of assembly line balancing. Our designed two-stage heuristic procedure isable to lower the ergonomic risks to an acceptable level without increasing either the number of worksta-tions or the cycle time for 50% of realistically generated instances. In addition, we present the ergonomicjob rotation scheduling as another effective tool in balancing ergonomic risks among workers. Althoughthe underlying problem is NP-hard and conventional rules of thumb deliver solutions with a large gapto optimum, the proposed heuristic algorithm finds optimal or near-optimal solutions in just a couple ofseconds of time for realistic instances.

4 - Capacitated Network Design - Multi-Commodity Flow Formulations, CuttingPlanes, and Demand UncertaintyChristian Raack, Optimization, Zuse Institute Berlin, Takustr. 7, D-14195, Berlin, Germany,[email protected]

This thesis is about methods in mathematical optimization to dimension networks at minimal cost. Theconsidered planning problems arise in the strategic design of telecommunication or public transport net-works and in logistics. One of the essential aspects studied in this work is the use of cutting planesto enhance solution approaches based on multi-commodity flow formulations. Providing theoretical andcomputational evidence for the efficacy of inequalities based on network cuts, we extend existing theoryand algorithmic work in different directions. We present the MCF-separator as implemented in the mixed-integer programming solvers Scip and Cplex. This separator generates cutset inequalities for general MIPsbased on a preceding network detection. With this approach we achieve speed-ups of more than -50% CPU

89

WC-17

time to solve pure network design problems and -18% to -30% for general MIPs if network structure isdetected. We further develop approaches to incorporate the uncertainty of traffic demand forecasts into thedesign of networks. We generalize cutset and flow-cutset inequalities to this context proving their theoreti-cal and computational strength. We also present a new routing scheme for robust network design, so-calledaffine routing. Providing a thorough and comprehensive classification, we compare it to the well-knownstatic and dynamic routing.

� WC-17Wednesday, 12:45-14:15New Orleans Room

Queueing systemsStream: Applied Probability and Stochastic Programming, ForecastingInvited sessionChair: René de Koster, Rotterdam School of Management, Erasmus University Rotterdam, RoomT10-56, PO Box 1738, 3000 DR, Rotterdam, Netherlands, [email protected]

1 - Queueing systems in a random environment with applicationsRuslan Krenzler, Department of Mathematics, University of Hamburg, Bundesstrasse 55, 20146,Hamburg, Germany, [email protected], Hans Daduna

In Operations Research applications queuing systems constitute an important class of models in very dif-ferent settings. But in many situations those parts of a complex system, which are modeled by queues (e.g.a production system), interact with other subsystems. These can often be considered as environment forthe queue.

An important example is an unreliable manufacturing system (machine, modeled by a queuing system)which can be interrupted due to random failures (failure and recovery process as environment).

We will introduce a general exponential model where the environment can simultaneously interrupt inputwith service of the queuing system and the queuing system can change the environment state. We willshow that the steady state distribution of this system has a product form, i.e., asymptotically the queuelength and the environment seem to decouple. From this property we derive some invariance and insensi-tivity properties of the systems. A generalization of the model will allow to model failures, maintenance,inventory, finite buffers, etc. as environment of the queuing systems.

2 - Integrated models for queueing-inventory systemsSonja Otten, Department of Mathematics, University of Hamburg, Bundesstrasse 55, 20146,Hamburg, Germany, [email protected], Hans Daduna

Over the last decades research on queueing systems with an attached inventory has gotten much attention,often in connection with research on integrated supply chain management. Usually, in these integratedqueueing-inventory models a continuous review policy is applied. The most simple supply chains of thisstructure consist of a production system with an inventory holding raw material. In most papers in the lit-erature it is assumed that one unit of raw material is taken from the inventory when the service commencesand to reorganize the inventory according to one of the established inventory policies (e.g. (r,Q)-policy,(r,S)-policy). With some suitable simplifications this leads to explicit steady state distributions for thequeueing-inventory process. The drawback is that these policies do not react on long queues. However, inmany situations it is more appropriate to reserve a piece of raw material from the inventory already whena demand (i.e. customer) arrives. We will investigate queueing-inventory models where this policy is inforce. We analyze the stability of these networks and the form of the steady state probabilities of the jointqueueing-inventory process.This will enable us to explicitly determine some important measures of systemperformance and to perform cost analysis for these models.

90

WC-17

3 - Analysis of a polling system for dynamic order pickingJelmer van der Gaast, Erasmus University Rotterdam, Netherlands, [email protected], René deKoster, Ivo Adan, Jacques Resing

Recent trends in distribution and manufacturing, like e-commerce, have increased the importance of ware-housing significantly. Fueled by the demand for greater customers’ responsiveness and by the availabilityof new picking aids that allow information to be conveyed in real time, a new way to organize the pickingprocess, called dynamic order picking, has gathered much interest recently. In a dynamic order pickingsystem (DPS), an order picker picks orders in batches that arrive in real-time during its picking cycle, whichsubsequently changes dynamically the order picker’s current picking route. Using the theory on pollingsystems, we model and analyze a DPS system and study the joint queue length distribution at arbitraryepochs. We further determine the waiting time distribution of an individual order line, as well as, the orderthroughput time. Finally, we conduct a numerical study to demonstrate the model under consideration.Results show that dynamic order picking can generally lead to short order throughput times and higheron-time service completion ratios compared to traditional batch picking systems.

91

WD-01


� WD-01Wednesday, 14:45-15:30Rotterdam Hall

Semi-plenary speaker: Miguel AnjosStream: Semi-plenary SessionsSemi-plenary sessionChair: Mirjam Duer, Mathematics, University of Trier, 54286, Trier, Germany, [email protected]

1 - Semidefinite Relaxations: The Cutting EdgeMiguel Anjos, Mathematics and Industrial Engineering & GERAD, Ecole Polytechnique deMontreal, Montreal, Quebec, Canada, [email protected]

See page 30 for bio & abstract

� WD-09Wednesday, 14:45-15:30Diamond Room 1

Semi-plenary speaker: Fred Espen BenthStream: Semi-plenary SessionsSemi-plenary sessionChair: Jörn Sass, Fachbereich Mathematik, TU Kaiserslautern, Postfach 3049, 67653, Kaiserslautern,Germany, [email protected]

1 - Weather, risk and energy marketsFred Espen Benth, Centre of Mathematics for Applications, University of Oslo, PO Box 1053Blindern, N-0316, Oslo, Norway, [email protected]


� WD-13Wednesday, 14:45-15:30Mees Auditorium

Semi-plenary speaker: Horst TempelmeierStream: Semi-plenary SessionsSemi-plenary sessionChair: Wilco van den Heuvel, Econometric Institute, Erasmus University Rotterdam, P.O. Box 1738,3000DR, Rotterdam, Netherlands, [email protected]

92

WD-16

1 - Dynamic lot sizing with random demandHorst Tempelmeier, Supply Chain Management and Production, University of Cologne,Albertus-Magnus-Platz, D-50923, Cologne, Germany, [email protected]


� WD-16Wednesday, 14:45-15:30Oscar Auditorium

Semi-plenary speaker: Henk ZijmStream: Semi-plenary SessionsSemi-plenary sessionChair: Rommert Dekker, Erasmus University Rotterdam, Econometric Institute, Erasmus UniversityRotterdam, P.O. Box 1738„ 3000 DR , Rotterdam, Netherlands, [email protected]

1 - The European Roadmap on Logistics and Supply Chain researchHenk Zijm, Faculty of Electrical Engineering &Mathematics and Computer Science, Universityof Twente, P.O. Box 217, 7500 AE Enschede, Enschede, Netherlands, [email protected]


93

WE-03


� WE-03Wednesday, 16:00-17:30Leeuwen Room 1

Enhanced Energy Network ModelsStream: Energy and EnvironmentInvited sessionChair: Katja Breitmoser, Mathematics, FAU Erlangen-Nürnberg, Cauerstraße 11, 91058, Erlangen,Germany, [email protected]

1 - Layout optimisation of decentralised energy systems under uncertaintyHannes Schwarz, Chair of Energy Economics, Karlsruhe Institute of Technology (KIT), 76187,Karlsruhe, Germany, [email protected], Valentin Bertsch, Wolf Fichtner

Electricity generation is currently being transformed with an emphasis on renewables and low-carbongeneration as well as the implementation of smart grid technologies leading to a fundamental, structuralrearrangement of the system. In this context, intelligent home energy management approaches making useof load flexibilities are discussed increasingly often. Especially, photovoltaic (PV) systems in combinationwith heat pumps and thermal storages have attracted attention in the recent past. Such systems can bemodelled as linear program (LP) or mixed-integer linear program (MILP). In the layout planning of suchsystems, many different design options can be chosen. The dimensioning of the individual components,such as the storage size, has an immediate impact on the economic profitability of the system as a whole.Moreover, the economic profitability is subject to manifold uncertainties, e.g. the uncertainty related to thefuture development of electricity prices or the development of the electrical and thermal load as well as thesolar PV generation profiles needs to be addressed adequately in the layout planning process. Therefore,we present an integrated approach which supports the generation of consistent ensembles of load andsolar PV profiles. The approach includes the fundamental relationships between weather and load as wellas solar PV generation. It should be noted that the focus is not on forecasting but on the generation ofa representative set of consistent load profiles, which are subsequently used for an optimisation underuncertainty. In addition to techno-economic optimisation parameters, the impact of cooperation betweenseveral households is discussed.

2 - Structural Properties of Solving the AC Linear Power Flow ProblemStephan Lemkens, Lehrstuhl II für Mathematik, RWTH Aachen University, Wüllnerstr. 5b,52062, Aachen, Deutschland(+49), Germany, [email protected], Arie Koster

In this talk we will present our current work on the linearized AC Power Flow Problem. In order to designa cost minimal power grid one needs to solve the corresponding power flow problem. The most popularmodels use the so called DC Power Flow Problem in order to determine a grids feasibility. In this work, wefocus ourselves on a linearized power flow which also models reactive flows and we will show similaritiesto the well known DC Power Flow formulation. We study the problems feasibility and solution space andcompare its properties to the properties of the DC formulation. In addition we will give an insight into thecomputational complexity of the corresponding design problems.

3 - Welfare Maximization of Autarkic Hybrid Energy SystemsKatja Breitmoser, Mathematics, FAU Erlangen-Nürnberg, Cauerstraße 11, 91058, Erlangen,Germany, [email protected], Bjoern Geissler, Alexander Martin

Hybrid energy systems become a promising way for electrification of off-grid rural areas. They usuallyconsist of two or more energy sources with at least one renewable source. Storage devices can help todecouple supply and demand and increase the efficiency of hybrid energy systems, but storage possibilitiesare limited and expensive. Thus, an optimal control strategy is essential. In our model, we consider anautarkic mini-grid of households equipped with local solar panels, diesel generators and energy storage

94

WE-04

devices. We optimize the power distribution within each household as well as the distribution of powerbetween them. With respect to the given market conditions as fuel prices, investment costs, operating costsand electricity demand, our aim is to find an energy distribution that maximizes the global welfare of thewhole system. We present an MINLP model for the hybrid energy system optimization problem togetherwith computational results that confirm the suitability of our approach.

� WE-04Wednesday, 16:00-17:30Leeuwen Room 2

Economic ModelingStream: Game Theory and Experimental EconomicsInvited sessionChair: Theo Driessen, Department of Applied Mathematics, University of Twente, P.O. Box 217, 7500AE, Enschede, Netherlands, [email protected]

1 - Influence of mutations in an evolutionary inspection game with three strategicplayersGünter Fandel, FernUniversität in Hagen, Fakultät für Wirtschaftswissenschaft, Lehrstuhl fürBetriebswirtschaftslehre, insb. Produktions- und Investitionstheorie, Universitätsstr. 41, 58084,Hagen, Germany, [email protected], Jan Trockel

Non-optimal decision situations in companies lead to higher costs. Based on the inspection game of Fan-del/Trockel (2013), who focused on the influence of changed bonuses and penalty payments of three strate-gic players in one company, now the Nash solution and the strategies of these three players are discussed byevolutionary aspects. It will be examined, whether the strategic behaviour of the players can be infiltratedby mutations. In this context it is analysed, which risk is given to the three players and their decision mak-ing situations to the company. In a first step we fix one of the three decision making variables. In this caseit is shown, that the company’s costs will be minimised, if the behaviour of the company management or ofthe level-one controller respectively is given. On the contrary, it cannot be shown, that a given probabilityof the manager’s behaviour leads to a Nash equilibrium in pure strategies. In a second step the evolutionaryinfluence of three strategic players is analysed. In this case mutations infiltrate the exactly calculated Nashsolution of Fandel/Trockel (2013).

Reference: Fandel, G., and Trockel, J., 2013. Avoiding non-optimal management decisions by applying athree-person inspection game. In: European Journal of Operational Research, 226 (1): 85-93.

2 - Risk-adjusted on-line portfolio selectionRobert Dochow, Saarland University, 66111, Saarbrücken, Germany, [email protected], EstherMohr, Günter Schmidt

In an on-line portfolio selection problem an investor allocates his wealth among a variety of assets forseveral trading periods. When new prices are announced the investor must decide how to rebalance hisportfolio. The objective is to maximize the terminal wealth. Existing on-line portfolio selection algorithmsdo not consider the associated risk of assets while learning. Moreover, these algorithms solely exploit firstorder information. It may be beneficial for an algorithm to consider the fact that the greater the asset growthrate the greater the risk of a trend reversal. We present a novel risk-adjusted portfolio selection (RAPS)algorithm which incorporates the maximum return fluctuation of the constant-rebalanced portfolio. Ourempirical results on historical markets provide strong evidence that this type of risk-adjusted algorithms canbeat universal portfolio algorithms. Empirically, our proposed RAPS algorithm is able to efficiently reducethe risk of on-line portfolio selection. The risk-adjusted terminal wealth achieved by RAPS converges tothe best risk-adjusted constant portfolio in hindsight.

95

WE-05

3 - Consistent conjectural variations equilibrium in a mixed oligopolyFelipe de Jesus Castillo Perez, Mathematics, ITESM, Campus Monterrey, Ave. Eugenio GarzaSada 2501 Sur, 64849, Monterrey, Nuevo Leon, Mexico, [email protected],Vyacheslav Kalashnikov, Vladimir Bulavsky, Nataliya Kalashnykova

We study a model of mixed oligopoly with conjectured variations equilibrium (CVE) with a not necessarilycontinuous demand function. The agents’ conjectures concern the price variations depending upon theirproduction output increase or decrease. We establish existence and uniqueness results for the conjecturedvariations equilibrium (called an exterior equilibrium) for any set of feasible conjectures. To introducethe notion of an interior equilibrium, we develop a consistency criterion for the conjectures (referred to asinfluence coefficients) and prove the existence theorem for the interior equilibrium (understood as a CVEwith consistent conjectures, or CCVE). On the base of our results related to the case of non-differentiabledemand functions, we also investigate the behavior of the consistent conjectures depending upon a param-eter representing the demand function derivative with respect to the market price. The latter results allowus to analyze the behavior of groups of consumers with different consumption abilities. The proposedtechniques permit one to develop a qualitative description of the dependence of the market price on theactive demand component.

It should also be especially noted that due to the non-smooth demand function, there is possibly a pathdependency and indeterminacy of equilibria in certain cases. This is, obviously, theoretically inconvenientresults (non-unique equilibria), but at the same time, it may actually very useful in applied work. It mayalso serve as a rationale for regulatory intervention to induce a change of equilibrium when total welfarecan be improved (as, for example, with the Keynesian stimulus).

4 - One-Convexity of Bankruptcy Games with distinct linear reward functionsTheo Driessen, Department of Applied Mathematics, University of Twente, P.O. Box 217, 7500AE, Enschede, Netherlands, [email protected]

The well-known bamkruptcy problem is augmented in that, besides the estate of the bankrupt firm andthe claims of a number of creditors, also reward functions for creditors are incorporated in the augmentedmodel. It is shown that the corresponding cooperative bankruptcy games possess the one-convexity prop-erty whenever the reward functions are linear and distinct from each other. In case of linear reward func-tions with identical slopes, the classical bankruptcy games show up possessing convexity property.

� WE-05Wednesday, 16:00-17:30Goudriaan Room 1

Facility Location in Healthcare Management SystemsStream: Health Care ManagementInvited sessionChair: Twan Dollevoet, Econometric Institute, Erasmus University of Rotterdam, BurgemeesterOudlaan 50, P.O. Box 1738, 3000 DR, Rotterdam, Netherlands, [email protected]

1 - The Roadside Healthcare Facility Location ProblemHarwin de Vries, Econometric Institute, Erasmus University, Burgemeester Oudlaan 50, 3062PA, Rotterdam, Netherlands, [email protected], Albert Wagelmans, J.j. van de Klundert

Providing long distance truck drivers with adequate access to prevention, testing, and treatment services forHIV, Sexually Transmitted Infections (STIs), Tuberculosis, and Malaria is an extremely effective way toreduce the burden and the spread of these diseases. Therefore, NGO North Star Alliance locates healthcarefacilities along the major African trucking routes. Choosing the locations of new facilities presents noveloptimization problems. This paper proposes a mixed-integer programming formulation for a general designproblem, called the Roadside Healthcare Facility Location Problem (RHFLP). The RHFLP is to select

96

WE-05

locations for a set of new facilities, and to choose for each of these facilities whether or not to add healthcareservices for one or more of the diseases HIV, STIs, Tuberculosis, and Malaria to the standard health servicepackage. The objective combines on the one hand the maximization of the utilization of these facilities(by locating them at places where high demand is expected), and on the other hand the maximization ofthe extent to which the truck drivers with one of the four diseases are covered with the needed healthcareservices. To quantify the latter, we develop three definitions of the coverage score of a truck driver withone of the four diseases. In each definition, the score depends on the travel time intervals between adjacenthealthcare facilities that offer the required services along the truck driver’s route. Results from a sensitivityanalysis show that the optimal solution is relatively insensitive with respect to errors in the input data.Computational experiments show that the solutions obtained through our model promise to yield significantbenefits over the solutions obtained through the NGO’s current location strategy.

2 - Dispatching of mobile breast cancer screening units by optimization methodsMichael Schroeder, Optimization, Fraunhofer Institute for Industrial Mathematics,Fraunhofer-Platz 1, 67663, Kaiserslautern, Germany, [email protected], MartinBerger

The German breast cancer screening program offers women in the age between 50 and 69 a two-yearlycost-free x-ray examination. Breast cancer in an early non-palpable stage can be detected with higherprobability and chances of recovery are significantly raised.

In rural areas the screening programm is often conducted by trailer units that provide a fully-equippedmobile roentgenologist’s office. Travel distances are significantly reduced for the clients since they can callat the mobile unit when it is stationed at a nearby location. This can positively influence the willingness toparticipate in the screening program.

From the perspective of cost-efficient operations of the mobile screening units the cost of re-locating thetrailers on the one hand and the additional earnings by increased participation reates due to shorter traveldistances on the other hand must be balanced.

We developed two optimization models to arrive a an optimal dispatching of mobile breast cancer screeningunits. The strategic model determines the best places to show up in a region and the associated areas withinwhich clients get an invitation letter and an appointment for the respective outreach. The tatctical modelrefines this plan and determines an optimal schedule spanning over two years that specifies during whattime a mobile unit is positioned at a certain location.

We used the model to improve the operations of two mobile screening units in a larger area in Germanycomposed of six rural districts.

3 - Solution Methods for the Net Optimization ProblemTwan Dollevoet, Econometric Institute, Erasmus University of Rotterdam, BurgemeesterOudlaan 50, P.O. Box 1738, 3000 DR, Rotterdam, Netherlands, [email protected], KristiaanGlorie

In this paper, we consider the flow of sterile instruments in hospitals. These instruments, required formedical procedures such as surgeries, have to be sterilized before they can be used. In order to improveoperational efficiency and to ease the sterilization process, the instruments are grouped in nets. Any com-bination of instruments that adhere to a weight and volume restriction gives a possible net configuration.A net of instruments can be sterilized at once. Therefore, larger nets reduce the total sterilization costs.However, larger nets also lower the efficiency of their utilization, as some of the instruments in largernets will not be required for all procedures that the net is assigned to. In this talk, we introduce the netoptimization problem (NOP) as the problem of determining both the composition of the nets and theirutilization. An important aspect of the NOP is the large amount of possible net configurations that have tobe considered. Finding solutions to the NOP is the main objective in this research project. We model theNOP as an integer programming problem with a large number of variables and constraints, correspondingto the many options to form nets. We develop a row-and-column generation algorithm to solve the linearrelaxation of this problem. When deriving the pricing problem, we consider the Lagrangian subproblemto obtain the reduced costs of a net and show that the pricing problem can be modeled as a mixed integerquadratic problem and can be solved by linearization. We develop a heuristic based on the linear relaxationand show that this heuristic can solve large-scale instances within a reasonable computation time.

97

WE-06

� WE-06Wednesday, 16:00-17:30Goudriaan Room 2

Text Mining and ClusteringStream: Information Systems, Neural Nets and Fuzzy SystemsInvited sessionChair: Ralf Stecking, Fakultät II - Institut für VWL und Statistik, Universität Oldenburg, A5 0-071,26111, Oldenburg, Germany, [email protected]

1 - Clustering for Data Privacy and Classification TasksKlaus Bruno Schebesch, Faculty of Economics, Vasile Goldis Western University Arad, Str.Mihai Eminescu Nr.15, 310086, Arad, Arad, Romania, [email protected], Ralf Stecking

The most convenient situation for predictive classification occurring in business intelligence and in datamining in general is to use a large database in an unrestricted way, which leads to supervised statisticallearning tasks employing LDA, LogReg or SVM in order to determine a separation functions betweenclasses. However, such classification tasks are often subject to privacy constraints. Such constraints maynot allow the modeler to use some of the existing feature variables, like e.g. race, religion, personal iden-tification or commercially sensitive data. More generally, a modeler may not be permitted to reveal theexplicit content of the data altogether. Both, Clustering and SVM can address this problem at differentstages. They have the means to produce a restricted number of representatives for large data sets. WhileSVM achieves this by forwarding support vectors, clusterings return cluster representatives. Support vec-tors are selected representative data points describing the boundaries between classes. Hence they revealsome original data in an unrestricted way. Cluster centers are often averaged quantities concealing originaldata, but they each come with a set of cluster members, which confer further information for adaptingrepresentatives. We propose to use Clustering and SVM step-wise and cooperatively, in applying alterna-tive privacy constraints to empirical data sets where all informations are available but where their use inlater steps is restricted. We compare the predictive performance of this approach to the reference instance,where the use of all data features would be unrestricted.

2 - Lexico-semantic Patterns for Information Extraction from TextFrederik Hogenboom, Erasmus University Rotterdam, Netherlands, [email protected],Flavius Frasincar, Uzay Kaymak, Franciska de Jong

The increasing amount of news data has led to many news personalization systems. Often, these systemsprocess data automatically into information, while relying on knowledge bases (ontologies), containingdomain-specific concepts and relations. Keeping these ontologies up-to-date is a time consuming andtedious process usually performed by domain experts.

Expert knowledge-driven methods have been a main topic of research for a long time, as large amounts ofdata are not always readily available, while domain knowledge is usually at hand. These methods requireless training data than statistical methods, and their results are more insightful. Various efforts have led todifferent pattern-languages for information extraction. Most of these are based on lexico-syntactic features,although more lexico-semantic languages are emerging. However, most languages are cumbersome in use,have a limited syntax, and do not make use of domain semantics (expressed in standard languages).

Therefore, we propose a semantics-based pattern language for learning ontology instances from text forknowledge base population. The language makes use of concepts that are defined in an ontology, al-lowing for inference. Additionally, the developed language supports syntactic categories, orthographicalcategories, logical operators, and repetition.

We have evaluated our pattern language on a financial data set of 500 news messages on 10 financial eventslike mergers, profit announcements, etc. for NASDAQ 100 companies. We found that the lexico-semanticpatterns are superior to lexico-syntactic patterns in efficiency and effectivity. When applied to news eventrecognition in the domains of finance and politics, our approach has a precision and recall of approximately80% and 70%, respectively.

98

WE-06

3 - Exploiting Discourse Structure for Sentiment Analysis of TextAlexander Hogenboom, Econometric Institute, Erasmus University Rotterdam, P.O. Box 1738,3000 DR, Rotterdam, Netherlands, [email protected], Flavius Frasincar, Uzay Kaymak,Franciska de Jong

As consumers nowadays generate increasingly more Web content describing their experiences with andopinions on, e.g., products and brands, information systems monitoring people’s sentiment with respect tosuch entities are crucial for today’s businesses - not in the least as one fifth of all tweets and one third ofall blog posts discuss products or brands. The exploration of the potential of such automated sentimentanalysis techniques has only just begun.

A typical automated sentiment analysis approach is to use frequencies of positive and negative words inorder to determine whether a text is predominantly positive or negative. Such an approach ignores structuralaspects of a text, whereas such aspects may contain valuable information. We hypothesize that it may notbe so much the sentiment-carrying words per se that convey a text’s overall sentiment, but rather the wayin which these words are used. Sentiment-carrying words in a conclusion may for instance contribute moreto the overall sentiment of a text than sentiment-carrying words in, e.g., background information.

In this light, we propose to guide automated sentiment analysis by a text’s discourse structure, as identifiedby applying Rhetorical Structure Theory on sentence level. We use the identified rhetorical roles to dis-tinguish important text segments from less important ones in terms of their contribution to a text’s overallsentiment. We subsequently weight the sentiment conveyed by the identified text segments in accordancewith their respective importance when determining a text’s overall sentiment. Weights optimized by a ge-netic algorithm yield significant improvements in sentiment classification performance in comparison to abaseline not taking into account text’s discourse structure.

4 - An Automated Approach for Product Taxonomy Mapping in E-commerceDamir Vandic, Erasmus University Rotterdam - Econometric Institute, WillemBarendszoonstraat 3, 3134 LT, Vlaardingen, Netherlands, [email protected], Flavius Frasincar

In recent years, the Web has increased dramatically in both size and range, playing an increasingly impor-tant role in our society and world economy. To address this problem, the Semantic Web was conceivedto make the Web more useful and understandable for both humans and computers. One of the goals ofthe Semantic Web is aggregating information across multiple heterogeneous Web sources. A commonaggregation problem in e-commerce is the mapping of one product taxonomy to another.

We introduce the ’Semantic Category Hierarchy for E-commerce Mapping Algorithm’ (SCHEMA) to beused for mapping between heterogeneous product taxonomies from multiple sources. The main objectivefor developing SCHEMA is to facilitate the aggregation of product information from different Web sourcesand thus reducing search failures when users shop online. In order to achieve this, SCHEMA employs wordsense disambiguation techniques, using WordNet, to find synonyms of the correct sense for the categoryname. Furthermore, it uses lexical similarity measures, such as the Levenshtein distance, together withstructural information, to determine the best candidate category to map to. In order to evaluate SCHEMA,its performance is compared on recall and precision with the PROMPT and Park & Kim algorithms.

The performance of our algorithm was tested on three real-life datasets and compared with the performanceof the PROMPT and Park & Kim algorithms. Our evaluation demonstrates that SCHEMA achieves aconsiderably higher average recall than the other algorithms, with a relatively small loss of precision. Theaverage F1-score was 55.10% for SCHEMA, against 20.75% for PROMPT, and 32.52% for the algorithmof Park & Kim.

99

WE-07

� WE-07Wednesday, 16:00-17:30Antwerp Room

Hinterland Container LogisticsStream: Maritime LogisticsInvited sessionChair: Rob Zuidwijk, Decision and Information Sciences, RSM Erasmus University, Postbus 1738,3000 DR, Rotterdam, Netherlands, [email protected]

1 - Synchronizing transport requests, containers, and trucks in the hinterland of acontainer terminalTobias Buer, Department of Business Studies & Economics, Chair of Logistics, University ofBremen, Wilhelm-Herbst-Str. 5, 28359, Bremen, – Please Select (only U.S. / Can / Aus),Germany, [email protected], Jörn Schönberger, Herbert Kopfer

Considered is a pickup-and-delivery problem (PDP) in the hinterland of a port where some of the pickup-locations as well as some of the delivery locations are unknown. There are traditional PDP requests whichrequire the transport of loaded containers. In addition, there are container storage-requests and containerprovide-requests. A storage-request requires the pickup of an empty container at a given customer locationwhile the delivery location is unknown. In contrast, a provide-request requires the delivery of an emptycontainer at given customer location while the pickup location is unknown. In this PDP scenario, we modelcontainers and requests as separate entities. To decide which container shall be used to fulfill which requestwe present a request-to-container assignment problem (RCAP).

The RCAP is used to solve the hinterland PDP at hand in a sequential and in a simultaneous manner. Forthe sequential approach, the RCAP is solved first to generate the missing pickup and delivery locations.With this data, a standard PDP with less-than-truckload requests can be solved. For the simultaneousapproach, RCAP and PDP models are integrated into a single optimization model.

We present results of a computational study with a commercial mixed integer programming solver. For this,we introduce and study different test instances for the problem at hand. The behaviour of the models underdifferent settings is studied and the performance of the simultaneous as well as the sequential approachesare compared.

2 - Port-Hinterland Barge Shipping Network DesignPanagiotis Ypsilantis, Decision and Information Sciences, RSM, Erasmus University ofRotterdam, 3062PA, Rotterdam, Netherlands, [email protected], Rob Zuidwijk, MortezaPourakbar

The problem of a barge operator that serves container transport demand among several terminals is con-sidered. We consider both strategic decisions regarding setting the fleet capacity and tactical decisionsregarding the rotations and routing of barges in seaport and inland terminals. We adapt the well studiedMIP Liner Shipping Network Design formulations to fit the special characteristics of our problem. Inparticular, we propose a multimodal formulation of the problem where demand can also be satisfied bytrucks. Moreover, we consider the circulation times of barges that consist of travel times, handling timesand delays at seaport and inland terminals. Delays are mainly caused by the low priority of barges in theberthing process of seaport terminals compared to the deep-sea vessels and constitute a big part of the totalcirculation time of barges. Each OD pair of demand can be satisfied by several barge rotations while wemeasure the service level of our solution as the number of services per week per OD pair. The model selectsthe optimal fleet of barges and sets their optimal weekly rotation plan under several feasibility and servicelevel constraints. Realistic size instances of about 20 nodes can be solved to optimality with commercialsolvers like CPLEX. We apply our model to a real case depicting the connection of four inland terminalsin the Brabant region with the 10 seaport terminals in the Maasvlakte and City areas of the Port Rotterdam.Our initial results show, first, that barge rotations can suggest the use of bigger barges compared to shuttleservices with the achievement of economies of scale, and second, that the service level of the proposedplans can be enhanced via more calls per circulation without considerable cost increases.

100

WE-08

3 - Placing ships in locks using exact and heuristic methods: a decision support toolJannes Verstichel, Computer Science, KAHO Sint-Lieven - KU Leuven, Gebroeders DeSmetstraat 1, 9000, Gent, Belgium, [email protected], Patrick De Causmaecker,Greet Vanden Berghe

Ships must often pass one or more locks when entering or leaving a tide-independent port. So do bargestravelling on a network of waterways. These locks control the flow and the level of inland waterways, orprovide a constant water level for ships while loading or unloading at the docks.

We consider locks with a single chamber or several (possibly different) parallel chambers, which cantransfer one or more ships in a single operation. When transporting ships through such a lock, threeproblems need to be solved: selecting a chamber for each ship, placing ships inside the chamber, andscheduling the lockages.

This abstract considers the ship placement problem, which constitutes a daily challenge for lock masters.This problem is closely related to 2D rectangular bin packing, and entails positioning a set of ships (rect-angles) into as few lockages (bins) as possible while satisfying a number of general and specific placementconstraints.

Both an exact decomposition approach and constructive heuristic are presented. Experiments on simu-lated and real-life instances show that the exact approach generates optimal solutions within acceptablecalculation times. The heuristic obtains high-quality results in just a few milliseconds.

Both solution methods are part of a decision support tool, which allows lock masters to compute andcompare several solutions for a set of arriving and departing ships. Live tests have shown that the tool’sflexibility and high solution quality may help the lock masters in making quick and informed decisions.

4 - The Value of Information in Container TransportRob Zuidwijk, Decision and Information Sciences, RSM Erasmus University, Postbus 1738,3000 DR, Rotterdam, Netherlands, [email protected], Albert Veenstra

Planning the transport of maritime containers from the sea port to final inland destinations is challenged byuncertainties regarding the time the container is released for further transport and the transit time from theport to its final destination. This paper assesses the value of information in container transport in terms ofefficiency and reliability. The analysis uses a stylized single period model where a decision maker allocatesreleased containers to two transport modes (slow, low price, no flexible departure times versus fast, highprice, flexible departure times), where the decision maker plans the departure time of the inflexible mode.We construct Pareto frontiers and the corresponding Pareto optimal decisions under various informationscenarios and show that the Pareto frontiers move in a favorable direction when the level of informationincreases. The mathematical results are explained and illustrated by means of a numerical example involv-ing barge transport. We also study the impact of erroneous information on efficiency and reliability basedon a numerical analysis.

� WE-08Wednesday, 16:00-17:30Penn Room 1

SchedulingStream: Scheduling and Project ManagementInvited sessionChair: Marc Uetz, Applied Mathematics, University of Twente, P.O. Box 217, 7500 AE, Enschede,Netherlands, [email protected]

101

WE-09

1 - Approximating the k-Splittable Capacitated Network Design ProblemEhab Morsy, University of Siegen, 57068, Siegen, Germany, [email protected] consider the k-splittable capacitated network design problem (kSCND) in a graph G with non-negativeedge weight w(e) for all edges in G. We are given a vertex s designated as a sink, a positive cable capacityand a source set S as a subset of the vertex set of G with non-negative demand q(v) for each v of S. For anyedge e we are allowed to install an integer number h(e) of copies of e. The kSCND asks to simultaneouslysend demand q(v) from each source v in S along at most k paths to the sink s. A set of such paths can passthrough a single copy of an edge in G as long as the total demand along the paths does not exceed the cablecapacity. The objective is to find a set P of paths of G that minimizes the total installing cost. In this paper, we propose a ((k + 1)/k +r)-approximation algorithm to the kSCND, where r is any approximation ratio achievable for the Steinertree problem.

2 - An IP formulation and algorithms for an E/T parallel machine scheduling problemRainer Xavier de Amorim, Institute of Computing, UFAM, 69077-000, Manaus, Amazonas,Brazil, [email protected], Rosiane deFreitas, Eduardo UchoaThis work presents an integer programming (IP) formulation for an earliness-tardiness (E/T) schedulingproblem on single and parallel machines. Some algorithms are developed involving Branch and Cut byCPLEX tool, and some heurisitics based on Local Search and Path Relinking techniques. The compu-tational experiments shows that our strategies are competitive, achieving good solutions in a reasonableprocessing times. For single machine, with tests performed on Tanakas instances, we got all optimal so-lutions in both exact and heuristics approaches. For multi-machine instances, with instances generatingbased on OR-Library instances, our method is also suitable, reporting the first results for this variation ofthe E/T scheduling problem

3 - Automated Mechanism Design for a Sequencing ProblemMarc Uetz, Applied Mathematics, University of Twente, P.O. Box 217, 7500 AE, Enschede,Netherlands, [email protected], Ruben HoeksmaWe show how an optimal mechanism for a classic sequencing problem can be computed in polynomialtime by linear programming techniques. The setting we consider is a single machine on which set of jobshave to be scheduled non preemptively. Given the jobs’ processing times as well as waiting costs (perunit time), it is well known that the total waiting costs are minimized by Smith’s rule - schedule jobs withhighest cost per unit time first. We here address the private information setting where the jobs’ processingtimes as well as waiting costs (per unit time) are private to the jobs. Given publicly known priors forjobs’ private data, we seek to find a scheduling rule and incentive compatible payments that minimize thetotal expected waiting costs of the jobs. Here, incentive compatible refers to a Bayes-Nash equilibrium.While this optimal mechanism design problem can be efficiently solved by standard techniques when jobshave single dimensional private data, we here address the problem with two dimensional private data.We show that the problem can be solved in polynomial time by linear programming techniques, therebysolving a problem left open by Heydenreich et al. [Proc. WINE 2008, LNCS 5385, 2008, 414-425].Our implementation is randomized and truthful in expectation. The main steps are a compactification of anexponential size linear program, and a combinatorial algorithm to decompose feasible interim schedules. Inaddition, in computational experiments with random instances, we generate some more theoretical insights.An extended abstract of this paper has appeared in [Proc. IPCO 2013, LNCS 7801, 2013, 242-253].

� WE-09Wednesday, 16:00-17:30Diamond Room 1

Meta-Heuristics and Multi-criteria ProgrammingStream: Production and Operations ManagementInvited sessionChair: Matthias Klumpp, Institut für Logistik & Dienstleistungsmanagement, FOM University ofApplied Sciences, Leimkugelstraße 6, 45141, Essen, Germany, [email protected]

102

WE-09

1 - Simulation-based optimization methods for setting production planning parame-tersMargaretha Gansterer, University of Vienna, Austria, [email protected],Christian Almeder, Richard Hartl

Manufacturing companies are often faced with challenging market situations concerning product com-plexity and changing demand situations. Their customers frequently emphasize on logistics performancein addition to product quality. In order to remain competitive various decisions, which are strongly influ-encing each other although referring to different planning levels, have to be made carefully. This is mainlyinduced by an increasing level of uncertainty involved in longer planning horizons. Coordination betweendecisions taken on different levels can be approached by implementing a hierarchical planning system. Achallenging task in this context is to determine good production parameter settings in order to benefit fromwell-known planning methods. We present a framework for hierarchical production planning in a make-to-order environment which we use to identify good settings for three planning parameters, namely plannedleadtimes, safety stock, and lotsizes. Within a discrete-event simulation which mimics the production sys-tem we use a mathematical optimization model for replicating the decision problem. This mathematicalmodel is solved to optimality using a standard optimization engine. We use data referring to four differ-ent demand market situations in order to derive general statements concerning the quality and sensitivityof the three analyzed planning parameters. For exploring the parameter space we follow the concept ofsimulation-based optimization. We compare the performance of 6 different optimization methods to a kindof systematic enumeration of parameter combinations. We show that a search procedure based on the ideaof Variable Neighborhood Search (VNS) leads to the best results in this context.

2 - Variable neighbourhood search for the CLSP with sequence dependent setupsDirk Van Eikenhorst, Economics Informatics and Social science, Molde University College,Britvegen 2, 6410, Molde, Norway, [email protected]

This research looks at the capacitated lot sizing problem with sequence dependent setup cost and time.Unlike research in this field, setup carryover is not considered in this study. Instead, the machine has tobe cleaned at the end of each period, after which any product can be produced the next period withoutrequiring a setup. This problem is common in non-continuous food production environments where themachine should not be left overnight with food remaining inside. The goal of this study is to improve onthe results of the mathematical modelling based relaxation methods that were developed for this problemby means of a variable neighbourhood algorithm. This study shows a brief overview of relaxation methodsfor the problem and then compares two neighbourhood structures based on mathematical modelling thatcan be used to improve the relaxation results. The first neighbourhood is a standard k-flip neighbourhood,which requires the solution of one TSP problem and a linear problem in the move evaluation function.The second neighbourhood structure is based on solving the problem with the binary variables fixed inall but k periods. Both neighbourhoods naturally extend to be used in a variable neighbourhood searchmeta-heuristic. For each neighbourhood type, we examine the different possibilities to improve the resultsof the heuristics. Results show that the right improvement heuristic outperforms the rate of improvementcompared to a commercial solver with warm start.

3 - Relativity determination for efficiency in DEAMatthias Klumpp, Institut für Logistik & Dienstleistungsmanagement, FOM University ofApplied Sciences, Leimkugelstraße 6, 45141, Essen, Germany, [email protected]

Data envelopment analysis has been widely used in multi-criteria input and output settings such as univer-sity production for example due to the advantage of non-parametric and relative efficiency analysis. Onemajor drawback of DEA applications in university settings as well as in other areas is the dependency ofefficiency results upon the number and configuration of included universities as decision making units. Forexample a given set of universities could identify one university as most efficient (100) - but as another,even more efficient university (DMU) is added all previous results may change and the newcomer mayreceives the singular 100-efficiency-leader-result. This cannot be avoided within the DEA method - but forpractical applications i.e. in university research and management an indicator or metaheuristics regardingthe impact quantification of such variation effects could be usefull in order to judge relative efficiencymeasures obtained from DEA calculations.

Therefore this research contribution proposes to explore possible indicators for the result "stability" ofDEA calculations by performing an iterative DEA scheme with more than 100 runs with university input

103

WE-10

and output data, starting with just 10 university DMUs and increasing the number up to 110. A secondrun will be performed with another add sequence of university DMUs in order to check for variancesaccording to the sequence. Data is provided for European universities from the Times Higher EducationRanking as well as EUMIDA datasets. One such indicator to be tested could be the total value of allchanges in efficiency values for all calculated DMU or on average per DMU. The contribution will showthe prognostic value of such an added metaheuristic indicator for DEA calculations.

� WE-10Wednesday, 16:00-17:30Diamond Room 2

Inventory IIStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Marko Jaksic, Faculty of Economics, University of Ljubljana, Kardeljeva ploscad 17, 1000,Ljubljana, Slovenia, [email protected]

1 - Myopic Replenishment Policies for Inventory Systems with Lost-SalesMarco Bijvank, Rotterdam School of Management, Erasmus University, Netherlands,[email protected]

We study the problem of managing a single item inventory system under periodic review where excessdemand is lost. Demands are stationary and stochastic. We are interested in minimizing the long-runaverage total costs incurred per period. First, we consider a cost model that consists of a holding costproportional to the amount of inventory and a penalty cost proportional to the amount of lost sales. Next,we extend the model to include a fixed order cost.

When demand is backordered in case of a stock out, it is well known that an order-up-to policy is optimaland the inventory position (on-hand inventory plus inventory on order minus backorders) is used as mainindicator of the inventory status. The optimal replenishment policy for lost-sales inventory systems is notof such a simple form. Even when there are no fixed order costs, there are hardly any analytical results onthe optimal policy let alone when a fixed cost is incurred with each order.

Since it is computationally intractable to find the optimal replenishment policy for lost-sales inventorysystems, most papers propose heuristic replenishment policies. Such simplified policies are usually subop-timal with respect to total expected costs. In this paper, we develop myopic replenishment policies that onlygrow linear with the lead time such that the policy can be computed very efficiently, and they can be usedin settings without or with fixed order cost. In particular, myopic policies are very useful when demandis non-stationary or changing over time, since no steady-state analysis needs to be performed and only thedemand over a limited time horizon is taken in consideration. We numerically illustrate the performanceof such myopic policies in these general settings.

2 - Inventory Management with Transshipments under Fill Rate ConstraintsAndreas Serin, Technology and Operations Management, Universität Duisburg-Essen,Lotharstraße 65, Gebäude LB | Raum LB 215, 47057, Duisburg, Germany,[email protected], Bernd Hillebrand

Transshipments reduce system-wide inventories and costs without decreasing the required service levelsby sharing goods between different stocking locations. The research aims a model for emergency lateraltransshipments between three identical retail outlets with negligible transshipment lead times. We con-sider a 2-echelon supply chain of One Warehouse, N Retailer type under the periodic review inventorymanagement. The supply chain is composed of the external manufacturer, the central warehouse and threeretail outlets serving the end-customer demand. Any demand that cannot be satisfied immediately or aftertransshipment is lost or backlogged. Shipment lead times are assumed to be deterministic as long as thesender is not out of stock. We use a kind of "greedy transshipment policy" (Nonås/Jörnsten (2007)) to

104

WE-11

determine the quantities to transship in each period. In case the CW is out of stock, the CW utilizes therisk balancing allocation rule equalizing the myopic stockout probability during the next demand period.Numerical simulations are used to achieve the moments of transshipment quantities under generally i.i.d.demand and fill rate constraints. We also provide an estimation of the cost benefit enabled by transship-ments, all other things being equal. A quick-and-dirty approximation is provided. We verify the fit of themodel. The effect of varying lead time demand probability distributions on several performance measures(e. g. the holding resp. transshipment costs, expected inventories resp. transshipment quantities, and fillrates) should be investigated.

3 - Inventory Control with Supply BackorderingMarko Jaksic, Faculty of Economics, University of Ljubljana, Kardeljeva ploscad 17, 1000,Ljubljana, Slovenia, [email protected], Jan C. Fransoo

We study the inventory control problem of a retailer working under stochastic demand and stochastic lim-ited supply. We assume that the unfilled part of the retailer’s order is fully backordered at the supplierand replenished with certainty in the following period. As it may not always be optimal for the retailer toreplenish the backordered supply, we also consider the setting in which the retailer has a right to cancelthese backorders, if desired. For both policies, we show that the structure of the optimal inventory policyis a base-stock policy where the base-stock levels are equal in both cases. The optimal base-stock levelsare thus independent of the starting inventory position and the backordered supply from the previous pe-riod. In addition, we characterize the threshold inventory position above which it is optimal to cancel thereplenishment of the backordered supply and place a new order instead. We carry out a numerical analysisto quantify the benefits of supply backordering and the value of the cancelation option, and reveal severalmanagerial insights.

� WE-11Wednesday, 16:00-17:30Penn Room 2

Supply Chain Optimization IIStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Joachim Arts, School of Industrial Engineering, Eindhoven University of Technology, PO box513, 5600MB, Eindhoven, Netherlands, [email protected]

1 - Optimal Capacity and Safety Stock Allocation for Assembly SystemsStefan Woerner, IBM Research - Zurich, Switzerland, [email protected], Marco Laumanns,Stephan Wagner

Problems in supply chain management are usually classified according to their planning horizon as strate-gic, tactical, or operational. A common approach is to consider each class independently under simplifyingassumptions about the other classes. This can obviously lead to suboptimal decisions since they do not ex-ploit possible synergies.

Here we study the joint problem of optimal capacity and safety stock allocation, thus combining tacticaland operational decisions. We consider assembly systems with base stock policies, periodic review, andcapacity constraints. Multiple assembly systems can be connected via budget constraints for capacityallocation. Our objective is to minimize holding costs while satisfying beta-service level constraints forend items and joint budget constraints for capacity allocation. The assembly systems are evaluated viasimulation.

We present an algorithm to approximate the jointly optimal capacity allocation and reorder points. Tothis end we introduce a set of convex approximations for the non-convex optimization problem and showhow sample path derivatives can be computed analytically via infinitesimal perturbation analysis. The

105

WE-11

derivatives are used to compute the optimal solution of the convex approximations. By iteratively adaptingthe approximations of the problem we achieve good reorder points for the original problem.

We demonstrate our approach on problem instances motivated by the semiconductor manufacturing net-work of IBM Microelectronics Division.

2 - Optimal Procurement under Market Price and Repair UncertaintiesGuangyuan Yang, Econometric Institute, Erasmus University Rotterdam, Netherlands,[email protected], Adriana F. Gabor, Jan-Kees van Ommeren, Rommert Dekker

Advanced capital equipment and products with long service life span and complex configurations requirehigh level services such as maintenance, repair and overhaul, in order to ensure operational reliability andmaximize availability. However, with rapid technology developments and decreasing product life cycles,increasing number of products cease production long before the end of service life span. As a result,procurement of repairable service parts in a closed-loop supply chain is challenging, due to fluctuationsand uncertainties in demand, repair costs and market prices over a long time horizon. This paper models theproblem as a stochastic dynamic programming problem, and finds the optimal procurement policy undermarket price and repair costs uncertainties. Differing from conventional methods, our method takes intoaccounts not only demand and market price uncertainties but also repair costs uncertainties. Our resultsshow that order-up-to level policy is optimal for operations with either finite horizon or infinite horizon.Specifically, our results answer the question when it is optimal to procure service parts from the secondarymarket and how many service parts should be purchased, given a certain repair cost upper threshold. Wesuccessfully apply our results in performance-based aircraft maintenance, repair and overhaul servicesprovider.

3 - Small is Beautiful: a Framework for Evaluating and Optimizing Compact StorageSystemsNima Zaerpour, Department of Management of Technology and Innovation, RSM ErasmusUniversity, Rotterdam School of Management (RSM), Erasmus University RotterdamDepartment of Management of Technology and Innovation, P.O. Box 1738, 3000 DR,Rotterdam, Netherlands, [email protected]

Warehouses are important nodes in supply chains. They decouple supply from demand. Warehouses areparticularly needed in densely populated areas while in many of these areas space has become short. Inorder to address this issue, enterprises are moving toward new storage systems, namely compact storagesystems. In this study we propose a framework to help warehouse managers in their decision makingprocess. This framework includes response time analysis of different system configurations, investmentand operational cost comparisons, and energy consumption and CO2 emission calculations.

4 - A system approach to repairable stocking and expediting under fluctuating de-mandJoachim Arts, School of Industrial Engineering, Eindhoven University of Technology, PO box513, 5600MB, Eindhoven, Netherlands, [email protected], Geert-Jan van Houtum

We consider a single stock that stock a large variety of repairable spare parts for both condition based andcorrective maintenance of several fleets of equipment. We assume demand each repairable is a Markovmodulated Poisson process and that the modulating chain can be observed directly. This allows us tomodel the non-stationarity of demand due to condition based maintenance as well as anticipation throughmaintenance planning. The repair shop of repairables allows a certain fraction of demand to receive anshorter repair lead time, and refer to this as expediting. We condider the joint stocking and expeditingpolicy decision for the entire assortment of repairable spare parts so as meet a required service level. Wecharacterize optimal expediting policies for a single repairable under a given base-stock level. For thecomplete problem, we derive a tight lower bound via a column generation algorithm. A feasible solution isobtained by applying the feasibility pump to a relaxed version of the problem. Numerical results indicatethat the non-stationarity of demand as well as the possibility to request expedited lead times can be used toreduce stock investment by around 10%. We also present some case study results.

106

WE-12

� WE-12Wednesday, 16:00-17:30Van Oldenbarnevelt Room

Production and Distribution PlanningStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Jens Hartmann, BIBA - Bremer Institut für Produktion und Logistik GmbH at the University ofBremen, Hochschulring 20, 28359, Bremen, Bremen, Germany, [email protected]

1 - Integration of revenue and production management for a manufacturer of multipleproducts from multiple raw materialsAlireza Kabirian, California State University, Northridge, United States,[email protected]

This paper models a supply chain problem and employs simulation-based optimization to analyze it. Themodel represents a manufacturer of multiple products from multiple raw materials that has control over theprice of the products. The decisions to be optimized in the model are ordering policies of raw materials,inventory control of finished goods, manufacturing capacity of each product, and prices set on the products.The uncertainties involved are lead times of ordering inventory and the demand of the products. Weconsider the case of periodic review of raw materials and finished goods inventories on discrete time. Theobjective is to find the best configuration of the system to maximize profit. We show how simulation-basedoptimization could find the best configuration through some examples and case studies.

2 - Combining Production Planning and Rail/Road Distribution Planning for a Multi-national Supply-NetworkFrank Meisel, Martin-Luther-University Halle-Wittenberg, Gr. Steinstr. 73, 06108, Halle,Germany, [email protected], Christian Bierwirth, Thomas Kirschstein

In this talk, we consider a combined problem of production planning and intermodal rail/road distributionof goods for a large multinational production-distribution network. Joining these problems is important asdeciding on where and when to produce products strongly determines the ability for consolidating cargoand exploiting economies of scale in rail transportation. We present a model that combines the determina-tion of production volumes at the sites of the production network, the consolidation of cargo at intermodalterminals, and the booking of road and rail transport capacities for multiple products and multiple periods.On this basis, we investigate the potential for reducing the cost of production and distribution and for es-tablishing environmentally friendly rail transportation. Pareto-optimal solutions that trade off these effectsare determined by the epsilon-constraint method. The approach is tested in a comprehensive case study forthe European branch of a large company from the chemical industry.

3 - Stochastic optimization of operational production planning for fisheriesKrystsina Bakhrankova, Applied economics, SINTEF - Technology and society, Box 4760Sluppen, S. P. Andersens veg 5, 7465, Trondheim, Norway, [email protected],Kjetil Midthun, Kristin Tolstad Uggen

The fishing industry is one of the main contributors to the national economy, value creation, and employ-ment in Norway. Furthermore, it is a significant source of export incomes. The fishing industry is alsoa well-known arena for applying operations research methodology. Traditionally divided in three mainparts, the works within this area have dealt with fish stock and harvesting, fish processing, and marketing.Recently, the focus has shifted to integrated planning, where fishing fleet operations are combined withplant processing. Currently, a broader view of the supply chain needs to be adopted as many companiesin this industrial sector are striving to improve their capacity utilizations, operational efficiency, and prof-itability. Thus, both upstream and downstream uncertainties need to be handled directly. While it has beenrecognized that decision flexibility can be used to manage supply chain uncertainty, no known stochasticmodeling formulations have explicitly accounted for it in fish processing.

107

WE-13

To address the described planning challenges, this paper develops a stochastic model, incorporating bothupstream (raw material quantities) and downstream (finished goods prices) uncertainties, while accountingfor fish quality deterioration and shelf-life restrictions. The model is used on a number of case studies,estimating the value of flexibility in the supply chain provided by the introduction of super-chilling tech-nologies and application of the described stochastic formulation. This way, it reflects a triangulation oftechnological development, operational efficiency, and market profitability. Thus, it is a unique opportu-nity to address the real-world complexity and enhance the body of knowledge in operations research.

4 - Quasi-real-time control of integrated production and transport operationsJens Hartmann, BIBA - Bremer Institut für Produktion und Logistik GmbH at the University ofBremen, Hochschulring 20, 28359, Bremen, Bremen, Germany, [email protected],Thomas Makuschewitz, Bernd Scholz-ReiterManufacturing supply chains source raw material on a global scale, utilize regional cost advantages forproduction and embed partners with outstanding expertise in their networks. Their goal is to create andto sustain a competitive and reliable network of distributed production facilities that are connected via anefficient transport system. To this end, decentralized operations need to be precisely synchronized in orderto materialize potential benefits. In practice, supply chains operate in a dynamic context, where process-ing times are stochastic and processes are subjected to disturbances. Nowadays, operational planning andcontrol activities are quite often performed in a disconnected manner, whereas an integrated view of pro-duction and transport processes enables further possibilities to mitigate the consequences of disturbancesin the overall system. This talk presents an approach for the quasi-real-time control of integrated manufac-turing supply chains on the operational level, where the execution of production and transport operationsfollows a given schedule. A signal based fault detection method monitors the system and recognizes crit-ical deviations of the systems parameters from a normal state, which make the execution of the scheduleinfeasible. The occurrence of a critical fault triggers a rescheduling routine that solves the integrated pro-duction and transport scheduling problem (PTSP) for the running system, i.e. it was designed to take thecurrent system status into account and reschedule only the remaining jobs. The methods are applied toa test scenario and the results of the numerical analysis show that the combination of fault detection andpartial rescheduling enables a control of integrated production and transport systems in quasi-real-time.

� WE-13Wednesday, 16:00-17:30Mees Auditorium

Railway Networks, Routing and SchedulingStream: Traffic and TransportationInvited sessionChair: Michael Römer, Juristische und Wirtschaftswissenschaftliche Fakultät,Martin-Luther-Universität Halle-Wittenberg, Universitätsring 3, 06108, Halle (Saale), Germany,[email protected]

1 - An interactive optimization framework for the vehicle scheduling problemHanno Schülldorf, GSV, Deutsche Bahn AG, Stephensonstr. 1, 60326, Frankfurt am Main,Germany, [email protected] vehicle scheduling problem in rail freight service is to assign a locomotive to each train of a giventimetable. We use a case study of this problem to show some aspects of bringing OR and practice together.In practice, the answer of OR that there is no better solution (because of the lower bound) often is notsufficient. Instead, a manager might be more interested in the question how to make a given goal feasible.On the other hand, a planner might not be able to do all the data work, in particular if default values are notsufficient. We developed an interactive optimization framework that reduces unnecessary data work andalso acts as a guide for planners to make minimal adjustments to the timetable for a much better vehicleschedule.

108

WE-14

2 - Analysis of Micro-Macro Transformations of Railway NetworksMarco Blanco, Zuse Institut Berlin, Germany, [email protected], Thomas SchlechteThe track allocation problem in railway networks can be formulated as the task of minimizing the totaltime needed for scheduling a set of fixed train routes through a network without incurring into capacityor operative conflicts. Efficient algorithms for solving this problem are of great importance since railwaysafety systems operate on the same principle all over the world.In the literature, very precise, so called microscopic, models for the track allocation problem have beendeveloped. However, microscopic models are inappropriate for applied mathematical optimization becauseof the size and high level of detail. Hence, most optimization approaches consider simplified, so calledmacroscopic, models. The challenging part is to construct a reliable macroscopic model for the associatedmicroscopic model and to facilitate the transition between both models.In this paper, we focus and analyze the effects of time discretization in micro-macro transformations fortrack allocation models from the literature. In particular, we consider the cumulative rounding algorithmintroduced by Schlechte et al., which performs well in practice but for which no thorough theoreticalanalysis has been carried out.We consider three natural optimality criteria for the track allocation problem, and prove that the cumulativerounding strategy satisfies two of them. Furthermore, as our main result we prove that no strategy existsthat satisfies all three optimality criteria, thus certifying the quality of cumulative rounding. We also presentsome cases that show that under extreme circumstances the cumulative rounding technique can performinconveniently with respect to the global optimality criterium, i.e., losing a factor of 2 or underestimatingthe optimal value.

3 - Crew scheduling and benefits from integration of the planning process at a freightrailway operatorLukas Bach, Aarhus University, Fuglesangs alle 4, 8210, Aarhus V, Arhus, Denmark,[email protected], Twan Dollevoet, Dennis HuismanWe investigate to what degree we can integrate a Train Timetabling / Engine Routing Problem and the CrewScheduling Problem, based on a case at a freight railway operator, DB Schenker Rail Scandinavia. Theplanning process at railway operators can be described in five levels; Line Planning, Timetabling, EngineRouting, Crew Scheduling and Crew Rostering In the Timetabling Problem we design a timetable for thedesired lines - from the Line Planning Problem - and fix the departure and arrival times. Also, we allocatetime-slots in the network to secure a feasible timetable. Next we assign engines in the Engine RoutingProblem to the lines in accordance with the timetable. We ensure that this allocation produces a feasibleplan. Often the objective would be to minimize cost. In this paper we consider an integrated version ofthe Timetabling and Engine Routing problem faced by a freight railway operator. We then integrate thesetwo phases with the crew scheduling; the overall goal is to compare whether it is favorable to consider thecrew scheduling at these earlier stages. The overall integration is achieved by obtaining an optimal solutionfor the Timetabling / Engine Routing Problem, while exploiting the fact that numerous optimal, and nearoptimal solutions exists. We extract the solutions where it is possible to alter the timetable while keepingengine routes intact. The Crew Scheduling model is allowed to use sub-optimal solutions if the additionalcost is outweighed by the crew savings. This information is implemented in a mathematical model for theCrew Scheduling Problem. The model is solved using a column generation scheme. Hereby it is possiblefor the Crew Scheduling algorithm to adjust the timetable, and achieve a better overall solution.

� WE-14Wednesday, 16:00-17:30Tokyo Room

Multi-objective optimizationStream: Decision Analysis and Multiple Criteria Decision MakingInvited sessionChair: Martin Josef Geiger, Logistics Management Department, Helmut-Schmidt-University,Holstenhofweg 85, 22041, Hamburg, Germany, [email protected]

109

WE-14

1 - An enumerative cutting plane approach to integer linear vector optimization prob-lems.Walter Habenicht, Business Administration, University of Hohenheim, Lst. Beschaffung undProduktion (580 B), 70593, Stuttgart, Germany, [email protected]

The approach presented in this paper is based on the concept of intersection cuts, which is well known inclassical integer linear programming. The method starts by generating an optimal basic solution of the lp-relaxation for some reasonable weighted sum of the objectives. The optimal basic solution is presented tothe decision maker and he is asked to define a region around the optimal solution in objective space. Infact, we assume this region will be a hypercube. The feasible integer points in the interior of this hypercubeare enumerated. The efficient ones are presented to the decision maker, and he is asked to choose the bestone out of them. Then, the hypercube serves as a cut-generating set. The generated cut cuts off only thosefeasible integer points that have been enumerated before. Comparing the best feasible integer solutionfound so far with the (non-integer) vertices of the facet generated by the cut allows us to decide whetherto stop the process or not. If the best feasible solution it preferred to the best vertex of the facet we canassume that under rather customary conditions the best solution is found. Otherwise, the best vertex of thefacet is used as the next basic solution of the lp-relaxation and another iteration is performed.

2 - Generalized Multiple Objective Bottleneck ProblemsKathrin Klamroth, Department of Mathematics and Informatics, University of Wuppertal,Gaussstr. 20, 42097, Wuppertal, Germany, [email protected], Jochen Gorski,Stefan Ruzika

We consider multiple objective combinatorial optimization problems where one objective has arbitrary typeand the remaining objectives are bottleneck or k-max objectives. While the objective value of a bottleneckobjective is determined by the largest cost value of any element in a feasible solution, the k-th largestelement defines the objective value of the k-max objective. An efficient algorithm for the generation ofthe complete nondominated set is developed, which is independent of the specific combinatorial problemat hand, and which is polynomial whenever the corresponding single-objective problem is. This impliespolynomial time algorithms for several important problem classes like shortest path, spanning tree, andassignment problems with bottleneck objectives which are known to be NP-hard in the general multipleobjective case.

3 - An Interactive Approach to the Bi-objective Inventory Routing ProblemMartin Josef Geiger, Logistics Management Department, Helmut-Schmidt-University,Holstenhofweg 85, 22041, Hamburg, Germany, [email protected], Sandra Huber, MarcSevaux

The talk presents an interactive approach for the bi-objective inventory routing problem (IRP). The IRPcombines delivery quantity decisions with a vehicle routing problem into a simultaneous model. In thisproblem setting, a considerable tradeoff between the minimization of the inventory levels on the one hand,and the minimization of the transportation effort (i.e. the traveled distances) on the other hand is present.A local search approach on the basis of a multi-point hillclimber is used for computing solutions. Aftereach solution construction phase, the decision maker is actively involved in the intermediate decisionmaking process. He/she is allowed to state personal preferences, thus guiding the subsequently derivedsearch direction. We investigate the effectiveness of a reference point-based approach for the problem athand. A prototypical implementation of the concept into a running system has been made available, andexperiments can and have been conducted on its basis. In addition to the study of the above mentionedlogistical problem, we also look into ideas of how to simulate a decision maker in silico. This related to ourmotivation of obtaining more general insights into the behavior of interactive search-and-decision-makingsystems for such multi-objective combinatorial optimization problems.

110

WE-15

� WE-15Wednesday, 16:00-17:30New York Room

Graphs and NetworksStream: Discrete and Combinatorial Optimization, Graphs and NetworksInvited sessionChair: Derek Verleye, Industrial Management, Ghent University, Technologiepark 903, 9052,Zwijnaarde, Belgium, [email protected]

1 - On the optimal layout of a parking lotMichael Stiglmayr, Department of Mathematics and Informatics, University of Wuppertal,Gaussstr. 20, 42119, Wuppertal, Germany, [email protected], Kathrin KlamrothWe consider the problem of finding a layout for a given size of a rectangular parking lot with the maximalnumber of individual parking spaces. A parking lot is assumed to have one exit in a corner of the rectangu-lar, and every individual parking space has to be connected to the exit by a street. In our model we dividethe parking lot into unit squares. Each unit square can either be a individual parking space or part of astreet. We present two different mixed integer programming formulations for this combinatorial optimiza-tion problem. The difference between these two models is in the way the ’connection to the exit’-constraintis modeled. The fist formulation uses additional integer variables, which measure the distance of a fieldto the exit along the streets. The second formulation uses a network flow model, where each unit squareproduces one unit of flow and only the exit field has a demand of flow. To compute lower bounds veryefficiently we use a heuristic construction scheme, which produces high quality solutions. Upper boundscan be obtained by the optimal ratio between street fields and parking spaces. We present preliminarycomputational results testing both model formulations.

2 - A copositive program for the Graph Isomorphism ProblemLuuk Gijben, Rijksuniversiteit Groningen, Netherlands, [email protected], Mirjam DuerThe Graph Isomorphism Problem is the problem of deciding whether two graphs are isomorphic. That is,whether they are the same graph or not, after a possible relabelling of the vertices. It is a problem withmany applications, most notably in computer science and chemistry. In the latter the Graph IsomorphismProblem is used to help classify chemical compounds by formulating them as graphs. It is also one ofthe older combinatorial problems whose complexity still has not been solved. That is, it still unknownwhether this problem is NP-complete or not. In this talk we will present a new approach to the graphisomorphism problem by formulating the problem as a copositive program. Then using several hierarchiesof cones, approximating the copositive cone, we will attempt to solve the graph isomorphism using bothlinear programming and semidefinite programming. In particular we will derive upper bounds on thelevel of precision needed in these approximation hierarchies to be able to definitively decide the graphisomorphism problem.

3 - On the number of supported nondominated points of the multiple objective span-ning tree problemStefan Ruzika, Department of Mathematics, University of Koblenz, 56070, Koblenz, Germany,[email protected], Florian SeippWe consider the multiple objective spanning tree problem: given a finite number of objective functions,we are looking for the set of Pareto optimal solutions of this problem, i.e., a set being minimal withrespect to the componentwise ordering. After reviewing the state-of-the-art including complexity andstructural results as well as algorithmic ideas, we state and prove our main result: the number of supportednondominated points of the multiple objective spanning tree problem is polynomially bounded in the inputsize. The proof relies on a collection of ideas such as the computation of supported solutions by weighted-sum problems, the solvability of the single objective spanning tree problem by greedy algorithms, andfinally the decomposition of the weight space into regions of equal edge ordering. Counting these regionsthen implies the result. It should be noted that this is the first result of its kind in multiple objectivecombinatorial optimization and that it can be transferred to more general matroid problems.

111

WE-16

4 - An extension of the Gradient method to find optimal solutions for multi-periodwater production and distribution planning in complex networksDerek Verleye, Industrial Management, Ghent University, Technologiepark 903, 9052,Zwijnaarde, Belgium, [email protected], El-Houssaine Aghezzaf

Nowadays, many drinking water companies rely on simulation software packages (EPANET, Infoworks)for the daily operation of their network. This software allows the user to make accurate and thus reliablesimulations of the network for a chosen configuration. However, the user-defined scenario’s do not nec-essarily lead to the lowest costs. Very often the pumps are working at a suboptimal point in their systemcurve, or water production centers are not being used efficiently. The algorithm to find solutions to specificconfigurations for water supply networks in EPANET is the "Gradient method’ developed by Todini andPilati (1987) and later by Salgado (1988), which is basically an implementation of the Newton method fornonlinear equations. After an initialization based on the network design, the values of flow and head areupdated iteratively until some convergence criterion is reached. Recently, Simpson and Elhay (2011) pre-sented the corrected formulation for the Jacobian matrix, restoring quadratic convergence of the method.Here, we present a complete model including pumps and buffers, for which we compute the optimal so-lution. We start with the corrected formulation for the Gradient method, add an objective function thatminimizes production and energy costs and solve the model as a MILP. Although the values for the pres-sure in nodes may be inadequate in the first couple of iterations, we define rules to assure the final valuesto be within the limits. If we allow enough degrees of freedom, we are able to find global optimal solu-tions for multi-period drinking water distribution systems in . The algorithm is tested on several samplenetworks and results are thoroughly analyzed and discussed.

� WE-16Wednesday, 16:00-17:30Oscar Auditorium

Young Participant with Most Academic/Practical Impact AwardsStream: Young Participant with Most Impact AwardsInvited sessionChair: Johann Hurink, Department of Applied Mathematics, University of Twente, P.O. Box 217, 7500AE, Enschede, Netherlands, [email protected]: John Poppelaars, NGB (Dutch OR Society), Netherlands, [email protected]

1 - From structures to heuristics to global solversTimo Berthold, Optimization, Zuse Institute Berlin, Takustraße 7, 14195, Berlin, Germany,[email protected]

Primal heuristics are an important component of state-of-the-art codes for mixed-integer nonlinear pro-gramming (MINLP). In this presentation, we introduce two novel large-neighborhood search heuristics,Undercover and RENS, that we designed to be employed as start heuristics inside a global solver. Un-dercover explores a mixed-integer *linear* subproblem of a given MINLP. Therefore, an auxiliary vertexcovering problem is solved to identify a smallest set of variables to fix such that each constraint is lin-earized. RENS uses a sub-MINLP to exploit the set of feasible roundings of a given (fractional) solutionof a relaxation. We give theoretic motivations and discuss implementation details of both approaches.Computational results assess the ability of these heuristics to find feasible solutions and their impact onthe overall performance of the MINLP solver SCIP. In this connection, we introduce a new performancemeasure, the primal integral, which depends on the quality of solutions as well as on the points in timewhen they are found.

112

WE-17

2 - Applications of Operations Research to Chronic Disease ManagementMariel Lavieri, The University of Michigan, MI 48109-2117, Ann Harbor, United States,[email protected]

My research focuses on the application of operations research to healthcare, with a special interest indynamic programming, stochastic control, and partially observable state space models. Healthcare deliv-ery often involves sequential decisions that have long-term implications. Those decisions are based onhigh dimensional state spaces, which pose a problem for traditional modeling paradigms. In some keyinstances, transition probabilities might not be known, but instead are random variables that are learnedas new information becomes available. As a first step, I describe some of my ongoing research modelingscreening, monitoring and treatment decisions of patients with chronic conditions. The models are moti-vated by diseases such as glaucoma, coronary heart disease and cancer. Key to the models developed isthe incorporation of the individual patient’s disease dynamics into the parameterization of the stochasticmodels of the disease state evolution. Model conception and validation is described, as well as the role ofmultidisciplinary collaborations in ensuring practical impact of my work.

� WE-17Wednesday, 16:00-17:30New Orleans Room

Resource allocationStream: Applied Probability and Stochastic Programming, ForecastingInvited sessionChair: Ger Koole, Mathematics, VU University Amsterdam, De Boelelaan 1081 a, 1081 HV,Amsterdam, Netherlands, [email protected]

1 - The Stochastic Dynamic Distance Optimal Partitioning Problem and a generalizedk-server ProblemR. Hildenbrandt, Inst. Mathematik, TU Ilmenau, Germany, [email protected]

The Stochastic Dynamic Distance Optimal Partitioning Problem and a generalized k-server Problem TheStochastic Dynamic Distance Optimal Partitioning problem (SDDP problem) is a complex OperationsResearch problem. The SDDP problem is based on an industrial problem, which contains an optimalconversion of machines. Superordinately regarded, the SDDP problem is a DA (”decision after”) stochasticdynamic programming problem. If we disregard the given probability distribution for the SDDP problemwe obtain a generalized k-server problem with parallel requests where several servers can also be locatedon one point. In this talk we will introduce the problems present results and show relations between theseproblems.

2 - Solving a stochastic territory design problem via a stochastic load balancingframeworkMonica Elizondo-Amaya, Graduate Program in Systems Engineering, Universidad Autonoma deNuevo Leon, Mexico, [email protected], Roger Z. Rios-Mercado, David Morton,Erhan Kutanoglu

Territory design can be defined as the problem of grouping small geographic units into larger geographicclusters called territories in such a way that the latter satisfy certain planning criteria. This work proposesa stochastic version of a commercial territory design problem in the presence of uncertain demands. Astochastic integer programming model is introduced. The objective is to minimize the territorial imbalanceswith respect to product demand subject to planning criteria such as territory connectivity, compactness,and balance with respect to the number of customers. Given its structural relationship with our model, westudy makespan minimization involving stochastic scheduling on parallel machines. In these problems, anumber of identical machines operating in parallel are to be used to process a batch of jobs with stochastic

113

WE-17

processing times so as to minimize the expected makespan. It has been established in the literature that onlyunder certain restrictive conditions the strategy of prioritizing jobs with the longest expected processingtime (LEPT) minimizes the expected makespan. We study our problem from the perspective of thesegeneral stochastic scheduling rules, and we provide a heuristic solution for the commercial territory designproblem with uncertain demands in the framework of stochastic load balancing by adopting the LEPTrule. Graham’s Rule (which allocates randomly ordered jobs to the machine which currently has thesmallest load) combined with an LEPT-based policy provide a natural heuristic for this problem. Themain contribution of this work is the study of the LEPT policy for a class of stochastic multiple machineproblems at solving more general models.

3 - Managing the Risks of Staff Planning PoliciesSusanne Wruck, Distribution and Supply Chain Logistics, VU University Amsterdam, DeBoelelaan 1105, 1081 HV, Amsterdam, Netherlands, [email protected]

Staff planning plays a vital role in many industries, as labor is typically accompanied with high costand restricted by specific conditions and statutory rules. In situations with uncertain demand, workforceplanning becomes crucial and a complex task. Specific working contracts, skills, and productivity of theworkers might complicate the staffing decisions. We consider a staff planning problem for commercialwarehouses. In e-commerce setting, the amount of customer orders and returns to be processed mightbe highly fluctuating. Uncertainties of workforce capacities and external influences, such as short-noticepromotion offers, further shape the stochastic environment of warehouse staffing problems. We develop arisk management tool which incorporates a stochastic optimization model and the choice between severalrisk control strategies. Classical approaches, which rely on the replacement of the stochastic parametersby expected values will only lead to good performance when the realizations corresponds approximatelywith the expectation and staffing policies which prepare for the mean case might fail to provide reasonableperformance for the majority of possible scenarios. In line with previous research on similar staffingproblems our decision tool is based on stochastic programming. Stochastic programs allow flexibility toreact on realizations of stochastic parameters within the planning horizon. We aim to provide a decisiontool, which allows to account for risk protection in the optimization. In numerical experiments we study theimpact of specific risk control variants for common scenarios. Doing so, we are able to provide guidance forpractitioners to choose appropriate risk control strategies to achieve high performance and robust planning.

4 - Adaptative Blending in Call CentersGer Koole, Mathematics, VU University Amsterdam, De Boelelaan 1081 a, 1081 HV,Amsterdam, Netherlands, [email protected], Benjamin Legros, Oualid Jouini

We examine a model with inbound calls and emails. There is an infinite source of emails, calls arriveaccording to a non-homogeneous Poisson process. Non-preemptive priority is given to calls, emails areonly assigned if the number of available agents exceeds a certain threshold level. The objective is tomaximize the email throughput while satisfying a service level constraint for the inbound calls. The optimalthreshold depends on the parameters of the model. We propose different adaptive methods to change thethreshold and report on experiments.

114

TA-03

Thursday, 9:00-10:30

� TA-03Thursday, 9:00-10:30Leeuwen Room 1

Energy Storages: Location and OperationStream: Energy and EnvironmentInvited sessionChair: Katrin Schulz, Faculty of Management and Econcomics, Chair of Operations and Research andAccounting, Ruhr-University Bochum, Universitätsstraße 150, GC 1/145, 44780, Bochum, Germany,[email protected]

1 - Comparison of heuristics towards approaching a scheduling and capacity plan-ning MINLP for hydrogen storage in chemical substancesSimon Schulte Beerbühl, Institute of Industrial Production, Karlsruhe Institute of Technology,Hertzstrasse 16, 76187, Karlsruhe, Germany, [email protected], Magnus Fröhling,Frank Schultmann

OR scheduling and capacity planning problems for electricity storage technologies such as compressed airor pumped hydro storage usually rely on linear modelling. However, hydrogen as energy storage optionrequires non-linear modelling due to strong load-dependent electrolyser efficiency. Hydrogen conceptsproducing base chemicals — such as ammonia and methanol — for capturing the produced hydrogen re-quire further sophisticated modelling, especially of approximated linear or non-linear as well as compositefunctions for plant operating characteristics. Such models therefore consist of non-linear as well as non-differentiable parts in its problem formulation, leading to mixed integer non-linear programs (MINLP).We present a combined scheduling and capacity planning model for a hydrogen - ammonia concept, whichhas been developed together with ThyssenKrupp Uhde, an ammonia plant engineering company. Themodel is formulated as a MINLP. Due to convergence problems of the MINLP two heuristic approachesfor approximation have been developed and will be presented. These approaches decompose the MINLPto refrain from one of the two aspects, solve the simplified program as either differentiable non-linearor mixed-integer linear problems and feature a post-optimisation adjustment for meeting the previouslyallowed discrepancies. By using best- and worst-case scenarios for the simplifications, upper and lowerbounds of the MINLP objective value can be derived. Both approaches will be compared with regard tothe way they influence the simplified optimum and especially the post-optimal adjustments. Comparingthe upper and lower bounds in each modelling case (incl. best and worst-case simplifications) allows for aquantitative assessment of the two types of approaches.

2 - The need for electricity storage systems and their allocation in the German powersystemSonja Babrowski, Chair of Energy Economics (IIP), KIT, Germany, [email protected],Patrick Jochem, Wolf Fichtner

Due to the increasing share of renewable energy the need for storage systems arises in the German en-ergy sector. To estimate the future need for daily storage systems as well as the best allocation for themwe enhanced the existing long-term energy system model PERSEUS-NET. The investment and dispatchmodel minimizes the system relevant costs and includes a DC approach of the German transmission grid.Considering technical and economic constraints the electricity demand at each grid node has to be satisfiedby either electricity transfer via the transmission grid or electricity generation in existing or newly buildpower plants. Because of the ambitious target of the German government of more than 60 % generationout of renewables in 2040 and the resulting capacities the generation of renewables has either to be cutin hours with a higher supply than demand or the electricity has to be stored. Therefore, we integratedstorage extension options (chemical batteries and potential hydro storages) at each grid node. To preventthat storage systems store and generate at the same time, using their efficiency to destroy electricity in

115

TA-03

order to minimize the costs of load changes by thermic generation units binary variables are needed. As ofthe high computing times that comes with the implementation of such a high number of storage systemsthe optimization model is solved twice. Once as a linear problem with ideal storages (100 % efficiency)to preselect possible storage allocation and the second time as mixed integer problem with real storages.For the second run the installation of the ideal storages is considered as an upper limit to the economicinstallation of real storages. Additionally, the load shift potential of electric vehicles is considered as analternative.

3 - Combining Optimization and Simulation for planning Water Tanks in Water SupplySystemsCorinna Hallmann, DS&OR Lab, University Paderborn, Warburger Strasse 100, 33098,Paderborn, Nordrheinwestfalen, Germany, [email protected]

In Germany, the water consumption has decreased in recent years. This causes the German municipalutilities to face several problems. One of those problems is that the water supply systems were built tohandle increasing water consumption, and so the components of water supply systems do not have theright dimensions to work efficiently anymore. So the German municipal utilities have to think aboutreplacing the old components by newer ones which may have improved dimensions. This work focuseson the optimization of water tanks in water supply systems. The main goal is to minimize the cost of thewater tanks. This includes the operational cost as well as the investment cost of new tanks. Aside fromthat, the optimal location and the optimal dimension of new tanks are decided. In addition, it is decided ifolder tanks may be replaced by newer ones. The decision making is done by running an application whichcombines network reduction, mathematical optimization and hydraulic simulation. The combination ofthose techniques is discussed in this talk.

4 - Influence of fluctuating electricity prices due to renewable energies on heat stor-age investmentsKatrin Schulz, Faculty of Management and Econcomics, Chair of Operations and Research andAccounting, Ruhr-University Bochum, Universitätsstraße 150, GC 1/145, 44780, Bochum,Germany, [email protected], Matthias Schacht, Brigitte Werners

Under the German Energy Act the feed-in of renewable energies into the grid is prioritised and leads to avolatile and stochastic residual power load. The resulting load pattern highly influences electricity pricesand constitutes a major challenge for energy providers, especially for municipal supply companies whichalso provide their customers with district heating. Efficient and variable combined heat and power (CHP)plants are used to fulfil the unsteady, usually asynchronous loads for heat and residual power wherebyelectricity may also be traded on the spot market. Additional flexibility can be gained by the use ofheat storages to decouple production from demand. This allows a power price-oriented operation of theCHP plant to realise profits. Revenues from electricity sales in times of high power prices and savings ofgeneration costs in times of low prices determine the added value of a heat storage. In order to supportthe investment decision of a municipal energy provider we quantify the influence of the value drivers fora heat storage which can be determined by analysing the conditions for its use in day-to-day operations.Integrating a well-known linear model we optimise the different plant operations and power trade in ageneration portfolio with and without heat storage. Using historical data from the German power marketand the corresponding renewable energy feed-in we simulate several weekly scenarios. Focussing onweeks with high impact on the profitability of heat storage devices we identify specific demand patternsand price developments. Results quantify the value of the storage depending on the extent and duration offluctuations in the feed-in of renewable energies and corresponding prices.

116

TA-04

� TA-04Thursday, 9:00-10:30Leeuwen Room 2

Option Pricing and Market ModelsStream: Financial Modeling, Banking and InsuranceInvited sessionChair: Bernd Heidergott, Econometrics and Operations Research, VU Amsterdam, De Boelelaan,Amsterdam, Netherlands, [email protected]

1 - Numeraire portfolios and utility based price systems under proportional transac-tion costsJörn Sass, Fachbereich Mathematik, TU Kaiserslautern, Postfach 3049, 67653, Kaiserslautern,Germany, [email protected]

Even in a complete discrete-time financial market model like the binomial CRR model it might happenthat, when facing proportional transaction costs, the replication price for a financial derivative is higherthan the superhedging price. In such a case, the price system based on replication is not consistent, it maylead to arbitrage. With costs, consistent prices are bounded by the superhedging price from above and bythe maximum amount one can borrow against the claim from below. How can we find a consistent pricesystem? In an incomplete market without transaction costs, consistent prices for financial derivatives canbe obtained by taking expectation of the claim with respect to some martingale measure. The numeraireportfolio allows to replace this change of measure by a change of numeraire. A well known approach isto find the growth optimal portfolio as candidate for the numeraire portfolio, but the numeraire portfolioin the strict sense (i.e. generating a martingale measure) might not exist. Using methods from dynamicprogramming, in a discrete-time market with independent stock returns it can be shown that a numeraireportfolio under proportional costs exists which leads to a consistent price system. The same approachworks for so called numeraire pairs derived from optimal strategies for maximizing expected power utilityfrom terminal wealth. This again leads to consistent price systems. The dynamic programming approachis to some extent constructive and we present several examples.

2 - Options Sensitivities with respect to Pay-Off Regions BoundariesBernd Heidergott, Econometrics and Operations Research, VU Amsterdam, De Boelelaan,Amsterdam, Netherlands, [email protected]

In this talk we propose a unified approach to sensitivity analysis of option premiums with respect to somearbitrary parameter. The classical framework assumes that the underlying parameter is some intrinsicparameter of the model, e.g., interest rate, volatility or time to maturity, in which case sensitivities are alsoknown as "Greeks". While intrinsic parameters only induce variability in the dynamic of the stock-price(s)under consideration, the present approach allows the parameter under discussion to induce variability inthe payoff function and also in the exercise rule of the option. Our approach essentially relies on thereduction of the differentiation of surface integrals to that of differentiation of one-dimensional integralsby means of conditioning. We develop a general Monte Carlo integration technique for evaluating suchintegrals that relies on a conditioning approach, where the conditioning is used in order to smooth out thenon-differentiability. This approach is known in the literature as smoothed perturbation analysis (SPA) andthe resulting estimators are of the SPA type. In this talk we will focus on an application of our techniqueto sensitivity analysis of Parisian options.

117

TA-05

� TA-05Thursday, 9:00-10:30Goudriaan Room 1

Ambulance planningStream: Health Care ManagementInvited sessionChair: Rob van der Mei, Centrum Wiskunde & Informatica, 1098 XG, Amstedam, Netherlands,[email protected]

1 - Probabilistic Ambulance Location Model in a Time-dependent EnvironmentPieter van den Berg, Delft Institue for Applied Mathematics, TU Delft, Netherlands,[email protected]

We study the ambulance location problem for emergency medical services (EMS). Although travel timesand demand rates in general vary over the day, most ambulance location models are time-independent.We discuss a probabilistic location model with time-dependent travel times and demand rates, in whichwe maximize the expected coverage. We compare this model with time-independent models and see thattaking time-dependency into account might result in better solutions.

2 - Near-optimal policies for a Dynamic Ambulance Management modelThije van Barneveld, Stochastics, Center for Mathematics and Computer Science, Science Park123, 1098 XG, Amsterdam, Netherlands, Netherlands, [email protected]

We study the Dynamic Ambulance Management (DAM) problem in which one tries to retain the coverageof a region based on the number of available ambulances and possibly their locations. The road network ismodeled as a graph with two types of vertices: nodes in which an accident can occur and nodes in whichthis cannot happen. We develop a Markov Decision Process (MDP) that can solve this problem exactly,under some assumptions. For large problem instances, we combine the MDP with simulation and learningalgorithms to obtain nearly optimal policies.

3 - Ambulance redeployment policies and the impact of city structureCaroline Jagtenberg, CWI, Netherlands, [email protected]

We address the problem of dynamic ambulance redeployment in urban regions, where the goal is to mini-mize the percentage of ambulances arriving later than a certain threshold value. We model the problem asa Markov decision problem, and develop a discrete event simulation model that captures many aspects ofthe uncertainties that occur in practice. We use techniques from ordinal optimization to gain insight intothe difficulty of the problem. In particular, we investigate how different city structures and road networksinfluence good policies.

4 - On Optimal Base Locations and Staffing of Ambulance ServicesMartin van Buuren, Centrum Wiskunde & Informatica, Netherlands, [email protected]

Ambulance care is of paramount importance in today’s society. For good service, Dutch ambulances shouldbe at an incident location within a response time threshold of 15 minutes for high urgency calls. Appliedmathematicians can help ambulance providers to achieve a higher service with the same capacity. Quitesoon, the two bases in Amsterdam will close down their service for high urgency calls. Instead, three newones will be built and one of the existing bases will change its function such that it can only be used for lowurgency BLS calls only. This yields that the staff should be relocated over the new bases. The re-staffingshould respect all kinds of labor laws. We made a binary integer programming (BIP) formulation whichsolves this problem for the ambulance provider. Our approach is also quite adaptable for similar problems.In this talk I will give an overview about the process towards the BIP-formulation, show the results anddiscuss how it is being applied in practice.

118

TA-06

� TA-06Thursday, 9:00-10:30Goudriaan Room 2

SAMS-solversStream: Software Applications and Modelling SystemsInvited sessionChair: Michael Bussieck, GAMS Software GmbH, Eupener Str 135-137, 59033, Cologne, Germany,[email protected]

1 - LocalSolver: toward a full mathematical programming solver based on localsearchFrédéric Gardi, LocalSolver, 24 avenue Hoche, 75008, PARIS, France, [email protected],Thierry Benoist, Julien Darlay, Bertrand Estellon, Romain Megel

We present LocalSolver (http://www.localsolver.com), model-and-run solver integrating pure local-searchtechniques. It can handle very large nonlinear problems with millions of 0-1 decisions. LocalSolver offerssimple APIs as well as an efficient modeling language for fast prototyping. It is used in real-life applica-tions by several companies through the world, but the product remains free for academics. As example ofits strength compared to classical MIP, CP, SAT softwares, LocalSolver was the sole general-purpose opti-mization solver to be qualified for the final tour of the Google ROADEF/EURO 2012 Challenge, thanks toa 100-line model.

LocalSolver is based on the fast exploration of multiple variable neighborhoods to provide high-quality(hopefully optimal) feasible solutions quickly. Different kinds of neighborhoods are employed during thesearch: small (basic moves), medium (compound moves), large (tree-search moves). Smaller are the neigh-borhoods, faster is their evaluation through incremental computations. The selection of the neighborhoodto explore is dynamically adapted during the search through learning to ensure the fastest convergence,while a simulated annealing heuristic with reheatings and restarts allows to escape local optima and to di-versify widely the search. On the other hand, LocalSolver provides an optimality gap based on propagationand relaxation techniques.

Having reviewed these current technical features and detailed some benchmarks to assess the performanceof LocalSolver, we will outline our current work to extend this technique for continuous or mixed-variableoptimization. More generally, we will present the roadmap of the LocalSolver project toward an all-in-onemath programming solver.

2 - A Heuristic for MIPs with Embedded NetworksChristopher Maes, Gurobi Optimization, Inc., 167 Beacon St. #8, 02116, Boston,Massachusetts, United States, [email protected]

We present a heuristic for MIPs containing an embedded network design problem. In this problem, thereis a fixed cost for opening an edge in the network. The heuristic begins with a feasible solution and triesto improve it by exploring a set of moves in the network. Each move corresponds to a single pivot of thenetwork-simplex algorithm. We find a cycle in the network, push flow around it, and compute the resultingchange in objective. However, we must modify the network-simplex algorithm to handle fixed costs. If animprovement in the network is found, we attempt to extend it to the underlying MIP.

Often the embedded network is quite small compared to the underlying MIP. Thus, this heuristic allowsus to quickly explore many feasible solutions. We evaluate the performance of this heuristic, within theGurobi Optimizer, on a set of MIPs with embedded networks.

3 - Open-source Quality Assurance and Performance Analysis ToolsMichael Bussieck, GAMS Software GmbH, Eupener Str 135-137, 59033, Cologne, Germany,[email protected], Steve Dirkse, Stefan Vigerske

119

TA-07

Until recently, much of the math programming community has focused primarily on performance testingand benchmarking, while the general commercial environment has emphasized reliability over perfor-mance. Around 10 years ago we introduced the PAVER platform (Performance Analysis and Visualizationfor Efficient Reproducibility) to aid in both QA and performance analysis of solver software. We willpresent new and enhanced QA and performance tools implemented in the second generation of the PAVERplatform.

� TA-07Thursday, 9:00-10:30Antwerp Room

Scheduling Applications IStream: Scheduling and Project ManagementInvited sessionChair: Grzegorz Pawlak, Institute of Computing Science, Poznan University of Technology, ul.Piotrowo 2, 60-965, Poznan, Poland, [email protected]

1 - Single truck railway scheduling problemGrzegorz Pawlak, Institute of Computing Science, Poznan University of Technology, ul.Piotrowo 2, 60-965, Poznan, Poland, [email protected], Jacek Blazewicz,Dawid Kowalski, Gaurav Singh

In the paper we present the Single Track Railway Scheduling Problem (STRSP). Trains are traveling formthe source station to the destination station traveling through the transiting stations on the single track.Trains can wait at particular stations, the number of the waiting trains depends on the station capacity. Thetarget station has unlimited capacity. Optimization criterion is to maximize the number of trains runningfrom the origin to the final destination and back, in the particular time window. There are considered theconstrains such as: the maximum train waiting time on the station, the safety distance between trains,the time limit on the train traveling time in two-ways. A mathematical model for STRSP was formulatedand algorithms were proposed for the for particular cases where the unit traveling time between stationswas assumed. The computational experiments and simulations models were conducted and results werepresented. The motivation for consideration of such an optimization problem was taken from the actualreal system of Australian Railways. From the one point of view the problem is similar the well-knownand elaborated in the literature trains timetabling problem on the other hand there is the unique schedulingproblem which depends on the constraints and the considered criterion function. In the real system theadditional constraints can be added like minimum stop time on the station, different speed-up time betweenthe empty and loaded train. This will be the subject of the future research.

2 - CSS sprite optimization for web pagesJakub Marszałkowski, Institute of Computing Science, Poznan University of Technology,Poland, [email protected], Maciej Drozdowski, Jan Mizgajski, DariuszMokwa

CSS sprite is a well-known technique of putting a collection of images into a single picture for web siteoptimization. In order to display proper parts of the picture in appropriate areas of a web page CSS rulesare applied. In this way numerous HTTP requests are reduced to a single one. Thus, the server is offloaded,communication overhead is reduced and displaying the web page is accelerated. CSS sprites may be con-structed manually, which however, will be time consuming and may lead to inferior quality results. Severalapplications based on ad hoc approaches for semiautomatic CSS sprites generation are available. Analysisof those tools reveals numerous drawbacks: bad use of PNG and JPEG format capabilities, manual selec-tion of images to these formats, single picture output minimizing file size only, and even undocumentedlossy compression to improve results. We propose a new approach to this problem with in-depth scien-tific analysis of its intricacies and use of combinatorial optimization. We assume that images should be

120

TA-07

automatically divided into JPEG and PNG picture classes, the number of CSS sprites should be found bythe algorithm, and the objective function should minimize the time of loading pictures. For this end, weevaluate options of both image formats as well as empirically determine how different ways of combiningimage tiles into pictures affect file sizes. Then, we analyze how connection time and transfer speed affectoptimality of CSS sprites. We propose objective function allowing inclusion of parallel download capa-bility. A combinatorial divide and merge heuristic performing CSS sprite optimization is described. Thealgorithm is tested on over 30 real world test sets, as well as compared with the already available solutions.

3 - Complex Multi-Item Shopping over The Internet with Additional Complicated Dis-countsJedrzej Musial, Institute of Computing Science, Poznan University of Technology, Poland,[email protected], Jacek Blazewicz, Pascal Bouvry

Previous research (covered by publications and co-tutelle Ph.D. thesis by Jedrzej Musial) investigated amultiple-item multiple-shop optimization problem, in which total expenses of a customer associated withbuying a given set of items should be minimized over all available offers. The Internet Shopping Opti-mization Problem (ISOP) is defined in a formal way and a proof of its strong NP-hardness is provided.Moreover, heuristic algorithm (greedy) was developed and experimentally tested against Price Compar-ison Sites algorithms. Furthermore new Internet Shopping Optimization Problem with Price SensitiveDiscounts, denoted as ISOPwD was introduced. It has been proved that problem ISOPwD is NP-hard inthe strong sense if not every shop has all the required products, or if every shop has all the required productsand product standard prices are distinct.

The aim of current research is to model and formulate new ISOP problem - Internet Shopping Optimiza-tion Problem with discounts and with additional conditions such as to consider price sensitive shippingcost, incomplete offer from shops. The model will be mathematical and theoretical well founded. De-veloping efficient lightweight heuristic based algorithms, considering approximation factors, greedy basedalgorithms will be important part of the project. Investigate pseudo-parallel cellular based optimizationalgorithms will be one the interests.

Moreover, very important task will be to verify the solution quality of the developed solutions for the ISOPproblem through application of real-world data. Extend the model to Cloud computing Platform-as-a-Service (PaaS) model.

4 - Real Parameters Values Optimization in Multi-Agent SystemSzymon Wasik, Institute of Computing Science, Poznan University of Technology, ul. Piotrowo2, 60-965, Poznan, Poland, [email protected], Mateusz Leszner, Jacek Blazewicz

The use of multi-agent systems is a modeling technique that utilize the concept of interacting agents andthat have been proved to be very successful in solving many real problems. Currently it constitutes a well-known and versatile methodology that is used in various areas, for example scheduling, business processesmodeling, logistics and many more.

During our research we have modeled one of such systems using a multi-agent system that consists ofvery limited number of interactions but it contains hundreds of thousands of similar agents. Such complexsystems are sometimes called massively multi-agent systems (MMAS) and are quite difficult to implementbecause of computational complexity of their simulation. However in our case the most difficult problemwas finding values of some real parameters existing in the model.

In our model we know some expected results of its simulation which are known from physical observa-tions. These results depend on some real parameters which are unknown and which values should beoptimized. There are already some approaches that propose methods for finding these values however theylack comparison of time required to find them. In case of MMAS when evaluating of each set of parame-ters takes a lot of time it is a big limitation. To find the method that finds the solution in acceptable timewe compared several of them. First we proposed the approach based on inverted simulation method thattries to find values of parameters using some metaheuristic. Then we compared some implementations ofGenetic Algorithms and Gaussian Adaptation and considered several definitions of the optimized function.As a result we achieved an algorithm that not only finds acceptable values of parameters but also solvesthe problem in relatively short time.

121

TA-08

� TA-08Thursday, 9:00-10:30Penn Room 1

Novel problems and methodological advances in response toneeds of practitionersStream: Scheduling and Project ManagementInvited sessionChair: Alena Otto, University of Siegen, 57068, Siegen, Germany, [email protected]

1 - Scheduling Setups on Multiple Machines with a Single Setup OperatorDaniel Schnitzler, Operations Research, University of Siegen, Hölderlinstr. 3, 57068, Siegen,NRW, Germany, [email protected]

We consider multiple machines with a predetermined sequence of jobs to be processed by each machine.For each job, the machine has to be set up by a setup operator. The setup time is hereby task-dependentand given. There is only one setup operator present, thus only one machine can be set up at a given time. Ifthe operator has to move from one machine to another, he needs a certain time for the travel. Furthermore,there are breaks for the operator, in which he cannot set up a machine. The goal is to find a schedule forthe setups, which minimizes the time when the latest job finishes. The problem is formulated as a MIPand solved using CPLEX and Gurobi. Due to the problem’s complexity, both solvers were only able tosolve small problem instances (2 machines, 56 tasks). Real world instances, however, have more machinesand tasks. Therefore, a genetic algorithm and a tabu search algorithm were developed. Different variantsof the methods are tested on the one hand with the help of random instances were a lower bound wascalculated before. On the other side, they were tested with the help of instances where the optimal solutionis already known. We can find feasible solutions which are very close to the lower bound /known optimumfor instances with up to 100 machines and 1000 tasks.

2 - Column generation based lower bounds for project scheduling problemsTobias Paetz, Operations Management Group, Clausthal University of Technology,Julius-Albert-Straße 2, 38678, Clausthal-Zellerfeld, Germany, [email protected],Christoph Schwindt

We consider project scheduling problems for which the objective function can be expressed as a linearfunction in the durations of the precedence order’s feasible antichains. Examples of such objective func-tions are the project duration and certain resource leveling criteria. The problem can be stated as a huge LPwith side constraints. By relaxing the side constraints one obtains a linear program, which can be solvedvia column generation and whose solution provides a lower bound on the minimum objective functionvalue of the original problem. Depending on the type of project scheduling problem, the pricing problemrepresents a multi-constraint binary knapsack problem or a convex-weight stable set problem on a com-parability graph. To tighten the relaxation, different types of preprocessing methods may be applied. Forexample, the lower bounds for resource leveling problems can be improved by identifying antichains whichmust be executed in any feasible schedule. Those antichains can be computed by generating the cliquesof an interval graph. We report on computational experience with project scheduling problems containinggeneralized precedence relations.

3 - Adaptive Large Neighborhood Search for the Resource-Constrained ProjectScheduling ProblemWolfgang Summerauer, Department of Business Administration, University of Vienna, BruennerStrasse 72, 1210, Vienna, Vienna, Austria, [email protected], ChristianAlmeder, Richard Hartl

We present an adaptive large neighborhood search (ALNS) for the single-mode resource-constrainedproject scheduling problem (RCPSP), where a project consists of activities that have to be scheduled underresource and precedence constraints. The objective is to find a schedule that minimizes the makespan ofthe project. ALNS is based on the principle that neighbors of a given solution are generated by applying

122

TA-09

different destroy and repair operators. The operators are selected randomly using a bias based on theirperformance in previous iterations. The destroy neighborhoods of our ALNS remove activities from theschedule mainly focusing on removing predecessor and successor activities at the same time. This is basedon the idea that a schedule can only be sufficiently altered if the predecessors and/or successors of an ac-tivity can also be moved to other positions. To successfully reconstruct a partly destroyed schedule, ourrepair neighborhoods either rely on some state-of-the-art priority rules, a random selection or the currentresource utilization of a schedule. Moreover, our ALNS is embedded into a simulated annealing frame-work, which governs the probability of accepting newly generated solutions. To evaluate the performanceof our ALNS, we conducted tests on the well-known data sets j30, j60, and j120 from the PSPLIB. Theobtained results show that our algorithm outperforms every state-of-the-art algorithm for the smaller testinstances, and is highly competitive for the medium and large test instances. The results also show that ourproposed method performs consistently well over all data sets for a problem class that is usually dominatedby evolutionary algorithms.

4 - Quantified Combinatorial OptimizationThorsten Ederer, Mathematics, TU Darmstadt, Dolivostraße 15, 64293, Darmstadt, Hessen,Germany, [email protected], Ulf Lorenz, Thomas Opfer

Mixed-integer linear programming (MILP) is the state-of-the-art technique for computer-aided optimiza-tion. However, companies observe an increasing danger of disruptions that prevent them from acting asplanned. One reason is that input data is often assumed to be deterministic, but in reality, they are oftenafflicted with uncertainties. Incorporating uncertainty in existing models, however, is tedious and oftenpushes the complexity of problems that are in P or NP, to the complexity class PSPACE.

Quantified mixed-integer linear programming (QMILP) is an extension of MILP with variables being eitherexistentially or universally quantified. This gives the opportunity to carry traditional MILP formulationsto the realm of uncertainty in a very natural way. The pure integer variant (QILP) is PSPACE-complete.With the help of QILP, it is possible to model games like Gomoku etc. as well as traditional combinatorialOR problems under uncertainty.

In this talk, we present how to extend the model formulation of classical scheduling problems by variousuncertain influences. In principle, it is possible to solve the resulting QILPs by generating deterministicequivalent problems (DEP) and using standard MILP software. However, the size of the DEP growsexponentially in the number of universal variables. Therefore, we develop a specialized QILP solver that isinpired by techniques known from QSAT solvers, branch and cut ILP frameworks, stochastic programmingand game tree search. Our current implementation can solve multi-stage binary programs to optimalitywith up to several hundred existential and two dozens of universal variables. Here, we present somecomputational results for the presented combinatorial optimization problems.

� TA-09Thursday, 9:00-10:30Diamond Room 1

Planning & control in the process industryStream: Production and Operations ManagementInvited sessionChair: Simme Douwe Flapper, Technische Universiteit Eindhoven, Netherlands, [email protected]

1 - Inventory control of a continuous production-inventory systemSteffen Klosterhalfen, GVM/S, BASF SE, B009, 67056, Ludwigshafen, Rheinland-Pfalz,Germany, [email protected], Falk Holzhauer

In the process industry various products are produced continuously around the clock. Due to prohibitivelylarge costs for stopping and starting the relevant machines, such an action is only taken when absolutelynecessary, e.g., for maintenance or cleaning purposes. The only feasible adjustment of the production

123

TA-09

process refers to the run rate of the machine, which can be varied within a certain range. The run-rate ad-justment of the particular machine that motivated this research is made at regular intervals only. For such aproduction system a simple model is to be developed that can be easily used by the planner to compute anappropriate safety stock level in order to satisfy a predefined service-level target. We show that under thesimplifying assumption of a production time of zero the above-described system can actually be interpretedas a periodic-review order-up-to level inventory system with a so-called order band. The order band definesthe lower and upper bounds on the replenishment (production) quantity. To obtain the optimal order-up-toand thus safety stock levels existing contributions use simulation optimization methods, in particular In-finitesimal Perturbation Analysis. For our purposes this method is rather cumbersome to implement anduse. Therefore, we propose a different solution algorithm based on a modification of the moment-iterationmethod by de Kok (1989), "A moment-iteration method for approximating the waiting-time characteris-tics of the GI/G/1 queue’, Probability in the Engineering and Informational Sciences, 3, 273-287. Thisapproach is easily implementable, e.g., in Visual Basic for Applications such that it is available in Excel.From a numerical study we find that the solution algorithm performs very well on a large set of parameters.

2 - Optimizing the tactical planning in the FMCG industry considering both economicand environmental factorsMartijn van Elzakker, Eindhoven University of Technology, Den Dolech 2, 5612 AZ,Eindhoven, Netherlands, Netherlands, [email protected], Leonardo Kenji Komay Maia,Neha Raikar, Hans Hoogland, Edwin Zondervan

In the last decades, an ever-growing pressure from government regulators, non-governmental organizationsand the market itself towards more environmental-friendly supply chain operations has led to an increasein research on the subject. The Fast Moving Consumer Goods (FMCG) industry, due to its high volumesof products and complex distribution chain, may take a considerable advantage from this research field.In this work, we study the optimization of the tactical planning for an FMCG company while consideringboth economic and environmental performance metrics. Many FMCGs, such as ice cream and yoghurt,need to be refrigerated after production and while being stored or transported.

In previous work a cost minimization MILP model was developed for optimizing the tactical planning inthe FMCG industry. The costs included in this model are: procurement, production, distribution, inven-tory, safety stock and missed sales costs. The supply chain consists of suppliers, factories, warehouses,distribution centers and retailers. In this work we additionally consider the environmental impact due totransportation and due to manufacture. Different modes of transportation are considered, as well as dif-ferent locations of factories. The corresponding energy mixes for each location are taken into account,leading to different emission profiles.

The environmental impact is considered using the Environmental Priorities Strategies (EPS) system. APareto analysis is performed in order to evaluate the trade-off between the economic and environmentalobjectives. The Pareto curve is generated using the e-constraint method. The Euclidian compromise solu-tion reduces the environmental impact by 4.3% at a 1.1% cost increase compared to the minimal economiccost solution.

3 - The Application of Sampling Plan Methods : A Case of Sugar CompanyEndy Suwondo, Agroindustrial Technology, Universitas Gadjah Mada, Indonesia, Jl Flora 1,Bulaksumur, Yogyakarta, Indonesia, 55281, Indonesia, [email protected], HenryYuliando, Adi Djoko Guritno

The sampling plan is a substantial aspect of the quality control problem, where inspection is an importanttask. In many cases, the inspector is unable to detect all defective items in a batch. As found in the practiceof quality control of bags used in a sugar company, the inspection is carried out based on each lot samplingand reject it according to undefined standard. Here, we propose a study of sampling plan using severalmethods that effective for decision making in the quality control of bags used in a sugar company. Thisproduct is made of circular weaved polypropylene with a quality attribute on its strength (up to 100 kg) andwaterproof. The data was taken from 25 random samples for several methods applied including VariableSingle Sampling, Quality Index Sampling, and Attribute Proportion Sampling. By using 15% of AQL, and40% LQL, 5% for both produser’s risk and 10% of consumer’s risk as a standard, tested by the parameterof probability of acceptance ( Pa ), average outgoing quality (AOQ) and average total inspection (ATI), itwas found that the best sampling method is Variable Single Sampling (VSS). This method gives significant

124

TA-10

result for all tested parameter and suggest an efficient way for the company in doing the inspection andquality control.

4 - Planning and control of production with recycling via mixingSimme Douwe Flapper, Technische Universiteit Eindhoven, Netherlands, [email protected]

In literature quite some attention has been paid to recycling, i.e., the reuse of materials. Recycling isbecoming more important among things due to limited material availability with related high purchasecosts. In a lot of companies producing base materials like plastics, metals, as well as in the food andbeverage industry, recycling requires mixing with virgin materials when those materials are transformedinto desired products. So far little attention to this type of recycling has been paid in literature. In thepaper attention is paid to production planning and control in situations with and without time restrictionswith respect to recycling via mixing, as well as limitations with respect to the percentage of material thatcan be recycled via mixing with virgin materials. Mathematical models supporting the above productionplanning and control are presented, as well as the results of numerical experiments showing the potentialsof the option.

� TA-10Thursday, 9:00-10:30Diamond Room 2

Coordination and Demand FulfillmentStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Jaime Cano Belmán, Institute for Supply Chain Management, EBS Universität,Konrad-Adenauer-Ring 15, 65187, Wiesbaden, Germany, [email protected]

1 - Coordinating a multi-echelon supply chain facing demand uncertainty: Spanningvs. pair-wise revenue sharing contractsAzarm Yeganehfallah, Planning and design department, Iranian Telecominucation Company,Papengracht 10, 2311 TW in Lieden, Modares Street, Babol, Iran, 2311, Leiden, Netherlands,[email protected], Hamid Mashreghi

Nowadays, competitions between supply chains forces members to participate in strategic partnershipsextending the isolated firm’s competitive advantages. One of the applicable partnership mechanisms iscoordinating contracts. The ability of contract mechanisms to achieve coordination has been dramaticallyanalyzed in the literature for two-echelon supply chains. In particular, revenue sharing contracts are onethe most powerful contracts to achieve coordination for two-echelon and three-echelon supply chains. Onthe other hand, under demand uncertainty the effects of the echelons’ number should be considered. Fur-thermore analyzing revenue sharing contracts can be developed through two approaches in multi-echelonsupply chains. One way is to install contracts between all pairs of adjacent entities with pair-wise contractsand the other one is implementing based on the concept of spanning contracts. In this research we aimto compare the effectiveness of a pair-wise and a spanning revenue sharing contract in a three-echelonchannel facing demand uncertainty. Then, the obtained results would be utilized to develop this approachmathematically for a multi-echelon (n-echelon) supply chain. Finally, we provide concluding remarks toimplement such contracts for coordinating multi-echelon partnerships in telecommunication industries.

2 - Supply chain coordination under demand uncertainty: Analysis of general contin-uous quantity discountsHamid Mashreghi, Industrial Engineering Department, Tarbiat Modares University, Center ofTIM, Leiden University, Jalale Ale Ahmad Ave, Tarbiat Modares University, Tehran, Iran, 2311,Leiden, Netherlands, [email protected], Mohammad Reza Amin Naseri

125

TA-11

Quantity discounts (QDs) are contract mechanisms that are commonplace in practice with different com-plexities. As a whole we can consider two general classes for QD contracts: continuous vs. discreteschemes. With continuous QD schemes there exists a general differentiable wholesale price function whichis decreasing with respect to order quantities. While in discrete QD schemes there is only a price list withdecreasing wholesale price levels (often constant wholesale price levels) that can be interpreted in differentways such as all-unit discount or incremental discounts. Thus, in this paper we aim to analyze the structureof general continuous quantity discounts to coordinate a two-tier supply chain under additive demand un-certainty. After demonstrating the general coordination conditions based on joint-optimization of pricingand ordering decisions, the specific case of continuous linear QD is analyzed and the sufficient conditionsto achieve supply chain coordination are presented. We also develop our findings to introducing some ap-plicable non-linear continuous QD schemes. Finally, considering prerequisites to achieve coordination bycontinuous QDs, managerial insights are developed to perform better negotiations between supply chainpartners.

3 - Advanced stochastic real-time order promising in Make-and-Ship-to-StockJian Cui, Supply Chain and Network Management, TU Darmstadt, Pfefferstr. 4, 70619,Stuttgart, Baden-Württemberg, Germany, [email protected], Herbert Meyr

In the real world situation, the uncertainties and the decision process are stochastic due to their evolutionover time. Two-stage stochastic programming with recourse (2SSP) is one of the most promising methodsto model this dynamic stochastic process, since it explicitly models the uncertain information and generatesplanning and scheduling policies before the uncertainty occurs, these decisions may be modified as timepasses with the realization of uncertainty. Meanwhile, it is a good compromise between rigorous solutionand less computational efforts for the feasibility of real-time online optimization with considerable sce-narios. A make-and-ship-to-stock (MSTS) demand fulfillment policy by considering networks (multi-site)with different types of transportations and customer heterogeneity is modeled in a disaggregated fashionand extended to 2SSP settings with multi-period demand uncertainties which lead to two-stage stochasticlinear programming models. Consequently, a novel two-stage stochastic MSTS framework with intercom-municated two layers, allocation and consumption, is proposed in this contribution.

4 - Advanced Demand Fulfillment in NetworksJaime Cano Belmán, Institute for Supply Chain Management, EBS Universität,Konrad-Adenauer-Ring 15, 65187, Wiesbaden, Germany, [email protected], Herbert Meyr

If enterprise’s production capacity is exceeded by demand, companies must decide whether to accept acustomer order. If accepted, an efficient fulfillment must be found, regarding some criterion normally re-lated with cost or revenue. This work deals with such short-term demand fulfillment decisions within anetwork approach (multi-site network). In our scope, this network consists of the last distribution ware-houses of a company in the supply chain and heterogeneous customers located on different sales-regions(Make-and-Ship-to-Stock environment). Following Revenue Management ideas to efficiently manage cus-tomer heterogeneity, existing single-location approaches are extended in order to handle real-time orderpromising in networks through two mathematical programming models: an Allocation Planning Modeland a Consumption Planning Model. The models also include basic transport considerations such likedifferent transport options, different lead times, or different transport capacities.

� TA-11Thursday, 9:00-10:30Penn Room 2

Robust Supply Chain ManagementStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Ralf Gössinger, Business Administration, Production and Logistics, University of Dortmund,Otto-Hahn-Str. 6, 44227, Dortmund, Germany, [email protected]

126

TA-11

1 - A Simulation Study of Supply Chain Disruptions ModellingAsır Genç, Statistic, Selcuk University, Konya, Turkey, [email protected], Tarik Yilmaz

Nowadays firms reengineer and reconstruct their internal processes to present more value to their customerswith lower costs.

Recently in business environment, the competition is not only between the individual firms but also thenetworks of the companies. Supply chain management is the management of material, information andmonetary flows of a network that is including suppliers, distributors and customers. However, in a highlycompetitive environment, firms must also improve their supply chains. But the weather, strikes, earth-quakes and so on, affect and disrupt supply chains in many cases.

A model was generated to optimize the supply chain taking into account the uncertainty environment. Thecases were evaluated with simulation study.

2 - Robust strategic capacity planning in closed-loop supply chainsSandra Hallek, Business Administration, Production and Logistics, University of Dortmund,Germany, [email protected], Ralf Gössinger

The task of strategic capacity planning is to set up and to adapt capacity in a medium-term time horizon(e.g. product life cycle). An analysis of approaches belonging to closed-loop supply chains turns out that(1) capacity for processing returned goods and manufacturing final products are planned independently,(2) information about the backflow is usually modelled as a stochastic variable with a known probabilitydistribution, and (3) there is no consideration of the risk-attitude of decision makers. On this basis the spe-cific interactions between forward flow, backflow and capacity utilization in closed-loop supply chains, andthe implications of uncertainty can only be rudimentarily taken into account. To diminish these deficits,in the planned paper a two-stage stochastic model with recourse is developed to create robust plans. Thepreference of robust plans can be interpreted as a form of risk-averse decision behaviour, which is relevantfor medium-term planning of production and logistic systems. The solution approach (1) plans the man-ufacturing capacity and the backflow processing capacity simultaneously, whereby investments in flexibleresources (e.g. assembly/disassembly lines) are possible, (2) contains a backflow quantity forecast withupdates, and (3) considers different possible future states by means of scenarios. In sum a stochastic MIPmodel is developed which supports strategic capacity planning under uncertainty. The model is imple-mented in AIMMS and tested in a rolling horizon by means of the solver CPLEX. In order to find out towhat extent the utilization of flexible resources increases the planning robustness, a numerical analysis iscarried out.

3 - Robust order promising integrating customer behaviorSonja Kalkowski, Business Administration, Production and Logistics, University of Dortmund,Germany, [email protected], Ralf Gössinger

During the past decades an individualization of customer demand that causes increasing production re-quirements has strengthened the interest in planning approaches for generating reliable delivery promises.In the context of supply chains capable-to-promise approaches are suggested, which calculate on-timedelivery promises with respect to available production capacities and resources. Since customers’ andsuppliers’ interests are to be balanced in the process of bargaining for delivery dates, there is a need forintegrating approaches of production planning and marketing planning. In the literature related to orderpromising the idea of considering different stakeholders has not been adequately taken into account. Anapproach that considers customer behavior and makes allowance for adjustment measures to enhance ro-bustness is developed and analyzed in the planned paper. In detail planning robustness shall be providedin such a way that confirmed delivery dates can be met even though the available information is incom-plete and changes over time. Based on a capable-to-promise approach (Chen/Zhao/Ball 2001) a two-stagedecision model is derived. At the first stage the common way of rejecting orders is applied if an order ful-fillment is not possible during the required delivery interval. In contrast to the usual practice of a bindingrejection, alternative delivery proposals considering anticipated costumers’ reactions are generated at thesecond stage. Using real-world data of a manufacturer the described planning approach is analyzed numer-ically by means of the AIMMS environment. Implications for the reliability of delivery promises causedby adjustment measures are studied as well as the influence of the second planning stage on performanceand solution quality.

127

TA-12

� TA-12Thursday, 9:00-10:30Van Oldenbarnevelt Room

Crowdedness, Evacuation and Inland ShippingStream: Traffic and TransportationInvited sessionChair: Stefan Minner, TUM School of Management, Technische Universität München, Arcisstrasse 21,80333, Munich, Germany, [email protected]

1 - Capacity, Crowding and Passenger InformationPaul Bouman, Decision and Information Sciences, Rotterdam School of management, ErasmusUniversity, Burgemeester Oudlaan 50, 3062 PA, Rotterdam, Netherlands, [email protected],Leo Kroon, Gabor Maroti, Peter Vervest

One of the key factors that determines the comfort experienced by a passenger in public transport is thecrowdedness of the vehicle. Over time, crowding can be an important reason for a passenger to performtime shifting or change the mode of travel. We previously explored how we can use the ideas proposed byminority games to evaluate the impact of information policy and available capacity for the crowdednessexperienced by the passengers. We constructed an agent-based model that can be evaluated under differentinformation policies for a simplified public transport situation. Our initial results show that exposing moreinformation to the agents increases the number of agents travelling by public transport, but decreasescustomer satisfaction. This tradeoff is influenced by the ratio of potential customers to the capacity of themodel instance.

There are many possible extensions to this model. The first extension considers the fact that public transportsystems usually have many stops where passengers travelling between different locations can enter andleave the vehicle leading to fluctuations of the crowding during a journey. We introduce this concept intothe model by expanding it to a line situation, where we have multiple stops and a single vehicle movingback and forth the line. The second extension considers noise during the exchange of information, asthis may decrease the correlation of the behaviors of the agents. Finally, a remaining question is whichinformation policy is best for the operator: suppose he knows the preferences of the agents, can he influencetheir behavior positively by sending the right information to the right person?

During our presentation, we explain the initial model and results and discuss the progress on these exten-sions.

2 - Evaluation and Planning Models for Evacuation SitesTakehiro Furuta, Nara University of Education, Takabatake-cho, 630-8528, Nara, Nara, Japan,[email protected], Ken-ichi Tanaka

In Japan, many local governments designate elementary schools and junior high schools as evacuationsites. So head teacher and faculty members of the schools perform a key administrative role on behalfof local governments in actual disaster situations. It is required for them not only to educate schoolersabout evacuation behaviors but also to prepare adequate refuge spaces in their school, although they arenot expert in evacuation planning. To support their work, we develop two mathematical models to evaluateevacuation sites and routes and apply the model to the actual geographical data and demographic data.

First, we develop a mathematical model for evaluating spatial mismatch between supply and demand. Inour model, the supply is the capacity of evacuation sites that is calculated based on area of evacuation spaceand the calculation method is determined by Japanese law. The demand is estimated numbers of evacueesin a region. Most cities in Japan estimate the number of evacuees some sizes and source of earthquakes.When maximum walking distance at the time of disaster is given, each evacuees may have more than oneevacuation site within the distance. So our model takes into account the spatial distributions of evacuationsites and evacuees to analyze the mismatch.

Second, we focus on evacuation routes to evacuation sites, when we evacuate at the time of disaster. In aregion, degree of collapse risk is different in different locations and some roads may not be adequate for

128

TA-13

evacuation. So, the shortest route from a point to the site may not be the best route. Our model decides theroute so as to minimize the total risk of evacuation. We use the model to evaluate a city with the estimateddegree, population data and road networks.

3 - Methods for lock scheduling with ship-dependent handling timesWard Passchyn, Research Center for Operations Research & Business Statistics, KU Leuven,Belgium, [email protected], Dirk Briskorn, Sofie Coene, Frits SpieksmaFreight transport on inland waterways has great potential as a reliable, inexpensive and environmentallyfriendly mode of transportation. An important role on these inland waterway networks is reserved forlocks, which allow control over the water level while, at the same time, allowing ships to move from oneside of the lock to the other. Depending on the position of a lock and the decisions that govern the lockbehaviour, ships may experience some waiting time before they can pass the lock. We are interested infinding an operating schedule for a lock so that the total waiting time of all ships is minimized. The goal isthus to find the moments in time when the lock should switch positions, and to decide which ships shouldbe handled at each of these times. The basic problem for a single lock is polynomially solvable. Weinvestigate two relevant extensions. We first consider ship-dependent handling times; we assume that thetime needed to change the lock increases with each ship that is in the lock, and that this handling time foreach of the ships may be different. We show that the problem becomes NP-hard under this setting. Next,we restrict ourselves to first-come first-served policies: when multiple ships are waiting at either side of thelock, we require that these ships must be handled in the order of their arrival. We show that this particularcase does admit a polynomial time algorithm. As a consequence, a special case of our first extension alsobecomes polynomially solvable, namely the case where the handling time for all ships is equal.

� TA-13Thursday, 9:00-10:30Mees Auditorium

Integer programmingStream: Discrete and Combinatorial Optimization, Graphs and NetworksInvited sessionChair: Frits Spieksma, Operations Research and Business Statistics, Katholieke Universiteit Leuven,Naamsestraat 69, B-3000, Leuven, Belgium, [email protected]

1 - An Integer Programming Model for the Hospitals Resident Problem with CouplesIain McBride, School of Computing Science, University of Glasgow, Lillybank Gardens,Hillhead, Glasgow, United Kingdom, [email protected] scale allocation processes can be modelled as matching problems involving sets of agents who mayexpress preferences over members of other sets. Centralised matching schemes, which use algorithms tosolve the underlying matching problems, are often employed in such allocation processes. The NationalResident Matching Program (NRMP) matches graduating medical students to hospitals in the US, matching25,526 students in 2012. Similarly, in Scotland medical graduates are matched to Foundation Programmeplaces via the Scottish Foundation Allocation Scheme (SFAS). These schemes may be modelled by a clas-sical combinatorial problem, the Hospitals / Residents Problem (HR). Centralised matching schemes havehad to evolve to accommodate couples who wish to be allocated to (geographically) compatible hospitals.This extension, modelled by the Hospitals / Residents Problem with Couples (HRC), has been in operationwithin the NRMP for a number of years and has also been applied more recently in the SFAS context. TheGale-Shapley algorithm solves the HR problem by finding a so called stable matching. We prove that, evenunder very severe restrictions, the problem of deciding whether a stable matching exists, given an instanceof HRC, is NP-complete. These complexity results drive the search for alternative methods of resolvingsuch problems. We describe an Integer Programming model of HRC which produces exact, optimal so-lutions in larger instances where previously only heuristics, which are not guaranteed to terminate, havebeen applied. We prove the correctness of the model and demonstrate the empirical performance of animplementation over a number of randomly generated datasets in addition to anonymised real data fromthe SFAS context.

129

TA-13

2 - The Red-Blue Transportation ProblemDries Goossens, Operations Research and Business Statistics (ORSTAT), KU Leuven,Naamsestraat 69, 3000, Leuven, Belgium, [email protected], WimVancroonenburg, Federico Della Croce, Frits Spieksma

This contribution considers the Red-Blue Transportation Problem (RBTP), a generalization of the trans-portation problem where supply nodes are partitioned into two sets and so-called exclusionary constraintsare imposed. We encountered a special case of this problem in a hospital context, where patients needto be assigned to rooms. We establish the problem’s complexity, and we compare two integer program-ming formulations. Furthermore, a maximization variant of RBTP is presented, for which we propose aconstant-factor approximation algorithm. We conclude with a computational study on the performance ofthe integer programming formulations and the approximation algorithms, by varying the problem size, thepartitioning of the supply nodes, and the density of the problem.

3 - Polyhedral Results for the Hypergraph Assignment ProblemOlga Heismann, Optimization, Zuse Institute Berlin, Germany, [email protected], RalfBorndörfer, Achim Hildenbrandt

The hypergraph assignment problem (HAP) generalizes the assignment problem from bipartite graphs tobipartite hypergraphs. It is motivated by applications in railway vehicle rotation planning. The HAP is NP-hard and APX-hard even for small hyperedge sizes and hypergraphs with a special partitioned structure.However, an algorithmically tractable model providing a strong LP relaxation which implies all cliqueinequalities can be derived from a suitable extended formulation of polynomial size. Further, an analysis ofthe polytope of feasible solutions for a small hypergraph using the facet classification framework HUHFAleads to the development of interesting new facet classes, which generalize odd set inequalities for thematching problem.

4 - HUHFA: A Framework for Facet ClassificationAchim Hildenbrandt, Computer Science, Ruprecht-Karls-University Heidelberg, ImNeuenheimer Feld 368, Room 104, 69120, Heidelberg, Baden-Würtenberg, Germany,[email protected], Olga Heismann, Francesco Silvestri,Gerhard Reinelt, Ralf Borndörfer

The polyhedral approach to combinatorial optimization problems studies the structure of their associatedpolytopes. One way is to compute complete linear descriptions of small polytopes in order to generalizethe equations and inequalities. Usually complete linear descriptions of polytopes consist of an enormousnumber of facet-defining inequalities already for very small problem sizes. However, there are also oftenmany symmetries implied by the combinatorial structure of the problem which can be used to classify thefacets. These symmetries act on the feasible solutions and naturally form a group. In their representationas maps on the variable values they can be extended to symmetries acting on the polytope. In this talk,we describe a method for dividing the inequalities into equivalence classes without resorting to a normalform. To do this classification, one applies the symmetries to the facet-defining inequalities and thenchecks whether any two facets can be transformed into each other and hence belong to the same class.Often, this check is not so easy as two linear expressions describing the same facet might differ by thesum of multiples of several equalities from the problem description. The main idea is to identify everyfacet defining inequality with the vertices of the polytope which satisfy it with equality. With this method,complete descriptions of polytopes can be analyzed to divide the facets into equivalence classes accordingto groups generated by given symmetry mappings. We propose an algorithm for the classification andillustrate its efficiency on a broad range of combinatorial optimization problems including the TSP andthe Linear Ordering Problem. We will also show that HUHFA is a useful component of the new polytopeanalyzer PORTA III.

130

TA-14

� TA-14Thursday, 9:00-10:30Tokyo Room

Vehicle Routing with Time WindowsStream: Traffic and TransportationInvited sessionChair: Alf Kimms, Mercator School of Management, University of Duisburg-Essen, Lotharstr. 65, LB125, 47057, Duisburg, Germany, [email protected]

1 - Granular Tabu Search for the Vehicle Routing Problem with Time WindowsMichael Schneider, DB Schenker Stiftungsjuniorprofessur BWL: Logistikplanung undInformationssysteme, TU Darmstadt, Hochschulstraße 1, 64289, Darmstadt, Germany,[email protected], Fabian Schwahn

One way to improve the runtime of local-search based metaheuristics without compromising solution qual-ity is the use of granular neighborhoods. So-called sparsification methods are applied to restrict the neigh-borhoods to include only elements which are likely to be part of high-quality solutions. The effectivenessof this approach has been shown in earlier works, e.g. for the Capacitated Vehicle Routing Problem.

This paper studies the application of granular neighborhoods for the Vehicle Routing Problem with TimeWindows (VRPTW). We develop a high-quality tabu search (TS) method and investigate the effectivenessof several sparsification methods. Here, both heuristic measures that trade off travel distance and waitingtime as well as reduced cost measures based on solutions of different assignment problems and networkrelaxations are studied. Moreover, we analyze the impact of the granular neighborhood size, which isdefined as a percentage of the complete neighborhood, on the solution quality and runtime of our method.Finally, we investigate how dynamically altering neighborhood sizes can be used to assist diversificationand intensification of the search.

With the right combination of sparsification method and neighborhood size, our TS is able to find highquality solutions for the notoriously hard VRPTW test instances of Solomon, achieving the best-knownvehicle number of 405. The solution quality in terms of traveled distance is only marginally worse com-pared to the TS with regular, complete neighborhood (the average gap is below 1%), while runtime isreduced threefold.

2 - Branch-and-Price on the Split Delivery Vehicle Routing Problem with Time Win-dows and Alternative Delivery PeriodsHeiko Breier, Institute for Material Handling and Logistics, Karlsruhe Institute of Technology,Gotthard-Franz-Str. 8, 76131, Karlsruhe, Baden-Württemberg, Germany, [email protected],Timo Gossler

In this paper we address the Split Delivery Vehicle Routing Problem with Time Windows and alternativePeriods. The consideration of multiple delivery periods per customer and the possibility of splitting de-liveries across different periods makes it a relaxation of the Split Delivery Vehicle Routing Problem withTime Windows. The opportunity for freight forwarders is to plan more efficient tours by exploiting alterna-tive delivery periods instead of a single Time Window. The introduction of only a few delivery alternativesalready allows relevant savings to be achieved. The contribution of this work is to prove the potential ofthis approach for cost savings and to demonstrate the decomposition of this problem in a demand focusedmaster problem and period related subproblems. The problem is solved by a branch-and-price method. Wewill present the following: - literature on related problems - the problem formulation as a compact problemand the decomposition in a Master Problem and multiple Pricing Problems - our branching rules for theBranch-and-Price procedure - Results proving the efficiency of splitting deliveries across different periods- results on the computational effort - A conclusion and an outlook for development of this approach

131

TA-15

3 - Branch and Price for the pollution routing problemSaid Dabia, Den Dolech 2, 5600 MB , Eindhoven, Netherlands, [email protected], Tom VanWoenselThe pollution-routing problem with time windows (PRPTW) is an extension of the vehicle routing prob-lem with time windows. The PRPTW concerns the routing of a number of vehicles to serve a set ofgeografically scattered customers, and deciding on the speed for each route. We propose an exact solutionalgorithm for the PRPTW based on a branch-and-price algorithm. The purpose is to determine the set ofroutes that minimizes the sum of fuel consumption and total route duration. Delaying departure time atthe depot reduces route duration. Considering speed and departure time at the depot as a decision variabletogether with the objective function considered notably complicates the pricing problem, which is solvedby means of a tailored labeling algorithm. Computational results support the value of the approach.

4 - The Vehicle Routing Problem with Time Windows, Collaboration between FreightCarriers, and a Rolling Planning HorizonChristoph Reiners, Mercator School of Management, University of Duisburg-Essen, Lotharstr.65, 47048, Duisburg, Germany, [email protected], Alf KimmsWe describe a framework for the Vehicle Routing Problem with Time Windows to model the aspect ofcollaboration between freight carriers. This framework enables them to exchange orders whereby thelength of the trip of the vehicles is decreased and their utilization is improved. Additionally, the frameworkis extended to handle a rolling planning horizon to add customer requests during a tour, swap them betweencarriers, and update the current routes. The proposed solution procedure is a Genetic Algorithm whichcomputes the shortest tour for every vehicle through a Constraint Programming model. To deal with alreadyvisited and a priori unknown nodes in a rolling planning horizon, the chromosomes must be modified toprotect visited nodes from being rearranged and to append new nodes to the tours. The rolling planninghorizon scenario is implemented by re-optimizing the tours periodically. Computational studies showthat the exchange mechanism reallocates several nodes during the planning periods and that collaborationimproves total travel costs. The algorithm can solve large-scale instances.

� TA-15Thursday, 9:00-10:30New York Room

Robust communication networksStream: Discrete and Combinatorial Optimization, Graphs and NetworksInvited sessionChair: Arie Koster, Lehrstuhl II für Mathematik, RWTH Aachen University, Templergraben 55, 52062,Aachen, Germany, [email protected]

1 - ROBUKOM: Robust Optimization of Communication NetworksArie Koster, Lehrstuhl II für Mathematik, RWTH Aachen University, Templergraben 55, 52062,Aachen, Germany, [email protected] BMBF project "ROBUKOM - Robust Communication Networks" is devoted to the design of networksunder uncertainties like link failures and traffic volumes. In this talk, we give a brief overview of the resultsof the project, and highlight one particular application: mixed-line-rate optical networks. In mixed-line-rate networks, signals of different speeds (and thus capacity) are transmitted. We adapt, the concept ofrobust optimization as proposed by Bertsimas and Sim (2003,2004) to this situation. First of all, thenominal and peak dmeand values are determined subject to the line rates. Based on this, a detailed integerlinear programming formulation for the problem is derived. Computational results show the benefits ofrobust optimization beyond a deterministic optimization approach.

The highlighted application is joint work with Svetoslav Duhovnikov, Manuel Kutschka, Franz Rambach,and Dominic Schupke.

132

TA-15

2 - Optimization of flexible multi-layer telecommunication networks under demanduncertaintyUwe Steglich, Chair of Communication Networks, Chemnitz University of Technology,Reichenhainer Str. 70, 09126, Chemnitz, Sachsen, Germany, [email protected],Thomas Bauschert

We present a mixed integer linear program optimization formulation for multi-layer telecommunicationnetworks under demand uncertainty. Objective is the minimization of the overall cost for network deploy-ment (CAPEX). The cost can be expressed by the base cost of the nodes and the corresponding interfacessummarized over all network layers (respective technologies). Two type of interfaces have been consid-ered: interfaces for interconnecting nodes of different networks layers and interfaces linking the nodeswithin one layer. Our optimization model considers important issues of multi-layer telecommunicationnetworks like flexible layer bypassing (e.g. for router offloading) and traffic multiplexing/grooming. Re-garding the layer interconnections two different options (top-bottom and variable) yielding different de-gree of flexibility are introduced and evaluated. The "Top-bottom’ option provides maximum flexibility bypotentially allowing shortcuts between all layers whereas the "variable’ option only exploits a subset re-flecting the capabilities of current technologies (e.g. IP-over-DWDM). Traffic fluctuations are consideredin our model by applying the gamma-robustness approach. With that, all traffic demands are expressed bymean and peak values and at most gamma demands are allowed to take their peak values at the same time(whereas all other demands remain on their mean value). We report the results of optimization runs thathave been carried out for different network topologies and different levels of demand uncertainty (gammafactors). Furthermore, the scalability of the robust multi-layer network optimization model is discussed.

3 - 0-1 Multiband Robust OptimizationFabio D’Andreagiovanni, Department of Optimization, Zuse-Institut Berlin (ZIB), Takustrasse7, 14195, Berlin, Germany, [email protected], Christina Büsing, Annie Raymond

We provide an overview of new theoretical results that we obtained continuing our investigations aboutMultiband Robust Optimization [BuDA12a,b], a new model for Robust Optimization that we recently pro-posed to tackle uncertainty in Mixed-Integer Linear Programming. This new model extends and refinesthe classical one proposed by Bertsimas and Sim [BeSi04] and is particularly suitable in the common caseof arbitrary non-symmetric distributions of the uncertainty. Here, we focus on uncertain 0-1 Programs(BP) and we analyze their robust counterparts when the uncertainty is represented through a multibandset. We show that if the uncertainty just affects the objective function, a robust optimal solution can becomputed by solving a polynomial number of nominal BPs with modified objective function. Tractabilityand approximability of the nominal BP are maintained. Moreover, we derive families of (strong) valid in-equalities that can sensibly strengthen the compact robust counterpart of the nominal problem. Finally, wepresent computational experiments over a set of instances of real-world uncertain optimization problemsstudied in collaboration with industrial partners in past and ongoing research projects.

REFERENCES

[BeSi04] D. Bertsimas, M. Sim, The Price of Robustness, Operations Research, 52 (1), 35—53, 2004

[BuDA12a] C. Büsing, F. D’Andreagiovanni, New Results about Multi-band Uncertainty in Robust Opti-mization, Proc. of SEA 2012, LNCS 7276, 63-74, 2012

[BuDA12b] C. Büsing, F. D’Andreagiovanni, Robust Optimization under Multiband Uncertainty - Part I:Theory. Submitted (2012) (preprint: Optimization Online 13-01-3748)

4 - Constructing Multi-Band Uncertainty Sets for Robust Optimization ProblemsChristina Büsing, Operations Research, RWTH Aachen University, Kackertstraße 7, 52072,Aachen, Berlin, Germany, [email protected], Fabio D’Andreagiovanni, Arie Koster,Manuel Kutschka

The essence of Robust Optimization is to find optimal solutions that remain feasible under all the deviationsincluded in a specified uncertainty set that should reflect the risk aversion of the decision maker. Tough thedefinition of the uncertainty set constitutes a crucial issue, the question of how building it has been highlyneglected in literature.

133

TA-17

In this talk, we focus on uncertainty sets built through the new Robust Multi-band Uncertainty Set. Thisset extends the classical concept of Bertsimas and Sim, where a single band for each uncertain parameteris given and the number of deviating values is limited. The new set breaks the single band in into multiplesub-bands to get a higher modeling resolution.

First, we derive probability bounds of constraint violation: a robust optimal solution is completely pro-tected against deviations included in the uncertainty set, but might still be infeasible for uncertainty real-ization that lie outside the set. In contrast to the classical ones, our bound has the desirable property of be-ing strongly data driven, since it exploits past observations about coefficient deviations that are commonlyavailable in real applications. Then, we devise good-practice rules that should guide the practitioners inthe definition of the uncertainty set. Finally, we present computational experiments on realistic instancesof real network design problems, to assess the efficacy of the construction rules that we propose.

� TA-17Thursday, 9:00-10:30New Orleans Room

Nonlinear Optimization IStream: Continuous OptimizationInvited sessionChair: Mirjam Duer, Mathematics, University of Trier, 54286, Trier, Germany, [email protected]

1 - On the optimality conditions for Stochastic Dynamic OptimizationJan Brinkhuis, econometric institute, erasmus university rotterdam, burg oudlaan 50, 30oodr,rotterdam, Afghanistan, [email protected]

Tikhomirov and coauteurs have unified the conditions of optimality for static and dynamic continuous op-tomization. This is done by means of a mixed smooth-convex theory of extremal problems. The presentlecture reports on ongoing research to extend this unification to stochastic dynamic continuous optimiza-tion. An additional aim is to make the optimality conditions for stochastic dynamic continuous optimiza-tion more accessible to practitioners.

2 - A New Method for Parameter Estimation of the GNL Model Using Real-Coded GAYasuhiro Iida, Industrial Management, Waseda University, Hakusann, 4-22-6, 112-0001,Bunkyo, Tokyo, Japan, [email protected], Kei Takahashi, Takahiro Ohno

This paper proposes a new parameter estimation method for the generalized nested logit (GNL) modelvia real-coded genetic algorithms (GA). When we do the parameter estimation of the GNL model, thelog likelihood maximization is usually used. In the algorithm of the log likelihood maximization, Quasi-Newton method or steepest descent method is widely used. However, the log likelihood function of theGNL model is multimodal. Therefore, the result of the parameter estimation depends on its initial settingwhen we use those methods and it is possible that we find local optima. In this study, we improve real-codedGA to the parameter estimation of the GNL model in order to find quasi-optima in each estimations fromdifferent initial setting. Real-coded GA has three operators, 1) crossover operator, 2) selection operator and3) mutation operator. In the first operator, we apply simplex crossover (SPX). However, when we applySPX to the GNL model which has constraints in allocation and logsum parameters, there is a possibility ofmaking offsprings which violate constraints. Therefore, we propose a method to recalculate if offspringsdo not satisfy constraints. In the second operator, we propose reJGG based on just generation gap (JGG).This method does not throw out the parent which has good utility and makes the convergence fast. In thelast step, we propose to use mutation operator which is not usually used in real-coded GA in order to keepdiversity and not to find local optima. In the numerical experiments, the log likelihood of our method iscompared to the Quasi-Newton method and normal real-coded GA which use SPX, JGG and do not usemutation operator using actual point of sales data. As a result, we found that our method is able to find thehigher log likelihood.

134

TA-17

3 - A new error bound result for Generalized Nash Equilibrium Problems and its algo-rithmic applicationMarkus Herrich, Technische Universität Dresden, Germany, [email protected],Axel Dreves, Francisco Facchinei, Andreas Fischer, Andreas Fischer

The local convergence theory of several algorithms is based on local error bounds. For nonlinear systemsof equations, optimization problems and variational inequalities there are well-known conditions implyingsuch an error bound. However, standard results are not suitable if applied to generalized Nash equilibriumproblems (GNEPs) so that new techniques are needed.

In this talk a sufficient condition for a local error bound for the Karush-Kuhn-Tucker system of a GNEPis provided and discussed. In particular, this condition implies neither local uniqueness of a solution norstrict complementarity. Based on the error bound we analyze convergence properties of a hybrid algorithmfor the solution of GNEPs which combines the robustness of a potential reduction algorithm and the localquadratic convergence of a recently introduced LP-based Newton method.

135

TB-02

Thursday, 11:00-12:30

� TB-02Thursday, 11:00-12:30Veder Room

Capacity ControlStream: Revenue Management and PricingInvited sessionChair: Thomas Vossen, University of Colorado at Boulder, United States, [email protected]

1 - A single-leg airline revenue management problem in continuous-timeJ.b.g. Frenk, Faculty of Engineering and Natural Sciences, Sabanci University, Orta MahalleUniversite caddesi no 27, Tuzla, Istanbul, 34956, Istanbul, Turkey, [email protected],Semih Sezer, A. Muzaffer Arslan

We consider a single-leg airline revenue management problem in continuous-time with overbooking andcancellations. In our model, booking requests arrive according to a non-homogeneous Poisson process.Each accepted request may later be canceled or the passenger may not show up at the departure time.In this setup, the objective of the airline is to dynamically allocate the available seats to different fareclasses in order to maximize the expected net revenue obtained from the given flight. Earlier work oncontinuous-time single-leg revenue management generally studies the Hamilton-Jacobi-Bellman equationcorresponding to a given formulation whenever the value function is differentiable and is a solution tothis equation. Here, we don’t assume that the value function is differentiable. Instead, we solve theproblem using a probabilistic sequential construction technique, which uses a continuous-time discrete-event dynamic programming operator. We provide a rigorous analysis of the problem and its solution.As a by-product of our approach we show that the value function is not always differentiable; therefore,one should be careful in using the Hamilton-Jacobi-Bellman equation in general. In our analysis, we alsodiscuss how the overbooking level should be determined in practice.

2 - Computing tight upper bounds for network revenue managementDavid Sayah, Chair of Logistics Management, Gutenberg School of Management andEconomics, Johannes Gutenberg University Mainz, Jakob-Welder-Weg 9, Haus Recht undWirtschaft I, 55128, Mainz, Rheinland-Pfalz, Germany, [email protected]

The network revenue management problem consists of selling a network of resources, e.g., flight legs inan airline network, with limited capacity to customers from different market segments who are randomlyrequesting bundles of resources. As the resulting dynamic program is computationally intractable, thereis a strong need for value function approximations that may serve as a basis for constructing heuristicbooking controls (e.g., a bid price control) and as a measure for assessing the quality of suboptimal solu-tions. A recent approach is based on value function decomposition using Lagrangian relaxation. Whileexisting theoretical and computational results are in support of Lagrangian relaxation, in solving the La-grangian dual problem, one potentially runs into convergence problems that might prevent from finding abest Lagrangian upper bound on the value function. We analyze the performance of different algorithmsfor solving the Lagrangian dual problem. Amongst others, we employ stabilization of Lagrangian dualsolutions to speed up convergence and report computational results on their performance.

3 - Reductions of Approximate Linear Programs for Network Revenue ManagementThomas Vossen, University of Colorado at Boulder, United States,[email protected], Dan Zhang

We consider equivalent reductions of approximate linear programs (ALPs) for network revenue manage-ment by exploring the relationship between ALPs and appropriate Dantzig-Wolfe reformulations. Dra-matic reduction of ALP problem sizes are shown for (i) separable piecewise linear approximation underindependent demand model, and (ii) affine approximation with customer choice behavior.

136

TB-03

� TB-03Thursday, 11:00-12:30Leeuwen Room 1

Environmental ManagementStream: Energy and EnvironmentInvited sessionChair: Florian Denz, Chair of Business Administration, Production & Supply Chain Management,Augsburg University, Germany, [email protected]

1 - Techno-economic analysis and evaluation of recycling measures for iron and steelslagsChristoph Meyer, Institute of Automotive Management and Industrial Production, TechnischeUniversität Braunschweig, Germany, [email protected], Matthias Gerhard Wichmann,Thomas Spengler

With a total of 1.5 billion tons of crude steel in 2012 worldwide steel production has reached its highestlevel to date. This also leads to the necessity of dealing with large amounts of by-products. An essentialgroup of by-products in the iron and steel industry are slags. In Germany the production of 42.5 milliontons of crude steel lead to 13.5 million tons of iron and steel slags in 2012. Slags perform importantmetallurgical tasks and are inevitable for iron and steel production processes. Although slag productionis inevitable, slags are not considered waste and can be used as secondary resources. For example, slagsare recycled to produce road construction material, cements and fertilizers. In order to use the potential ofrecycling slags there is a variety of alternative recycling measures. The potential of a recycling measurestrongly depends on a multitude of technical, legal and economic variables. For example the chemical andmechanical composition of slag products is subject to laws and also influences the attainable product price.Due to the actual concurrence of technical, legal and economic variables a general statement concerningthe advantage of one specific recycling measure cannot be given. In short term production planning thisleads to the question how slags are to be recycled. An approach to production planning taking into accountall relevant variables is not known. This contribution introduces a production planning approach for slagrecycling considering technical, legal and economic variables. The planning approach is based on anactivity analysis focusing operating points of possible recycling processes.

2 - Supporting Investment Decisions in Green Manufacturing Production SystemsSoumia Ichoua, College of Business, Embry-Riddle Aeronautical University, Daytona Beach,Florida, United States, [email protected], Agnes Pechmann

With the growing demand for energy around the globe, industrialized countries such as the United Statesand Germany are taking the lead in promoting the use of renewable energy sources as clean and safealternatives to fossil and nuclear power traditionally used. For these countries, engaging small or mediumsized manufacturing plants is especially crucial as they typically represent a substantial part of the industrialsector, which accounts for a major part of the overall energy consumption. However, for a small or mediumsized manufacturing plant, investing in a Virtual Power Plant (VPP) is a major capital investment that needsto be examined with caution. In this talk, we present a decision support model to assist small and mediumsized enterprises in deciding whether a VPP is a worthy investment or not. Based on historical data onenergy levels from wind and solar sources, our model simulates and balances the energy demands andsupplies over a given production planning horizon in a cost effective manner without exceeding the storagecapacity limits within the VPP. The balancing process is based on an optimization procedure which seeksthe best way to satisfy power demand and allocate excess of a given power type among storage, transferto the other type or to the grid when profitable. The optimization procedure takes into account the storagecapacity limit of each power type within the VPP, pricing policies and energy cost projections over theplanning time horizon. The present value of yearly cost reductions obtained from using the VPP over theplanning horizon is then calculated and the total net present value of investing in the VPP is derived takinginto account installation and procurement costs.

137

TB-04

3 - Implications of energy demand and supply characteristics for energy-orientedschedulingFlorian Denz, Chair of Business Administration, Production & Supply Chain Management,Augsburg University, Germany, [email protected]

As the establishment of sustainability in production is a major concern, approaches aimed at the improve-ment of energy efficiency are crucial for the environmental performance of modern production processes.The integration of energy related objectives into scheduling is a promising but neglected research field. Inorder to encourage prospective, more comprehensive research on the incorporation of energy aspects intoscheduling approaches, the article at hand analyses basics for their adequate consideration in optimizationapproaches. The investigation supposes a characterization of jobs by their temporal varying demand fordifferent energy sources (e.g. electric current, pressurized air or heating gas) in order to be capable to ad-dress energy optimization expediently. Furthermore, the effect of both technical as well as organizationaldifferences in the supply of these energy sources is examined. The impacts of scheduling decisions onthe resulting (cumulated) demand for energy sources are examined for several machine environments. Inconsequence, the study provides detailed insights about the interdependencies of jobs’ demands for energysources, the temporal as well as the machine related accumulation of these demands and the technicalstructures of energy supply systems. As a result of this analysis, guidelines are given how these differentcircumstances can be adequately integrated into quantitative planning approaches. An investigation aboutthe relationships of energy oriented objectives and traditional scheduling objectives concludes the study.

� TB-04Thursday, 11:00-12:30Leeuwen Room 2

Financial Modeling, Optimization and RiskStream: Financial Modeling, Banking and InsuranceInvited sessionChair: Ursula Walther, FB 1, Berlin School of Economics and Law, Badensche Str. 50-51, 10825,Berlin, Germany, [email protected]

1 - Efficient dimension reduction in interest rate modelling by global optimizationRalf Werner, Institut für Mathematik, Universität Augsburg, Universitätsstr. 14, 86159,Augsburg, Germany, [email protected]

We investigate the behaviour of different formulations for the calibration problem of low parametric modelsto real world yield curves. We highlight problems using a simple approach based on standard optimizationroutines currently followed in financial industry. We improve the standard approach by using a suitabledeterministic adaptive global optimization routine based on sparse grids, while keeping computation timeswithin reasonable limits. We close the talk with a description of the benefits based on an application withinthe risk capital model of an insurance company.

2 - DEA Modelling for Efficiency Optimization of Indian Banks using Negative DataSetsPankaj Kumar Gupta, Centre for Management Studies, JMI University, Jamia Nagar, New Delhi,110088, New Delhi, Delhi, India, [email protected], Seema‘ Garg

Indian banking has experienced exponential growth after reforms of 1990’s that helped to improve the prof-itability, performance and efficiency. However, still there are conflicting concerns of operating efficiencyand risk management across the major groups comprising of public sector banks and private sector banksparticularly after the global financial crisis. We have used Data Envelopment Analysis (DEA) for measur-ing the efficiency of a set of decision making units (DMUs) which traditionally assumes that all the inputand output values are non-negative. Quantitative measures of bank performance like net profits, growth

138

TB-05

rates and default portfolios frequently show negative values for output variables. We draw motivation fromsome studies done in other developing countries for handling the negative data sets. We cross examine theapproaches for dealing with variables that are positive for some DMUs and negative for others. We testthe validity of Range Directional Measure Model(RDM) for examining the cases when some inputs and/oroutputs can take negative as well as positive values. We find some support for the RDM in handling datanegative sets without the need for any transformation (conversion of the negative values with small positivevalues) as a measure of efficiency akin to the radial measures in traditional DEA. Our preliminary investi-gation indicates no significant difference between the operational efficiency and profitability of public andprivate banks modelled for negative data and undesirable output. We also attempt to optimise efficiencyvalues while obtaining the improvement index that can possible suggest the determination of interest ratestructures for banks.

3 - How to improve credit risk models? A conceptional design at graph theory andsemantic webTorsten Rex, TU Dresden, Germany, [email protected], Mario Straßberger

Credit risk models are applied to estimate value at risk (VaR) and economic capital of a banks credit port-folio. Common models like Credit Metrics or Credit Risk+ usually neglect micro structural relationshipsbetween obligors.

The aim of this work is to improve such models using semantic web and graph theory, in particular thetriple algorithm and Bayesian networks.

The established models typically use highly structured data for modelling stochastic dependencies of oblig-ors profit and loss variables. They operate with sector correlations. This proceeding does not reflect actualmicro structural relations between obligors, those who arise from a supply chain for example. So, themodels do inadequately take into account a) sudden tremendous changes of an obligors situation itself andb) situations when a supplier or a client of a banks borrower will get fraught with problems.

Furthermore, a second deficit of the current models is eliminated. Parameter estimation is based not onlyon highly structured data but also on data that are at least 6 - 12 months of age. Latest information hasoften been left out so far. "Breaking news’ are regularly unstructured data. In this work latest informationis integrated with semantic web approaches.

First attempts to map these micro structural relationships were developed using contagion models. Multi-factor asset value models were proposed. However, these models are still incomplete. They form sectorsto estimate correlations and then they determine the degree of infection between sectors. This work showshow to dispense sector indices using graph theory and semantic web. Micro structural relationships be-tween obligors are recorded and models are improved.

� TB-05Thursday, 11:00-12:30Goudriaan Room 1

Healthcare modelling and optimizationStream: Health Care ManagementInvited sessionChair: Johann Hurink, Department of Applied Mathematics, University of Twente, P.O. Box 217, 7500AE, Enschede, Netherlands, [email protected]

1 - A broader view on health care system design and modellingNico Vandaele, Operations Management Dept., Katholieke Universiteit Leuven, Naamsestraat69, 3000, Leuven, Belgium, [email protected], Catherine Decouttere

139

TB-06

Many rigorous models have been developed to support health care system design. However, embeddingthese models in a broader stakeholder based framework, will substantially enhance the societal and humanimpact. Moreover, the acceptance of the proposed health care system (re)design suggestions will be moreevident. Building on the model of an NMR scanning department, we propose an integrated health caredesign approach to support the modelling, the stakeholder analysis, the generation of alternative scenario’sand the final design choice.

2 - Duty and workstation scheduling for physicians at hospitalsJens Brunner, University of Augsburg, Germany, [email protected]

The paper addresses a physician scheduling problem at hospitals. We present two new mixed integerlinear programming models — a duty-roster and a workstation-roster model. The objective is to maximizefairness subject to legal and hospital specific constraints. The models are implemented in a new softwaretool that has an intuitive user-interface. Furthermore, we present computational results that indicate theusability of the models in practice. High quality schedules are achieved. We compare the schedules withcommon practice at our cooperation hospital.

3 - Clustering clinical departments for wards to achieve a prespecified blocking prob-abilityTheresia van Essen, University of Twente, Postbus 217, 7500 AE, Enschede, Netherlands,[email protected], Johann Hurink

When the number of available beds in a hospital is limited, it can be beneficial to cluster several clinicaldepartments such that the probability of not being able to admit a patient is acceptably small. However,not all clinical departments can be clustered as e.g. surgical patients can easily get an infection from oneof the non-surgical patients. In addition, patients from one clinical department should not be spread outover the entire hospital as this complicates the process of doing rounds.

In this paper, we consider a situation where wards with a fixed amount of beds are given. The questionis how to cluster the clinical departments and to determine the assignment of these clustered departmentsto the available wards such that enough beds are available to guarantee a blocking probability below aprespecified percentage. We first give an exact formulation of the problem to be able to achieve optimalsolutions. However, computational experiments show that the resulting computation times for this methodare too long for this method to be applicable in practice. To reduce the computation time, we introducetwo heuristic solution methods. The first heuristic uses the same formulation as the exact solution method,however, the number of beds needed is approximated by a linear function. The resulting model is againsolved by an exact solver. The second heuristic uses a restricted version of the exact solution methodwithin a local search approach. Hereby, the local search is used to determine the assignment of clinicaldepartments to clusters and the exact method is used to determine the assignment of clusters to wards.

� TB-06Thursday, 11:00-12:30Goudriaan Room 2

SAMS-modellingStream: Software Applications and Modelling SystemsInvited sessionChair: Ovidiu Listes, Paragon Decision Technology, Paragon Decision Technology, Schipholweg 1,2034 LS, Haarlem, Netherlands, [email protected]

1 - An IBM-certified course on modeling with IBM ILOG CPLEX Optimization StudioClaudius Steinhardt, Chair of Analytics & Optimization, University of Augsburg,Universitätsstraße 16, 86159, Augsburg, Germany, [email protected],Melanie Reuter, Hans Schlenker, Robert Klein, Stefan Nickel

140

TB-06

For the past two years, the Karlsruhe Institute of Technology and the University of Augsburg have beenteaching a course on modeling with ILOG CPLEX Optimization Studio in cooperation with IBM. Thecourse complements existing theoretical bachelor courses on mathematical programming. In the course,students get a detailed introduction into the elements of OPL and learn how to efficiently model, implementand solve real-world optimization problems. Students who successfully pass the course receive an officialcertificate from IBM. Shortly, the material will be published, allowing other interested universities to givethe course and officially certify their students. As IBM ILOG CPLEX Optimization studio is part of theIBM Academic Initiative, the software can be used free of charge by the teaching staff as well as by thestudents attending the course. In this talk, we give an overview of the content of the course and show howthe course can successfully be included into existing study programs. We also explain how the course canbe continued on the master’s level. Furthermore, we give details on the official certification process.

2 - Recent Enhancements in GAMSLutz Westermann, GAMS Software GmbH, Eupener Str. 135-137, 50933, Koeln, Germany,[email protected]

From the beginning in the 1970s at the World Bank till today, GAMS, the General Algebraic ModelingSystem, has evolved continuously in response to user requirements, changes in computing environmentsand advances in the theory and practice of mathematical programing. We will outline several recent en-hancements of GAMS supporting efficient and productive development of optimization based decisionsupport applications.

3 - Object Oriented GAMS API: Java, Python and .NETClemens Westphal, GAMS Software GmbH, Germany, [email protected], MichaelBussieck, Lutz Westermann

The object oriented GAMS API provides convenient access to GAMS from within different program-ming languages. Implementations are currently available for .NET, Java and Python. The object orientedAPI offers a seamless integration of GAMS into other programming environments. The API extends thecapabilities of GAMS by the addition of the rich features provided by object oriented programming lan-guages and their execution and development environments. The powerful in-memory representation of aGAMS model can increase performance when solving a model multiple times with slightly changed databy performing the model creation only once. Convenient data structures allow random access to data aswell as data iteration. The recently added facility of domain controlled data structures lead to safer datapreparation.

4 - Advanced optimization modeling enabled by the Generated Math Program (GMP)library in AIMMSOvidiu Listes, Paragon Decision Technology, Paragon Decision Technology, Schipholweg 1,2034 LS, Haarlem, Netherlands, [email protected]

AIMMS GMP library offers a fine-grained control over the generation, manipulation and solution of aMath Program instance. This enables capabilities which go far beyond those offered solely by the high-level symbolic modeling. Column generation, asynchronous solver sessions, or finding the best K solutionsfor a MILP are examples of subjects which can be addressed using the standard GMP functions. DedicatedAIMMS GMP functionalities support MINLP solution by AIMMS Outer Approximation (AOA), NLPpre-solver, NLP multi-start, Stochastic Programming, Robust Optimization, or Benders decomposition forMILP. There are also smaller, but very useful model-related tasks enabled by the AIMMS GMP functions.We attempt here a concise synthesis based on illustrative examples.

141

TB-07

� TB-07Thursday, 11:00-12:30Antwerp Room

Scheduling Applications IIStream: Scheduling and Project ManagementInvited sessionChair: Sigrid Knust, Institute of Computer Science, University of Osnabrück, Albrechtstr. 28, 49069,Osnabrück, Germany, [email protected]

1 - A classification scheme for storage unloading and loading problemsJana Lehnfeld, Fachbereich 06 (Institut für Informatik), University of Osnabrück, Albrechtstraße28, Raum 31/326, 49076, Osnabrück, Germany, [email protected]

Storage unloading problems deal with the retrieval of items out of a storage area while storage loadingproblems consider the assignment of a position for incoming items. Both kinds of problems have beentackled from several practical views, e.g. container storage in container terminals, tram and bus storagein depots and steel slab warehousing. Integrated problems combine storage unloading and loading andhave been considered in container stowage planning, in particular. Although all these problems originatefrom different practical applications they basically treat the same problem class. In this talk, we present aclassification scheme which is able to unite this diversified problem field and provides a simple and clearsurvey. We will classify some problems existing in literature and point out the border between polynomiallysolvable and NP-hard problems.

2 - Storage planning in intermodal rail-road terminalsFlorian Bruns, Institute of Computer Science, University of Osnabrück, Germany,[email protected], Sigrid Knust

Storage planning defines storage positions in a potentially partially filled storage area of intermodal rail-road terminals which consists of one or multiple storage lanes for a set of load units that have to be stored.Load units can be stored at ground level and many load units (containers and some swap bodies) can also bestacked on top of other load units, but trailers can not be stacked. Different constraints like non-overlappingof load units, stacking restrictions or hard terminal policies have to be respected. The aim of the storageplanning is to minimize handling costs in the terminal. We consider transportation costs and costs forrepositioning (or reshuffling) load units when a lower load unit of the same stack has to be delivered.

We analyse different MIP formulations and their performance for real-world instances. Furthermore, wepropose a heuristic and present computational results.

3 - Two-stage scheduling in shelf-board production: A case studySigrid Knust, Institute of Computer Science, University of Osnabrück, Albrechtstr. 28, 49069,Osnabrück, Germany, [email protected], Stefan Waldherr

We consider a production planning problem arising at a company assembling shelf boards for kitchenelements. Different products have to be produced on circular production units with a synchronous trans-portation system, which leads to sequence-dependent production times. Additionally, changeover timesoccur due to changes of resources. The objective is to find a production schedule of given orders such thatthe number of late orders is minimized.

We propose a hierarchical approach which is split into a coarse and a fine planning stage. While the coarseplanning problem is solved as a network flow problem, the fine planning is done by an asynchronous team(A-team) approach. Some computational results for real-world data are presented.

142

TB-08

� TB-08Thursday, 11:00-12:30Penn Room 1

Job Shop SchedulingStream: Scheduling and Project ManagementInvited sessionChair: Han Hoogeveen, Department of Information and Computer Science, Utrecht University,Princetonplein 5, P.O. Box 80089, 3508 TB, Utrecht, Netherlands, [email protected]

1 - Optimizing NEH Schemes for Complex Flow Shops with Genetic ProgrammingChristian Rathjen, WINFOR (Business Computing and Operations Research), SchumpeterSchool of Business and Economics, University of Wuppertal, Gaußstraße 20, building M, room12.34, 42119, Wuppertal, Germany, [email protected], Stefan Bock

Permutation flow shop production systems in the process industry frequently pursue buffer-reducing JiT-policies, where tight time restrictions have to be met. Additionally setup operations have to be performedthat tend to be sequence-dependent and consume time or money. In order to construct suitable scheduleswith small makespan or tardiness costs for due-date related objectives, we optimize NEH procedures withGenetic Programming (GP). In contrast to previous publications in this field we consider setup times anddo not only focus on finding good initial sequences, but in addition we let GP determine good tie-breakingrules for the insertion stage of NEH. The efficiency of the GP-optimized heuristics is evaluated on a largeset of benchmark instances in direct comparison with existing solution approaches.

2 - Genetic Algorithm Approaches for a Job Shop Scheduling: A Real Case StudyTarik Yilmaz, Statistics, Science Institute, Selcuk Universitesi Alaaddin Keykubat Kampusu,Selcuklu, 42020, KONYA, Turkey, [email protected], Asır Genç

As is known, job shop scheduling is an important decision process in contemporary manufacturing systems.This study was done to industry-leading manufacturer of a calendar in Turkey. In this paper, we focusedon to finish jobs which were order in as short a time as possible. So the total weighted tardiness must beminimized. To achieve this goal we use a genetic algorithm to optimize the set of dispatching rules usedfor schedule construction. For users who do not have experience with job scheduling also use a softwarehas been developed. The superiority of the proposed approach is especially remarkable when the size ofthe scheduling problem is large.

3 - A GRASP approach for the job shop scheduling problem with the objective ofminimizing the total weighted tardinessJens Kuhpfahl, School of Economics and Business, Chair of Production and Logistics, MLUHalle-Wittenberg, Germany, [email protected], Christian Bierwirth

We consider the job shop scheduling problem with minimizing the total weighted tardiness as objective.This objective function is more and more relevant since deadlines and their fulfilment recently became im-portant in production scheduling. In our talk, we present a greedy randomized adaptive search procedure(GRASP) for solving this problem. Basically, this procedure consists of a construction and an improvementphase based on local search. The search is navigated with the help of adaptive components. In the con-struction phase, the algorithm looks like the Hybrid Scheduling Algorithm of Storer, Wu, Vaccari (1992).In the improvement phase, solutions to the problem are represented as a directed graph G. For this purpose,the graph is derived from modifications of the well-known disjunctive graph for the minimum makespanjob shop scheduling problem. Calculating all longest paths in G enables the identification of a set of crit-ical arcs and blocks that are subject to local search based neighborhood operators. In our approach, weuse four different neighborhood operators in order to provide a high diversity in the search process. Anadaptive selection of dispatching rules within the construction phase and a back jump tracking concept forintensifying the search in promising regions of the solutions space are included in the GRASP. We presentcomputational results to demonstrate the performance and high potential of our GRASP approach.

143

TB-09

4 - Exact solution methods for the flexible job shop problem using column generationHan Hoogeveen, Department of Information and Computer Science, Utrecht University,Princetonplein 5, P.O. Box 80089, 3508 TB, Utrecht, Netherlands, [email protected], RobinCijvat, J.m. van den AkkerThe standard job shop problem is defined as follows. There are n jobs that have to be executed. Eachjob consists of a chain of operations, which must be executed without interruption in the given order. Foreach operation, it is given by which machine it should be executed, and how long this takes. The goal isto minimize the time by which the last job has finished (makespan). We consider the flexible job shopproblem, in which each machine is replaced a station containing two or more parallel, identical machines;an operation then has to be executed by exactly one of these dedicated identical machines.We present an ILP-based solution method for this problem. We first prove a lower bound on the makespan,and then feed this information into an ILP to find an optimal feasible schedule. We have applied twomethods to find a lower bound. The first one is based on a destructive method: the lower bound is equal tothe smallest makespan value for which we cannot show that there does not exist a feasible solution. Thesecond lower bound method is based on solving the LP-relaxation of an ILP-formulation that is based onjob patterns. Given the resulting lower bound LB, we formulate the job shop scheduling problem as a timeindexed integer linear program, to which we add the additional constraint that the makespan is equal toLB. If we find a feasible solution, then we know it is optimal; if we can’t find one, then we know that wecan increase the lower bound to LB+1, and try again. Adding this constraint is crucial.If turns out that this method works very well, especially in combination with the second lower bound. Itdoes not work for the standard job shop problem.

� TB-09Thursday, 11:00-12:30Diamond Room 1

Maintenance Planning IIStream: Production and Operations ManagementInvited sessionChair: Geert-Jan van Houtum, Fac. of Technology Management, Eindhoven University of Technology,P.O. Box 513, 5600 MB Eindhoven, Netherlands, [email protected]

1 - Improving product design and maintenance through a better understanding ofstocks of spare parts and spare assetsRob Basten, Twente University, 7522NB, Enschede, Netherlands, [email protected], JorgeEduardo Parada Puig, Joachim ArtsOwners of fleets of assets, e.g., airplanes, ships, or busses, require a certain number of assets to be available(operational) at any given point in time. The total unavailability (downtime) of assets is the product of thenumber of downtimes, both planned and unplanned, and the duration of each downtime, where durationmay be split in the active maintenance time, or hands-on-tool-time, and logistic and administrative delays.If the asset design is given, then in order to guarantee with a high probability that there are sufficient assetsavailable, it is possible to invest in spare parts to lower logistic delays and to invest in additional (spare)assets. Finding the optimal investment turns out to be an interesting optimization problem. We give ana-lytical and numerical results for this problem and we show how these results may be used when solvingseveral related problems in product design and maintenance optimization. For example, when designingor upgrading assets, it may be possible to lower the active maintenance time or to reduce the number ofexpected downtimes. Using our results, we can determine the worth of both design improvements. An ex-ample in maintenance optimization is the so-called LRU definition problem. An LRU is a line replaceableunit, so the component that is replaced in the field. An LRU may be a small part with low holding costsand high associated active maintenance time, or a big module with the opposite characteristics. Our resultsmay be used to include the effect of the LRU definition on the required number of spare parts and spareassets.

144

TB-09

2 - An opportunistic maintenance policy for components in complex systems undercondition monitoringQiushi Zhu, Industrial Engineering, Eindhoven University of Technology, De Lismortel,Paviljoen E.14, 5600 MB, Eindhoven, Netherlands, Netherlands, [email protected], Hao Peng,Geert-Jan van Houtum

Due to the advanced sensor technologies nowadays, we can continuously monitor the degradation behav-ior of critical components in a system to prevent the unexpected failures by employing condition-basedmaintenance (CBM) policies. For complex engineering systems, usually not all of the components are ap-plicable for CBM. The other components in the system may subject to corrective maintenance or periodicpreventive maintenance. How to coordinate these different maintenance policies becomes a challengingproblem. In this research, we propose a new optimization model to determine the control limits of op-portunistic maintenance for a monitored component. It considers the chances to maintain this monitoredcomponent together with other components subject to corrective maintenance or periodic preventive main-tenance. A case study on lithography machines in semiconductor industry is provided, which shows theutilization of our model.

3 - A Condition-Based Maintenance Policy for a Continuously Deteriorating Multi-UnitSystem with Aperiodic InspectionsMinou Olde Keizer, Operations, University of Groningen, Nettelbosje 2, 9747 AE, Groningen,Netherlands, [email protected], Ruud Teunter

In condition-based maintenance (CBM), it is intended to perform maintenance right before a failure occursby estimating the pending moment of failure based on monitoring a certain condition, such as vibration ortemperature. We consider an existing CBM optimization approach that is advanced, compared to others,in that it optimizes both the inspection moments and the condition thresholds (on which the planning ofmaintenance actions is based) simultaneously. A discrete time system with two machines that operate inseries is considered, where the objective is to minimize the long run average maintenance cost per timeunit. We analyze an adapted version of the system where two units operate in parallel, and provide newinsights on CBM for systems with redundancy.

4 - The Loss and Failure Process of Sugar Production in Indonesia : A CaseHenry Yuliando, Agroindustrial Technology, Universitas Gadjah Mada, Indonesia, Jl Flora 1,Bulaksumur, Yogyakarta, Indonesia, 55281, Indonesia, [email protected], Adi Djoko Guritno,Endy Suwondo

The sugar produced in Indonesia is mainly from sugarcane. In its production, one of the problem is due tolosses during the process. As found in this study, a loss of sugar occurred mainly in the milling plant. Basedon data taken from a sugar plant in Yogyakarta, Indonesia, on May to June 2012, the left residue (bagasse)was 32% of total input. The main cause was identified stem from the old machines and facilities that areused to experience a failure, and the lack of concern to the maintenance activity. As a solution, the companyaddressed a plan for maintenance that was divided into major and minor maintenance . Here, this studypropose a measurement on loss, and mean time between failure (MTBF) for the case concerned. As thesituation of the milling plant of sugar company, this process is critical in order to determine the loss level.By employed the availabilty schedule of the milling plant for every period (month) reduced by maintenancetime that should be done in the middle of the process due to the encountered failure as estimated by MTBF,a test was done to see the significant correlation between loss and failure occurred. The result shows thatthe loss mainly caused by failure during the process caused by lack of proper maintenance schedule. Thisfinding discuss a need to put the problem in a platform of reliability management.

145

TB-10

� TB-10Thursday, 11:00-12:30Diamond Room 2

Demand ModelingStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Arnd Huchzermeier, Production Management, WHU - Otto Beisheim School of Management,Burgplatz 2, 56179, Vallendar, Germany, [email protected]

1 - Serving an Uncertain Season: When and How to Enter the MarketJochen Schlapp, Department of Business, University of Mannheim, Chair of Logistics,University of Mannheim, POB 103462, 68131, Mannheim, Germany,[email protected], Moritz Fleischmann

The standard newsvendor model is tailored to determine a firm’s optimal stocking quantity if a stochasticnumber of customers buys the product over a given selling season. In many industries, however, demandis not only uncertain in quantity, but also in the time of occurrence. In particular, for many productswhich depend on exogenous factors such as weather or fashion trends, the beginning and the end of theselling season are actually unknown by the firm. To account for this additional timing uncertainty, weinvestigate a newsvendor whose demand is uncertain in (i) quantity; (ii) the start of the selling season; and(iii) the length of the season. Since demand is uncertain in timing and quantity, the firm now has to maketwo crucial decisions upon launching its product. First, it has to decide on the optimal time for marketentry; and second, the optimal stocking quantity needs to be determined. Obviously, the firm faces a two-fold tradeoff between stocking too much and too little, and launching the product too early and too late.We show that the firm always stocks less than a standard newsvendor if the selling season is uncertain.Additionally, if the timing uncertainty exceeds a moderate level, the firm completely forgoes any sales andleaves the market. In contrast, if the firm decides to launch the product, then it either enters the market atthe very first opportunity, or it takes a substantial risk of being late. Apart from holding costs, we identifyopportunity costs as a crucial driver for this behavior. Lastly, we show that the firm’s optimal decision isnon-monotone in both, production and opportunity costs.

2 - Dynamic Lot-sizing Model with Advance Demand Information for Online RetailersYeming Gong, Emlyon Business School, 12 Rue Dunoir, 69003, Lyon, France,[email protected]

In an environment of time-varying demands, this paper studies the inventory replenishment planning prob-lems for online retailers who are able to obtain advance demand information (ADI). We incorporate ADIto dynamic lot-sizing (DLS) models to formulate the replenishment planning problems for online retailers.We consider three scenarios in this research. 1) Companies act as pure-play online retailers with customershomogeneous in demand lead time. 2) Online customers are heterogeneous in demand lead time withpriorities. Demand lead time of high-priority customers is shorter than that of low-priority customers. 3)Online retailers operate in a bricks-and-clicks structure, in which demands come from both online andoffline channels. These two channels can be either independent or interactive.

3 - Improving Demand Forecast Accuracy in a Continuous High—Low Pricing Envi-ronment with Multiple Customer SegmentsArnd Huchzermeier, Production Management, WHU - Otto Beisheim School of Management,Burgplatz 2, 56179, Vallendar, Germany, [email protected]

In the fast-moving consumer goods (FMCG) industry, the demand for traffic-generating products (e.g., dia-pers) is volatile—and especially so when retailers follow a high—low pricing strategy. Inaccurate demandforecasts reduce supply chain profits. Better forecasting requires retailers to recognize that customers areaware of promotion cycles and thus to factor customer behavior into demand forecasts. We present anefficient and accurate multi-segment demand forecasting model that accommodates continuous high—lowpricing schemes, accounts for the behavioral patterns of particular customer segments, and incorporates

146

TB-11

stock-out effects. As a hedge against any remaining unexplained forecast error, we also describe a profit-maximizing inventory policy for retailers. We validate the model by comparing its demand forecast toactual data on diaper sales by a German retailer.

� TB-11Thursday, 11:00-12:30Penn Room 2

Supply Chain Optimization IIIStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Joaquim Gromicho, Econometrics and OR, VU University, FEWEB, De Boelelaan 1105, 1081HV , Amsterdam, Netherlands, [email protected]

1 - Leveling the in- and outbound shipments in distribution centersStefanie Schlutter, Optimization, Fraunhofer SCS, Nordostpark 93, 90411, Nürnberg, Germany,[email protected], Thomas Bauer

This work is based on a practically motivated problem. A trading company was planning to centralize itsdistribution, but the picking volume tended to exceed the capacity limits of their distribution centers. Thiseffect was enhanced by seasonal fluctuations and typical weekly trend lines. In order to limit handlingcosts, an increase of warehouse staff should be avoided. Therefore a mixed-integer program was created tooptimize the throughput of shipments by exploiting freedoms in their order fulfillment periods. The level-ing model reduces the order peaks along with an improved resource utilization in periods of lower demand.Thus, the expected handling and storage expansion costs could be decreased significantly. An amount oftwo million shipments is a challenge for solving these problems in acceptable time. Five different levelingfunctions, which are motivated by the different customer problems, are going to be presented. The func-tions were tested with real world data based on periods of two months and half a year. The computationalresults will demonstrate the different objectives of these functions.

2 - Loading pallets and containers with practical restrictionsJoaquim Gromicho, Econometrics and OR, VU University, FEWEB, De Boelelaan 1105, 1081HV , Amsterdam, Netherlands, [email protected], Maria Teresa Alonso Martínez,Ramon Alvarez-Valdes, Jelke van Hoorn, Francisco Parreño, Gerhard Post

The weight of the load for a container or truck is in practice often significant. This leads to restrictionsthat are weight related to bind earlier than the container’s volume. Traditional GRASP based algorithmsare challenged by restrictions such as abeyance to load height distributions and axle weight bearings. Wepresent, and compare, two novel approaches for loading goods on pallets and placing the loaded pallets incontainers. One of these methods is based on GRASP but has the novelty that building a pallet anticipateson the effect this will have when placed in the container. The second method focuses on a global appre-ciation of the whole set of pallets built while placing them in the containers. Our computational resultsare based on an extensive set of real life loads from a large retailer in the United States and show verypromising results for both methods from both the quality and the running times point of views.

147

TB-12

� TB-12Thursday, 11:00-12:30Van Oldenbarnevelt Room

Facility Location and LogisticsStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Teresa Melo, Business School, Saarland University of Applied Sciences, Waldhausweg 14,D-66123, Saarbrücken, Germany, [email protected]

1 - Linear location-inventory models for service parts logisticsAbdullah Dasci, Faculty of Management, Sabanci University, Orhanli-Tuzla, Istanbul, Turkey,[email protected]

We present two integrated network design and inventory control problems in service-parts logistics sys-tems. Such models are complicated due to demand uncertainty and nonlinear time-based service levelconstraints. Exploiting unique properties of the nonlinear constraints, we provide an equivalent linear for-mulation under part-warehouse service requirements, and an approximate linear formulation under partservice requirements. Computational results indicate the superiority of our approach over the existingapproaches in the literature.

2 - Optimization of logistic networks for renewable resources from by-products con-sidering cascade utilizationMohammad Sadegh Taskhiri, Georg August Universität, 37073, Götingen, Götingen, Germany,[email protected]

Biomass sources play a significant role for greenhouse gas reduction and somehow substitute for fossilfuels. However it is essential to consider other strategies such as cascade utilization of biomass sourceswhich is the sequential use of biogenic raw materials to produce material and energy. The logistic conceptis crucial in utilization of the biomass resources and improving cascade utilization since the largest shareof costs in biomass material use or energy generation originates from logistic processes. One approachtowards maximizing the utilization of the biomass resources is to establish symbiotic networks betweenbiomass source (agriculture and forestry) and industries, considering the risks and uncertainties associatedwith each process, such that the collaboration of business entities will result in greater biomass utilizationbenefits through material and energy exchanges. This study will thus focus on developing a modelingframework for designing biomass logistic which considers the cascade utilization by determining optimumeconomically and environmentally solution. The result of this research will be applied in a case study inGermany for wood waste utilization. This work is a part of a research training group (RTG 1703) thatsupported by Germany Research Foundation (DFG) where 16 PhD students from forestry, agriculture,mathematic, information system and business administration are involved.

3 - A multi-period logistics network design problem with transportation mode selec-tion and outsourcing opportunitiesTeresa Melo, Business School, Saarland University of Applied Sciences, Waldhausweg 14,D-66123, Saarbrücken, Germany, [email protected], Maria Cortinhal, Maria JoãoLopes

We address the problem of redesigning a three-echelon logistics network comprising suppliers, plants,warehouses, and customer zones. To better cope with changes in customer demands, the geographic loca-tion of facilities (i.e. plants and warehouses), as well as their capacities, can be adjusted over a multi-periodhorizon. This involves opening new facilities at potential sites and selecting their capacities from a set ofavailable discrete sizes. Capacity expansion may occur more than once over the planning horizon both atnew locations and at existing facilities. The latter may also be closed. In addition, the operation of thenetwork is subject to logistical decisions involving procurement, production, distribution, and outsourcingof goods. Multiple types of products can be manufactured at plants by processing given raw materials.The latter can be procured from various suppliers taking into account their availability and cost. Finished

148

TB-13

products can be shipped from plants to warehouses or directly to customer zones. Distribution channelsare to be identified in each time period as well as the modes of transportation for raw materials and finishedproducts. Warehouses may also purchase finished products from external sources at additional cost whenthe network faces insufficient production and/or storage capacity. We propose a mixed-integer linear pro-gramming model to redesign the network so as to minimize the total cost. Valid inequalities are developedto enhance the original formulation. For test instances of reasonable size, we report on our computationalexperience with standard mathematical optimization software. In particular, useful insights on the impactof various factors on network redesign decisions are provided.

� TB-13Thursday, 11:00-12:30Mees Auditorium

Decomposition methodsStream: Discrete and Combinatorial Optimization, Graphs and NetworksInvited sessionChair: Stefan Minner, TUM School of Management, Technische Universität München, Arcisstrasse 21,80333, Munich, Germany, [email protected]

1 - Benders Decomposition for large-scale Linear Bilevel Problems with binary leadervariablesPirmin Fontaine, TUM School of Management, Technische Universität München, Germany,[email protected], Stefan Minner

Bilevel Problems are mathematical programming problems consisting of an optimization problem (leaderproblem) with nested optimization problems (follower problem) in the constraints. In practice these prob-lems can occur if decentralized or hierarchical decisions have to be taken. First the leader has to decideover a subset of the decision variables, which effects the feasible region of the follower. Afterwards thefollower has to decide over the other subset of decision variables, which effects the objective value of theleader. We propose a new fast solution method for Bilevel Problems in the special case with binary leadervariables and continuous follower variables. Additionally we are assuming the partial cooperation assump-tion, which allows the leader to select a for himself optimal follower decision among all optimal followerdecisions, if there exists more than one. Our approach reformulates the Bilevel Problem into a single levelproblem by using the Karush-Kuhn-Tucker conditions. This nonlinear model can be linearized because ofthe special structure achieved by the binary decision variables and subsequently solved by a Benders De-composition Algorithm to global optimality. Numerical tests will be provided to show the rigorous powerfor solving large instances of Bilevel Problems even though we are calculating the global optimum.

2 - Solving staircase structured problems via Dantzig-Wolfe ReformulationMartin Bergner, Operations Research, RWTH Aachen University, Kackertstr. 7, 52072, Aachen,Germany, [email protected], Marco Lübbecke

Many stochastic, lot sizing and resource allocation problems formulated as linear integer programs inhibita so called staircase or staged structure, which is widely exploited by various methods in the literature. Inthese problems, the stairs (or stages) share variables only with the previous or next stage. In our talk, wewill generalize an approach to exploit staircase structures using Dantzig-Wolfe reformulation, a methodapplied successfully so far to staircase structured resource allocation problems, to general mixed integerproblems (MIPs) from the literature. Each stage of the optimization problem will be solved in the pricingproblems keeping only the coordination of the stages in the master problem. We will analyze both exactand heuristic algorithms for staircase structure detection in arbitrary MIPs. In order to judge whether itpays off to detect these structures in arbitrary problems, we will finally analyze computational results forproblems both with and without known staircase structures in order to asses the quality of the reformulationfor general MIPs.

149

TB-14

3 - Evaluating the quality of a Dantzig-Wolfe decomposition via graph modularityAnnika Thome, Operations Research, RWTH Aachen University, Kackertstr. 7, 52072, Aachen,Germany, [email protected], Marco Lübbecke

An integer program (IP) can be Dantzig-Wolfe reformulated in different ways. Algorithms to obtain such areformulation automatically, and a posteriori measures of the ’quality of a decomposition’ (strength of theobtained relaxation or computation times to solve the reformulated model) have been recently proposedin the literature. In this presentation, we are interested in an a priori measure, and propose to evaluatecertain modularity parameters of a graph that exploit the structure of the coefficient matrix of the IP. Wepresent a computational study for IPs in general, and discuss whether and how ’good’ Dantzig-Wolfedecompositions can be obtained. We further investigate the specialization of our approach to the stable setproblem.

� TB-14Thursday, 11:00-12:30Tokyo Room

Airline ApplicationsStream: Traffic and TransportationInvited sessionChair: Taieb Mellouli, Business Information Systems and Operations Research, University of Halle,Universitätsring 3, 06108, Halle Saale, Germany, [email protected]

1 - Multi-Objective Mathematical Modelling for Aircraft Landing by Simultaneous Min-imization of Time and CostSeyedmirsajad Mokhtarimousavi, Civil Engineering, M.Sc Student of Transportation Planning,Science & Research Branch, Islamic Azad University of Tehran, Tehran-Iran, Sciences &Research Branch, Hesarak, Tehran ,I.R.Iran, 1477893855, Tehran, Tehran, Iran, IslamicRepublic Of, [email protected], Hossein Rahami, Mahmoud Saffarzadeh, EzazSadegh Vaziri

Due to an anticipated increase in air traffic during the next decade, air traffic control in busy airports is oneof the main challenges confronting controllers in the near future. The most important factors in the aircraftlanding modeling are time and cost. The Aircraft Landing problem (ALP) is a typical hard multi-constraintoptimization problem. In real applications, it is not most important to find the best solution but to providea feasible landing schedule in an acceptable time. Our study first presents a brief review of the aircraftlanding problem. Then we propose a formulation of the problem as a mathematical programming modelin order to reduce the number of constraints (which can positively reduce the computing time) and to givea more rigorous formulation. The model was introduced to solve this problem by using penalty functionsand MATLAB programming solutions and scenarios by which we can control then a model for the problemis then introduced, and possible solution approaches are discussed.

2 - Air freight transport in Baltic Sea region - Perspectives and necessary develop-mentsJoachim R. Daduna, Hochschule für Wirtschaft und Recht Berlin Berlin, Badensche Str. 52,10825, Berlin, Germany, [email protected]

Air freight transport is an essential element for the design of efficient and customer-oriented logisticsstructures. It is for this reason that in recent years an increasing number of airport locations for freightand passenger transport have been planned to be able to meet the existing and expected demand. However,in many cases the economic efficiency is not taken into account sufficiently, as often rather prestigiousaspects are of importance for local authorities. This problem is also present in the Baltic Sea region and theneighbouring hinterland. Due to the existing international economical and traffic-related interdependencies

150

TB-15

this situation can not be considered from the point of view of the individual countries. These problemsmust be rather analyzed and discussed in a trans-national cooperation. The question arising here is howmany and which of the available airports are required to cover the demand in airfreight transport on along-term basis for this region. In view of sustainable economic efficiency it would be appropriate torestrict the number of locations which are used to be able to develop airports that are not being subsidizedand hence provide economically feasible solutions. To be able to ensure full-coverage service even witha limited number of airports, it is essential to identify suitable site configurations as well as to developcorresponding concepts in order to guarantee a sufficient accessibility of the selected locations. FlyingTruck or Road Feeder Services could be interesting solutions for the connection of the airports and theirhinterland.

3 - Anticipation in Airline Crew Pairing Optimization: Results from a Real World CaseStudyMichael Römer, Juristische und Wirtschaftswissenschaftliche Fakultät,Martin-Luther-Universität Halle-Wittenberg, Universitätsring 3, 06108, Halle (Saale), Germany,[email protected], Taieb Mellouli

The short term airline crew planning process is typically divided into two steps: In crew pairing opti-mization, flights are combined into pairings (legal sequences of daily flight duties) starting and ending atthe same crew domicile. In the subsequent crew rostering step, these anonymous pairings are assignedto concrete crew members along with weekly rests and standby duties. However, due to the existence ofpreassignments such as training activities and vacation leaves and due to the complexity of the rules gov-erning roster legality, it often occurs that a subset of the pairings cannot be assigned during rostering. Inorder to arrive at a feasible rostering solution in such a case, either the pairings have to be modified locallyor the airline has to rely on expensive fall-back options such as hiring freelance crew members or payingregular crews for involuntary flying. Given these observations, it only seems natural to incorporate as muchrostering-related information as possible into the crew pairing step in order to enforce consistency with thesubsequent rostering step. In this study, we investigate the effects of different levels of rostering antici-pation in crew pairing optimization on the solution quality of the full short term crew planning process.To this end, we evaluate the solutions of different models for crew pairing optimization with an automaticrostering heuristic in a real world setting based on data sets from a medium sized German airline. It turnsout that while the computational effort for pairing optimization grows with the level of anticipation, severalindicators such as the number of unassigned flights and the incurred overtime costs show that it is indeedbeneficial to employ a high level of anticipation in crew pairing optimization.

� TB-15Thursday, 11:00-12:30New York Room

Branch-and-PriceStream: Discrete and Combinatorial Optimization, Graphs and NetworksInvited sessionChair: J.m. van den Akker, Information and Computing Sciences, Utrecht University, POBox 80089,3508TB, Utrecht, Netherlands, [email protected]

1 - Branch-and-Price for the k-Club Partitioning and Covering ProblemsStefan Irnich, Chair of Logistics Management, Gutenberg School of Management andEconomics, Johannes Gutenberg University Mainz, Jakob-Welder-Weg 9, 55128, Mainz,Germany, [email protected], Fabio Furini, Roberto Wolfler-Calvo

Relationships between objects can be modeled with graphs, where nodes represent the different objectsand edges express the relationship. Social network analysis is an example where clusters, e.g., formedby members of a community, are studied using cliques and clique relaxations. A clique is a subgraph

151

TB-15

with pairwise directly connected nodes, i.e., a subgraph with diameter one. Several relaxations have beendefined either in terms of distance (k-clique), degree (k-plex, k-core), diameter (k-club), or density. Themajority of the literature deals with identifying such subgraphs of maximum cardinality or weight. In thispresentation, we consider the problem of covering or partitioning a graph with a set of k-clubs. Note thata k-club is a subgraph with diameter at most k. For k=1, the clique partitioning problem results, whichis equivalent to the graph coloring problem in the complement graph. For k>1, however, covering andpartitioning are properly different problems because a subset of a k-club is not necessarily a k-club again.We present an exact solution approach to the k-club partitioning and covering problems based on columngeneration. Herein, the subproblem consists of finding a k-club of maximum weight. We present heuristicsand a new combinatorial branch-and-bound algorithm for its resolution. The most interesting part of thebranch-and-price is the branching scheme. We will derive branching rules that together guarantee integersolutions. Such branching rules are different and non-trivial for both cases (partitioning and covering)because a good rule should at the same time be compatible with the subproblem’s structure and solution,create a small number of branches, and generally improve the bound in all resulting problems.

2 - Combining Different Pricing Approaches for Branch-and-Price AlgorithmsMichael Bastubbe, Operations Research, RWTH Aachen University, Kackertstr. 7, 52072,Aachen, Germany, [email protected], Marco Lübbecke, Christian Puchert

When applying branch-and-price algorithms one often has a choice between different approaches to solvethe pricing problem. These approaches often differ in complexity, accuracy and capability to respect thebranching decisions. Furthermore for many problems there is an overwhelming number of possibilities toimprove or recombine (or repair) known (infeasible) solutions of the pricing problem. We want to givea general classification of different pricing approaches and propose a meaningful scheme that combinesthem.

In the beginning of this talk we discuss the influence of the design of the pricing scheme to the overallperformance for an IP formulation for a specific problem. This is rearranging the rows and the columnsof a matrix into doubly bordered block-diagonal form while minimizing the number of border rows andcolumns.

Our results are implemented in the generic branch-and-price solver GCG. We present several approachesfor the pricing problem arising when applying Dantzig-Wolfe decomposition to a binary program andsuggest a combining scheme.

3 - Column generation for the demand robust shortest path problemJ.m. van den Akker, Information and Computing Sciences, Utrecht University, POBox 80089,3508TB, Utrecht, Netherlands, [email protected], Paul Bouman, Han Hoogeveen,Denise Tonissen

We study the demand robust shortest path problem. Initially we are given the source but the sink is un-certain. We have to buy a set of arcs containing a path from the source to the sink. Arcs can be boughtwhen the sink is still unknown, or, at a higher price, after the sink has been revealed. Our approach isbased on recoverable robustness, i.e. we make an initial plan which is guaranteed to be recoverable to afeasible solution by a fast and simple recovery algorithm. We apply the technique of column generationto find solutions to our recoverable robust optimization problem. In an earlier paper, we have identifiedtwo types of column generation approaches: separate recovery and combined recovery, and have testedthese for a recoverable robust knapsack problem. For the demand robust shortest path problem, we presentan algorithm based on combined recovery. Our computational results indicate that the column generationstrategy in crucial for the performance of the algorithm.

152

TB-17

� TB-17Thursday, 11:00-12:30New Orleans Room

Conic OptimizationStream: Continuous OptimizationInvited sessionChair: Mirjam Duer, Mathematics, University of Trier, 54286, Trier, Germany, [email protected]

1 - Handelman’s hierarchy for the maximum stable set problemZhao Sun, Tilburg University, Netherlands, [email protected], Monique Laurent

The maximum stable set problem is a well-known NP-hard problem in combinatorial optimization, whichcan be reformulated as the maximization of a quadratic square-free polynomial over the (Boolean) hyper-cube. A popular approach to polynomial optimization problems is to build a hierarchy of convex tractablerelaxations, based on replacing the (hard to test) positivity condition for a polynomial by a tractable, suf-ficient condition for positivity. One may for instance search for positivity certificates in the form of acombination of products of constraints where the multipliers could be sums of squares of polynomials orsimply nonnegative numbers. This leads respectively to semidefinite and linear programming relaxations.Although semidefinite hierarchies are stronger, they are more difficult to analyze and computationally moreexpensive. This motivates our study of the linear programming hierarchy for the maximum stable set prob-lem, based on Handelman’s representation result for positive polynomials on a polytope. Using Bernsteinpolynomials Park and Hong (2012) could show some error bounds for the approximate solutions obtainedat any order in the hierarchy. We further investigate this hierarchy. In particular, we show a relation to frac-tional clique covers of graphs, we give bounds for the rank of the Handelman hierarchy (i.e., the smallestorder needed to obtain the true optimum), and we point out links to some other LP and SDP hierarchies.

2 - Linear cone programs: Structure and generic properties.Georg Still, Mathematics, University of Twente, P.O.Box 217, 7500 AE, Enschede, Netherlands,[email protected], Mirjam Duer, Bolor Jargalsaikhan

In this talk we consider linear cone programming, an important class of convex optimization problems. Itcontains linear programming (LP), semidefinite, and copositive programming as special cases. The talkaims to provide an introduction into the basic ideas and techniques of the topic. We introduce the dualityconcept and discuss the generic structure of conic programs. In our exposition we say that a propertyis generic if it holds for almost all problem instances and if the property is stable with respect to smallperturbations of the problem data. The following main points will be treated: - The generic structure of LPis compared with the structure of general conic programming. -It appears that strong duality is a genericproperty in general cone programming and that unicity and strict complementarity of the solutions holdsfor almost all such problems. - It will be shown how the generic structure of LP can be analysed by methodsfrom (smooth) differential geometry. - On the other hand, we also will see, that general cone programscannot be treated as mooth problems but have to be studied by techniques from (nonsmooth) geometricmeasure theory. We will mention some important results from this field.

3 - Order of solutions in conic programmingBolor Jargalsaikhan, SCAA, University of Groningen, Johann Bernoulli Institute, P.O. Box 407„9700 AK , Groningen, Netherlands, [email protected], Mirjam Duer, Georg Still

We discuss the order of solutions in general conic programming which contains linear, semidefinite andcopositive problems as special cases. The convergence of certain solution methods depends on this order.As conic problems can be seen as a special case of linear semi-infinite programs (LSIP), we study therelation between the properties of LSIP and conic programming in detail. In particular, we formulate thecondition of first order solution in terms of our conic programs. We also discuss higher order solutions.

153

TC-01

4 - Cutting planes for completely positive conic problemsMirjam Duer, Mathematics, University of Trier, 54286, Trier, Germany, [email protected]

Many combinatorial problems as well as nonconvex quadratic problems can be reformulated as completelypositive problems, that is, linear problems over the cone of completely positive matrices. Unfortunately,this reformulation does not solve the difficulty of the problem because the completely positive cone isintractable. A tractable relaxation is lies in substituting it by the semidefinite cone. In this talk, we showhow we can construct cutting planes that will sharpen the relaxation.

This is joint work with Abraham Berman, Naomi Shaked-Monderer and Julia Sponsel.

Thursday, 13:00-13:45

� TC-01Thursday, 13:00-13:45Rotterdam Hall

Roundtable Discussion (see page 28/29)Stream: Plenary SessionsPanel sessionChair: Marc Salomon, University of Amsterdam, Netherlands, [email protected]

154

TD-10

Thursday, 14:00-14:45

� TD-01Thursday, 14:00-14:45Rotterdam Hall

Semi-plenary speaker: GOR Science PrizeStream: Semi-plenary SessionsSemi-plenary sessionChair: Stefan Nickel, Institute for Operations Research (IOR), Karlsruhe Institute of Technology(KIT), Kaiserstrasse 12, 76131, Karlsruhe, Germany, [email protected]

� TD-09Thursday, 14:00-14:45Diamond Room 1

Semi-plenary speaker: Kalyan TalluriStream: Semi-plenary SessionsSemi-plenary sessionChair: Robert Klein, Chair of Analytics & Optimization, University of Augsburg, Universitätsstr. 16,86135, Augsburg, Germany, [email protected]

1 - Research trends in dynamic pricing and revenue managementKalyan Talluri, University of Pompeu Fabra, Barcelona, Spain, [email protected]


� TD-10Thursday, 14:00-14:45Diamond Room 2

Semi-plenary speaker: Andrea LodiStream: Semi-plenary SessionsSemi-plenary sessionChair: Erwin Pesch, FB 5, University of Siegen, Hoelderlinstr. 3, 57068, Siegen, Germany,[email protected]

1 - Performance Variability in Mixed-Integer ProgrammingAndrea Lodi, D.E.I.S., University of Bologna, Viale Risorgimento 2, 40136, Bologna, Italy,[email protected]


155

TD-12

� TD-12Thursday, 14:00-14:45Van Oldenbarnevelt Room

Semi-plenary speaker: Leo KroonStream: Semi-plenary SessionsSemi-plenary sessionChair: Leena Suhl, Int. Graduate School of Dynamic Intelligent Systems, University of Paderborn,Decision Support & OR Lab, Warburger Str. 100, 33098, Paderborn, Germany, [email protected]

1 - Public Transport: planning and real-time controlLeo Kroon, Rotterdam School of Management, Erasmus University Rotterdam, P.O. Box 1738,3000 DR, Rotterdam, Netherlands, [email protected]


� TD-13Thursday, 14:00-14:45Mees Auditorium

Semi-plenary speaker: Dick den HertogStream: Semi-plenary SessionsSemi-plenary sessionChair: Dennis Huisman, Econometric Institute, Erasmus University, Rotterdam, Netherlands,[email protected]

1 - Practical robust optimization: an introductionDick den Hertog, Econometrics and Operations Research, Tilburg University, Warandelaan 2P.O. Box 90153, 5000 LE, Tilburg, Netherlands, [email protected]


156

TE-02

Thursday, 15:15-16:45

� TE-02Thursday, 15:15-16:45Veder Room

Dynamic PricingStream: Revenue Management and PricingInvited sessionChair: Arnoud den Boer, Stochastics, TU Eindhoven, PO Box 513, 5600 MB, Eindhoven, Netherlands,[email protected]

1 - A dynamic customer-centric pricing approach for the product line pricing problemwith non-unit demandMichael Neugebauer, Chair of Analytics & Optimization, University of Augsburg,Universitätsstraße 16, 86159, AUGSBURG, Germany,[email protected]

In this paper, we address a service provider’s product line pricing problem for substitutable products. Incontrast to the prevailing approach in the product line pricing literature we depart from the assumption thatprices have to be static throughout the selling period and allow them to be adapted several times duringthe selling season such that the provider’s total profit is maximised. We use the standard assumption thatthe market is composed of different customer segments of various sizes. The customers have preferencerankings for all products and purchase the available product that ranks highest in their preference lists.The customers within a segment are assumed to be homogenous with respect to their preference lists butmay differ with respect to the amount of their demand. We introduce a linear mixed-integer programmingformulation for this problem which is solved by means of IBM ILOG CPLEX. We conduct several compu-tational experiments and present some preliminary results. Potential practical applications for this modelare, for example, television advertisement pricing and attended home delivery in E-fulfillment.

2 - Optimality conditions in Cartel Equilibrium Optimization ModelMichal Fendek, Department of Operations Research and Econometrics, University ofEconomics Bratislava, Dolnozemská cesta 1/b, 85235, Bratislava, Slovakia, Slovakia,[email protected], Eleonora Fendekova

A cartel is a specific case of oligopoly with an unspecified number of buyers but only a small numberof sellers. It should be noted that each action of a particular firm in an oligopoly affects the behaviorof other firms on the market. Price lowering of one firm will likely decrease a market share of otherfirms on the production of a sector. In other words, responses of the competitors in oligopoly may have asignificant effect on a result of managerial decision making on an oligopoly market. It is therefore clearthat a decision making about optimal price and supply in an oligopoly is far more complicated than inother market structures. Methodological problems of formulating oligopoly models rise from the greatdiversity of ways in which firms can interact and conclude agreements on the distribution, market sharesand market prices. Simply said, there is no general model of oligopoly. However there are models whichanalyze oligopoly decision making based on certain assumptions about interactions between firms. Inthis paper we examine a general scheme of oligopoly equilibrium model on which we present specificaspects of mutual relations between oligopoly subjects in a process of setting an equilibrium price andsupply of oligopoly. In the paper we show how a method of setting an optimal price and supply of a cartellimits the laws of competition. We study the properties of a cartel profit optimization problem consideringvarious marginal costs of homogeneous production of the cartel subjects and point out some interesting,economically interpretable implications of Kuhn-Tucker optimality conditions in a cartel subject’s optimaldecision making problem in a context of its behavior on the market of imperfect competition.

157

TE-03

3 - Risk-averse Dynamic PricingJochen Gönsch, Department of Analytics & Optimization, University of Augsburg,Universitätsstraße 16, D-86159, Augsburg, Germany, [email protected],Michael Hassler, Robert Klein

Considering a risk-averse decision maker is particularly appropriate when regarding rare events or actingas part of a volatile and fast changing environment. This talk presents an approach for risk-averse dynamicpricing based on stochastic dynamic programming where the Conditional Value-at-Risk (CVaR) servesas risk measure. Although the use of CVaR is intuitively appealing and it possesses desirable theoreticalproperties, its integration into dynamic programming is quite complicated. To the best of our knowledge,we present is the first dynamic pricing model using a time-consistent formulation of the CVaR suitablefor evaluation of multi-period revenue streams. Furthermore, on the basis of an illustrative numericalexample, we compare control policies resulting from using the expected revenue as target criterion withpolicies resulting from our approach.

4 - Dynamic pricing of low-priority users in wireless networksArnoud den Boer, Stochastics, TU Eindhoven, PO Box 513, 5600 MB, Eindhoven, Netherlands,[email protected]

We study a dynamic pricing problem arising in wireless networks.

In cognitive radio networks, primary users are licensed to use a certain amount of spectrum capacity ofa wireless communication system. In many applications, these primary users are not continuously usingtheir dedicated channel capacity. Since spectrum capacity typically is a scarce resource, network operatorsallow so-called secondary users to use the spectrum whenever the primary user is absent. However, if theprimary user returns when a secondary user is transmitting data, a costs is incurred due to degraded serviceto the primary user.

This leads to a dynamic pricing problem for access of secondary users to the channel, where the chargedprice dynamically varies with the expected time that the primary user will remain absent. We show struc-tural properties of the optimal dynamic pricing policy, and discuss settings where model-primitives are notknown in advance but may be learned on-the-fly.

� TE-03Thursday, 15:15-16:45Leeuwen Room 1

Modeling Gas Transportation: Yesterday, Today, and TomorrowStream: Energy and EnvironmentInvited sessionChair: Claudia Stangl, Mathematics, University of Duisburg-Essen, Butlerstrasse 10, 47058, duisburg,Germany, [email protected]

1 - Network optimization at germany’s leading gas transmission companyUwe Gotzes, Open Grid Europe GmbH, Kallenbergstrasse 5, 45141, Essen, Germany,[email protected]

Open Grid Europe is germany’s leading gas network operator. Its pipeline system covers some 12,000 km.The liberalisation of gasmarkets led to far reaching changes over recent years and network and capacityplanning became more and more complex tasks. Amongst others, Open Grid Europe’s network planningdepartment is responsible for the determination of most favorable network expansions to increase the net-work’s transportation capacity if needed. Furthermore, the amount of so called freely allocable capacitieshas to be maximised and when a customer requests freely allocable capacity at some network point, theplanners have to decide whether this capacity is available or not. If such a capacity request is denied,congent justifications have to be given. In the past, from a practitioners perspective, all those tasks boiled

158

TE-04

down to the generation and simulation of meaningful scenarios and to draw the correct conclusions fromresults obtained by simulation. On the other hand, regulation requires that best available technology isapplied for capacity planning and the like. From a mathematician’s point of view, the aforementionedtasks lead to extremly complicated very large-scale mixed-integer nonlinear stochastic programs. There-fore Open Grid Europe initiated research cooperations with research institutes and universities to makeuse of modern mathamatics in its daily workflow aiming at less manual effort, better quality and leadingthe way in defining "best available technology" in gas transportation. We report on our experiences withresearch cooperations and on its benefits for the company.

2 - Optimal Looping of Pipelines in Gas NetworksRalf Lenz, Optimization, Zuse Institute Berlin, Takustraße 7, 14195, Berlin, Germany,[email protected], Robert Schwarz

Given a gas network, gas transmission companies are faced with the task to transport specified amounts offlow from sources to sinks while accounting for physical and operational restrictions. In order to enablefeasible operations, gas transmission companies frequently need to extend the network. In practice, acommon method is to build new pipes in parallel to existing ones, called looping. Compared to the originalpipe, a loop provides the possibility of transporting more gas and leads to a reduction of the pressure loss.The decisions to be taken comprise the selection of the pipes to be looped with continuous loop lengthsas well as the appropriate choice of loop diameters out of a discrete set. Due to the nonlinear pressureloss and flow relationship, these diameters do not explicitly correspond to flow capacities. The problem offinding cost-minimal looping extensions that enable feasible gas operations results in the formulation of anMINLP model. Since we are able to determine best looping diameters a priori, we can efficiently reduceits problem size. In this talk, we present the structure of the resulting MINLP model and conclude with thedemonstration of computational results.

3 - Checking Feasibility in Stationary Models of Gas Transportation NetworksClaudia Stangl, Mathematics, University of Duisburg-Essen, Butlerstrasse 10, 47058, duisburg,Germany, [email protected], Rüdiger Schultz

Checking the feasibility of transportation requests belongs to the key tasks in gas pipeline operation. In itsmost basic form, the problem is to decide whether a certain quantity of gas can be sent through the networkfrom prescribed entries to prescribed exit points. In the stationary case, the physics of gas flow togetherwith technological and commercial side conditions lead to a pretty big (nonlinear, mixed-integer, finitedimensional) inequality system. The approach presented in this talk relies on transforming nonlinearitiesinto a more accessible form, reducing the problem dimension of the underlying NLP. This transformationapproach is embedded into a heuristic procedure for finding promising switching decisions.

4 - Some Remarks on Stochastic Programming in EnergyRüdiger Schultz, Mathematics, University of Duisburg-Essen, Lotharstr. 65, D-47048, Duisburg,Germany, [email protected]

The talk addresses some recent developments in stochastic optimization in energy. These involve riskaversion in the objective and the constraints of an optimization problem with uncertain data.

� TE-04Thursday, 15:15-16:45Leeuwen Room 2

Topics in Covered Bond ModelingStream: Financial Modeling, Banking and InsuranceInvited sessionChair: Ralf Werner, Institut für Mathematik, Universität Augsburg, Universitätsstr. 14, 86159,Augsburg, Germany, [email protected]

159

TE-04

1 - Towards stochastic modeling of German Covered BondsManuela Spangler, University of Augsburg, Germany, [email protected], Ralf Werner

Until today, no single German Covered Bond has ever defaulted. Before the financial crisis, GermanCovered Bonds were considered to be nearly risk-free, and their spreads to government bonds were of-ten interpreted as a pure liquidity premium. This has now changed as investors have, in general, becomemore sensitive regarding potential losses. To model such losses for German Covered Bonds, it is impor-tant to understand their mechanics and the cash flows arising under different scenarios. We show how aone-period model by Sünderhauf can be extended to a multi-period setting which incorporates the mostimportant features of German Covered Bonds. If time allows, we present simulation results from a firstimplementation.

2 - Extended JLT ModelYang Yan, Hochschule München, Germany, [email protected]

In 1997 Jarrow, Lando and Turnbull published their well-known Markov chain model, called JLT model,for the pricing of credit risk. The standard JLT model assumes a constant rating transition matrix, constantrecovery rate and deterministic market price of risk, which leads in turn to constant spreads provided thecredit rating does not change. However, on the other side, continuously changing markets require differentapproaches to appropriately consider credit risk. We therefore review extended JLT models, which cancapture the dynamics of credit spread. We will demonstrate how this framework can be used to price riskybonds and credit derivatives and is especially very well suited for modeling the asset side of the balancesheet of a German Pfandbriefbank.

3 - On the Valuation of German Covered Bonds - One-Period ApproachesMax Hughes, Hochschule München, Germany, [email protected], Manuela Spangler, RalfWerner

We examine events of default occurring due to asset-liability mismatches in a bank’s balance sheet, in par-ticular, of German Pfandbrief banks, concentrating on the mortgage-Pfandbrief. Since the recent financialcrises this has become an ever more greater concern to investors even in the segment of the, as consideredto be relatively safe, covered bond market. The valuation of the German Pfandbrief with efficient MonteCarlo methods is the key for applying advanced modeling approaches, especially in a risk neutral setup.The novelty of the underlying scheme is to use copulas for capturing dependencies between asset positions,adopted from established CDO modeling techniques. Consequently, the underlying model has substantialadvantages over Sünderhauf’s one period model concerning numerical efficiency and degrees-of-freedomof parameters while retaining full modeling flexibility.

4 - Preemption, leverage, and financing constraintsMichi Nishihara, Graduate School of Economics, Osaka University, 1-7, Machikaneyama,Toyonaka, 560-0043, Osaka, Japan, [email protected], Takashi Shibata

This paper investigates the interactions between preemptive competition and leverage. We find that thesecond mover always leaves the duopoly market before the first mover, although the leader may exit be-fore the follower’s entry. We also see the leverage effects of debt financing increasing firm values andaccelerating investment, even in the presence of preemptive competition. In addition to the case with opti-mal capital structure, we analyze a case with financing constraints that require firms to finance investmentcosts by debt. Notably, financing constraints can delay preemptive investment and improve firm values inpreemptive equilibrium. Indeed, the leader’s high leverage due to the financing constraints can lower thefirst-mover advantage and weaken preemptive competition. Especially with strong first-mover advantage,the financing constraint effects can dominate the leverage effects. These findings are almost consistentwith empirical evidence that high leverage leads to competitive disadvantage and mitigates product marketcompetition.

160

TE-05

� TE-05Thursday, 15:15-16:45Goudriaan Room 1

Home healthcare routing; ambulance planningStream: Health Care ManagementInvited sessionChair: Stefan Nickel, Institute for Operations Research (IOR), Karlsruhe Institute of Technology(KIT), Kaiserstrasse 12, 76131, Karlsruhe, Germany, [email protected]

1 - A comparative study of computational solution methods for the Home Health CareRouting and Scheduling Problem with Interdependent ServicesDorota Slawa Mankowska, Martin-Luther-University Halle-Wittenberg, Grosse Steinstrasse 73,06108, Halle, Germany, [email protected], Frank Meisel, ChristianBierwirth

Health Care Companies offer a broad range of on-site services to persons in need. For this purpose, theyemploy individually qualified staff. The staff members are typically equipped with cars, enabling them tomove among patients’ homes and the central office. The patients require certain types of services, whichmust be performed by suitably qualified staff members, preferably within a given time window. Possibleinterdependencies among the services can occur, if, for example, a temporal separation of two servicesis needed as is required if drugs have to be administered a certain time before providing a meal. Otherservices like handling a disabled person may require two staff members at a patient’s home at the sametime. The planning problem, also known as the Home Health Care Routing and Scheduling Problem withInterdependent Services, is to route staff members and to schedule service operations including interde-pendent services according to the individual service requirements of a given set of patients. In this talk, wepresent a comparative study of computational methods for solving this problem.

2 - Data driven ambulance optimization considering dynamic and economic aspectsDirk Degel, Fac. of Management and Economics, Ruhr-University Bochum, 44780, Bochum,Germany, [email protected], Lara Wiesche, Brigitte Werners

Providing high quality emergency medical services (EMS) and ensuring accessibility to these servicesfor the public is a key task for health care systems. Given a limited budget the available resources, e.g.ambulances, have to be used economically in order to ensure a high quality coverage. Emergency vehicleshave to be positioned and repositioned such that emergencies can be reached within a legal time frame.

Empirical studies have shown temporal and spatial variations of emergency demand as well as variations oftravel times during a day. The numbers of emergency calls with respect to the 24 hours differ significantlybetween night and day and show peaks especially during rush hours. Existing models do not considertime-dependency of important model parameters as demand and travel times for EMS-vehicles. This leadsto inefficient use of resources especially during night periods. We provide a data driven model consideringtime and spatial dependent degrees of coverage. This allows a simultaneous optimization of empiricalrequired coverage and minimizing the number of used ambulances, respectively costs. Therefore utilizationand quality criteria are to be implemented.

A mixed-integer linear program is formulated using time periods in order to model time-dependent demandand time-dependent travel times. On the basis of a simulation tool, it is shown on large empirical datarecords that the presented dynamic model outperforms existing static models with respect to coverage andutilization of resources.

3 - Ambulance Planning with and without Region BordersMelanie Reuter, Institute for Operations Research, Karlsruhe Institute of Technology (KIT),Germany, [email protected], Theresia van Essen, Stefan Nickel, Johann Hurink

161

TE-06

Border regions are still a major challenge for Emergency Medical Service (EMS) systems today. Whilein border regions between countries assuring a certain level of support in case of urgent need is desired,within countries efficient planning of the services is crucial. It often happens that at a border betweentwo different federal states of a country there are ambulance stations on both sides even though one atthe border could serve both sides simultaneously. This talk addresses the research gap on the influenceof having borders within a country when determining the locations and numbers of ambulances and ofambulance bases, neglecting of course possible legal regulations to determine the general potential. Inaddition, we divide the planning problem into a strategic and a tactical level and compare the impact onsolutions solving the two problems simultaneously and separately, considering exact approaches as wellas heuristics. We present the results of a case study using data generated for the Netherlands and solvingdifferent scenarios concerning the number and the arrangement of states and regions.

� TE-06Thursday, 15:15-16:45Goudriaan Room 2

Public Transport and Bike SharingStream: Traffic and TransportationInvited sessionChair: Dirk Christian Mattfeld, Business Information Systems, Universitaet Braunschweig, DecisionSupport Group, Mühlenpfordtstraße 23, 38106, Braunschweig, Germany, [email protected]

1 - Cyclic Crew SchedulingMarcel van Kooten Niekerk, Computer Science, Universiteit Utrecht, Princetonplein 5 K506,3584CC, Utrecht, Utrecht, Netherlands, [email protected], J.m. van den Akker, HanHoogeveen

The first step of building a public transport timetable is usually the creation of a timetable for one or morehours, which is repeated during the day in order to get a regular timetable which is easy to remember.When designing a timetable pattern, we should already take the cost and quality into account. This costconsist of vehicle cost and crew cost, which both consist of a fixed cost per vehicle or employee and avariable cost per kilometer or hour. The quality of a timetable and timetable pattern is determined by thetravel times and reliability of this travel times of the passengers. In literature, we see that the vehicle costand quality of a timetable pattern are investigated thoroughly, but we cannot find any work on crew cost oftimetable patterns. We introduce the Cyclic Duty, this is a crew duty with cyclic nature, it repeats every nhours. This Cyclic Duty has infinite length, but has meal breaks so that when we cut out a part of it, it willgive a valid duty. We developed an algorithm to determine the optimal set of Cyclic Duties and hence theminimal cost. Computational experiments show that the algorithm is capable of determining the crew costof real-life timetable patterns in tens of seconds

2 - Planning for robust and cost-efficient operations in public transport: Integratedvehicle and crew scheduling with time windows for scheduled tripsBastian Amberg, Information Systems, Freie Universitaet Berlin, Garystr. 21, 14195, Berlin,Germany, [email protected], Boris Amberg, Natalia Kliewer

This contribution deals with robust and cost-efficient vehicle and crew scheduling in public transport. Wepresent an integrated vehicle and crew scheduling approach with the possibility to shift scheduled tripswithin defined time windows. In addition to minimizing vehicle and crew costs the approach aims atbuilding robust schedules by minimizing the expected overall propagation of delays through the transportnetwork. Two concepts are combined: The consideration of mutual dependencies between vehicles anddrivers by integrated scheduling allows inserting buffer times at appropriate positions into the vehicle andcrew schedules to minimize delay propagation. Further, the possibility to shift scheduled trips within smalltime windows offers additional degrees of freedom to distribute buffer times into the vehicle and crew

162

TE-07

schedules without increasing costs. The integrated problem with time windows is solved with columngeneration in combination with Lagrangian relaxation. We test our approach on real-world instances andevaluate the results with regard to planned costs and delay propagation.

3 - Optimization approaches to the resource allocation in bike sharing systemsDirk Christian Mattfeld, Business Information Systems, Universitaet Braunschweig, DecisionSupport Group, Mühlenpfordtstraße 23, 38106, Braunschweig, Germany, [email protected],Patrick Vogel

Bike sharing systems have recently enabled public but likewise individual means of mobility at unattendedstations. Imbalances in the distribution of bikes affect the service quality, i.e., the provision of bikes andfree bike racks. These imbalances are caused by one-way use and spatio-temporal variation of bike rentalsfollowing typical traffic flows in the course of day. In order to ensure a certain service quality withinsystem operation, imbalances can be handled by means of bike allocation and relocation. The initialallocation of bikes at the beginning of the day compensates varying bike demand in the course of day.Relocation of bikes from rather full to rather empty stations maintains a certain service quality. Decisionsare interdependent: Reasonable allocation of bikes may reduce relocation effort, whereas high relocationeffort may compensate insufficient bike allocation. We propose an integrated approach ensuring a certainservice quality. The optimization model determines the optimal initial bike allocation at stations whileminimizing the expected costs of relocation. In particular, we discuss three approaches to the modelingof relocations. The deterministic linear program neglects restrictions on the frequency and volume ofrelocation, but is computationally fast. The deterministic mixed-integer program anticipates lot sizes bypooling of bikes, but is computationally intractable for big instances. On the contrary to the deterministicapproaches, the stochastic linear program yields a robust solution taking different demand scenarios intoaccount. The approaches are exemplified based on operational data from Vienna’s BSS "Citybike Wien’.

� TE-07Thursday, 15:15-16:45Antwerp Room

Scheduling Applications IIIStream: Scheduling and Project ManagementInvited sessionChair: Corinna Kaufmann, Mathematics, TU Kaiserslautern, Paul-Ehrlich-Straße 14, 67663,Kaiserslautern, Germany, [email protected]

1 - Quality-Oriented Scheduling of Projects with a Flexible Project StructureCarolin Kellenbrink, Institut für Produktionswirtschaft, Universität Hannover, KönigswortherPlatz 1, 30167, Hannover, Germany, [email protected]

In projects with a flexible project structure, the activities that have to be scheduled and implemented arenot completely known beforehand. Instead, scheduling the project includes the decision whether to carryout particular activities at all. In such projects, the project structure is therefore ex ante partially unknown.

In application fields such as the regeneration of complex capital goods, alternative project structures oftengo along with varying quality characteristics or functional features. For this reason, the different projectstructures typically affect not only the capacity consumption but also the revenues of the project. This canhave a major effect on the eventually chosen project structure and schedule. Hence, when planning flexibleprojects, the quality of the project should also be considered.

In this talk, the established resource-constrained project scheduling problem (RCPSP) is hence extendedby a highly general model-endogenous decision on the flexible project structure. Thereby, the qualityaspect of the project’s outcome is modeled explicitly. In addition, a genetic algorithm to solve the modelis presented.

163

TE-08

2 - Work-in-process in multi-criteria scheduling: Some resultsNatalia Prischepov, Industrial Engineering, Chair of Business Administration and OperationsManagement, University of Duisburg-Essen, University of Duisburg-Essen, Bismarckstraße 90,47057, Duisburg, Germany, [email protected]

Scheduling, as a decision-making process, plays an important role in most manufacturing systems and inservice organizations. It deals with the allocation of resources to tasks over given time periods and its goalis to optimize one or more objectives. Therefore, it is part of the manufacturing planning hierarchy. Shopfloor control, in contrast, controls short-term job processing after job release while also sequencing jobson the shop floor. Several researches propose that scheduling criteria should have a reflection of shop floorcontrol performance. However, for most of the settings such relationships cannot be theoretically proved.Therefore, it cannot be assured that schedules resulting from scheduling optimization are optimal or evengood if evaluated based on shop floor control criteria. In this presentation, we analyse a linkage betweenthe scheduling criterion quotient of flowtime and makespan, and shop floor performance measure WIP asoutcome. Subject to some rather moderate assumptions, this linkage can be interpreted in terms of Little’sLaw in a manufacturing scheduling context. However, WIP as quotient of flowtime and makespan is al-most never regarded explicitly, although the minimization of capital lockup is a highly relevant shop floorobjective in practice. Therefore, the analysis of WIP in manufacturing scheduling makes sense from botha scientific and from an application point of view. Additionally, we present some approaches for identi-fying bounds for WIP in manufacturing scheduling context, give an intuitive interpretation for modeling aproblem and show some first results of experimental analysis for small flowshop scheduling instances.

3 - Application of Scheduling Theory to the Bus Evacuation ProblemCorinna Kaufmann, Mathematics, TU Kaiserslautern, Paul-Ehrlich-Straße 14, 67663,Kaiserslautern, Germany, [email protected]

We present a scheduling problem which can be used for the bus evacuation problem. In this problem weare given a fixed number of evacuees waiting in a set of gathering points to be evacuated to a fixed set ofemergency shelters. Each shelter has a maximum capacity of evacuees which must not be exceeded. Thetransportation of evacuees is done by a set of buses traveling along shortest paths from gathering pointsto shelters and back. The goal is to find an allocation of evacuees to buses and for each bus a tour thatminimizes the maximum evacuation time. We model the uncapacitated bus evacuation problem, i.e., theproblem where shelters have unlimited capacity, as a Pm | s(i,j) | C(max) scheduling problem. That means,we model the bus evacuation problem by scheduling n jobs (= evacuees) on m parallel identical machines (=buses) where jobs have sequence-dependent setup times. All jobs have to be processed without preemption,are ready for processing at time zero and independent of each other, such that there are no precedenceconstraints. Although this problem has been extensively studied in the literature, existing IP formulationscan only solve efficiently by commercial IP solvers very small instances. Since research has concentratedon finding good heuristics, there are no other exact methods for this scheduling problem. Therefore,a customized branch-and-bound method is developed for the scheduling version of the bus evacuationproblem and compared to solutions from commercial solvers and heuristics. To include capacities onshelters the branch-and-bound method as well as the existing IP formulations and heuristics are modifiedand solutions are again compared.

� TE-08Thursday, 15:15-16:45Penn Room 1

Resource Constrained SchedulingStream: Scheduling and Project ManagementInvited sessionChair: Roland Braune, Institute for Production and Logistics Management, Johannes KeplerUniversity, Altenbergerstrasse 69, 4020, Linz, Austria, [email protected]

164

TE-08

1 - A hybrid CP-SAT approach for the exact solution of the multi-mode resource-constrained project scheduling problemAlexander Schnell, Dept of Management, University of Vienna, Austria,[email protected], Richard Hartl

The problem of assigning starting times to a number of jobs subject to resource and precedence constraintsis called the resource-constrained project scheduling problem (RCPSP). In this talk, we present a new ex-act algorithm for the multi-mode RCPSP (MMRCPSP). The MMRCPSP has a mode-assignment and ascheduling part.The mode-assignment part guarantees that the capacities of the non-renewable resourcesare not exceeded and the scheduling part integrates precedence constraints and resource constraints forthe renewable resources. Our exact algorithm aims at finding an exact solution of this problem whichminimises the project makespan. For the single-mode RCPSP (SMRCPSP), the most successful exactalgorithms combine Branch and Bound-based optimisation search with principles from Constraint Pro-gramming (CP) and Boolean Satisfiability Solving (SAT). Our exact approach generalises the CP and SATtechniques for the SMRCPSP to the MMRCPSP. To realise our exact approach, we implemented a newconstraint handler for the solver SCIP which captures the above generalisations. Our approach is evaluatedon instances of the MMRCPSP with 20, 30, 50 and 100 jobs. We outperform the state-of-the-art exactapproach from the literature on the instances with 30 jobs. Moreover, to our knowledge, we are the firstto exactly solve the instances with 50 and 100 jobs. Furthermore, we are able to improve the best-knownheuristic solutions for 61 and 90 instances with 50 and 100 jobs. Finally, we conclude that our approachis easily usable via SCIP and can be easily applied to generalisations of the MMRCPSP containing moregeneral linear objective functions and precedence constraints.

2 - Solving the resource-constrained multi project scheduling problem with flexibleresource usageTorben Schramme, DS&OR Lab Paderborn, University of Paderborn, Warburger Straße 100,33098, Paderborn, Germany, [email protected], Leena Suhl

The resource-constrained project scheduling problem (RCPSP) has been discussed in literature for decades,although there is still a gap between the needs of practical project scheduling on the one side and availableproject scheduling methods on the other side. Especially methods dealing with flexible resource allocationfor activities together with multiple projects in one plan are still rare and not practicable.

In our research we developed a planning model which incorporates these features and makes further as-pects possible. Due to the extremely augmented decision space compared to the standard RCPSP, solvinginstances of that model needs new solution methods which can deal with the new aspects and producesolutions in reasonable computing time.

We will talk about the special characteristics of that of that problem and present a hybrid approach consist-ing of a meta-heuristic for determining activity times and durations and an embedded LP for calculatingthe resource distribution. At the end we will provide computational results to show the effectiveness ofthat approach.

3 - Scheduling on Multi-Capacitated Resources with Flexibility and Variable Task Pro-filesRoland Braune, Institute for Production and Logistics Management, Johannes Kepler University,Altenbergerstrasse 69, 4020, Linz, Austria, [email protected], Karl Doerner

We consider a scheduling problem originating from polyolefine industries. Test specimens of final productshave to pass a series of tests in a chemical laboratory. The technological sequence of test activities or tasksis described by precedence constraints forming an out-tree structure. Each task may require multipledifferent resources to be executed, where some of these requirements are subject to flexibility. Resourcecapacities are considered in an aggregate manner, based on time "buckets" corresponding to time intervalsranging between two days and a week. Consequently, each resource may process more than one task at atime. Although the total amount of work is specified for each task, its duration is non-constant and dependson the amount of capacity allocated during processing. A task’s capacity allocation profile may thus varyover time as long as it remains within a predefined range.

Despite some similarities to scheduling models from the areas of multi-capacitated or cumulative job shopscheduling, energy scheduling and multi-mode resource-constrained project scheduling, to the best of au-thors’ knowledge, the above scenario in its entirety has not yet been discussed in literature. Therefore we

165

TE-09

first present a mixed integer programming formulation for the scheduling problem at hand under the aspectof minimizing maximum completion time. We then propose several different lower bounding schemes,some of which are based on relaxations of the given MIP formulation. Due to the NP-hardness of theproblem and in view of large-scale real-world instances, we further develop and analyze local search basedmeta-heuristics, partially relying on constraint propagation techniques such as (extended) energetic rea-soning.

� TE-09Thursday, 15:15-16:45Diamond Room 1

Production and Inventory controlStream: Production and Operations ManagementInvited sessionChair: Jan Necil, Information Process Engineering, FZI Research Center for Information Technology,Haid-und-Neu-Straße 10, 76131, Karlsruhe, Baden-Württemberg, Germany, [email protected]

1 - Optimizing the production planning at a sawmill with linear programming and sim-ple heuristicsMaria Anna Huka, Department of Economics and Social Science, Institute of Production andLogistics, University of Natural Resources and Life Sciences, Feistmantelstrasse 4, 1180,Vienna, Vienna, Austria, [email protected], Manfred Gronalt

The European sawmilling market is faced with various competitors who provide timber for the regionaland international market. Moreover, wood as raw material is more in demand than it used to be due tothe problem of summed up excess capacity on the European market. As a result of the competition forthe resource it is essential to optimize the timber production and gain an advantage over competitors.Different approaches for optimizing the efficient resource scheduling have been developed and presentedin various papers. In this work different methods to optimize the multi-period sawmill production planningproblem are investigated, while optimally using the allocated resources. We compare different modellingapproaches and consider both a profit margin and cost oriented objective function. In doing so both decisionmaking models are formulated as mixed integer programs where not only the net revenue but also thevariable costs of the production, inventory costs for raw material and products, the purchasing price forraw material and the backlog costs for products are considered. In addition some special assumptions(e.g. minimum production values) have to be made to make the model applicable in practice. The resultsof a multi-period planning problem are again compared to the solutions of simple heuristics applied byplant managers. Due to computational reasons and convenience of handling, heuristics are important forthe optimization in real life. We develop new heuristics and show their performance compared to anoptimization based approach.

2 - Determining optimal discount policies for a supplier in B2B relationshipsViktoryia Buhayenko, Business and Social Sciences, Aarhus University, Tåsingegade 3, 1, 8000, Aarhus C, Denmark, [email protected], Dirk Van Eikenhorst

This research studies which discounts a supplier needs to offer to give incentive to his customers to changetheir order patterns in a way that minimizes the supplier’s total cost. The problem of when and how muchdiscount to offer is not so often addressed in the literature. This approach is very different from the yieldmanagement in the fact that if price changes in one period it will affect the demand in other periods. Inthis research a limited number of large customers is considered. Given the cost and demand informationof each customer, the response to a discount can be computed. The supplier can use this to calculate whatdiscounts to offer in order to regulate when the customers demand is ordered so it matches the producer’sproduction schedule. Savings for the producer arise from reduction of setup and inventory cost. A heuristicsolution is given by separating the problem when orders should be placed and how much discount should

166

TE-10

be offered to achieve this. First the problem for each customer is solved, giving their cost. The suppliers’problem is now to minimize its cost, plus discounts offered to the customers. The discounts are a lump sumthat would compensate the customer for ordering at the periods requested. The discount offered in periodswith production can now be calculated to achieve the wanted order pattern. This approach can be usefulin situations when the buyer represents a retailer with his own customers which is unwilling to introducediscounts to them. Prices of the retailer can be fixed by the government as well. In this case despite gettingitems for lower price from his supplier the buyer has no incentive to increase his total demand.

� TE-10Thursday, 15:15-16:45Diamond Room 2

Closed-loop supply chainsStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Irena Okhrin, Juniorprofessur in Information & Operations Management, European UniversityViadrina, Grosse Scharrnstrasse 59, 15230, Frankfurt (Oder), Germany, [email protected]

1 - Optimal control of a hybrid production system for customized products with com-bined manufacturing and remanufacturingLeena Steinke, Institute for Operations Research and Information Systems, Hamburg Universityof Technology (TUHH), Schwarzenbergstraße 95 D, 21073 , Hamburg, Germany,[email protected], Kathrin Fischer

Often, companies can achieve a competitive advantage by customizing their products and by respondingquickly to changing customer needs. Consequently, companies tend to design their products in a modularway, using commonalities between components to delay product differentiation. Moreover, today manycompanies interact more intensely with their customers, and they also focus on more local productionagain. In this context, the remanufacturing of returned products can be an attractive sourcing option.Locally producing companies with a high level of costumer participation have more information about thereturned product, especially about its condition. Moreover, the modular design of the product facilitatesthe disassembly and the re-integration of remanufactured items into the production system.

In order to control inventories and to plan production for such systems, multistage models are required.Hence, a discrete Markov decision modeling approach for an open tandem system with two stages, i.e.component production and final assembly, is proposed in this work. It consists of single inflexible ma-chines at both stages, with a buffer at the first stage and component inventory holding at the second stage.Remanufacturing and manufacturing are considered as sources for the second stage, on which the machineworks demand-driven, thus starting to work only when demand arrives and if enough components are instock. The objective of the approach is to find a strategy that coordinates the input of the system suchthat expected total costs are minimized. This strategy determines whether to accept an arriving remanufac-turable product, and when to restock component inventory by (re-)manufacturing.

2 - Collection of Recyclable Materials with Unknown Filling RateMaria Elbek, Economics and Business, Aarhus University, Fuglesangs Allé 4, 8210, Aarhus V,Denmark, [email protected], Sanne Wøhlk

We consider an approach for scheduling the multi-period collection of recyclable materials. Citizens candeposit glass and paper for reuse in cubes located at several collection points. To establish a high servicelevel for the citizens, the cubes must not be over-filled. A vehicle carrying two containers is used to emptythe cubes. Glass and paper are therefore collected jointly, but must be handled separately, due to differentprocedures for reuse of the materials. Once the containers are filled, they are transported to the finaldestinations where the collected materials will be processed for reuse. This problem has many similaritiesto the Inventory Routing Problem (IRP) and waste collection problems with node-routing. However, the

167

TE-11

solution methods for these problems cannot tackle multi-period routing problems with combined collectionin multi-compartment vehicles. In addition, our problem is complicated by the stochastic accretion rateof material in the cubes. We examine how scheduling of emptying cubes and related transportation tofinal destinations should be done in order to minimize the operation cost while ensuring that the cubes arenot over-filled and capacity constraints of the containers on the vehicle are not exceeded. We develop aheuristic for solving the multi-period collection problem within a rolling horizon framework. We apply aconstruction heuristic once and reoptimize the solution every subsequent period. We estimate which cubesthat need to be emptied and insert them in routes in the forthcoming planning period based on an objectivefunction value. We investigate if an inclusion of a future cost in the objective function could lead to betterplanning and minimization in the total cost.

3 - Dynamic lot sizing with minimum order quantity and remanufacturingIrena Okhrin, Juniorprofessur in Information & Operations Management, European UniversityViadrina, Grosse Scharrnstrasse 59, 15230, Frankfurt (Oder), Germany,[email protected], Knut Richter

This paper explores a single-item lot sizing problem with minimum order quantity (MOQ) and remanufac-turing option for both capacitated and uncapacitated cases. Incorporating remanufacturing option makesthe problem highly relevant for practice, as many companies are integrating returns back into the pro-duction process being motivated by increasing environmental awareness as well as by social and economicpressure. The minimum order constraint, on the other hand, plays the role of minor set-up cost and specifiesthat in every period one can produce either nothing or at least as much as MOQ. For the problem at handwe develop the mixed-integer dynamic programming model and elaborate on the solution algorithm basedon the dynamic programming principle. The very important property of manufacturing-remanufacturingsystems was proven by Teunter et al. (Dynamic lot sizing with product returns and remanufacturing, IJProd Research 44, 4377—4400, 2006) for the uncapacitated problem with joint set-up cost and withoutMOQ constraint. According to the property, the remanufacturing is a preferred option and one should man-ufacture only in the case, when it is not possible to efficiently satisfy the demand by only remanufacturing.We prove this result also for the capacitated lot sizing problem with the minimum order quantity restriction.By making use of this property, we solve the problem by disassembling it into two parts and consideringthem separately. In the first step, we find the aggregated manufacturing-remanufacturing values for allperiods of the planning horizon with the help of published algorithms. In the second step, we disseminatethose values into separate remanufacturing and manufacturing lots and calculate those numbers explicitly.

� TE-11Thursday, 15:15-16:45Penn Room 2

Real Life Applications in Supply Chain ManagementStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Bruce Hartman, College of Business, University of St Francis, Intermodal TransportationInstitute, 500 Wilcox St, 60435, Joliet, Illinois, United States, [email protected]

1 - An Interactive Approach to a Real-World Sales Territory Design ProblemMatthias Bender, Information Process Engineering, FZI Research Center for InformationTechnology, Haid-und-Neu-Str. 10-14, 76131, Karlsruhe, Germany, [email protected], WernerHeid, Jörg Kalcsics, Anne Meyer, Stefan Nickel

In this talk we deal with a real-world sales territory design problem identified in cooperation with PTVGroup, a commercial provider of logistics planning software. The problem consists of partitioning a setof customers into sales territories, each of which is served by precisely one salesman. Territories aresupposed to be geographically compact, contiguous and balanced with respect to a quantitative measure.

168

TE-11

The locations of the salesmen’s offices are given in advance and must be taken into account in order toachieve easily accessible territories.We use a classical IP formulation (Hess et al., 1965) for assigning customers to salesmen and enhance it bynew real-world requirements such as the consideration of salesmen’s skills, fixed customer assignments andgiven customer groups. Our computational results show that real-world problems with several thousandcustomers can be solved in reasonable running times using a general purpose solver in default settings.However, a drawback of this approach is that, from a human planner’s point of view, solutions may haveundesirable properties that significantly decrease decision-maker acceptance, e.g. territories overlappingeach other.Thus, we provide an overview of some solution properties that are undesirable in practice and propose anapproach capable of reducing their occurrence. The approach aims at locally improving an existing solutionin an interactive manner: A planner manually specifies the subarea to be improved; all customers containedin this subarea are then considered as candidates for reassignments and reoptimization is performed. Wepresent computational results for this approach and give an outlook on the automatic recognition of suchundesirable properties.

2 - Determining Product Frequencies and Produce-Up-To Levels in a Single FacilityMulti-Item Production System with Stochastic DemandsPhilipp Zeise, Lehrstuhl für Quantitative Planung, Universität Siegen, Hölderlinstr. 3, 57076 ,Siegen, NRW, Germany, [email protected] consider the production process of a pharmaceutical company. In general, such a production processcomprises three stages but we will focus on the packaging of final products which is the last stage. Essentialcharacteristics of this process are multiple products divided into product families, sequence-dependentsetup times and stationary stochastic demands. The main goal is to reduce finished product inventory costswhile a certain alpha service level has to be guaranteed.The production policy used by the company so far bases on a cyclic pattern with a frequency of one for eachproduct per cycle while the cycle length must be in a given interval. Using a sequence that minimizes totalsetup time each product variant is produced up to its’ predetermined produce-up-to level (gated policy). Inorder to keep the actual cycle length in a predetermined interval idle time may be inserted or productionmay be cut short. Simulating this scenario we find that an extremely high produce-up-to level for severalproducts is necessary or the service level of some products cannot be reached at all. We present someideas how this policy can be extended, i.e. determining product frequencies in combination with severalsequencing strategies to reduce total costs per unit of time. Furthermore, we give first insights (such as theoptimal cycle length and produce-up-to levels) yielded by conducting computational experiments based onpractical instances.

3 - Optimization of Sales & Operations Planning at Shell Chemicals EuropeThijs Van Dongen, ORTEC Consulting Group, ORTEC, 2700AB, Zoetermeer, Netherlands,[email protected], Dave van den Hurck, Nort Thijssen, Pascal BrakonierIn the chemical industry, planning and scheduling are labour-intensive, complex, rolling processes. Interde-pendent decisions have to be made around different stages within the supply chain (purchases, production,distribution, exchanges, storage levels and sales). In taking these decisions the overall enterprise marginneeds to be maximized across the global supply chain. To support making these decisions the chemicalssupply chain has been modeled using GMOS/NetSim, an AIMMS-based network optimization tool jointlydeveloped by Shell Global Solutions and ORTEC. Years of extensive collaboration with various customershave made GMOS/NetSim a proven tool for strategic supply chain studies. For this project a module wasdeveloped to calculate accumulated costs/margins throughout the supply chain. The outcomes are usedto do detailed margin analyses. The key challenge was to integrate the model into the monthly Sales &Operations Planning (S&OP) at Shell Chemicals Europe (SCE). Input data needs to be obtained from 15+people around the world from various fields of expertise on a regular basis, as market conditions constantlychange. Moreover, actual data is used for model validation purposes and margin analyses for past months.The key outcomes from the optimization are shared with the user community twice every month. The mainbenefit of this project is that we are able to establish a unified global base plan and a unified approach forfact-based decision making. The complex mathematical model behind this approach includes a great levelof detail reflecting reality in everyday SCE business. This improves both the quality and speed of businessdecisions at 3 months (S&OP) and multi-year (business plan) horizons across the global supply chain.

169

TE-12

4 - Chassis Pooling and Profit Allocation at the Chicago Dry Port ComplexBruce Hartman, College of Business, University of St Francis, Intermodal TransportationInstitute, 500 Wilcox St, 60435, Joliet, Illinois, United States, [email protected]

A shipping container requires a wheeled chassis for every over the road move. In the USA, approximately30To make chassis available and ensure compliance several ocean carriers have been involved in the develop-ment of third party chassis pools. This paper discusses the chassis pooling situation at the dry port complexin the Chicago area, which is a major hub for intermodal traffic transit, with over 3 million container liftsper year. While pooling increases profit, as in any inventory situation with transshipment, it is not clearhow to fairly share the profit. We present a model of profit sharing for the chassis inventory pooling prob-lem with transshipment, and prove that allocation by initial order quantity, and a pricing scheme based onthose units, is not stable– some members will have an incentive to leave the pool and form a smaller one oftheir own. Other pricing schemes also suffer from this defect. In fact, we can only identify one method thatoffers some promise for efficient and stable allocation. This result means that pools contain the seeds oftheir own destruction unless the profit sharing method is chosen carefully, and all clients understand howit works for them on average.

� TE-12Thursday, 15:15-16:45Van Oldenbarnevelt Room

WarehousingStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: René de Koster, Rotterdam School of Management, Erasmus University Rotterdam, RoomT10-56, PO Box 1738, 3000 DR, Rotterdam, Netherlands, [email protected]

1 - Class-based Storage with Space Sharing in Unit Load WarehousesXiaolong Guo, School of Management, University of Science and Technology of China, 96Jinzhai Road, 230026, Hefei, Anhui, China, [email protected], René de Koster, Yugang Yu

In this paper, we study the effect of item space sharing in class-based storage warehouses on unit-loadretrieval time. The warehouse is operated by a lift truck in a single-command mode and each operationbegins and ends at the depot. We first develop travel-time models for such a warehouse for two differentaisle configurations: parallel aisle and fishbone aisle layouts. We show that more storage classes do notnecessarily result in a shorter travel time. Although space within a class is shared among items, eachitem within a class needs a certain amount of space, which increases if fewer items share the class (i.e. ifwe distinguish more storage classes). Due to this effect, an optimal number of classes exist, minimizingthe unit load retrieval time. The results also show that, compared to a parallel aisle layout, the fishboneaisle layout brings shorter retrieval time but needs more storage space. For a given number of items inthe warehouse, our aim is to determine which aisle configuration should be chosen and how many classesshould be adopted to minimize annual operational cost. The optimal class boundaries can be found throughdynamic programming. Including cost of space, the optimal number of classes is smaller than when onlytravel-time would be taken into account. Furthermore, fishbone aisles are superior to parallel aisles forlarge-size warehouses or in case of low annual costs of storage space. The findings of this paper may helpwarehouse designers and managers in selecting a proper layout and product to storage class assignment.

2 - Analyzing alternate vehicle allocation strategies in mobile rack-based order-picksystemsRené de Koster, Rotterdam School of Management, Erasmus University Rotterdam, RoomT10-56, PO Box 1738, 3000 DR, Rotterdam, Netherlands, [email protected], Debjit Roy, ShobhitNigam

170

TE-13

Warehouses are increasingly employing automation technologies to reduce operational costs, increase cus-tomer satisfaction and improve operational efficiencies by managing processes and resources efficiently. Inparticular, the Kiva warehouse-management system creates a new paradigm for pick-pack and ship ware-houses that significantly improves worker productivity. The Kiva system uses movable storage shelves thatcan be lifted by small, autonomous robots. By bringing the item to the worker, productivity is increasedby a factor of two or more, while simultaneously improving accountability and flexibility. In this regard,closed queuing network models provide a rich framework for rapidly evaluating the performance of alter-nate warehouse configurations. We develop blocking protocols within the aisles and cross-aisles of a tierand analyse the system with dedicated as well as pooled vehicles. Using product-form approximations forclosed queuing networks, we evaluate the network and determine the performance measures of interest.

3 - Dual-load retrieval in puzzle-based storage systemsMasoud Mirzaei, Rotterdam School of Management, Erasmus University Rotterdam,Burgemeester Oudlaan 50, 3062 PA, Rotterdam, Netherlands, [email protected], René de KosterPuzzle-based storage systems are very compact storage systems, in which each unit load can be retrievedindependently, and which can solve problems caused by lack of space in some industries and areas. Thefirst unit-load retrieval time models for puzzle-based storage systems were developed for systems whichwork with a single unit-load empty space. Such systems have much in common with the solving thewell known 15-tile sliding puzzle in a minimum number of moves. In this paper we extend these initialmodels, by proposing a method to retrieve multiple unit loads in a minimum number of moves. In thecase two items need retrieval, the method uses a systematic coupling procedure to bring two items togetherefficiently, after which they are retrieved simultaneously. The presented method can reduce the retrievaltime with up to 30% in some cases compared to individual load retrieval. We additionally study retrievingmore than two items, using approximate analysis. Knowledge of the makespan for retrieving multipleitems can help to determine optimal system dimensions.

� TE-13Thursday, 15:15-16:45Mees Auditorium

Combinatorial OptimizationStream: Discrete and Combinatorial Optimization, Graphs and NetworksInvited sessionChair: Rosiane deFreitas, Institute of Computing, UFAM / UFRJ, Brazil, [email protected]

1 - An Integer Programming Approach to the Hospital/Residents Problem with TiesAugustine Kwanashie, School of Computing Science, University of Glasgow, UniversityAvenue, G12 8QQ, Glasgow, United Kingdom, [email protected], DavidManloveThe classical Hospitals/Residents problem (HR) models the assignment of junior doctors to hospitals basedon their preferences over one another. In an instance of this problem, a stable matching M is sought whichensures that no blocking pair can exist in which a resident r and hospital h can improve relative to Mby becoming assigned to each other. Such a situation is undesirable as it could naturally lead to r and hforming a private arrangement outside of the matching.The original HR model assumes that preference lists are strictly ordered. However in practice, this maybe an unreasonable assumption: an agent may find two or more agents equally acceptable, giving rise toties in its preference list. We thus obtain the Hospitals/Residents problem with Ties (HRT). In such aninstance, stable matchings may have different sizes and the problem of finding a maximum cardinalitystable matching is NP-hard.In this talk, we provide an Integer Programming (IP) model for HRT. We show that an exploitation of thestructure of the problem can be used to greatly reduce the size of the model it produces. We also discussseveral ways of improving optimisation performance of the IP solver. Finally we provide results obtainedfrom an empirical evaluation of the IP model based on real-world and randomly generated data.

171

TE-13

2 - Sequencing problems with uncertain parameters and the OWA criterionAdam Kasperski, Wroclaw University of Technology, Institute of Industrial Engineering andManagement, ul. Wybrzeze Wyspianskiego 27, 50-370, Wroclaw, Poland,[email protected], Pawel Zielinski

We consider a class of sequencing problems, in which we seek a feasible order of jobs to minimize agiven cost function. For each job some parameters, such as a processing time, a due date and a weightare specified. The cost function depends on job completion times and can be, for example, the weightedsum of completion times, the maximum weighted tardiness, or the weighted number of late jobs. In reallife the precise values of the parameters are often unknown before a solution is computed. To model thisuncertainty, we can introduce a scenario set containing a finite number of distinct vectors of the parameters,called scenarios. We can then apply the popular robust optimization framework to choose a solution. Inrobust optimization we seek a solution minimizing the cost in the worst case, i.e. under the worst scenariowhich may happen. We then typically apply the minimax criterion which is, however, often regardedas too pessimistic or conservative. To overcome this drawback we propose to use the ordered weightedaveraging aggregation operator (shortly OWA), introduced by Yager in 1988. The OWA operator containsthe maximum, minimum, average, median, and Hurwicz criteria as special cases. It is thus more flexibleand allows us to model different attitudes of decision makers towards the risk. Our aim is to show somecomplexity and approximation results for a class of sequencing problems with uncertain parameters andthe OWA criterion. We identify some special cases which are polynomially solvable and which are NP-hard. We also investigate the approximability of the NP-hard problems by showing that some of them areapproximable within some factor (constant or depending on the number of scenarios), while some otherare not at all approximable.

3 - Moving bins from conveyor belts onto pallets using FIFO queuesFrank Gurski, Computer Sciences, Heinrich Heine University, Universitätsstraße 1, Building25.13 Level/Room 02.35, 40225, Düsseldorf, Germany, [email protected],Jochen Rethmann, Egon Wanke

In delivery industry, bins have to be stacked up from conveyor belts onto pallets. Given k sequences oflabeled bins and a positive integer p. The goal is to stack up the bins by iteratively removing the first bin ofone of the k sequences and put it onto a pallet located at one of p stack-up places. Each of these pallets hasto contain bins of only one label, bins of different labels have to be placed on different pallets. After allbins of one label have been removed from the given sequences, the corresponding place becomes availablefor a pallet of bins of another label. In this work we show that the FIFO stack-up problem is NP-completein general, even if the sequences together contain at most 6 bins per pallet. The problem can be solved inpolynomial time, if the number k of given sequences is fixed, or if the number p of given stack-up placesis fixed.

4 - Constraint programming algorithms to determine the position of atoms in amoleculeClarice Santos, Universidade Federal do Amazonas, Manaus, Amazonas, Brazil,[email protected], Rosiane deFreitas

Given a simple graph with some arbitrary weighted edges, to represent a set of atoms and a subset oftheir distances in a molecule, in this work we discuss about algorithms for determing the 3D structureof it, solving the Molecular Distance Geometry Problem (MDGP), a NP-hard computational problem foran incomplete and arbitrary set of distance constraints. Continuos and discrete mathematical approachesto solve MDGP is revised, based on the analysis of two types of calculating sphere intersection: solvingsystems from interatomic Euclidean distance equations, or solving internal coordinate systems using ma-trix multiplication techniques. We adapted the Branch-and-Prune (BP) method considering four-spheresintersection, for these two ways of approaches. Computational experiments using PDB instances are per-formed.

172

TE-14

� TE-14Thursday, 15:15-16:45Tokyo Room

Vehicle Routing Applications and Network DesignStream: Traffic and TransportationInvited sessionChair: Rommert Dekker, Erasmus University Rotterdam, Econometric Institute, Erasmus UniversityRotterdam, P.O. Box 1738„ 3000 DR , Rotterdam, Netherlands, [email protected]

1 - On the modeling of recharging stops in context of vehicle routing problemsStefan Frank, Faculty of Traffic and Transportation Sciences, Institute for Logistics andAviation, Technical University of Dresden, Germany, [email protected], HenningPreis, Karl Nachtigall

Causing by regulations regarding to climate protection, battery electric vehicles (BEVs) offer great oppor-tunities in context of ecological compatibility of urban transport systems. Therefor, operating models incontext of vehicle routing are required. Because of the BEVs more restrictive driving range in comparisonto vehicles with an internal combustion engine (ICE) and due to the fact of a less-developed network ofservice stations, model formulations have to include the possibility of recharging at dedicated locations.So additional restrictions in formulations are needed to handle the maximum range depending on batterycapacity. Over the past few years there were published only a small number of articles addressed to includeenergy consumption, battery range and possible recharging stops in mixed-integer programming (MIP) for-mulations in the underlying practice relevant Vehicle Routing Problem with Time Windows (VRPTW). Sowe describe different MIP-formulations for an enhanced VRPTW, considering capacity restrictions con-cerning to freight and energy, customer time windows and including of the capability of charging stops.Effects of these formulations are shown for small-sized problems, while a column generation approach ispresented for more realistic large-sized problem instances.

2 - Fuel consumption costs of routing uncertaintyStephan Unger, Finance, University of Vienna, Brünnerstraße 72, 1210, Vienna, Vienna, Austria,[email protected], WIlliam Cheung

We solve a car driver’s routing decision problem under uncertainty in terms of fuel consumption costs.Suppose a car driver can estimate his fuel consumption for a given route beween A and B. We studythe optimal decision which route to take, given the possibility of transferring between A and B by usingdifferent routes where each route is characterized by stochastic uncertain fuel consumption due to unknowntraffic at the time of decision. We show that the cost of fuel consumption decreases significantly whentaking routes with uncertain knowledge about prevailing traffic. We also identify parameters under whichone would be indifferent in taking an alternative route or not.

3 - Service network design for an intermodal container network with flexible duedates/times and the possibility of using subcontracted transportBart Van Riessen, Econometric Institute, Erasmus University Rotterdam, Netherlands,[email protected], Rudy Negenborn, Rommert Dekker, Gabriel Lodewijks

An intermodal container transportation network is being developed between Rotterdam and several inlandterminals in North West Europe: the EUROPEAN GATEWAY SERVICES (EGS) network. This networkis developed and operated by the seaports of EUROPE CONTAINER TERMINALS (ECT). To use thisnetwork cost-efficiently, a centralized planning of the container transportation is required, to be operated bythe seaport. In this paper, a new mathematical model is proposed for the service network design. The modeluses a combination of a path-based formulation and a minimum flow network formulation. It introducestwo new features to the intermodal network-planning problem. Firstly, overdue deliveries are penalizedinstead of prohibited. Secondly, the model combines self-operated and subcontracted services. The servicenetwork design considers the network-planning problem at a tactical level: the optimal service schedulebetween the given network terminals is determined. The model considers self-operated or subcontracted

173

TE-15

barge and rail services as well as transport by truck. The model is used for the service network designof the EGS network. For this case, the benefit of using container transportation with multiple legs andintermediate transfers is studied. Also, a preliminary test of the influence of the new aspects of the modelis done. The preliminary results indicate that the proposed model is suitable for the service network designin modern intermodal container transport networks. Also, the results suggest that a combined businessmodel for the network transport and terminals is worth investigating further, as the transit costs can bereduced with lower transfer costs.

� TE-15Thursday, 15:15-16:45New York Room

Traveling Salesman Problem and Branch-and-BoundStream: Discrete and Combinatorial Optimization, Graphs and NetworksInvited sessionChair: Bruce Golden, Decision & Information Technologies, University of Maryland, 4339 VanMunching Hall, College Park, MD 20742, 20742, College Park, MD, United States,[email protected]

1 - White-box optimization from historical dataQing Chuan Ye, Erasmus University Rotterdam, Burg. Oudlaan 50, 3062 PA, Rotterdam, SelectU.S. States, Netherlands, [email protected], Sicco Verwer, Yingqian Zhang

It is challenging to construct a mathematical model describing the properties of a system. This holds inparticular for when the structure of the system cannot be fully determined from the hypotheses at hand. Insuch cases, machine learning techniques can be used to replace (parts of) a mathematical decision model.However, such models have so far only been used in a black-box fashion, e.g. as fitness functions orparameters. We propose a white-box optimization by mapping a learned regression tree model to a mixedinteger linear program that can be used for optimization. In a regression tree each leaf represents a valueof the dependent variable, whereas the path from root to the leaf illustrates the values of the independentvariables. Consequently, the learned models properties are visible as constraints to a mathematical problemsolver, that can then use sophisticated branching and cutting techniques on these constraints when findingsolutions, which are impossible in black-box optimization.

We illustrate our approach using a sequential auction design problem. The objective is to maximize the so-cial welfare by ordering the resources in sequential auctions. This social welfare depends on combinatorialpreferences and budget constraints of the bidders. Since these are unknown to the auctioneer, we proposeto learn them from historical data. We use an auction simulator and design several experiments to test theperformance of our novel method. Our experiments show that optimization based on historical data resultsin high social welfare. Furthermore, we compare our white-box approach with a black-box best first searchapproach and show its advantages over the black-box method.

2 - Computing an Upper Bound for the Longest Edge in an Optimal TSP-SolutionHans Achatz, University of Passau, D-94030, Passau, Germany, [email protected], PeterKleinschmidt

A solution of the traveling salesmen problem (TSP) with n nodes consists of n edges which might havedifferent lengths. In our approach we compute an upper bound u for the longest edge which could be inan optimal solution. This means that every edge longer than this bound cannot be in an optimal solution.The quantity u can be computed in polynomial time. We have applied our approach to different problemsof the TSP-LIB (library of sample instances for the TSP). Of course this does not mean that this boundsubstantially improves the fastest TSP-algorithms. However, the reduction of the number of edges mightbe useful for certain instances.

174

TE-17

3 - The Hierarchical Traveling Salesman Problem: Some Worst-Case ResultsBruce Golden, Decision & Information Technologies, University of Maryland, 4339 VanMunching Hall, College Park, MD 20742, 20742, College Park, MD, United States,[email protected], Kiran Panchamgam, Yupei Xiong, Benjamin Dussault, EdwardWasil

The distribution of relief aid is a complex problem where the operations have to be managed efficiently dueto limited resources. We present a routing problem for relief operations whose primary goal is to satisfydemand for relief supplies at many locations taking into account the urgency of each demand. We havea single vehicle of unlimited capacity. Each node (location) has a demand and a priority. The priorityindicates the urgency of the demand. Typically, nodes with the highest priorities need to be visited beforelower priority nodes. We describe a new and interesting model for humanitarian relief routing that wecall the hierarchical traveling salesman problem (HTSP). We compare the HTSP and the classical TSP interms of worst-case behavior. We obtain a simple, but elegant result that exhibits the fundamental tradeoffbetween efficiency (distance) and priority and we provide several related observations and theorems. Next,we try to generalize these worst-case results, with only partial success.

� TE-17Thursday, 15:15-16:45New Orleans Room

Nonlinear Optimization IIStream: Continuous OptimizationInvited sessionChair: Mirjam Duer, Mathematics, University of Trier, 54286, Trier, Germany, [email protected]

1 - Vector duality for convex vector optimization problems by means of the quasi in-terior of the ordering coneSorin-Mihai Grad, Faculty of Mathematics, Chemnitz University of Technology, 09107,Chemnitz, Sachsen, Germany, [email protected], Emilia-Loredana Pop

We define weakly minimal elements of a set with respect to a convex cone by means of the quasi interior ofthe cone and characterize them via linear scalarization, generalizing the classical weakly minimal elementsfrom the literature. Then we attach to a general vector optimization problem a dual vector optimizationproblem with respect to (generalized) weakly efficient solutions and establish new duality results. Byconsidering particular cases of the primal vector optimization problem we derive vector dual problems withrespect to weakly efficient solutions for both constrained and unconstrained vector optimization problemsand the corresponding weak, strong and converse duality statements.

2 - On Farkas’ Lemma and the Hahn-Banach TheoremLenka Plohakova, University of Ostrava, 30. dubna 22, 70200, Ostrava, Czech Republic,[email protected]

First we recall the classical algebraic Hahn-Banach theorem. Then we recall Farkas’ lemma within a vectorspace and a linearly ordered vector space over a linearly ordered field. Using the latter result, we formulatea new version of Dax’s theorem of the alternative and then we derive a Hahn-Banach type theorem. Inspiredby those results, we derive a Universal Theorem of the Alternative and a yet more general Hahn-Banachtype theorem.

175

TE-17

3 - Farkas’ Lemma, Gale’s Theorem, and Linear Programming: the Infinite Case in anAlgebraic WayDavid Bartl, Department of Mathematics, University of Ostrava, 30.dubna 22, 701 03, Ostrava,Czech Republic, [email protected]

We study a problem of linear programming in the setting of a vector space over a linearly ordered (possiblyskew) field. The dimension of the space may be infinite. The objective function is a linear mapping intoanother linearly ordered vector space over the same field. In that algebraic setting, we recall known results:Farkas’ Lemma, Gale’s Theorem of the alternative, and Duality Theorem for linear programming withfinite number of linear constraints. Given that "semi-infinite’ case, finite systems of linear inequalitiesin an infinite-dimensional space, we are motivated to consider the infinite case: infinite systems of linearinequalities in an infinite-dimensional space. Given such a system, we assume that only a finite numberof the left-hand sides is non-zero at a point. We shall also assume a certain constraint qualification (CQ),presenting counterexamples violating the (CQ). In the described setting, we then formulate an infinitevariant of Farkas’ Lemma along with an infinite variant of Gale’s Theorem of the alternative. Finally, weformulate the problem of an infinite linear programming, its dual problem, and the Duality Theorem for theproblems. In conclusion, we present our main motivation: the question whether the theory can be appliedto some problems of infinite linear programming, whose solution is already known; we aim to establish anew approach to resolution of those problems. We mention an infinite variant of the transportation problemin particular.

176

FA-02

Friday, 9:00-10:30

� FA-02Friday, 9:00-10:30Veder Room

Revenue Management in Non-Service IndustriesStream: Revenue Management and PricingInvited sessionChair: Jochen Gönsch, Department of Analytics & Optimization, University of Augsburg,Universitätsstraße 16, D-86159, Augsburg, Germany, [email protected]

1 - Revenue Management in a Regenerate-To-Order EnvironmentFelix Herde, Institut für Produktionswirtschaft, Leibniz Universität Hannover, KönigswortherPlatz 1, 30167, Hannover, Germany, [email protected]

Remanufacturing restores the functionality of a used good for another life cycle. In the context of complexdurable goods the term regeneration is used as a synonym for the term remanufacturing. These goods,such as aircraft engines, are owned by the customers of regeneration service providers. Different regen-eration modes can be applied in order to regenerate those goods. On the one hand, these modes can bedistinguished based on resource consumptions, durations and costs. On the other hand, these modes havea different impact on the good’s functional features, e.g. energy efficiency. Depending on the customer-specific business model the appreciation of selected modes varies in different customer types. Hence, thewillingness to pay and simultaneously the service providers contribution margins depend on the ability toaddress the customers specific needs. Similar to Revenue Management (RM) applications in a make-to-order (MTO) environment, regeneration service providers have to decide whether to accept or to declineincoming regeneration requests. The reason is that they have only limited short-run capacities. Hence,providers are confronted with the problem of pricing and capacity control. Where the afore-mentioned RMapplication in a MTO environment deals primarily with new products, in the regenerate-to-order (RGTO)environment used products with individual conditions are addressed. These conditions are not known withcertainty in advance. In addition workloads are of stochastic nature. Taking the characteristics of theRGTO environment into account, a bid-price-based capacity control with network capacities is presented.Different conditions of the goods as well as different regeneration modes are integrated into randomizedlinear programs.

2 - The Impact of Consumer Returns on a Manufacturer Multichannel StrategyPaolo Letizia, Management of Technology and Innovation, ERASMUS UNIVERSITY, 3062PA, Rotterdam, Netherlands, [email protected]

Products sold over Internet, thanks to product variety and customization, can offer higher marginal valueto the consumer than the ones sold at the store. However, there is uncertainty about product fit which mighttrigger consumer returns. The manufacturer has to strategically control, through prices and a returns policy,the flow of consumers buying his products through the on-line channel vs. the retail channel. Our analysisshows that the manufacturer will sell his product through the retail channel only if consumers expecthigher product marginal value there and the manufacturer can extract a limited salvage value from returnedproducts. The presence of the online channel, however, increases the manufacturer’s profits even thoughno sales take place online, as the retailer, threatened by the competition of the online channel, decreases theselling price to the benefit of the manufacturer’s profits.Further, the retail channel is the optimal recipientfor consumers who have high uncertainty about product fit as the manufacturer not efficient in productrecovery would offer a strict returns policy at the online channel. We extend our analysis to include thephenomenon of free-riding, where consumers use the retail channel only to get more information about theproduct characteristics, but then purchase the product online.

177

FA-03

3 - Optimal operation of a combination of energy storage and renewable energysourcesMichael Hassler, Department of Analytics & Optimization, University of Augsburg,Universitätsstrasse 16, 86159, Augsburg, Deutschland, Germany,[email protected], Jochen Gönsch

Over the last decade, renewable energy sources (RES) have been increasingly used to replace large con-ventional power stations. By now, the energy production from RES in Germany occupies a significant partof the total capacity of the energy distribution system. However, due to the volatile nature of energy pro-duction from RES, this raises the issue of power grid stability. The use of energy storage systems providesone means to cushion the impact that the increasingly volatile production has on the power grid. In orderto show how energy storage could help facilitate the integration of RES, we regard a combination of RESwith energy storage systems and derive a model for the optimal control of energy storage capacity. As thisproblem resembles the problem of capacity control in revenue management, we use a similar methodol-ogy to formulate our model based on stochastic dynamic programming. Furthermore, on the basis of anillustrative numerical example, we show the practical applicability of our approach.

4 - Integrating Keyword Advertising and Revenue Management in the Online TravelIndustryThomas Görtz, Information Systems & Business Administration, JohannesGutenberg-Universität Mainz, Haus Recht und Wirtschaft I, Jakob-Welder-Weg 9, 55128,Mainz, Germany, [email protected], Jella Pfeiffer, Henning Schmidt

Keyword advertising is an important marketing instrument of e-commerce companies to generate trafficon their website originating from search engines. Empirical studies in recent years have shown that toppositions lead to more clicks on the advertisers’ ad placement but also to higher costs. An alternative to toppositions in the organic search results are top positions in the sponsored section of search engines becausethat firstly creates awareness among Internet users and secondly boosts their sales and revenue. Usually,the costs for keyword advertising are determined in an auction that is conducted for every single searchquery entered by a user. The companies’ bid determines the position of their ad and the price that they needto pay to the search engine whenever a customer actually clicks on the ad (cost-per-click). In contrast toexisting models that solely set the focus on finding the optimal bidding strategy for a portfolio of keywordsin order to maximize revenue under a budget constraint, we additionally try to find the optimal price ofadvertised products. Thus, we propose an integrated keyword advertising model that combines finding theoptimal bidding strategy with finding the optimal pricing strategy for the offered products under a budgetand capacity constraint. In this way, the model provides a simultaneous control of bidding and revenuemanagement decisions. Furthermore, in a field experiment, we compare the revenue contribution of thepure bidding model with the state-of-the-art optimization of a large online market place.

� FA-03Friday, 9:00-10:30Leeuwen Room 1

Simulation and System DynamicsStream: Simulation and System DynamicsInvited sessionChair: Henk Akkermans, Dept of Information Management, Tilburg University, Warandelaan 2, POBox 90153, 5000LE, Tilburg, Netherlands, [email protected]

1 - Protecting Critical Infrastructures: Computational Red Teaming for Agent-BasedSimulationsSilja Meyer-Nieberg, Department of Computer Science, Universität der Bundeswehr München,85577, Neubiberg, Germany, [email protected], Erik Kropat

178

FA-03

The modern, interconnected world is vulnerable against disturbances or attacks. Global supply chains spanseveral continents but can take considerable harm by removing only a few nodes of the network. Localblackouts and their cascading effects can cause disruptions to energy and transportation networks as wellas production on a much larger scale. The protection of critical infrastructures, therefore, is of utmostimportance.

Simulation models can be used in this area to assess the safety of an installation w.r.t. attacks by a humanplayer or nature. Red Teaming, the developing of attacker strategies, is an important tool which enablesto identify weaknesses in structures or strategies. Red teaming is used for instance with great success innetwork security to assess the robustness against attacks. Traditional red teaming, however, is based onhuman expertise. While it is very successful, it comprises an inherent danger: Are there weaknesses orstrategies that have not been thought of?

Computational red teaming coupled with natural computing presents an innovative means to circumventthis problem. Up to now, there are only few attempts to evolve and adapt an agent’s behavior in redteaming. This paper explores the use of evolutionary algorithms and swarm-based methods examining amaritime security scenario.

2 - Wisdom of Crowds in Social Networks with Structural ShocksJia-Ping Huang, VU University Amsterdam, De Boelelaan 1105, 1081 HV, Amsterdam,Netherlands, [email protected], Bernd Heidergott, Ines LindnerSocial networks are becoming increasingly important in the dispersion of information, opinions and be-haviors. The interest in the modeling and analysis of social networks lies in understanding phenomenathat are typical for these networks. Communication in a social network often involves repeated transfers ofknowledge among large numbers of agents, and fully rational learning becomes infeasible. Nevertheless,agents may use fairly simple updating rules for their beliefs and still arrive at outcomes like those achiev-able under fully rational learning. In the seminal paper of Golub and Jackson (2010), simple updating rulesare introduced based on DeGroot (1974) such that an individual places her trust on the current beliefs ofher neighbors in the social network and forms her own belief for the next period by taking the weightedaverage of her neighborhood. They formulated conditions about network structure under which the beliefsconverge to the true state when the network is growing in size. Such a social network is said to be wise.One shortcoming of their paper is that they assumed that the network grows according to fixed pattern.In this research, we relax this assumption and study the case when the network structure is occasionallyperturbed by a shock, i.e. we study mixtures of wise networks and non-wise networks. As we will show,a wise network remains wise under mixture with an arbitrary (non-wise) network provided the rate of ashock satisfies certain conditions. Simulation results are obtained for the preferential attachment model ofscale-free networks introduced by Barabási and Albert (1999) to support our findings.

3 - The effects of customer misclassification on cross-training in call centersAndreas Schwab, Julius-Maximilians-Universität Würzburg, Germany,[email protected], Burak BukeWe observe prevalent trends towards multi-type call centers, which are a main driver of customer misclas-sification, i.e. a caller declares to face a certain problem, e.g. a hardware problem, where in fact anotherproblem persists, e.g. a software problem. In call centers that apply no cross-training, misclassified callsneed to be rerouted to agents who are able to serve the true problem, whereas cross-training enables agentsto serve different problem types, what reduces cycle times of the calls. Two-type queueing models havebeen developed in order to experiment on the effects of customer misclassification on cross-training incall centers. Accelerating the time to identify calls as misclassified by adequate technology can be a sig-nificant cost saver, since similar performance can be achieved by partial cross-training as compared of acompletely cross-trained workforce. If only a third of the agents is cross-trained, high increases in modelperformance can be confirmed, whereas little benefit is added by higher amounts of cross-training. In termsof customer abandonment, partial cross-training was found to outperform a completely flexible workforce.We conclude, that partial cross-training outperforms full cross-training from different perspectives. Theassignment policy of calls in partially cross-trained systems strongly influences customer waiting times.Similarly, this result was achieved for a Markovian loss model with no waiting space. When accountingfor misclassification events, we show that a proof of preferred call admission to dedicated servers will mostlikely fail, since simulation results display significant performance improvement for primary admission tocross-trained agents, when assuming descent misclassification probabilities.

179

FA-04

4 - Steady as she goes? A case study of growth robustness analysis in shipbuildingHenk Akkermans, Dept of Information Management, Tilburg University, Warandelaan 2, POBox 90153, 5000LE, Tilburg, Netherlands, [email protected] is a risky business. Periods of high growth are frequently followed by major declines. Behindevery business success lurks failure. Projects can run out of control, ship designs can lose their popularity;customers can become dissatisfied, financial performance can become dismal. The interrelations and thetime delays involved are very complex to grasp, too complex for most management teams. Model-buildingand simulation can come to the rescue. This paper reports on a model-building study where the man-agement team of a highly successful shipbuilding unit was aided in its strategy assessment process witha "growth robustness analysis’. In two so-called group model-building workshops, a system dynamicssimulation model of the team’s business context was developed. Potential future growth trajectories wereevaluated and more and less robust policies to assure this future growth were analyzed by management withthe aid of the simulation model. Policy recommendations ranged from organizational culture to marketing,from human resources to operations management.

� FA-04Friday, 9:00-10:30Leeuwen Room 2

Portfolio OptimizationStream: Financial Modeling, Banking and InsuranceInvited sessionChair: Olaf Menkens, DCU, Ireland, [email protected]

1 - The Efficient Frontiers of the Mean —Risk Portfolio Selection ProblemsMarius Radulescu, Mathematical Statistics, Institute of Mathematical Statistics and AppliedMathematics, Casa Academiei, Calea 13 Septembrie, nr. 13, 050711, Bucharest, Romania,[email protected], Sorin Radulescu, Constanta Zoie RadulescuIn the paper are defined the notions of efficient frontier set and efficient frontier function of a parametricoptimization problem. We formulate several portfolio selection problems which are nonlinear program-ming problems. Two of them are minimum risk type problems and the other two are maximum expectedreturn type problems. Taking into account various hypotheses on the risk and return functions a dualitybetween minimum risk type problems and maximum expected return type problems is investigated. Weare interested when the efficient frontier sets of the minimum risk problems and of the maximum expectedreturn problems are equal. Some applications to the crop planning problems are suggested.

2 - Robustness Analysis of Evolutionary Algorithms to Portfolio Optimization againstErrors in Asset MeansOmar Rifki, Economic Engineering, Kyushu University, Afghanistan,[email protected] Mean-Variance (MV) optimization is a well-studied model for Portfolio Optimization. Although themain focus is primarily in finding the best efficient portfolios, MV model is known to be extremely sensitiveto perturbations in assets means. This paper investigates the robustness of MV optimization when solvedby Evolutionary Algorithms (EA), in the case of linear constraints, i.e., budget constraint and holdingconstraints. To this end, comparisons were made on Quadratic Programming (QP), Genetic Algorithms(GA) and Evolution Strategies (ES). In order to identify, for EA, robust portfolios, which are supposed toexhibit low sensitivity to small changes in assets means, we proceed by exploiting the population aspectof EA and computing the performances of some selected ’good’ individuals under multiple runs subjectsto perturbations. Comparisons of portfolios follows two procedures, the first measures the loss in terms ofutility functions, while the second is more practical enabling the decision maker to incorporate a preferredlevel of robustness. The experimental results using real-world data show that EAs have stronger robustnessthan QP; many individuals of EAs population outperform the QP-based optimal portfolio.

180

FA-05

3 - Costs and Benefits of Crash HedgingOlaf Menkens, DCU, Ireland, [email protected]

The worst-case scenario portfolio problem which has been introduced by Korn and Wilmott (2002) willbe considered in this talk. In the setting of Korn and Wilmott, approximations for the optimal crashhedging strategy will be derived. Furthermore, the costs and benefits of using the optimal crash hedgingstrategy instead of the classical optimal portfolio strategy will be calculated. Additionally, we compute theefficiency of the optimal crash hedging strategy and derive the break even crash size, that is the crash sizewhere the investor is indifferent in either using the optimal crash hedging strategy or the classical optimalportfolio strategy.

� FA-05Friday, 9:00-10:30Goudriaan Room 1

Healthcare planning and schedulingStream: Health Care ManagementInvited sessionChair: Ger Koole, Mathematics, VU University Amsterdam, De Boelelaan 1081 a, 1081 HV,Amsterdam, Netherlands, [email protected]

1 - Nursing home staffing: gaining more insight iunto scale and skill-mix efficienciesDennis Moeke, Department of Mathematics, VU University Amsterdam, De Boelelaan 1081a,1081 HV , Amsterdam, Netherlands, [email protected], Ger Koole

Care workers account for a significant proportion of the total health expenditure in nursing homes and areby far the largest controllable resource. Therefore determining the appropriate number and type of careworkers required plays an important role in the search for more efficiency. This study provides insights inhow and why ’scale of scheduling’ and the enlargement of care workers’ jobs (blending tasks of differentqualification levels) effect the number and type of staff required to meet the preferences (in terms of dayand time) of nursing home residents. The scheduling data for plannable care activities of three separatedecision making units within a single Dutch nursing home have been analyzed. The results show that in>75% of the examined cases substantial scale and skill-mix economies can be achieved. We also found thatthe correlation between the demand patterns of different types of care tasks is of considerable importancewhen it comes to possible scale and skill-mix efficiencies.

2 - Discharge Planning to Mitigate Bed Block and ER OvercrowdingRene Bekker, VU University Amsterdam, Netherlands, [email protected], Jonathan Helm, MarkVan Oyen

Bed block in hospitals leads to long waits in Emergency Department, patients being placed in the hallwayson stretchers, poor quality of care, high costs and mortalities. The timing of discharges from inpatient bedssignificantly impacts bed block, as the mismatch between arrivals and discharges over the day typicallyleads to a mid-day spike. This research models the intra-day census process and discharge timing andinvestigates how hospital characteristics impact bed blocking.

3 - Patient Appointment Scheduling in HospitalsSarah Kirchner, Operations Research, RWTH Aachen, Templergraben 64, 52062 , Aachen,Germany, [email protected], Marco Lübbecke, Christina Büsing

181

FA-06

For our patient appointment scheduling problem we consider the following hospital setup: The hospitalhas resources to perform tests and treatments. These resources have fixed capacities and they are onlyavailable during their opening hours. Every treatment can be processed on a subset of these resources andhas a fixed processing time. We consider two different patient types - outpatients who come to the hospitalfor diagnosis or treatment but don’t stay overnight and admitted patients. All appointment requests have tobe scheduled such that resource capacities and opening hours are respected and every patient has at mostone treatment at a time. As objective we consider the mean length of hospitalization, as the settlementamount for a patient is determined by his diagnoses and received treatments.

Today in every hospital unit a local scheduler assigns appointments sequentially to incoming treatmentrequests disregarding the patients calendar and the length of hospitalization of a patient.

An additional challenge in patient appointment scheduling is the highly dynamic environment of a hospital.Over time new patients are admitted to the hospital and new appointment requests are made. Thereforescheduling decisions should be in some sense “robust” against changes due to new appointment requests.

This is the motivation for a joint project with the university hospital in Aachen and Inform with the goalto develop a decision support system for patient appointment scheduling in hospitals. For a first “proofof concept” we developed solution approaches based on heuristics and integer programming that can copewith the highly dynamic environment of a hospital. We report on experiments with real data from theuniversity hospital in Aachen.

4 - Nurse assignment to endoscopy unit rooms at the Centre hospitalier universitairede SherbrookeMélisende Brazeau, MATHEMATICAL AND INDUSTRIAL ENGINEERING, ÉcolePolytechnique de Montréal, 207, Principale, J7X 1A1, Les Coteaux, Québec, Canada,[email protected], Alain HertzNurse scheduling is a complex task that is still often done by hand. In the endoscopy unit at the Centrehospitalier universitaire de Sherbrooke (Quebec, Canada), the assistant head nurse (AHN) must assigneach nurse to a specific room in order to make a monthly schedule. This task is time consuming and doesnot ensure the best possible schedule.

To solve this assignment problem, a mixed integer programming model was developed. Each month, thedata required for the model are collected, such as nurses’ availability and doctors’ room assignment. Anew instance of the problem is therefore created and solved with a free solver, the COIN Branch and Cut(CBC).

The model involves boolean decision variables with three indices in order to decide if a nurse is assignedto a room at each period of time. It also takes into account all the constraints that have to be considered bythe AHN. Some of these constraints are hard, such as the assignment of a nurse to a room where a doctor isworking, and they must be respected. Other soft constraints are used to encourage or avoid some situations.According to their relative importance, each of these situations is weighted in the objective function.

The model was tested a posteriori on four instances of the problem in order to generate optimal schedulesthat can be compared to the schedules done by hand. The purpose of this project is to develop a programthat will be used by the AHN to replace the manual method. The program will provide a faster and userfriendly method that will give an optimal schedule in terms of the criteria defined by the AHN.

� FA-06Friday, 9:00-10:30Goudriaan Room 2

SAMS-decision-supportStream: Software Applications and Modelling SystemsInvited sessionChair: Kostja Siefen, DS&OR Lab, University of Paderborn, 33098, Paderborn, Germany,[email protected]

182

FA-06

1 - The history of System Dynamics for strategic planning in the electric powerindustry- Implications for the energy transition in GermanyArmin Leopold, Department for Computer Science, Universität der Bundeswehr München,Werner-Heisenberg-Weg 39, 85577, Neubiberg, Bavaria, Germany, [email protected],Bo Hu

Generally speaking System Dynamics modeling has been used for strategic energy planning and policyanalysis already for more than forty years. The story begins with the world modeling projects conducted inthe early 1970s by the System Dynamics Group at the Massachusetts Institute of Technology. During thelate 1970s John Sterman, an MIT Ph.D. student, was hired by Roger Naill to work with a team to modifyand extend the FOSSIL1 model into the FOSSIL2 model. In the early 1980s, system dynamicist GeorgeRichardson met a British petroleum analyst who claimed that the "amount of oil in the world is increas-ing." Naill and Hubbert’s work formed the basis for Sterman, Richardson and Davidsen’s synthetic dataexperiments on analyzing techniques for forecasting the ultimately recoverable amount of oil in the worldand in the United States, while knowledge of the weaknesses in the FOSSIL2 model caused Sterman toinvestigate the dynamics of energy-economy interactions during the energy transition. Fiddaman’s recog-nition that, although the source constraints on the energy-economy system had been investigated by energymodelers, sink constraints had not, lead to the creation of the FREE model. The world modeling projectsalso stimulated the study of the US electric power industry by Andrew Ford and the subsequent EPPAMmodels and their offshoots. Since Ford’s pathbreaking work, system dynamics has been used extensivelyby utility managers for strategic planning. A detailed analysis of the existing literature and comparingthe variety of System Dynamics Models in the field of electric power industry will provide a variety ofimplications for the energy transition in Germany.

2 - Resource Scheduling Services — A Fusion of Cloud Computing and AdvancedPlanningChristian Gahm, Chair of Production & Supply Chain Management, Augsburg University,Universitätsstraße 16, 86159, Augsburg, Germany, [email protected],Bastian Dünnwald

Cloud Computing (CC) and Advanced Planning (AP) are two emerging topics in research as well as prac-tice. Nevertheless, there exists neither a structured approach nor any other scientific contribution thatmerges both in order to exploit the benefits resulting from Advanced Planning Systems (APS; software sys-tems supporting AP) based on the CC paradigm. Therefore, the contribution at hand presents the conceptof "Resource Scheduling Services’ (RSSs) as a first approach. RSSs are designed to tackle the applicationdomain of operational production planning (i.e. lot-sizing and scheduling) as well as project scheduling.In other words, the problem domain of allocating scarce resources to jobs or tasks (under the considerationof different technical and/or organizational constraints) is addressed. Against this background, we firstdiscuss the potentials of merging CC and AP and emphasize central aspects concerning the provisioningof RSSs. In this context, the service model "Result as a Service’ (RaaS) is introduced. We further presenta comprehensive scheme to specify RSSs and the corresponding resource scheduling tasks and problems.This scheme includes general problem specific aspects (like planning process characteristics or temporaland factual interdependencies) and integrates and combines well known problem classification schemesfor project and production scheduling in order to provide a domain specific ontology. This ontology formsthe base for the description of RSSs (by service providers) and thus enables the selection of appropriateservices by consuming organizations. Furthermore, it can be used to classify and catalog solution methodsfor an efficient selection and application or adaption and application.

3 - Optimal timetables at RWTH Aachen UniversityFlorian Dahms, Operations Research, RWTH Aachen, Templergraben 64, 52056, Aachen,Germany, [email protected], Gerald Lach, Marco Lübbecke

In the Winterterm 2013/14 RWTH Aachen University will be hit by a wave of increased enrollment num-bers due to changes in the German education system. This, alongside with already sparse room capacitiesand many lectures serving multitudes of different curricula, led to the decision to break with establishedmanual timetabling processes and to introduce a software based timetabling system. Introducing a new,prototypical system into established structures (sometimes grown over many decades) comes with many

183

FA-07

challanges. Data has to be collected, digitalized and maintained, which so far only existed in legacy sys-tems, paper form or - in the worst case - only in the heads of a few people. Different stake holders needto be pursuaded and their interests balanced against each other, something that cannot be easily incorpo-rated in a simple objective function. And existing processes need to be changed in order to incorporate thenew work flow. This talk will briefly cover the software, models and algorithms we used. But the mainfocus will be on the actual implementation of the system, as the scientific part would not be of much valuewithout it.

4 - Requirements Engineering for Mathematical Optimization in Service Environ-mentsKostja Siefen, DS&OR Lab, University of Paderborn, 33098, Paderborn, Germany,[email protected], Florian Stapel, Lars Beckmann

From a development perspective, the design and integration of mathematical optimization models as wellas the selection and adaptation of solution algorithms have many similarities to the development of com-plex software systems. Models and algorithms have to meet diverse stakeholders’ requirements and deliverexpected results within specified time limits. We present a requirements driven approach for developingmathematical programming models in a business environment. We generalize the project team membersinto organizational roles and describe the target solution workflow in a service-oriented software archi-tecture. We define a new specification language which allows for automated validation of model aspects,explicit integration of data semantics and the preservation of tacit knowledge through structured changemanagement. We demonstrate the integration of our approach into a service environment and elaborate onthe advantages and further research perspectives.

� FA-07Friday, 9:00-10:30Antwerp Room

Container Terminal OperationsStream: Maritime LogisticsInvited sessionChair: Frederik Schulte, Institute of Information Systems, University of Hamburg, HAMBURG,Germany, [email protected]

1 - New research directions in Yard Operations in Container TerminalsIris F.A. Vis, Faculty of Economics and Business, Dep. of Operations, University of Groningen,P.O. Box 800, 9700 AV , Groningen, Netherlands, [email protected], Hector Carlo, Kees JanRoodbergen

Storage yards at containers terminals are used to temporarily store inbound and outbound containers. Acombination of container demand increase and storage yard capacity scarcity create complex operationalchallenges for storage yard managers. We present an in-depth overview of relevant literature on storageyard operations, including the material handling equipment used, and highlight current industry trends anddevelopments. Current operational paradigms on storage yard operations are discussed and challenged.Lastly, we identify new avenues for academic research based on current trends and developments in thecontainer terminal industry.

2 - Integrated Scheduling Optimization of Multiple Straddle Carriers and a GantryCrane Loading and Unloading Trains at a Deep-sea TerminalAmir Hossein Gharehgozli, Rotterdam School of Management, Erasmus University Rotterdam,Postbus 1738, 3000 DR, Rotterdam, Netherlands, [email protected], Leo Kroon, René deKoster

184

FA-07

We study the difficult operational problem of loading and unloading trains at a container terminal. Trainsare served by a gantry crane which spreads over all of them on parallel tracks next to each other at theterminal landside. Multiple straddle carriers are available to move containers from the stacking area toa drop off lane parallel to the trains and vice versa. We focus on how to sequence the gantry crane tocarry out all containers, where to drop off containers on the drop off lane by the gantry crane and straddlecarriers, and finally how and in which sequence containers should be assigned to straddle carriers. Theobjective function is to minimize the total earliness and tardiness of trains with different time windows.The problem is formulated as a mixed integer programming model which cannot be solved in reasonabletime. Therefore, we decompose the problem into three subproblems the first of which determining the dropoff locations, the second scheduling the straddle carriers, and the last obtaining the schedule of the gantrycrane. These problems are solved sequentially until no improvement in the objective value can be obtained.Due to the complexity of each individual problem, we propose a heuristic algorithm to solve each problem.Extensive numerical experiments are performed to show the efficiency of the solution method.

3 - Strategies to deal with uncertainty in the berth allocation problemNitish Umang, Transport and Mobility Laboratory (Transp-OR), Ecole Polytechnique FederaleDe Lausanne, Avenue Tir Federale 92, 1024, Ecublens, Vaud, Switzerland,[email protected], Michel Bierlaire, Alan Erera

We study pro-active and reactive strategies to deal with stochastic variability in the arrival times and han-dling times of the vessels in the berth allocation problem (BAP).

The proposed methodology for schedule recovery in real time is based on the re-optimization of the unas-signed vessels in the berthing schedule in the events of disruption using a set-partitioning method, andan alternative heuristic based smart greedy algorithm. The uncertainty is modeled by making appropriateassumptions about the probability distributions of the unknown parameters. The objective is to minimizethe total realized cost of berthing the vessels along the quay. A simulation studies is carried out to assessthe solution performance and efficiency of the proposed algorithms. Results indicate that the proposedalgorithms can significantly reduce the total realized costs of the berthing schedule as compared to theongoing practice of reassigning vessels at the port.

To develop a robust formulation for the BAP, we focus on the much simpler variant of the problem analo-gous to the single machine scheduling problem, as this permits us to do a more elegant and useful analysis.The measure of robustness is the maximum absolute deviation of the realized outcome from the corre-sponding optimal solution over the set of all potential outcomes. We discuss some interesting properties ofrobust solutions and propose a heuristic method based on variable neighborhood search that can be used toobtain good, near-optimal solutions to the robust scheduling problem.

4 - Simulation Requirements for (Empty) Container Repositioning in Liner Shippingwith Sustainability AspectsFrederik Schulte, Institute of Information Systems, University of Hamburg, HAMBURG,Germany, [email protected]

Liner shipping has seen an immense growth over the last few decades and is expected to grow similarlyin the future while global trade continues to increase. This development leads to several sustainabilitychallenges: Most obviously, container movements cause fuel consumption, various kinds of emissionsand congestion. Currently, about one fifth of all ocean container movements involve the repositioningof empty boxes. In order to reduce the environmental impacts of container repositioning, dynamic andstochastic planning problems need to be considered. For these problem classes simulation models can beapplied. This work aims to provide a holistic simulation framework to evaluate sustainability strategies forempty container repositioning in liner shipping. The framework is designed in a generic way to supportits application for related problems, such as terminal operations or inter-terminal transport (ITT). First,the environmental impacts of liner shipping processes are analyzed, problems are identified and respectivesustainability strategies are presented. Second, problem, goal and strategy classes are derived and theirsimulation requirements are discussed. While we have a framework of these classes, we transfer theapproach to the problem of ITT and thus deduce simulation requirements for sustainable ITT operations.

185

FA-08

� FA-08Friday, 9:00-10:30Penn Room 1

Staff SchedulingStream: Scheduling and Project ManagementInvited sessionChair: Marco Schulze, Operations Research Group, Clausthal University of Technology, Germany,[email protected]

1 - A Column Generation Approach to Home Care Staff Routing and SchedulingSusumu Morito, Industrial and Management Systems Engineering, Waseda University, Bldg 51Room 13-01, 3-4-1 Ohkubo, 169-8555, Shinjuku, Tokyo, Japan, [email protected], Chen YingWen, Hiroki Hayashi, Atsushi Torigoe, Shigeo Okamoto, Yuki Matsukawa, Nao Taniguchi

To cope with soaring cost of health care, our government tries to shift toward home health care and morehome care stations are introduced. Moderate-size care stations cope with 100-200 weekly home visits with10-20 nurses and therapists. One of the difficulties of their management is the generations of staff routingand scheduling, which is the theme of the paper. The problem calls for determination of daily routes of staffmembers to cover all patients’ visits for a week. Constraints to be considered include staff working hours,time window for each visit, means of transportation (bicycle/car), maximum allowable idle time betweenvisits, lunch time, patient/staff compatibility, among others. Since it is desired to have compact routes in thegeographically scattered area, the minimization of route length is used. Daily route generation problemsshare the basic structure of the vehicle routing problem with time window (VRPTW), and is solved by thecolumn generation approach. The column generation subproblem becomes a resource constrained shortestpath problem, solved by a variant of a labeling algorithm. Computational results based on real data willbe presented. To further reduce computation time, preprocessing of input data is performed to reduce thesolution space by narrowing the time window of visits and by limiting candidate staff members who couldbe assigned to a specific visit. The preprocessing is performed by solving two smaller 0-1 programs. It isshown how the preprocessing cuts down the computation time of the column generation algorithm. Finally,we describe how daily route generation problems are integrated with a weekly schedule generation in sucha way that weekly working hours of the staff members fall within the target level.

2 - Workload Balancing in Transportation Crew RosteringGuvenc Sahin, Faculty of Engineering and Natural Sciences, Sabanci University, SabanciUniversity, FENS, Orhanli, Tuzla, 34956, Istanbul, Turkey, [email protected], FardinDashty Saridarq

In personnel scheduling and crew rostering, balanced workload allocation is a critical issue and an impor-tant planning phenomenon that affects both the quality of duty schedules and satisfaction of the personnel.We focus on workload balancing in crew rostering problems of transportation systems where deadheading(repositioning with no duty) of crew is also possible. Deadheading is a viable option to idle waiting andis frequently used to increase the utilization of human resources in both airlines and railways. Althoughthe research interest on particularly workload balancing issue is not vast, we realize that the existing workhas only focused on systems where deadheading is not considered. However, the existing approach is notdirectly suitable to be employed in crew scheduling with deadheading as the deadheading option could bejust used unnecessarily for the sake of balancing the workload among team members. A solution approachbased on a network flow formulation of the problem is first adapted from the current work which does notconsider deadheading. Based on this formulation, we develop an optimal solution method. We comparethe computational performance and the solution quality of the optimal solution method with those of atwo-phase heuristic approach and the exact solution of the problem. We present the results of our compu-tational experiments with well-known problem instances from the crew scheduling literature and data setsthat are representative of largest crew regions in Turkish State Railways.

186

FA-09

3 - Refinements for solving the Vehicle Routing and Truck Driver Scheduling ProblemAsvin Goel, School of Engineering and Science, Jacobs University Bremen, Germany,[email protected], Thibaut Vidal

The Vehicle Routing and Truck Driver Scheduling Problem (VRTDSP) is the problem of determining a setof routes for a fleet of vehicles, such that each customer is visited within given time windows, that eachdriver can comply with applicable hours of service (HOS) regulations, and that transportation costs areminimized. In the last years detailed models of several HOS regulations world wide have been presentedand truck driver scheduling methods were developed which can be used within algorithms for optimisingvehicle routes. Due to the complexity of finding schedules complying with HOS regulations, developingan effective and efficient algorithm for solving the VRTDSP goes far beyond combining sophisticated al-gorithms for the vehicle routing problem with truck driver scheduling methods. This contribution presentsand evaluates various refinements and speedup techniques which help improving the performance of algo-rithms for the VRTDSP. With these refinements it was possible to develop a hybrid genetic algorithm forthe VRTDSP which outperforms all previous methods.

4 - Combined staff and machine scheduling in the context of underground miningMarco Schulze, Operations Research Group, Clausthal University of Technology, Germany,[email protected], Jürgen Zimmermann

We consider a German potash underground mine where crude salt is mined using a room and pillar miningmethod. The excavation is based on conventional drilling and blasting techniques. This kind of under-ground mining is characterized by different consecutive production steps (operations) such as filling blastholes with explosive substance or loading broken material. Each production step requires one tracklessmachine (from a set of heterogeneous machines) and a mine worker who has the corresponding skill. Thedaily workforce scheduling problem forms the bottom level of a hierarchical planning approach. In or-der to generate reasonable shift schedules, the overlying planning levels provide input concerning whichamount has to be mined per shift/day and which parts of the mine should be excavated with higher priority.Therefore, our problem consists of specifying the assignment and scheduling of the planned operations tothe resources that are available in the respective shift. We formulate the problem as a mixed-integer linearprogram, subject to constraints regarding the priority of the operations and precedence constraints derivedfrom the mining process. Moreover, travel times for the trackless machines as well as the skills from theminers are taken into account. From a practical point of view, the minimization of a weighted sum of com-pletion times and travel times depicts a sophisticated objective. In a comprehensive performance analysiswe use CPLEX to solve medium-scale instances derived from real-world data.

� FA-09Friday, 9:00-10:30Diamond Room 1

Queuing and SchedulingStream: Production and Operations ManagementInvited sessionChair: Tim Lamballais Tessensohn, Management of Technology and Innovation, Rotterdam School ofManagement, Zocherstraat 42, 3067 AT, Rotterdam, Netherlands, [email protected]

1 - Performance analysis of closed-loop assembly lines with general processingtimes and finite buffer spacesSvenja Lagershausen, Department of Production Management, Leibniz Universitaet Hannover,Koenigsworther Platz 1, 30167, Hannover, Germany,[email protected]

187

FA-09

In this paper, we analyze flow lines with converging and diverging material flow, limited buffer sizes, gener-ally distributed processing times and a constant number of workpieces as a closed assembly or disassemblyqueueing network. We use a decomposition approach in which each subsystem is modeled as a G/G/1/Kqueueing system. The population constraint is enforced by requiring that the sum of the expected numberof customers in the subsystems is equal to the total number of workpieces. The results of a simulationexperiment indicate that the proposed approximation provides accurate results and that it performs betterthan other approaches.

2 - Performance analysis of heterogeneous queueing systems under time-dependentconditionsGregor Selinka, Chair of Production Management, University of Mannheim, Schloss, 68131,Mannheim, Germany, [email protected], Raik Stolletz

Queueing systems are often characterized by inhomogeneities as well as stochastic and time-dependentinput parameters.

This work analyzes an N-shaped queueing system that includes jobs of two heterogeneous job classesthat arrive with stochastic inter-arrival times. The system consists of two heterogeneous service facilities.One facility is able to handle only one job class, whereas the other one has the capabilities to serve bothjob classes. Both facilities include multiple servers that are working with stochastic processing times.Arriving jobs are routed to one of the service facilities directly upon arrival dependent on the actual stateof the system.

Analytical methods for the time-dependent performance evaluation of such a queueing system are devel-oped. The considered methods are based on the stationary backlog-carryover (SBC) approximation. Inorder to speed up calculation times, the basic version of this approach is modified by applying simplesteady-state models to approximate the performance of the more complex N-shaped queueing system.A numerical study is conducted to validate the quality of the performance approximations under severalstate-dependent routing policies by comparison to simulation results.

3 - Hot strip mill scheduling under consideration of energy consumptionKaren Puttkammer, Institute of Automotive Management and Industrial Production, TechnischeUniversität Braunschweig, Institut für Automobilwirtschaft und Industrielle Produktion,Lehrstuhl für Produktion und Logistik, Katharinenstraße 3, 38106, Braunschweig, Germany,[email protected], Matthias Gerhard Wichmann, Thomas Spengler

In steel industry hot rolling is an energy intensive process as steel slabs need to be heated to about 1250Cbefore being rolled on the hot strip mill. They are typically charged into furnaces at a temperature between0C and 400C. The temperature results from the previous continuous casting process. Slabs leave thecasting process at a temperature of about 1000C and cool down during the following transportation andstorage processes. The cooling curve is non-linear and thus the temperature loss is highest in the first hoursafter casting. In any case, the colder a slab enters the hot strip mill the more energy is required to reheatit. The processing start of a slab on the hot strip mill and therewith the energy required for reheating isdetermined by the production schedule. Since the required fuel is expensive, decision support is necessaryfor the hot strip mill scheduling problem (HSMSP) under consideration of energy consumption. There area few models known from literature that do consider energy consumption in the HSMSP. Nevertheless,they neglect the continuous, non-linear cooling curve in favor of highly aggregated cooling assumptions.In this contribution a mixed integer problem formulation of the HSMSP under consideration of energyconsumption is presented. It incorporates the interdependencies between the schedule, the processing startof a slab on the hot strip mill as well as its charging temperature and energy requirement for heating. Theeffect of including energy consumption as a scheduling objective is shown by using a small test instance.

4 - Estimating performance of a warehouse with a robotic fulfillment systemTim Lamballais Tessensohn, Management of Technology and Innovation, Rotterdam School ofManagement, Zocherstraat 42, 3067 AT, Rotterdam, Netherlands, [email protected]

The rapid growth of E-Commerce calls for new methods to operate warehouses. Online demand fluctuatesheavily, grows unpredictably and consist of small sized orders. Traditionally workers handled pallets butE-Commerce requires workers to handle single items, which is detrimental to efficiency. Robotic mobile

188

FA-10

fulfillment systems have emerged as a response to this challenge. One example would be Kiva systems,which from the start was designed to maximize warehouse productivity in an E-Commerce environment.The items are stored on pods which are lifted and moved around by the robots. This in effect means that thestorage hall is a flexible and dynamic whole as pods are stored at different storage locations throughout theday. This leads to new, interesting problems. At which location should a robot store his pod to minimizetravel time and which pod should he fetch next? His storage decision influences how far other robots musttravel that need the pod later and also the storage location is temporarily no longer available to other pods.The aim of our research is to estimate the long term performance of warehouse with a robotic fulfillmentsystem. At the start of the day there are orders waiting that need to be fetched before the truck going totheir destination leaves. During the day new orders come in. Order throughput times are approximatedwith a queuing network and strategies for assigning pods and orders to robots are evaluated. These areintegrated into a model that optimizes travel distances while keeping the probability of missing due datesunder a threshold value. This model will help warehouse managers to decide whether they should investin a robotic mobile fulfillment system.

� FA-10Friday, 9:00-10:30Diamond Room 2

Perishability in Supply Chain ManagementStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Jacqueline Bloemhof, Operations Research and Logistics, Wageningen University,Hollandseweg 1, 6700 EW, Wageningen, Netherlands, [email protected]

1 - A one-warehouse, multi-newsvendor distribution systems with resupply and vehi-cle routingBelma Turan, Department of Business Administration, University of Vienna, Chair forProduction and Operations Management, Brünner Straße 72, 1210, Vienna, Austria,[email protected], Richard Hartl, Stefan Minner

We study a problem that is inspired by a real-world application of delivering perishable products (e.g.milk, bread, meat) from a retailer to stores. Products with a short shelf-life pose a challenge for findingthe balance between the costs of over- and understock. Improvements can be achieved by partly postpon-ing decisions after some additional information has revealed, i.e. by allowing second delivery for someretailers. As there are typically several outlets that are served by the same vehicle in the retail distribu-tion, inventory and routing decisions have to be taken in an integrated way. The purpose of our model isto make four decisions: to determine the time when the second delivery has to take place, to assign thecustomers to the given routes, and to determine delivery quantities for the first and the second delivery.For handling the routing decisions, we use an algorithm that is inspired by ideas of Variable NeighborhoodSearch. We use a dynamic programming algorithm to solve the inventory allocation problem for a givendelivery plan. The optimal time point for the second delivery is determined by using numerical search. Weassume that demand follows a stochastic distribution and introduce a tractable demand model based on aCompound Renewal process. For the purpose of testing the performance of our algorithm, we generated aset of numerical problems based on benchmark data from the literature. Numerical results show that theprofit significantly increases when the second delivery is allowed.

2 - Incentive schemes to improve order quantity decisions for perishable goods underuncertain demand in a profit center organizationGuido Voigt, Operations Management, Otto-von-Guericke University Magdeburg,Universitätsplatz 2, PF 4120, 39108, Magdeburg, Germany, [email protected], BarbaraSchoendube

189

FA-10

We consider a decentralized organization. A principal hires an agent to be head of a profit center. Theagent’s only task is to choose the periodic order quantity for a perishable product in the presence of uncer-tain demand. One out of two possible demand distributions materializes. By assumption the agent knowswhich distribution is present, but the principal does not. Moreover, the principal aims at maximizing longterm firm value while the agent maximizes short run profit. Given this setting an adverse selection problemis present. In order to reduce agency costs we discuss two compensation schemes to be offered by theprincipal. The first one combines a profit sharing element (i.e., compensation as a percentage of operatingprofit) with a bonus paid when all demand is fulfilled. The second scheme directly links the profit shareparameters to the agents’ order size decision. A numerical study shows that, for high profit products, bothcompensations schemes provide an efficient means of incentive alignment. However, for low profit prod-ucts, the service level scheme outperforms the fixed order size scheme, both in terms of overall efficiencyand in terms of agency costs. Intuitively the latter allows for more flexibility so that the agent can tailor hisorder size to the compensation scheme.

3 - The inventory management approaches using inventory balance and decision ofinventory across supply chain of fresh vegetablesAdi Djoko Guritno, Agroindustrial Technology, Universitas Gadjah Mada, Indonesia, Jl.Flora 1,Bulaksumur, Jl. Pengok Kidul 25, Baciro, 55281, Yogyakarta, Yogyakarta, Indonesia,[email protected], Endy Suwondo, Henry Yuliando

Purpose - The research objectives are determine the opportunity loss and cost excess of inventory, calcu-late the optimal inventory level, and decide the inventory types of using distribution lead-time approaches.Design/methodology/approach - Research was based an in-depth interviews concerning to develop modelhierarchy of IPSF (Integrated Planning Support Framework). An extensive, six-month supply chain track-ing from some point of sales was conducted to validate the proposed IPSF model and to conduct realcase studies. Findings - The research result shows an inefficient supply chain caused by an imbalanceof customer services and inventory levels in each tier of supply chain in the IPSF model. The researchresult indicates the dissimilar of inventory size influences the inventory decision in each tier along theexisting supply chain. The research concludes the losses of fresh vegetables in term of cost and qualitycould be reduced using inventory balance and decision types of inventory approaches. Research limita-tions/implications - A wider application of the IPSF model could offer interesting opportunities in termsof managerial practice and provide a more practical basis for real case studies benchmarking. Original-ity/value - Various authors have attempted to capture the real picture of inventory in the real conditionacross a supply chain, but combination of balance-decision of inventory and IPSF model could providecomprehensive multi-tiers of fresh vegetables supply chain.

4 - Sustainable Food Supply Chain ManagementJacqueline Bloemhof, Operations Research and Logistics, Wageningen University,Hollandseweg 1, 6700 EW, Wageningen, Netherlands, [email protected], Jack vander Vorst

Sustainable Food Supply Chain Management focuses on optimizing the performance of logistical activities,driven by economic, environmental, social and quality requirements. In food supply chains the percentageof waste / spoilage is very high due to limited lifetime of products. Therefore, efficiency and effectivenessare extremely important. Furthermore, as most ingredients arise from farming, environmental and socialimpacts play a large role in the performance in the supply chain. Inevitably, multiple objectives by multiplestakeholders need to be balanced to increase sustainable performance of the supply chain. We reviewcurrent literature on quantitative models in sustainable food supply chain management and claim thatspecific requirements from practice need to be taken in to consideration. We present an iterative approachto assess, evaluate and improve the sustainability of food chains and apply this approach for i) soy andbeef chains from Latin America to Europe and ii) food manufacturers and logistic service providers inNord-West Europe.

190

FA-12

� FA-11Friday, 9:00-10:30Penn Room 2

Supply Chain DesignStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Behzad Hezarkhani, Department of Industrial Engineering, Eindhoven University ofTechnology, 5600 MB , Eindhoven, Netherlands, [email protected]

1 - Solution Method for the Inventory Distribution ProblemTakayuki Shiina, Chiba Institute of Technology, 2-17-1 Tsudanuma, 275-0016, Narashino,Chiba, Japan, [email protected]

The approach to supply chain issues in recent years has been for suppliers to seek to improve service levelsand at the same time to reduce inventory amounts and their associated expenses. However, there is a trade-off between inventory volume and service levels. To improve both at the same time, a supply chain mustbe carefully constructed from the planning stage, which may involve a large investment. Lateral trans-shipments between retail bases are viewed as effective method for improving both inventory volume andservice levels, and has come into use in some operating businesses. Two inventory transfer policies havebeen investigated in previous research on distribution between bases: preventive lateral transshipment andemergency lateral transshipment. Each has its own advantages and disadvantages, and so it is reasonableto expect that combining these will allow higher service levels to be provided. For this reason, examiningthe combination of these policies is the focus of the present study. Specifically, a mathematical program-ming problem is formulated with demand as a random variable, and the policy of combined preventiveand emergency lateral transshipment is examined for its effectiveness and solution methods for the formu-lated problem are examined for their efficiency. The algorithm using Benders decomposition method isdeveloped and the numerical experiments show that the proposed algorithm is quite efficient. Finally, theadvantage using both of transshipments is shown.

2 - On optimal quantities in symmetric transshipment coalitionsBehzad Hezarkhani, Department of Industrial Engineering, Eindhoven University of Technology,5600 MB , Eindhoven, Netherlands, [email protected], Wieslaw Kubiak, Bruce Hartman

Through transshipment, supply chain agents utilize leftover inventories at some locations to satisfy the un-met demand at other locations. We study the transshipment coalitions in symmetric setting and analyse theeffect of coalition size on optimal order quantities. We show that optimal quantities gets closer to distri-bution mean as the size of coalition grows. However, the convergence point depends on the transportationcost. We discuss that the inventory anomaly, that is the increase of order quantities in transshipping coali-tions when optimal quantity for a single newsvendor is larger than mean, would be disappeared as thecoalition size gets large enough.

� FA-12Friday, 9:00-10:30Van Oldenbarnevelt Room

Routing IIStream: Supply Chain Management, Logistics and InventoryInvited sessionChair: Goos Kant, FEB, Tilburg University, Warandelaan 2, 5037AB, Tilburg, Netherlands,[email protected]

191

FA-12

1 - Combining transportation flows and networksMarjolein Veenstra, Operations, Rijksuniversiteit Groningen, Nettelbosje 2, 9747AE,Groningen, Netherlands, [email protected]

In a growing number of regions the population is declining. In these low-density regions the costs of severalservices, such as public and freight transportation, are relatively high compared to the costs in high-densityregions. A way to decrease the costs of services is to decrease the underlying transportation costs. Theintegration of transportation flows and networks is a way to achieve this goal. Methods for optimizationand design are developed to support the integration of transportation flows and networks.

2 - The Vehicle Routing Problem with Private Fleet and Common CarrierSybren Huijink, Econometrics and Operations Research, Tilburg University, Netherlands,[email protected], Goos Kant, René Peeters

In practice, many package transportation companies lower their costs by hiring outside carriers to servecustomers that cannot be served efficiently by their own trucks. The problem which takes this outsourceoption into account is the Vehicle Routing Problem with Private Fleet and Common Carrier (VRPPC). Forthis variant of the Vehicle Routing Problem, customers are either served by the private fleet or outsourced toa common carrier in such a way that the sum of the delivery costs and the outsourcing costs is minimized.Similarly as in the classical Capacitated Vehicle Routing Problem, there are different construction andimprovement methods. Notice that in the VRPPC, besides making routes, one also has to decide whichcustomers are served by the private carrier and which by the common carrier. The goals of this paper are tofirst give an overview of existing literature and second to find a robust heuristic which is able to find goodsolutions on instances with different characteristics. We compare different methods for the constructionphase such as (1) pre-selection of private fleet served customers (2) classical construction methods ofroutes like saving and insertion, and (3) the final selection of outsourced customers. Finally, we examinethe influence of different construction methods on various improvement heuristics.

3 - Optimizing Transport Logistics by Workload BalancingGoos Kant, FEB, Tilburg University, Warandelaan 2, 5037AB, Tilburg, Netherlands,[email protected]

Low margins characterize the logistics industry. In addition to that, legislation rules, service requirements,fuel and labor costs and the pressure on sustainability increases. Therefore optimizing transportation re-mains a crucial topic. Here we focus on balancing the work load over the drivers over a period of time,which shows significant additional benefits.

To balance the work load over multiple days the delivery day should have flexibility, for example at Vendor-Managed Inventory (VMI). VMI implies that the delivery day is flexible, as long as the inventory staysbetween the minimum and maximum storage capacity. We present a heuristic approach for this problemby formulating it as an extension of the Periodic Vehicle Routing Problem (PVRP). For a case study at agas company, we show over 26% decrease in distance travelled and 6% increase in volume delivered.

This ’peak shaving’ method can also be applied to the cyclic situation, where the goal is to create a balanceddelivery schedule for a period, without the possibility to postpone deliveries outside this period. Analysisin the retail industry shows 10 — 20% efficiency improvements against existing methods. Similarly, wecan use the method to balance the workload load for one day, e.g., for computing delivery areas for driversin the parcel and express business.

Based on the general applicability and obtained savings, the method is implemented in a commercialAdvanced Planning System of ORTEC. The method is applied at TNT Express for balancing the deliveryareas for drivers, which was one of the fundamentals for winning the Franz Edelman Award 2012. Themethod is implemented as well in various industries, including retail, field service and the gas industry.

192

FA-13

� FA-13Friday, 9:00-10:30Mees Auditorium

Railway OptimizationStream: Traffic and TransportationInvited sessionChair: Thomas Schlechte, Optimization, Zuse-Institute-Berlin, Takustrasse 7, 14195, Berlin, Berlin,Germany, [email protected]

1 - Railway timetable rescheduling in case of large disruptionsLucas Veelenturf, Department of Decision and Information Sciences, Rotterdam School ofManagement, Erasmus University, Postbus 1738, NL-3000 DR, Rotterdam, Netherlands,[email protected], Martin Kidd, Valentina Cacchiani, Dennis Huisman, Leo Kroon, PaoloToth

On a daily basis railway operations have to deal with disruptions in the network characterized by partialor complete blockages of open track sections. These disruptions often continue for some hours. Trainservices need to be cancelled and others need to be delayed in order to compensate for the loss of networkcapacity. In particular, such disruptions typically lead to infeasibilities in the planned timetable, the rollingstock allocations and the crew duties. This requires the railway operator and the infrastructure managerto reschedule the timetable, and the rolling stock and crew schedules. In this research we focus on therescheduling of the timetable at a macroscopic level. An integer programming model is formulated forthe purpose of solving the timetable rescheduling problem. The formulation minimizes the number ofcancelled and delayed trains while adhering to capacity constraints on the tracks and within the stations.It also ensures that there is rolling stock available for all trains in the resulting timetable. Results ofcomputational tests on a large part of the Dutch railway network are promising. Computation times arejust a few seconds if the model is solved by CPLEX. This allows the approach to be applicable for real-timerescheduling in practice.

2 - Modelling Delay Propagation in Railway NetworksFabian Kirchhoff, Clausthal University of Technology, Germany,[email protected]

Our goal is to optimise timetables with regard to their stability under moderate delays. One criterion weuse is the total expected arrival delay of passengers. This includes delays that are propagated while waitingfor delayed connections. We do not use Monte Carlo methods to estimate the delays, instead we want todetermine the delay distributions analytically from given source delays.

In order to include accumulation and propagation of delays, the source delay distribution must belong toa family of distributions that is closed under appropriate operations. This is the case if we can representthe distribution functions by so called theta-exponential polynomials. A drawback to this representation isthe increasing number of parameters needed to describe the results of the operations. A combination withmoment approximations allows to solve this problem with sufficient accuracy.

We consider a network that represents the relations between feeder and connection lines. Generally, thecalculation of propagated delays requires a topological sorting of arrival and departure events. In contrastto other models we can handle circles in this network. We present a new iterative method to approximatethe long run delay distributions in circles.

In this talk details of the model are given together with some experimental results.

3 - Optimization of a railway line systemJoel van ’t Wout, Netherlands Railways, Netherlands, [email protected], Bart de Keizer,Pieter Fioole

193

FA-14

The line system is the basis of the timetable for a railway company. It defines start- and end-point oftrains, their routes, the stopping stations and the frequency trains will run. Based on an OD-matrix anda (schematic) railway network, the aim is to provide good travel options for all passengers on one handand to do this in a cost-efficiënt way for the railway company on the other hand. We propose a geneticalgorithm approach to solve this problem. We will present the potential benefits of this model both in termsof service improvement and cost reduction.

4 - Robust online optimization at motorail terminalsPascal Lutter, Fac. of Management and Economics, Ruhr-University Bochum, Chair ofOperations Research and Accounting, 44780, Bochum, Germany, [email protected], BrigitteWerners

Motorail transportation deals with long distance transportation of passengers along with their vehicles ona train. Vehicles are loaded onto special motorail transportation wagons under several specific constraints,e.g. weight, width, height and other limitations. Due to an increased capacity demand of vehicles, for ex-ample due to growing popularity of SUVs during the past 50 years, the present capacities of transportationwagons become a critical issue. Especially the combination of increased vehicle weights and heights raisesthe complexity of the vehicle loading problem. Much research effort has been spent on container terminaloptimization. Motorail terminals differ from these terminals regarding layout, storage space and transportgoods. The main goals in motorail terminal optimization consist of determining a feasible loading for a setof vehicles on a given train and the minimization of shunting operations at the terminal. Vehicle specifi-cations are known, but the order and time of vehicle arrivals at the terminal are uncertain in advance. Asstorage spaces are limited and shunting operations become more time consuming with increasing terminalutilization, it is preferable to load arriving vehicles on the train as soon as possible. Vehicles have to beassigned in a way minimizing capacity waste such that later arriving vehicles can still be loaded, even in thecase of deviations from prespecified characteristics. We will present a new optimization approach findinga compromise between early loading and minimal capacity waste adapting the relevance automatically tothe current situation. As the loading process is critical in terms of time we will provide and evaluate primalheuristic solution methods combined with local search procedures.

� FA-14Friday, 9:00-10:30Tokyo Room

Distribution, Pickup&Delivery, Dial-a-RideStream: Traffic and TransportationInvited sessionChair: René de Koster, Rotterdam School of Management, Erasmus University Rotterdam, RoomT10-56, PO Box 1738, 3000 DR, Rotterdam, Netherlands, [email protected]

1 - Joint distribution of distances to the first and the second nearest facilitiesMasashi Miyagawa, Regional Social Management, University of Yamanashi, 4-4-37, Takeda,400-8510, Kofu, Yamanashi, Japan, [email protected]

This paper derives the joint distribution of the distances to the first and the second nearest facilities. Fa-cilities are represented as points of regular and random patterns, and distance is measured as Euclideanand rectilinear distances on a continuous plane. An application of the joint distribution can be found in afacility location problem with non-closest facility service where the distance to the second nearest facilityis also important. The joint distribution allows us to examine the first and the second nearest distancessimultaneously, and will thus be useful to evaluate the reliability of facility location when some of the ex-isting facilities are closed. The joint distribution of the road network distances is also obtained to confirmthat the joint distribution on a continuous plane gives a good estimate for that on road networks.

194

FA-14

2 - A Comparison of Different Column-Generation Formulations for the Pickup-and-Delivery Problem with Static and Dynamic Time WindowsTimo Gschwind, Johannes Gutenberg University Mainz, Jakob-Welder-Weg 9, 55128, Mainz,Germany, [email protected]

The Pickup-and-Delivery Problem with Static and Dynamic Time Windows (PDPSDTW) is a VRP withpairing and precedence, capacities, and static and dynamic time windows. It is the prototypical VRPwith temporal intra-route synchronization constraints. The most natural column-generation formulationfor the PDPSDTW uses master program variables representing PDPSDTW-feasible routes. The resultingsubproblem to generate such routes, however, is challenging. An alternative approach is to formulate themaster program in relaxed routing variables that may violate one (or more) types of constraints relatingto a single route. These constraints must then be handled by adding valid inequalities to the master pro-gram. The easier to solve subproblem in the latter approach comes at the cost of weaker master programlower bounds. Thus, it is a priori not clear which formulation is preferable. Our contribution is twofold:First, we devise a new labeling procedure to solve the natural PDPSDTW-subproblem allowing for an in-teger column-generation algorithm where all route constraints are handled in the subproblem. Second, wecompare this algorithm to approaches based on alternative formulations in a computational study. Theseformulations use subproblems relaxing some route constraints and, hence, can be solved using more effec-tive labeling algorithms. Preliminary results indicate that additional effort in the subproblem pays off inthe overall algorithm.

3 - The Value of Inaccurate Advanced Information in the Dynamic Dial-a-ChauffeurProblemNiels Agatz, Decision and Information Science, RSM University, Burg. Oudlaan 50, 3000DR,Rotterdam, Netherlands, [email protected], Jordan Srour, Johan Oppen

This contribution considers a vehicle routing problem in which the jobs are customers requiring a chauffeurfor a drive home in their own cars. It is one faced by several businesses that provide drivers to customerswho require that both they and their cars arrive home safely after drinking too much. In this setting,customers often pre-announce their desire for a ride in advance without committing to a specific timeyet. Later, during operations, when the pre-announced customer actually requires service, they call thecompany for pick-up. In this contribution, we focus on the value of such advanced inaccurately timewindow information for this dynamic pickup and delivery routing problem. Specifically, we introduce amultiple scenario route selection approach to incorporate the inaccuracy in the route planning. Based on anumerical study based on real-life data, we show the strengths of this strategy relative to one which ignorespre-announced information and one that includes pre-announced information in a naive manner.

4 - A Collaborative Approach in City DistributionDerya Eren Akyol, Department of Industrial Engineering, Dokuz Eylul University, Buca, 35160,IZMIR, Turkey, [email protected], René de Koster

To reduce the negative effects of urban freight transport, in many cities local governments implement timewindows which limit the time during which freight vehicles are allowed to enter city centers to short periodsand an attractive city for the residents and the shopping public is tried to be achieved. The delivery timewindow for each city is determined by the city authority itself and the retailer complies with the restrictionsimposed. However, the time window restrictions that a particular city applies affect not only the deliveriesto the retailers’ stores in that city, but also the store deliveries of the retailers in the neighbouring cities.In this paper, we question whether it is possible to find time windows better than the current used timewindows using a game theoretic collaborative policy where a coalition of the cities cooperate to determinetheir time windows together. Using the municipal time preferences for truck deliveries, a regression basedmodel is developed to represent the city satisfaction function which takes into consideration the interactionbetween the time windows of the neighbouring cities. Using a multi-objective optimization approach, thetime windows resulting from the possible coalitions are evaluated. A case study is presented to illustratethe approach and the results.

195

FA-15

� FA-15Friday, 9:00-10:30New York Room

Mixed Integer ProgrammingStream: Discrete and Combinatorial Optimization, Graphs and NetworksInvited sessionChair: Tobias Achterberg, CPLEX Optimization, IBM, Zuse Institute Berlin, Takustr. 7, 14195, Berlin,Germany, [email protected]

1 - Compact Mixed-Integer Formulations for Multi-Entity and Multi-Period ProblemsMernout Burger, Delft Center for Systems and Control, Delft University of Technology,Mekelweg 2, 2628CD, Delft, Netherlands, [email protected] scheduling and routing problems often mixed-integer linear programs (MILPs) are used to formulatethe problem mathematically. Solving such MILPs can be done by standard solvers, or using more spe-cialised algorithms based on e.g. decomposition methods. Independent of the algorithm, in general themore binary variables are needed to describe the problem, the longer it takes to compute the (optimal)solution.

At the basis of many problems is the Travelling Salesman Problem (TSP). Key elements of the TSP arethe decision variables that define the order in which the cities are visited. These variables use two indicesto represent whether or not the salesman will travel from one city (the first index) to another (the secondindex). As a result the amount of binary variables grows quadratically with the number of cities in theproblem.

When considering multiple salesmen starting from (and returning to) multiple locations one has to distin-guish the routes of the individual salesmen. Often this is done by adding an extra index to the decisionvariable that identifies a salesman. This results in a cubic growth of the number of binary variables, therebymaking the problems inefficient to solve.

For problems with multiple entities (people, vehicles, etcetera) or multiple periods the three-index formu-lation of the multi-depot TSP (MTSP) is often at the basis of the MILP formulation. Recently, we haveshown that the MTSP can be rewritten as a problem using two-index decision variables plus N continuousvariables (with N the amount of cities and salesmen) with much lower computation times to find the opti-mal solution. Now we show that the same method can be applied to other three-index MILP problems aswell, such as vehicle routing problems and variants.

2 - Heuristic Approaches for Staircase-Structured Mixed Integer ProgramsChristian Puchert, Operations Research, RWTH Aachen University, Kackertstr. 7, 52072,Aachen, Germany, [email protected], Marco LübbeckeIn the field of production and planning, one is often encountered with a multi-level optimization problem,such as e.g. lot sizing or resource allocation. When solving such a problem as a MIP, the multiple stagesare then often reflected by the MIP formulation. Its constraint matrix then often bears a particular structure,a so-called staircase structure.

In this talk, we transfer heuristic approaches for solving multi-level problems to their MIP model, andgeneralize these approaches to arbitrary staircase-structured MIPs. These heuristics involve consecutivelyfixing parts of the variables while leaving others free, thus breaking down the MIP into several smallerones which are expected to be easier to solve.

3 - Recent Improvements in IBM ILOG CPLEX Optimization StudioTobias Achterberg, CPLEX Optimization, IBM, Zuse Institute Berlin, Takustr. 7, 14195, Berlin,Germany, [email protected] present new features that have been recently added to IBM ILOG CPLEX Optimization Studio andgive detailed benchmarking results that demonstrate the continuous performance improvements in CPLEXsince version 6.5.

196

FA-17

� FA-17Friday, 9:00-10:30New Orleans Room

Nonlinear Optimization and ApplicationsStream: Continuous OptimizationInvited sessionChair: Mirjam Duer, Mathematics, University of Trier, 54286, Trier, Germany, [email protected]

1 - Computing accurate solutions for linear programming models from computationalbiologyAmbros Gleixner, Optimization, Zuse Institute Berlin (ZIB), Takustr. 7, 14195, Berlin,Germany, [email protected], Joshua Lerman, Matthias Miltenberger

Linear programming is becoming a popular tool to model biological systems, in particular for bacterialgrowth (see, e.g., Lewis et al. 2012). The computational demands involved are increasing as problemsare expanding both in size and scope. One example are unifying models including metabolism and geneexpression (Lerman et al. 2012, Thiele et al. 2012), which take as input the nutritional environment andproduce predictions for the cell’s maximum growth rate, inputs and outputs to the cell, metabolic fluxes,and gene expression levels. One core difficulty is that these models describe events at multiple scales (bothin terms of concentrations and rates) that span several orders of magnitude.

As a result, standard linear programming solvers using double-precision struggle to produce solutions ofsufficient accuracy or even to reliably decide feasibility. The condition number of the basis matrices mayrange up to $1010$ while the variables may take physiologically relevant values as small as $10-10$. Wewill discuss methods to overcome these numerical difficulties using an iterative refinement procedure withextended floating-point precision. This produces highly accurate solutions without the need for matrixfactorization in rational arithmetic.

2 - Optimal pricing policy of a Monopolistic Two-Sided NetworkFalk Meyerholz, Optimization, Fraunhofer SCS, Nordostpark 93, 90411, Nuremberg, Germany,[email protected]

In a two-sided market, two user groups interact with each other by means of a shared platform. In amonopolistic market, there is one dominating platform, which holds the majority of market share, such asthe internet auction website EBay. The so-called platform sponsor provides this shared platform and canmake profit by charging user fees. In return the users are provided with the possibility to create added valueby interacting with each other via this platform. The platform sponsors profit depends on the number ofusers, the pricing policy and on the life cycle of business. The number of users depends on the number ofalready participating users and on the pricing policy. Thus, the platform sponsor must choose the optimalpricing policy in order to maximize profit. From a mathematical point of view, the optimal pricing policycan be found by formulating this problem as an optimal control problem and solving it. When applying afinite time horizon, the optimal pricing policy causes a rapid increase of fees at the end of life cycle in orderto maximize profit. Consequently, the user numbers decline rapidly. One way of obtaining a sustainableoptimal pricing policy is to apply an infinite time horizon. However, a major difficulty, that arises whendealing with infinite horizon optimal control problems, is that the transversality condition of the PontryaginMaximum Principle does not commonly apply. Without the transversality condition, it is difficult to finda valid solution. Fortunately, the optimal pricing policy can be found by extending the solution of a finitetime horizon optimal control problem. Another approach is finding the steady state of the market. Bothapproaches will be outlined in this talk.

3 - Advances in Modeling, Optimization and Control of Stochastic Dynamics - Appli-cations in Finance, Economics, Biology and EnvironmentGerhard-Wilhelm Weber, Institute of Applied Mathematics, Middle East Technical University,ODTÜ, 06531, Ankara, Turkey, [email protected], Erdem Kilic, Yeliz Yolcu Okur, FatmaYerlikaya Ozkurt, Azar Karimov, Nadi Serhan Aydin

197

FA-17

This presentation introduces into some recent research achievements in continuous—time models of thefinancial sector and related fields, supported by mathematics. Stochastic Optimal Control plays an increas-ingly important role in science, economics and the sectors of environment and finance, and is extensivelyused in various applications. We present applications of Stochastic Hybrid models in biology, ecology,monetary systems and finance to account for regime switching dynamics. Stochastic models with a motionpart and additionally a jump part are able to capture abrupt fluctuations that are a usual phenomenon ingenetic and environmental networks and in financial markets. These kinds of models allow for more re-alistic investigation of portfolio optimization and utility maximization in financial markets and in genetic,metabolic and ecological interaction. The models comprise portfolio optimization with optimal invest-ment and consumption strategies. Explicit consideration of risk aversion in an optimal investment andconsumption problem allows for optimality conditions that are related to specific risk types in a market. Amore general model for portfolio and gene-environment optimization is established afterwards. In anotherstudy, we develop a new theory of estimating Hurst parameter using conic multivariate adaptive regressionsplines (CMARS) method. Stochastic Differential Equations (SDEs) generated by fractional Brownianmotion (fBm) with Hurst parameter, H, are widely used to represent noisy and real-world problems. Thereason why fBm is preferred in modeling, to other Markov processes is its property of capturing the de-pendence structure of observations. It is, therefore, a more realistic model compared to Markov processes.

4 - Keeping a Learned Society YoungGustav Feichtinger, Institute of Mathematical Methods in Economics, Vienna University ofTechnology, Argentinierstraße 8 /E105-4, IWM/ORCOS, 1040, Wien, Austria,[email protected], Vladimir Veliov, Joshua R. Goldstein, Herbert Dawid

The ageing of learned societies is a problem in many national academies. Faced with rising life expectancy,particularly for older persons, the average age of academy members increases. Another reason for "over-ageing" is an increase in the age at election. In a constant sized organization the annual intake is strictlydetermined by the number of deaths and a statutory retirement age. This leads to a fundamental dilemmaof two conflicting goals of many learned societies: to keep a young age structure while guaranteeing ahigh recruitment rate. We derive an optimal recruitment policy which is bi-polar, i.e. it shifts the recruit-ment partly to younger ages and partly to older ages, decreasing in this way the middle-age recruitment.Although the optimization problem explicitly involves only the average age and the recruitment rate asobjectives, the methodology allows to take implicitly into consideration all other objectives (formal orinformal) used in the election practice in reality.

198

FB-12

Friday, 11:00-11:45

� FB-01Friday, 11:00-11:45Rotterdam Hall

Semi-plenary speaker: Rommert DekkerStream: Semi-plenary SessionsSemi-plenary sessionChair: Rob Zuidwijk, Decision and Information Sciences, RSM Erasmus University, Postbus 1738,3000 DR, Rotterdam, Netherlands, [email protected]

1 - Maritime Transportation sails on the waves of ORRommert Dekker, Erasmus University Rotterdam, Econometric Institute, Erasmus UniversityRotterdam, P.O. Box 1738„ 3000 DR , Rotterdam, Netherlands, [email protected]


� FB-09Friday, 11:00-11:45Diamond Room 1

Semi-plenary speaker: Stein W. WallaceStream: Semi-plenary SessionsSemi-plenary sessionChair: Rüdiger Schultz, Mathematics, University of Duisburg-Essen, Lotharstr. 65, D-47048,Duisburg, Germany, [email protected]

1 - Stochastic Programming: Do we really need it?Stein W. Wallace, Management Science, Lancaster University Management School, LA1 4YX,Lancaster, United Kingdom, [email protected]


� FB-12Friday, 11:00-11:45Van Oldenbarnevelt Room

Semi-plenary speaker: Nelly LitvakStream: Semi-plenary SessionsSemi-plenary sessionChair: Flavius Frasincar, Econometric Institute, Erasmus University Rotterdam, Netherlands,[email protected]

199

FB-13

1 - Algorithms for Web graphs - can we handle big data?Nelly Litvak, Applied Mathematics, University of Twente, P.O.Box 217, 7500 AE, Enschede,Netherlands, [email protected]


� FB-13Friday, 11:00-11:45Mees Auditorium

Semi-plenary speaker: Erik DemeulemeesterStream: Semi-plenary SessionsSemi-plenary sessionChair: Florian Jaehn, Sustainable Operations and Logistics, Universitätsstr. 16, 86159, Augsburg,Germany, [email protected]

1 - Robust project schedulingErik Demeulemeester, Decision Sciences and Information Management, Katholieke UniversiteitLeuven, Naamsestraat 69, 3000, Leuven, Belgium, [email protected]


Friday, 12:00-13:00

� FC-01Friday, 12:00-13:00Rotterdam Hall

Closing and Plenary: Steve Sashihara (see page 27)Stream: Plenary SessionsPlenary sessionChair: Albert Wagelmans, Econometric Institute, Erasmus University Rotterdam, POBox 1738, 3000DR , Rotterdam, Netherlands, [email protected]

200

AUTHOR INDEX

AAchatz, Hans TE-15Achterberg, Tobias FA-15Adan, Ivo WC-17Afsharian, Mohsen WC-14Agatz, Niels FA-14Aghezzaf, El-Houssaine

WE-15Ahn, Heinz WC-14Akkermans, Henk FA-03Almeder, Christian TA-08,

WE-09Alonso Martínez, Maria

Teresa TB-11Althoefer, Ingo WB-17Alvarez-Valdes, Ramon TB-

11Alvelos, Filipe WC-05Amberg, Bastian TE-06Amberg, Boris TE-06Amin Naseri, Mohammad

Reza TA-10Anjos, Miguel WD-01Araújo, Olinto WC-13Arslan, A. Muzaffer TB-02Arts, Joachim TB-09,

WE-11Awasthi, Anjali WB-12Aydin, Nadi Serhan FA-17

BBabrowski, Sonja TA-03Bach, Lukas WE-13Bakhrankova, Krystsina

WE-12Ballestin, Francisco WC-08Bartl, David TE-17Basten, Rob TB-09, WC-09Bastubbe, Michael TB-15Bauer, Thomas TB-11Baumann, Philipp WC-08Bauschert, Thomas TA-15Beckmann, Lars FA-06Bekker, Rene FA-05Bender, Matthias TE-11Benoist, Thierry TA-06Benth, Fred Espen WD-09

Berger, Martin WE-05Berger, Theo WB-17Bergner, Martin TB-13Berthold, Timo WE-16Bertsch, Valentin WE-03Bichler, Martin WB-04Bierlaire, Michel FA-07Bierwirth, Christian TE-05,

TB-08, WE-12Bijvank, Marco WE-10Bispo, Carlos WC-10Blanco, Marco WE-13Blazewicz, Jacek TA-07Block, Joachim WC-04Bloemendal, Chagiet WB-

12Bloemhof, JacquelineFA-10Bock, Stefan TB-08Borenstein, Denis WC-13Borndörfer, Ralf TA-13,

WB-13Bouman, PaulTA-12, TB-15Bouvry, Pascal TA-07Boysen, Nils WB-07Brakonier, Pascal TE-11Brandenburg, Marcus WC-

11Braune, Roland TE-08Brazeau, Mélisende FA-05Breier, Heiko TA-14Breitmoser, Katja WE-03Breitner, Michael H. WB-03Brinkhuis, Jan TA-17Briskorn, Dirk WB-07, 08,

TA-12Broekmeulen, Rob WB-10Brunner, Jens TB-05Bruns, Florian TB-07Buer, Tobias WE-07Buhayenko, Viktoryia TE-

09Buke, Burak FA-03Bulavsky, Vladimir WE-04Burger, Mernout FA-15Büsing, Christina FA-05,

TA-15

Bussieck, Michael TA-06,TB-06

Butsch, Alexander WC-11

CCacchiani, Valentina FA-13,

WB-15Cano Belmán, Jaime TA-10Cardoso Dias, BrunoWB-15Carlo, Hector FA-07Castillo Perez, Felipe de Je-

sus WE-04Cattrysse, Dirk WB-13Cárcaba, Ana WB-05Chauhan, Satyaveer S. WB-

12Cheung, WIlliam TE-14Cijvat, Robin TB-08Coene, Sofie TA-12Constantino, Miguel WC-05Cortinhal, Maria TB-12Crama, Pascale WB-08Crama, Yves WB-08Cui, Jian TA-10

DD’Andreagiovanni, Fabio

TA-15Dabia, Said TA-14Daduna, Hans WC-17Daduna, Joachim R. TB-14Dahms, Florian FA-06Darlay, Julien TA-06Dasci, Abdullah TB-12Dashty Saridarq, Fardin FA-

08Davarzani, Hoda WC-12Dawid, Herbert FA-17De Causmaecker, Patrick

WE-07de Jong, Franciska WE-06de Jonge, Bram WC-09de Keizer, Bart FA-13de Koster, René FA-07, TE-

12, FA-14, WC-17de Mare, Rutger WC-07de Vries, Harwin WE-05

201

AUTHOR INDEX

Decouttere, Catherine TB-05

deFreitas, Rosiane WE-08,TE-13, WB-15

Degel, Dirk TE-05Dekker, Rommert FB-

01, WC-07, WB-11, WE-11, TE-14

Della Croce, FedericoTA-13Dellnitz, Andreas WC-14Demeulemeester, Erik FB-

13den Boer, Arnoud TE-02den Hertog, Dick TD-13Denz, Florian TB-03Desrosiers, Jacques WB-15Dewilde, Thijs WB-13Dieter, Weninger WC-11Dinges, Andreas WB-14Dirkse, Steve TA-06Dochow, Robert WE-04Doerner, Karl TE-08Dokka, Trivikram WB-08Dollevoet, Twan WE-05,

WE-13Doppstadt, ChristianWB-11Dreves, Axel TA-17Driessen, Theo WE-04Drozdowski, Maciej TA-07Duer, Mirjam WE-15,

TB-17Dünnwald, Bastian FA-06Dussault, Benjamin TE-15

EEderer, Thorsten TA-08Ekelhart, Andreas WC-06Elbek, Maria TE-10Elizondo-Amaya, Monica

WE-17Emde, Simon WB-07Eren Akyol, Derya FA-14Erera, Alan FA-07Estellon, Bertrand TA-06

FFacchinei, Francisco TA-17Fadaei, Salman WB-04Fandel, Günter WE-04

Fedtke, Stefan WB-07Feichtinger, Gustav FA-17Fendek, Michal TE-02Fendekova, Eleonora TE-02Fichtner, Wolf TA-03,

WE-03Fioole, Pieter FA-13Fischer, Andreas TA-17Fischer, Kathrin TE-10Flapper, Simme Douwe TA-

09Fleischmann, BernhardWB-

09Fleischmann, MoritzTB-10,

WC-12Fontaine, Pirmin TB-13Forrer, Salome WC-08Frank, Stefan TE-14Fransoo, Jan C. WE-10Frasincar, Flavius WE-06Frenk, J.b.g. TB-02Fröhling, Magnus TA-03Furini, Fabio TB-15Furuta, Takehiro TA-12

GGabor, Adriana F. WC-10,

WE-11Gahm, Christian FA-06Gamrath, Gerald WC-11Gansterer, Margaretha WE-

09Gardi, Frédéric TA-06Garg, Seema‘ TB-04Gauthier, Jean-Bertrand

WB-15Geier, Sebastian WB-09Geiger, Martin Josef WE-14Geissler, Bjoern WE-03Geldermann, Jutta WC-06Genç, Asır TB-08, TA-11Gharehgozli, Amir Hossein

FA-07Ghosh, Diptesh WC-15Gijben, Luuk WE-15Gleixner, Ambros FA-17Glorie, Kristiaan WC-05,

WE-05Goel, Asvin FA-08

Goeree, Jacob WB-04Golden, Bruce TE-15Goldstein, Joshua R. FA-17Gong, Yeming TB-10Gönsch, Jochen FA-02,

TE-02Gonzalez, Eduardo WB-05Goossens, Dries TA-13Gorski, Jochen WE-14Görtz, Thomas FA-02Gössinger, Ralf TA-11Gossler, Timo TA-14Gotzes, Uwe TE-03Grad, Sorin-Mihai TE-17Grill, Bernhard WC-06Gromicho, Joaquim TB-11,

WB-12Gronalt, Manfred TE-09Grothmann, Ralph WB-06Gschwind, Timo FA-14Guedes, Pablo WC-13Guo, Xiaolong TE-12Gupta, Alok WC-06Gupta, Pankaj KumarTB-04Guritno, Adi Djoko TA-09,

TB-09, FA-10Gurski, Frank TE-13

HHabenicht, Walter WE-14Hagen, Hans WB-14Hahn, Gerd J. WC-12Haijema, René WB-10Hallek, Sandra TA-11Hallmann, Corinna TA-03Hao, Zhen WB-04Hartl, Richard TA-

08, TE-08, WE-09,FA-10

Hartman, Bruce FA-11,TE-11

Hartmann, Jens WE-12Hassler, Michael FA-02,

TE-02Hayashi, Hiroki FA-08Heid, Werner TE-11Heidergott, Bernd FA-03,

TA-04Heismann, Olga TA-13

202

AUTHOR INDEX

Helm, Jonathan FA-05Hendrix, Eligius M.T. WB-

10Herde, Felix FA-02Herrich, Markus TA-17Hertz, Alain FA-05Hezarkhani, Behzad FA-11Hildenbrandt, Achim TA-13Hildenbrandt, R. WE-17Hillebrand, Bernd WE-10Hoeksma, Ruben WE-08Hoffmann, Kirsten WB-16Hofmann, Petar WB-17Hogenboom, Alexander

WE-06Hogenboom, Frederik WE-

06Holzhauer, Falk TA-09Hombach, Laura Elisabeth

WB-03Hoogeveen, Han TE-06,

TB-08, TB-15Hoogland, Hans TA-09Hu, Bo FA-06Huang, Jia-Ping FA-03Huber, Sandra WE-14Huchzermeier, Arnd TB-10Hughes, Max TE-04Huijink, Sybren FA-12Huisman, Dennis WC-07,

FA-13, WE-13Huka, Maria Anna TE-09Hurink, Johann TB-05,

TE-05, WB-08

IIchoua, Soumia TB-03Iida, Yasuhiro TA-17Inderfurth, Karl WC-10Irnich, Stefan WC-13,

TB-15

JJagtenberg, Caroline TA-05Jaksic, Marko WE-10Jargalsaikhan, Bolor TB-17Jochem, Patrick TA-03Jouini, Oualid WE-17Ju, Wanrong WC-10

Juenger, Michael WB-15

KKabirian, Alireza WE-12Kalashnikov, Vyacheslav

WE-04Kalashnykova, Nataliya

WE-04Kalcsics, Jörg TE-11,

WC-11Kalkowski, Sonja TA-11Kant, Goos FA-12Karimov, Azar FA-17Kasperski, Adam TE-13Kaufmann, Corinna TE-07Kaymak, Uzay WE-06Kellenbrink, Carolin TE-07Ketter, Wolfgang WC-06Kharoufeh, Jeffrey WC-09Kidd, Martin FA-13Kiesling, Elmar WC-06Kiesmüller, Gudrun WC-10Kilic, Erdem FA-17Kimms, Alf WB-11, TA-14Kirchhoff, Fabian FA-13Kirchner, Sarah FA-05Kirschstein, Thomas WE-12Klamroth, Kathrin WE-14,

15Klein, Robert TE-02, TB-06Kleine, Andreas WC-14Kleinschmidt, Peter TE-15Kliewer, Natalia TE-06Klimentova, Xenia WC-05Klimm, Max WC-16Klingenberg, Warse WC-09Klosterhalfen, Steffen TA-

09Klumpp, Matthias WE-09Knust, Sigrid TB-07Koberstein, Achim WB-11Kok, Leendert WB-12Komay Maia, Leonardo

Kenji TA-09Koole, Ger FA-05, WE-17Kopfer, Herbert WE-07Koster, Arie WE-03, TA-15Koukal, André WB-03Kowalski, Dawid TA-07

Kozeletskyi, Igor WB-11Krenzler, Ruslan WC-17Kroon, Leo FA-07,

TA-12, TD-12, FA-13, WB-13

Kropat, Erik FA-03, WC-04Kubiak, Wieslaw FA-11Kuefer, Karl-Heinz WB-14Kuemmel, Sherko WB-14Kuhpfahl, Jens TB-08Kutanoglu, Erhan WE-17Kutschka, Manuel TA-15Kwanashie, Augustine TE-

13

LLach, Gerald FA-06Laengle, Sigifredo WC-04Lagershausen, SvenjaFA-09Lamballais Tessensohn, Tim

FA-09Lange, Anne WB-09Lange, Stefan WB-03Laporte, Gilbert WC-11Larsen, Christian WB-03Larsen, Jesper WC-07Laumanns, Marco WE-11Laurent, Monique TB-17Lavieri, Mariel WE-16Legros, Benjamin WE-17Lehnfeld, Jana TB-07Lemkens, Stephan WE-03Lenz, Ralf TE-03Leopold, Armin FA-06Lerman, Joshua FA-17Leszner, Mateusz TA-07Letizia, Paolo FA-02Levit, Vadim WC-15Li, Dong WC-12Lieder, Alexander WB-08Liers, Frauke WB-15Liiv, Innar WC-14Lindner, Ines FA-03Lino, M.Pilar WC-08Listes, Ovidiu TB-06Litvak, Nelly FB-12Lodewijks, Gabriel TE-14Lodi, Andrea TD-10,

WB-15

203

AUTHOR INDEX

Lohmann, Christian WB-05Lopes, Maria João TB-12Lorenz, Ulf TA-08Loyola, Gino WC-04Lu, Yixin WC-06Lübbecke, MarcoFA-05, 06,

TB-13, FA-15, TB-15, WB-15

Lusby, Richard WC-07Lutter, Pascal FA-13

MMaculan, Nelson WB-15Maes, Christopher TA-06Mahvash, Batoul WB-12Maillart, Lisa WC-09Makuschewitz, Thomas

WE-12Mandrescu, Eugen WC-15Mankowska, Dorota Slawa

TE-05Manlove, David WC-05,

TE-13Maroti, Gabor TA-12,

WB-13Marszałkowski, Jakub TA-

07Martin, Alexander WE-03Mashreghi, Hamid TA-10Matsukawa, Yuki FA-08Mattfeld, Dirk ChristianTE-

06Mayer, Stefan WB-04McBride, Iain TA-13Megel, Romain TA-06Meisel, Frank TE-05,

WE-12Mellouli, Taieb TB-14Melo, Teresa TB-12Menkens, Olaf FA-04Meyer, Anne TE-11Meyer, Christoph TB-03Meyer-Nieberg, SiljaFA-03,

WC-04Meyerholz, Falk FA-17Meyr, Herbert TA-10Midthun, Kjetil WE-12Miltenberger, Matthias WC-

11, FA-17

Minner, Stefan FA-10,WB-10, TB-13

Mirzaei, Masoud TE-12Mishra, Nishant WC-12Mitra, Subrata WC-13Miyagawa, Masashi FA-14Mizgajski, Jan TA-07Moeke, Dennis FA-05Mohr, Esther WE-04Mokhtarimousavi, Seyed-

mirsajad TB-14Mokwa, Dariusz TA-07Morito, Susumu FA-08Morsy, Ehab WE-08Morton, David WE-17Mulder, Judith WC-07Musial, Jedrzej TA-07

NNachtigall, Karl TE-14Nagai, Ko WC-12Naundorf, Jessica WB-06Negenborn, Rudy TE-14Neugebauer, Michael TE-02Nickel, Stefan TE-05, TB-

06, TE-11, WC-11Nigam, Shobhit TE-12Nishihara, Michi TE-04Nossack, Jenny WC-16

OOhno, Takahiro WC-12,

TA-17Ohst, Jan Peter WB-15Okamoto, Shigeo FA-08Okhrin, Irena TE-10Olde Keizer, Minou TB-09OMalley, Gregg WC-05Opfer, Thomas TA-08Oppen, Johan FA-14Otten, Sonja WC-17Otto, Alena WC-16

PPaetz, Tobias TA-08Panchamgam, Kiran TE-15Pape, Christian WC-03Parada Puig, Jorge Eduardo

TB-09, WC-09

Parreño, Francisco TB-11Passchyn, Ward TA-12Pauls-Worm, Karin G.J.

WB-10Pawlak, Grzegorz TA-07Pérez, M. Angeles WC-08Pechmann, Agnes TB-03Pedroso, Joao Pedro WC-05Peeters, René FA-12Peng, Hao TB-09, WC-09Pennings, Clint WB-05Pfeiffer, Jella FA-02Plohakova, Lenka TE-17Pop, Emilia-Loredana TE-

17Post, Gerhard WB-08,

TB-11Pourakbar, Morteza WE-07Preis, Henning TE-14Preuß, Michael WC-04Prischepov, Natalia TE-07Puchert, Christian FA-15,

TB-15Puttkammer, Karen FA-09

QQuintanilla, Sacramento

WC-08

RRaack, Christian WC-16Radulescu, Constanta Zoie

FA-04Radulescu, Marius FA-04Radulescu, Sorin FA-04Radzik, Tadeusz WB-04Rahami, Hossein TB-14Raikar, Neha TA-09Rais, Abdur WC-05Rathjen, Christian TB-08Raths, Stephan WC-03Raymond, Annie TA-15Reindorp, Matthew WB-09Reinelt, Gerhard TA-13Reiners, Christoph TA-14Renatus, Fabian WC-06Resing, Jacques WC-17Rethmann, Jochen TE-13

204

AUTHOR INDEX

Reuter, Melanie TE-05,TB-06

Reuther, Markus WB-13Rex, Torsten TB-04Richter, Knut TE-10Rifki, Omar FA-04Rios-Mercado, Roger Z.

WE-17Rischke, Roman WB-16Rödder, Wilhelm WC-14Römer, Michael TB-14Roodbergen, Kees Jan FA-

07Roy, Debjit TE-12Ruediger, Patrick WB-14Ruzika, Stefan WE-14,

WB-15, WE-15

SSadegh Vaziri, Ezaz TB-14Saffarzadeh, Mahmoud TB-

14Sahin, Guvenc FA-08Santos, Clarice TE-13Santos, Nicolau WC-05Sass, Jörn TA-04Sawada, Kiyoshi WC-15Sayah, David TB-02Schacht, Matthias TA-03Schebesch, Klaus Bruno

WE-06Scheller-Wolf, Alan WB-09Scherrer, Alexander WB-14Schlapp, Jochen TB-10Schlechte, Thomas WB-13,

WE-13Schlenker, Hans TB-06Schlutter, Stefanie TB-11Schmidt, Daniel WB-15Schmidt, Günter WE-04Schmidt, Henning FA-02Schneider, Michael TA-14Schnell, Alexander TE-08Schnitzler, Daniel TA-08Schoendube, Barbara FA-10Scholz-Reiter, BerndWE-12Schönberger, Jörn WE-07,

WB-11Schramme, Torben TE-08

Schroeder, Michael WE-05Schülldorf, Hanno WE-13Schulte Beerbühl, Simon

TA-03Schulte, Frederik FA-07Schultmann, Frank TA-03Schultz, Rüdiger TE-03Schulz, Katrin TA-03Schulze, Marco FA-08Schwab, Andreas FA-03Schwahn, Fabian TA-14Schwarz, Hannes WE-03Schwarz, Robert TE-03Schwidde, Ilka WB-14Schwindt, Christoph TA-08Seipp, Florian WE-15Selinka, Gregor FA-09Sels, Peter WB-13Serin, Andreas WE-10Sevaux, Marc WE-14Sezer, Semih TB-02Shabalin, Pash WB-04Sharifyazdi, Mehdi WC-12Shibata, Takashi TE-04Shiina, Takayuki FA-11Siefen, Kostja FA-06Silvestri, Francesco TA-13Singh, Gaurav TA-07Smit, Gerard WB-08Spangler, Manuela TE-04Spengler, Thomas TB-03,

FA-09Spieksma, Frits WB-08,

TA-12, 13Spliet, Remy WC-07,

WB-11Srinivasan, Mahesh WC-10Srour, Jordan FA-14Stangl, Claudia TE-03Stapel, Florian FA-06Stecking, Ralf WE-06Steglich, Uwe TA-15Steinhardt, Claudius TB-06Steinke, Leena TE-10Stephan, Konrad WB-07Stiglmayr, Michael WE-15Still, Georg TB-17Stindt, Dennis WB-14Sting, Fabian WB-08

Stolletz, Raik WB-08,FA-09

Straßberger, Mario TB-04Strauss, Christine WC-06Stummer, Christian WC-06Suhl, Leena TE-08Summerauer, Wolfgang TA-

08Sun, Zhao TB-17Suwondo, Endy TA-09,

TB-09, FA-10

TTakahashi, Kei WC-12,

TA-17Talluri, Kalyan TD-09Tanaka, Ken-ichi TA-12Taniguchi, Nao FA-08Tanrisever, Fehmi WB-09Taskhiri, Mohammad

Sadegh TB-12Tempelmeier, Horst WD-13Teunter, Ruud TB-09,

WC-09Thijssen, Nort TE-11Thomas, Douglas WC-10Thome, Annika TB-13Tietz, Christoph WB-06Tilk, Christian WC-13Tinga, Tiedo WC-09Tolstad Uggen, Kristin WE-

12Tonissen, Denise TB-15Torigoe, Atsushi FA-08Toth, Paolo FA-13Trautmann, Norbert WC-08Trockel, Jan WE-04Tubertini, Paolo WC-05Turan, Belma FA-10Turkensteen, Marcel WB-03

UUbi, Evald WC-14Ubi, Jaan WC-14Uchoa, Eduardo WE-08Uetz, Marc WE-08Umang, Nitish FA-07Unger, Stephan TE-14

205

AUTHOR INDEX

VValls, Vicente WC-08van ’t Wout, Joel FA-13van Barneveld, Thije TA-05van Buuren, Martin TA-05van de Klundert, J.j.WC-05,

WE-05van den Akker, J.m. TE-06,

TB-08, TB-15van den Berg, Pieter TA-05van den Hurck, Dave TE-11van der Gaast, Jelmer WC-

17van der Klauw, ThijsWB-08van der Vorst, Jack FA-10,

WB-10van Dongen, Leo WC-09Van Dongen, Thijs TE-11van Donselaar, KarelWB-10Van Eikenhorst, DirkTE-09,

WE-09van Elzakker, MartijnTA-09van Essen, Theresia TB-05,

TE-05van Heck, Eric WC-06van Hoorn, Jelke TB-11van Houtum, Geert-Jan TB-

09, WE-11van Jaarsveld, Willem WB-

09van Kooten Niekerk, Marcel

TE-06van Ommeren, Jan-Kees

WC-10, WE-11van Oosterom, ChielWC-09Van Oyen, Mark FA-05Van Riessen, Bart TE-14Van Woensel, Tom TA-14Vancroonenburg, Wim TA-

13Vandaele, Nico TB-05Vanden Berghe, Greet WE-

07Vandic, Damir WE-06Vansteenwegen, Pieter WB-

13Veelenturf, Lucas FA-13

Veenstra, Albert WE-07Veenstra, Marjolein FA-12Veliov, Vladimir FA-17Ventura, Juan WB-05Verleye, Derek WE-15Verstichel, Jannes WE-07Vervest, Peter TA-12Verwer, Sicco TE-15Viana, Ana WC-05Vidal, Thibaut FA-08Vigerske, Stefan TA-06Vigo, Daniele WB-11Vilhelmsen, Charlotte WC-

07Vis, Iris F.A. FA-07Visentini, Monize Sâmara

WC-13Vogel, Patrick TE-06Vohandu, Leo WC-14Voigt, Guido FA-10Vossen, Thomas TB-02

WWagelmans, Albert WC-05,

WE-05Wagenaar, Joris WB-13Wagner, Stephan WE-11Waldherr, Stefan TB-07Wallace, Stein W. FB-09Walther, Grit WB-03Wanke, Egon TE-13Wasik, Szymon TA-07Wasil, Edward TE-15Wøhlk, Sanne TE-10Weber, Christoph WC-03Weber, Gerhard-Wilhelm

WC-03, FA-17Werner, Ralf TB-04, TE-04Werners, Brigitte TA-03,

TE-05, FA-13Westermann, Lutz TB-06Westphal, Clemens TB-06Wichmann, Matthias Ger-

hard TB-03,FA-09

Wiesche, Lara TE-05,WB-16

Woerner, Stefan WE-11

Wolfgang Pickl, StefanWC-04

Wolfler-Calvo, Roberto TB-15

Woll, Oliver WC-03Wruck, Susanne WE-17Wu, Yaozhong WB-08

XXavier de Amorim, Rainer

WE-08Xiong, Yupei TE-15

YYan, Yang TE-04Yang, Guangyuan WE-11Yanikoglu, Ihsan WB-17Ye, Qing Chuan TE-15Yeganehfallah, Azarm TA-

10Yerlikaya Ozkurt, FatmaFA-

17Yilmaz, Tarik TB-08, TA-11Ying Wen, Chen FA-08Yıldırım, Miray HanımWC-

03Yolcu Okur, Yeliz FA-17Ypsilantis, Panagiotis WE-

07Yu, Yugang TE-12Yuliando, Henry TA-09,

TB-09, FA-10

ZZaerpour, Nima WE-11Zeise, Philipp TE-11Zhang, Dan TB-02Zhang, Yingqian TE-15Zhu, Qiushi TB-09Zielinski, Pawel TE-13Zijm, Henk WD-16Zimmermann, Adrian WC-

08Zimmermann, Hans Georg

WB-06Zimmermann, Jürgen FA-08Zondervan, Edwin TA-09Zuidwijk, Rob WE-07

206

SESSION CHAIR INDEX

AAchterberg, Tobias FA-15Akkermans, Henk FA-03Arts, Joachim WE-11

BBerger, Theo WB-17Bichler, Martin WB-04Bloemhof, JacquelineFA-10Boysen, Nils WB-07Brandenburg, Marcus WC-

11Braune, Roland TE-08Breitmoser, Katja WE-03Briskorn, Dirk WB-08Bussieck, Michael TA-06

CCano Belmán, Jaime TA-10

Dde Koster, René TE-12,

FA-14, WC-17de Mare, Rutger WC-07deFreitas, Rosiane TE-13Dekker, Rommert TE-14,

WD-16den Boer, Arnoud TE-02Denz, Florian TB-03Dewilde, Thijs WB-13Dollevoet, Twan WE-05Driessen, Theo WE-04Duer, Mirjam WD-01,

FA-17, TA-17, TB-17, TE-17

FFlapper, Simme Douwe TA-

09Frasincar, Flavius FB-12

GGeiger, Martin Josef WE-14Ghosh, Diptesh WC-15Glorie, Kristiaan WC-05Golden, Bruce TE-15Gönsch, Jochen FA-02Gössinger, Ralf TA-11Gromicho, Joaquim TB-11

Grothmann, Ralph WC-06

HHahn, Gerd J. WC-12Hartman, Bruce TE-11Hartmann, Jens WE-12Heidergott, Bernd TA-04Hezarkhani, Behzad FA-11Hombach, Laura Elisabeth

WB-03Hoogeveen, Han TB-08Huchzermeier, Arnd TB-10Huisman, Dennis TD-13Hurink, Johann TB-05,

WE-16

IInderfurth, Karl WC-10Irnich, Stefan WC-13

JJaehn, Florian FB-13Jaksic, Marko WE-10

KKant, Goos FA-12Kaufmann, Corinna TE-07Kimms, Alf TA-14, WC-16Klamroth, Kathrin WB-16Klein, Robert TD-09Kleine, Andreas WC-14Klumpp, Matthias WE-09Knust, Sigrid TB-07Koberstein, Achim WB-11Kok, Leendert WB-12Koole, Ger FA-05, WE-17Koster, Arie TA-15Kropat, Erik WC-04

LLamballais Tessensohn, Tim

FA-09Listes, Ovidiu TB-06Lübbecke, Marco WB-15

MMattfeld, Dirk ChristianTE-

06Mellouli, Taieb TB-14

Melo, Teresa TB-12Menkens, Olaf FA-04Minner, Stefan WB-10,

TA-12, TB-13

NNecil, Jan TE-09Nickel, Stefan TD-01,

TE-05

OOkhrin, Irena TE-10Otto, Alena TA-08

PPawlak, Grzegorz TA-07Pesch, Erwin TD-10Poppelaars, John WE-16

RRaths, Stephan WC-03Römer, Michael WE-13

SSalomon, Marc TC-01Sass, Jörn WD-09Scherrer, Alexander WB-14Schimmelpfeng, Katja WB-

05Schlechte, Thomas FA-13Schulte, Frederik FA-07Schultz, Rüdiger FB-09Schulz, Katrin TA-03Schulze, Marco FA-08Siefen, Kostja FA-06Spieksma, Frits TA-13Stangl, Claudia TE-03Stecking, Ralf WE-06Suhl, Leena TD-12

TTrautmann, Norbert WC-08

UUetz, Marc WE-08

Vvan den Akker, J.m. TB-15van den Heuvel, Wilco WD-

13

207

SESSION CHAIR INDEX

van der Mei, Rob TA-05van Houtum, Geert-Jan TB-

09, WC-09van Jaarsveld, Willem WB-

09Verleye, Derek WE-15

Vossen, Thomas TB-02

WWagelmans, Albert FC-01,

WA-01Walther, Ursula TB-04Werner, Ralf TE-04

ZZimmermann, Hans Georg

WB-06Zuidwijk, Rob FB-01,

WE-07

208

Documents

SESSIONS AND TALKS DAY BY DAY - EURO...When the slower transportation mode is less-polluting, higher inventory levels need to be kept. Both the choice for a clean but slow transportation