5
978-1-4244-2289-0/08/$25.00 ©2008 IEEE Abstract— In this paper, we present a framework for collaborative networks in the long-haul truckload trucking industry. To help preserving members’ privacy, a distributed design is proposed for the network. Moreover, decisions about when to engage a collaboration and how much information to reveal is totally left to the network participants. The proposed framework integrates a dynamic optimization base that intents to achieve a good trade-off between local optimization representing each network participant personal goals and global optimization representing the network global goals. Issues related to the implementation of the proposed solution approaches in a dynamic setting are also discussed. Index Terms—Collaboration, Competition, Distributed design, Truckload trucking. I. INTRODUCTION he increased interest in just-in-time distribution systems has recently led to bigger and more frequent shipments for the trucking industry but it also has accentuated the problem of deadheads. In the trucking industry, deadheads are a common source of inefficiency. A deadhead occurs when a truck travels empty because a return shipment could not be scheduled resulting in a waste of energy, useless operating costs and unnecessary use of highways. According to National Statistics, 30% of all trucks on the roads are empty [5]. On the other hand, the U.S. Environmental Protection Agency indicates that for a typical long-haul truck up to 15,000 miles each year may be non-revenue empty miles, consuming over 2,400 gallons of diesel fuel and producing 24 metric tons of carbon dioxide, the most prevalent greenhouse gas. In the past few years, there has been an increased interest in building freight transportation networks that help overcoming these shortcomings (e.g. Transplace in the US). A freight transportation network is a web/information platform generally created by independent trucking companies that intent to increase their efficiency by optimizing their operation costs and resource utilization and by minimizing empty mileage through collaboration. In fact, collaboration allows gathering a larger demand (set of loads) and a larger capacity Manuscript received February 1, 2008. S. Ichoua is with the Department of Computer Science & Engineering, Johnson C. Smith University, Charlotte, NC 28216 USA (phone: 704-378- 1088, e-mail: [email protected]). (available drivers and trucks) which increases the number of possibilities to match loads to empty vehicles in order to minimize total empty miles. This kind of collaboration seem to be a judicious solution for the trucking industry given that 70% of the companies operate 6 or fewer trucks and spend as much as 20 % of their revenues on purchase of fuel in addition to facing a penury in drivers [5]. The emergence of these networks was possible because of the spectacular growth in information technology (e.g. exchange of electronic data and availability of real-time information at affordable costs). However most of the freight transportation networks that are actually implemented have the following shortcomings: The decision-making is centralized which compromises the independence of the shareholder companies, The decision of adhesion to the network is not easily reversible since shareholders invest in the optimization base and reveal strategic information to the platform manager. A significant percent of the transportation requests are typically pre-assigned by contracts and therefore, the relevance to take part in such a network is questionable. These factors discourage many companies from joining freight transportation networks which may prevent these networks from having a significant impact on carriers’ performance since gathering a significant capacity and demand is essential to the functioning of transportation networks. In this paper, we present an on-going work which proposes a collaborative distributed design for freight transportation networks that will accurately take into account the particularities and challenges of the freight transportation industry: Trucking industry is competitive by nature (carriers do not want to give the decision power to other hands. Moreover, a significant percent of the transport contracts are often pre-negotiated in advance). Collaboration helps increasing the efficiency by minimising empty miles through load exchanges and consequently improving resource utilization and lowering operating costs. On the other hand, it allows access to a larger market (share market). Our design integrates a dynamic optimization base that intents to find a good trade-off between local optimization representing each shareholder personal goals and global optimization representing the network global goals. A Collaborative-Distributed Framework for the Long-Haul Truckload Trucking Industry Soumia Ichoua T

[IEEE 2008 IEEE International Engineering Management Conference (IEMC-Europe 2008) - Estoril, Portugal (2008.06.28-2008.06.30)] 2008 IEEE International Engineering Management Conference

  • Upload
    soumia

  • View
    213

  • Download
    1

Embed Size (px)

Citation preview

Page 1: [IEEE 2008 IEEE International Engineering Management Conference (IEMC-Europe 2008) - Estoril, Portugal (2008.06.28-2008.06.30)] 2008 IEEE International Engineering Management Conference

978-1-4244-2289-0/08/$25.00 ©2008 IEEE

Abstract— In this paper, we present a framework for

collaborative networks in the long-haul truckload trucking industry. To help preserving members’ privacy, a distributed design is proposed for the network. Moreover, decisions about when to engage a collaboration and how much information to reveal is totally left to the network participants. The proposed framework integrates a dynamic optimization base that intents to achieve a good trade-off between local optimization representing each network participant personal goals and global optimization representing the network global goals. Issues related to the implementation of the proposed solution approaches in a dynamic setting are also discussed.

Index Terms—Collaboration, Competition, Distributed design, Truckload trucking.

I. INTRODUCTION

he increased interest in just-in-time distribution systems has recently led to bigger and more frequent shipments for the trucking industry but it also has accentuated the

problem of deadheads. In the trucking industry, deadheads are a common source of inefficiency. A deadhead occurs when a truck travels empty because a return shipment could not be scheduled resulting in a waste of energy, useless operating costs and unnecessary use of highways. According to National Statistics, 30% of all trucks on the roads are empty [5]. On the other hand, the U.S. Environmental Protection Agency indicates that for a typical long-haul truck up to 15,000 miles each year may be non-revenue empty miles, consuming over 2,400 gallons of diesel fuel and producing 24 metric tons of carbon dioxide, the most prevalent greenhouse gas. In the past few years, there has been an increased interest in building freight transportation networks that help overcoming these shortcomings (e.g. Transplace in the US). A freight transportation network is a web/information platform generally created by independent trucking companies that intent to increase their efficiency by optimizing their operation costs and resource utilization and by minimizing empty mileage through collaboration. In fact, collaboration allows gathering a larger demand (set of loads) and a larger capacity

Manuscript received February 1, 2008. S. Ichoua is with the Department of Computer Science & Engineering,

Johnson C. Smith University, Charlotte, NC 28216 USA (phone: 704-378-1088, e-mail: [email protected]).

(available drivers and trucks) which increases the number of possibilities to match loads to empty vehicles in order to minimize total empty miles. This kind of collaboration seem to be a judicious solution for the trucking industry given that 70% of the companies operate 6 or fewer trucks and spend as much as 20 % of their revenues on purchase of fuel in addition to facing a penury in drivers [5]. The emergence of these networks was possible because of the spectacular growth in information technology (e.g. exchange of electronic data and availability of real-time information at affordable costs). However most of the freight transportation networks that are actually implemented have the following shortcomings:

� The decision-making is centralized which compromises the independence of the shareholder companies,

� The decision of adhesion to the network is not easily reversible since shareholders invest in the optimization base and reveal strategic information to the platform manager.

� A significant percent of the transportation requests are typically pre-assigned by contracts and therefore, the relevance to take part in such a network is questionable.

These factors discourage many companies from joining freight transportation networks which may prevent these networks from having a significant impact on carriers’ performance since gathering a significant capacity and demand is essential to the functioning of transportation networks. In this paper, we present an on-going work which proposes a collaborative distributed design for freight transportation networks that will accurately take into account the particularities and challenges of the freight transportation industry:

� Trucking industry is competitive by nature (carriers do not want to give the decision power to other hands. Moreover, a significant percent of the transport contracts are often pre-negotiated in advance).

� Collaboration helps increasing the efficiency by minimising empty miles through load exchanges and consequently improving resource utilization and lowering operating costs. On the other hand, it allows access to a larger market (share market).

Our design integrates a dynamic optimization base that intents to find a good trade-off between local optimization representing each shareholder personal goals and global optimization representing the network global goals.

A Collaborative-Distributed Framework for the Long-Haul Truckload Trucking Industry

Soumia Ichoua

T

Page 2: [IEEE 2008 IEEE International Engineering Management Conference (IEMC-Europe 2008) - Estoril, Portugal (2008.06.28-2008.06.30)] 2008 IEEE International Engineering Management Conference

2

The rest of the paper is organized as follows. The second section presents a brief literature review dedicated to freight transportation networks. The third section describes the problem considered in this study. The fourth section presents the proposed distributed design and describes the optimization base. Finally, the fifth section concludes and discusses future and on-going work.

II. LITERATURE REVIEW

In the past few years, different type of groupings of transportation companies have emerged to help carriers facing the new challenges of the market. Châteauvert [6] classifies these type of groupings in five major categories including blackboards, electronic auctions, Third Party Logistics Providers (3PL), collaborative networks and joint ventures. The author also discusses the advantages and shortcomings of each one of them. Boyles [1] and Normandeau [13] discuss several case studies illustrating how some freight transportation groupings were created and how they are managed. All these studies identify achieving critical mass of demand and resources as a critical success factor for freight transportation groupings. However, the existing type of groupings fail to attract many carriers and shippers because of adhesion costs, privacy, impartiality and independence issues besides the high competitive nature of the market ([1], [6]). To help overcoming these problems, some authors have proposed new frameworks and designs for managing collaboration in the freight transportation sector. Fisher et al. ([8], [10]) propose a multi-agent system to model cooperative scheduling based on negotiations within a society of shipping companies in a Less-Than-Truckload trucking (LTL) context where many loads can be consolidated onto the same vehicle. Shipping companies and trucks are represented as agents. Vehicle agents minimize their transportation costs by sequential insertion of orders handled by a contract-net interaction protocol. The insertion of a single service request may require many rounds of negotiations because of capacity constraints. Each company agent can then improve its truck tours based on an auction procedure that exchanges loads between trucks. As mentioned in [12], this approach requires a significant overhead in computation due to more expensive agent communications compared to a centralized solution. This issue is intensified in a dynamic setting where decisions must be made in a changing environment. Moreover, a centralized fast heuristic approach performed by the company agent is likely to result in a better solution. The authors also present a framework for bilateral cooperation where companies can exchange loads to deal with unforeseen events like traffic jams. This is achieved through a peer-to-peer negotiation based on a parameterized bargaining process. It is worth noticing that these parameters, like the desired profit for an order and the minimal profit accepted by an agent need to be determined with caution. However, this issue was not addressed in these papers. The authors reported results on a set of static problems where a single company was considered. However, no experimental results were reported on cooperation between companies.

Burckert et al. ([2], [4]) present TELETRUCK and TELETRUCK-CC, an extended re-implementation of the multi-agent system proposed in [8] and [10] for cooperative scheduling within a single company and bilateral cooperation between companies, respectively. In the proposed re-implementation, the physical transportation objects such as drivers, trucks and trailers are modeled by basic agents which join together to form holons or holonic agents (i.e., agents composed of sub-agents that act in a corporate way). Hence, holonic agents represent for example vehicles with their drivers. Each vehicle holon is headed by a planning agent that coordinates the formation of the vehicle holon and plan its tour. The vehicle holons in turn join to form the company holon which is headed by a company agent that coordinates the overall optimization process. Finally, an additional agent in introduced to coordinate bilateral cooperation between companies. The model proposed in [3] and [9] includes a decommitment strategy where companies can forego an agreement at any time for another more profitable one, at the cost of a pre-negotiated penalty. A similar strategy is also proposed in [11] in case of on-line bidding by agents in a multi-company. A major drawback in using decommitment strategies in the transportation field is the risk to fail meeting customers’ time windows especially in a just-in-time context and in a highly competitive market where the quality of service is critical to companies’ survival. Figliozzi [7] proposes a collaborative mechanism (CM) for shipping companies in a Truckload trucking (TL) context where loads cannot be consolidated onto the same vehicle. Given an estimate of the cost of servicing any shipment, each carrier who sends a shipment to the CM specifies a maximum value that he is willing to pay for this shipment service. The other participants send a minimum value that they are willing to charge for servicing that shipment. The CM then selects the best allocation of carriers to shipments assuming that a maximum of two shipments are received. The implementation of the proposed mechanism is not discussed. In fact, the proposed framework does not take into account the time needed for communication between participants or the time required for performing the necessary computations.

III. PROBLEM DESCRIPTION

We consider a society of freight transportation companies operating in a long-haul truckload trucking (TL) framework. In this context, the problem consists in dynamically assigning vehicles of limited capacity to loads which unfold randomly over time. Each load is characterized by its origin, its destination and its time windows. Hence, a demand may be rejected if it cannot be serviced within a reasonable time. Furthermore, at any time, a vehicle can be either empty or carrying a single load. Therefore, minimizing deadheads or empty miles is critical to lowering operating costs. To survive in a highly competitive market that has low profit margins, companies also need to focus on respecting customers’ time windows to maintain a good quality of service. This may be hard to achieve in a dynamic setting which is prone to

Page 3: [IEEE 2008 IEEE International Engineering Management Conference (IEMC-Europe 2008) - Estoril, Portugal (2008.06.28-2008.06.30)] 2008 IEEE International Engineering Management Conference

3

unexpected events like accidents, vehicle breakdowns, traffic jams, etc… Clearly, collaboration will help these companies minimizing their empty miles, lowering their operating costs, improving their resource utilization and facing unexpected events through load exchanges. However, the trucking industry is competitive by nature since 80 to 90% of the transport contracts are often pre-negotiated in advance. Moreover, the net profit margin in this industry is around 2% and 3 % [5]. Hence, a collaborative framework that would attract enough companies to reach the required critical mass of demand and resources must take into account the particularities and challenges of this industry. The key is to achieve a good trade-off between collaboration and competition. In the next section, we propose a distributed design for a freight transportation network composed of small or medium sized competing truckload trucking companies that collaborate through load and resource exchanges to improve their performance.

IV. NETWORK ARCHITECTURE

In the proposed design, each company is represented by an agent controlled by the company. This agent manages the company internal network composed of transport contracts that were previously pre-negotiated directly with customers (we recall that this kind of loads represents 80 to 90% of the market demand). This agent also manages loads obtained through the exchanges that take place in the network. A coordinator agent collects sporadic shipping requests from erratic customers (we recall that this kind of loads represents up to 20% of the market demand) as well as shipping requests sent by the network participants. A company may send to the coordinator agent information about one of its vehicles which are traveling empty or about one of its shipping requests which cannot be satisfied because of an unexpected event (i.e., an accident, a vehicle breakdown, etc…). The coordinator agent applies a Contract Net Protocol procedure [14] to assign the received loads to network participants that have reported an idle vehicle. Communications between the coordinator agent and the company agents is achieved through an internet connection. In the proposed framework, each agent includes the following modules:

A. Communication module

This module includes sensors that provide information about the environment (e.g. GPS-system which gives the current position of vehicles at any time, notification about the arrival of new service requests, etc…). Moreover, interactions between the company agents and the coordinator agent is handled by a Contract Net Protocol which specifies broadcasting and bidding for loads.

B. Knowledge base module

This module contains information about the transportation requests that are handled by the agent (e.g. time windows and locations), its resources (e.g. vehicles and drivers) and in the case of a company agent, data about its internal network (e.g. a map).

A company agent has two additional modules: Namely, the reactive module and the optimization base module.

C. Reactive module

This module is responsible for updating different parameters as time goes by (e.g. vehicle positions, travel times, vehicle status (idle or in service), etc…). It is also responsible for implementing the plans elaborated by the optimization base module (e.g. next customer to be served by an idle vehicle). Furthermore, the reactive module identifies loads that can no longer be serviced because of an unexpected change and notifies the optimization base module accordingly. In addition, it decides for which loads to bid based on some fast and simple on-line strategies. The selected items are then transmitted to the optimization base module for further refinements.

D. Optimization base module

This module is responsible for dynamically dispatching available vehicles to incoming transportation requests while minimizing empty miles and satisfying vehicle capacity and customers’ time window constraints. It is also responsible for using the bidding mechanism implemented in the network to minimize its empty miles and to maximize the utilization of its resources. These goals are achieved through the use of on-line strategies and optimization procedures that are adapted to cope with time pressure which is inherent to a dynamic setting. Basically, fast on-line strategies are first used to include an incoming event in the solution. A more sophisticated local re-optimization procedure is then performed, until the occurrence of the next new event, to improve the solution quality. These strategies and procedures are the subject of an on-going work. They are briefly outlined in the following. We consider three types of new events: (i) When a local customer request occurs, a fast procedure is first executed to include it in the solution. Then, a more elaborated re-optimization procedure is run, until the occurrence of the next event, to improve the solution quality. If the customer request could not be added to the solution because of the time constraints, it is sent to the coordinator agent who initiates a bidding process and sends the best offer within a fixed amount of time. A previously accepted transportation request which cannot be serviced because of an unexpected event is also handled with the same bidding process. (ii) When a vehicle has finished servicing its current customer, a fast re-optimization procedure is performed to assign idle vehicles to loads that are not picked up yet. (iii) When a load is received from the coordinator, the company agent computes a bid based on its cost estimate of using an idle vehicle to service this load. If the coordinator agent accepts the offer, the load is actually assigned to the idle vehicle. A re-optimization procedure is then executed, until the occurrence of the next new event, to improve the solution quality. The solution approaches outlined above raise several issues: • Given the dynamic context, it is important to maintain the

consistency of the generated schedule with the current environment.

• When a given strategy or procedure is applied at some instant t, it requires some amount of time ∆t. Since the environment is dynamic, a decision based on the situation

Page 4: [IEEE 2008 IEEE International Engineering Management Conference (IEMC-Europe 2008) - Estoril, Portugal (2008.06.28-2008.06.30)] 2008 IEEE International Engineering Management Conference

4

at instant t when the procedure was started, may not reflect the state of the system at (t+ ∆t) when the decision becomes available.

• When computing the bid for a load received from the network coordinator, if the load pickup and delivery points correspond to the current location of the idle vehicle and its destination, respectively, the company agent may offer the service at a very cheap price and still have an acceptable profit margin. However, when the locations of the load and the idle vehicle do not match perfectly, opportunities to make an acceptable profit may still be possible but need to be considered with caution. Two key questions must then be tackled carefully:

o How far can an idle vehicle be diverted from its planned route in order to pick up a partner’s load?

o How long the vehicle driver can wait at the pickup location for the load to be ready?

• In the bidding mechanism, a network participant who does not intent to actually engage in the negotiation process may be tempted to send a false offer in order to gather information about the network. Thus, the bidding mechanism should include some policies that prevent this behavior without compromising the participants’ privacy.

These issues are currently addressed in an-on going work.

E. Illustrative Example

The following, example demonstrates the benefits that can be gained from collaboration for both transportation companies and shippers. It also shows some rules and parameters that need to be established in order to run the collaborative network. In this example, we consider two transportation companies: company C1 which is based in Halifax and has accepted a load L1 from Halifax to Vancouver, and company C2 which is based in Edmonton and has received a load L2 from Edmonton to Montreal. Assuming that the time window constraints for load L2 are wide enough, we investigate the possibility that company C2 delegates load L2 to company C1. Fig. 1 illustrates this load exchange. In this figure, dashed lines represent empty moves. Moreover, the first number on each link represents the distance (in kilometers) and the second number represents the cost (in dollars) that the company would charge for transporting a load on this link. These costs are calculated based on the same cost structure used in [6]. If company C2 services Load L2, the cost to be charged to the customer is $5771.5. The cost of using the collaborative network to service this load is 5% of this amount. We assume that company C1 accepts to make a detour to service load L2 if the total detour (i.e., distance(Vancouver, Edmonton) + distance(Montreal, Halifax)) is less than a maximum percent of the total empty mileage necessary to go back from Vancouver to Halifax (i.e. distance(Vancouver, Halifax)). In this example, we assume that 42% is an acceptable percent to make the detour. This detour will cost company C1 a total of $3358.08 in operating costs (i.e., 88% of the total cost that would have been charged for transporting loads on these links).

Using the collaborative network to service load L2 instead of servicing this load by company C2 will result in a total profit of $2124.84 = $5771.5 - $3358.08 - $288.575. This profit represents 37% of the cost that company C2 would have charged to the customer if the network was not used. Given that a transportation company has an average profit margin of 12%, the 37% percent of profit can be used as follows:

• Company C2 can keep 15% which is larger than its average profit margin while saving its resources for another use.

• Company C1 can make a 12% profit instead of a large empty mileage.

• The customer can benefit from a 10% discount. The distribution of the profit gained from using the network among participants is a strategic decision jointly made by the network members.

Fig. 1 shows load exchange between companies C1 and C2. In (a), company C1 services load L1 from Halifax to Vancouver and company C2 services load L2 from Edmonton to Montreal. In (b), company C2 delegates load L2 to company C1.

V. CONCLUSION

In this paper, we presented a framework for collaborative networks in the long-haul truckload trucking industry. To help preserving members’ privacy, a distributed design was proposed for the network. Moreover, decisions about when to engage a collaboration and how much information to reveal is totally left to the network participants. This is likely to encourage a large participation and therefore gather critical mass of demand and resources to significantly impact the participants’ performance. On the other hand, interactions between the agent companies are done through the coordinator agent which insures transparency and prevent malicious

Page 5: [IEEE 2008 IEEE International Engineering Management Conference (IEMC-Europe 2008) - Estoril, Portugal (2008.06.28-2008.06.30)] 2008 IEEE International Engineering Management Conference

5

bilateral sub-alliances that may harm the network. We are currently developing strategies that will discourage any participant from sending a false bid in order to gather information about the network. We are also in the process of developing the optimization algorithms and on-line strategies needed to run the optimization base module. In elaborating these algorithms, some important issues that arise in a dynamic setting are addressed (e.g. consistency of the schedules with the current situation and accuracy of the decisions made that must reflect the actual state of the environment). Future work will be aimed at assessing our design and solution approaches under different scenarios, on simulated data that are inspired from real-world applications. We are also in contact with a local distribution company that will provide real data to assess the proposed design.

REFERENCES [1] M.D. Boyle, “Business-to-business marketplaces for freight

transportation”, MS. Thesis , Engineering System Division, Massachusetts Institute of Technology, USA, 2000.

[2] H.J. Bürckert, K. Fischer, and G. Vierke, “Teletruck: A holonic fleet management system”, in Proc. 14th European Meeting on Cybernetics

and Systems Research, 1998, vol. 2, pp. 695-700. [3] H.J. Burckert, K. Fisher, and G. Vierke, “Holonic transport scheduling

with TELETRUCK”, Appl. Artif. Intell., vol. 14, no. 7, 2000. [4] H.J. Bürckert, P. Funk, and G. Vierke, “An Intercompany Dispatch

Support System for Intermodal Transport Chains”, in Proc. Hawaii Int. Conf. on System Sciences (HICSS-33), 2000.

[5] E.Carmel, “Trucking Industry”, Kogod School of Business, American University, Washington D.C.

[6] J.S. Chateauvert, “La collaboration dans le transport en chargements complets”, MBA. Thesis , FSA, Univ. Laval, Quebec, 2004.

[7] M.A. Figliozzi, “Analysis and evaluation of incentive compatible dynamic mechanisms for carrier collaboration”, Transportation Research Record, vol. 1966, pp. 34-40, 2006.

[8] K. Fisher, K., B. Chaib-draa, B, J.P. Muller, M. Pischel, and C. Gerber, "A Simulation Approach based on Negotiation and Cooperation between Agents: A case Study", IEEE Trans. on Systems, Man, and Cybernetics, vol. 29, no. 4, pp. 531-545, 1999.

[9] K. Fisher, J.P. Muller, and M. Pischel, “Cooperative transportation scheduling: An application domain for distributed AI”, Appl. Artif. Intell., vol. 10, no. 2, 1996.

[10] . K. Fischer, J.P. Muller, M. Pischel, and D. Schier, “A Model for Cooperative Transportation Scheduling”, in Proc. 1st Int. Conf. Multi-Agent Systems, San Francisco, CA, 1995, pp. 109-116.

[11] P. J. 't Hoen and J. A. La Poutre, “A decommitment strategy in a competitive multi-agent transportation setting”, in Agent-Mediated Electronic Commerce V (AMEC-V), P. Faratin, D. Parkes, J. Rodriquez-Aguilar,Eds. Berlin, Germany: Springer-Verlag, 2004, pp. 56-72.

[12] N. Neagu, K. Dorer, D. Greenwood, and M. Calisti, “LS/ATN: Reporting on a Successful Agent-Based Solution for Transport Logistics Optimization”, in Proc. IEEE Workshop on Distributed intelligent Systems: Collective intelligence and Its Applications (Dis'06), vol. 00, Washington, DC, 2006.

[13] G. Normandeau, “Les nouvelles technologies de l’information et des communications et la concentration des entreprises de transport routier des marchandises”, Univ. Paris Dauphine, France, Tech. Rep. 24, 2003.

[14] R. G. Smith. “The Contract Net Protocol: High-Level Communication and Control in a Distributed Problem Solver”, IEEE Trans. On Computers, series C-29, no. 12, pp. 1104-1113, 1980.