A methodology to estimate travel time using dynamic traffic assignment (DTA) under incident conditions

Transportation Research Part C 19 (2011) 1215–1224

Contents lists available at ScienceDirect

Transportation Research Part C

journal homepage: www.elsevier .com/locate / t rc

A methodology to estimate travel time using dynamic trafficassignment (DTA) under incident conditions

Camille N. Kamga a,⇑, Kyriacos C. Mouskos b, Robert E. Paaswell c

a University Transportation Research Center, The City College of New York, New York, NY 10031, United Statesb Institute for Transportation Systems, The City College of New York, New York, NY 10031, United Statesc Department of Civil Engineering and Director Emeritus, University Transportation Research Center, The City College of New York,New York, NY 10031, United States

a r t i c l e i n f o a b s t r a c t

Article history:Received 26 July 2010Received in revised form 2 February 2011Accepted 4 February 2011

Keywords:Traffic delayDynamic traffic assignmentIncident managementTravel timeTraveler information

0968-090X/$ - see front matter � 2011 Elsevier Ltddoi:10.1016/j.trc.2011.02.004

Abbreviations: ATIS, advanced traveler informainformation system; HR, hour; ITS, intelligent transorigin–destination; SO, system optimal; STD, standaVH, vehicle hour; V/C, volume to capacity ratio; VIS⇑ Corresponding author.

E-mail addresses: [email protected] (C.N. Kamg

This paper presents results from a research case study that examined the distribution oftravel time of origin–destination (OD) pairs on a transportation network under incidentconditions. Using a transportation simulation dynamic traffic assignment (DTA) model,incident on a transportation network is executed under normal conditions, incident condi-tions without traveler information availability, and incident conditions assuming thatusers had perfect knowledge of the incident conditions and could select paths to avoidthe incident location. The results suggest that incidents have a different impact on differentOD pairs. The results confirm that an effective traveler information system has the poten-tial to ease the impacts of incident conditions network wide. Yet it is also important to notethat the use of information may detriment some OD pairs while benefiting other OD pairs.The methodology demonstrated in this paper provides insights into the usefulness ofembedding a fully calibrated DTA model into the analysis tools of a traffic managementand information center.

� 2011 Elsevier Ltd. All rights reserved.

1. Introduction

The development of dynamic traffic assignment (DTA) models enlarges the scope of transportation related studies andbridges the differences between traffic operations and transportation planning studies. DTA models can estimate and predicttime-dependent network conditions by capturing the temporal and spatial variations in dynamic traffic networks (Peeta andZiliaskopoulos, 2001). DTA models produce the time–space trajectory of each individual vehicle from its origin to its desti-nation. Each vehicle trajectory includes the departure time from the origin, the arrival time at the destination, the vehicle’schosen path, and the location of the vehicle at any time along this path. Dynamic traffic assignment (DTA) is particularlyappropriate for modeling highway incidents because the timing of incident occurrence, management, recovery, and theuse of alternate routes is critical to roadway performance and driver behaviors. Static methods based on average daily trafficwill fail to identify and test the short-term control actions necessary to manage non-recurring events such as crashes orinfrastructure failures (Wirtz et al., 2005).

. All rights reserved.

tion systems; DTA, dynamic traffic assignment; DUE, dynamic user equilibrium; GIS, geographyportation system; MIN, minutes; MSA, method of successive averages; NRD, non-recurring delay; OD,rd deviation; TDSP, time-dependent shortest path; TT, travel time; UE, user equilibrium; VEH, vehicle;TA, visual interactive system for transport algorithms; VMT, vehicle miles traveled.

a), [email protected] (K.C. Mouskos), [email protected] (R.E. Paaswell).

http://dx.doi.org/10.1016/j.trc.2011.02.004

mailto:[email protected]



http://dx.doi.org/10.1016/j.trc.2011.02.004

http://www.sciencedirect.com/science/journal/0968090X

http://www.elsevier.com/locate/trc

1216 C.N. Kamga et al. / Transportation Research Part C 19 (2011) 1215–1224

This study showcases the application of simulation-based dynamic traffic assignment (DTA) modeling on travel time dur-ing a traffic incident. Simulation-based DTA models are used to estimate time-varying network conditions by capturing traf-fic flow and route choice behavior. A basic characteristic of these models is the utilization of a traffic simulator to emulatetraffic conditions.

The objective of this study was to use DTA principles to:

(a) Analyze the travel time of OD pairs on the urban transportation network due to incidents.(b) Propose a methodology to estimates delays in the network due to incidents.

A case study using a subset transportation network for the greater Chicago, Illinois, is presented in this paper. A transpor-tation network model was developed and tested for the Chicago network to study travel time of OD pairs resulting from inci-dent management using a DTA simulation tool. More specifically, the visual interactive system for transport algorithms(VISTA) DTA based software was used to analyze the travel time of OD pairs and assess the effectiveness of providing trav-elers information to vehicles in the Chicago area.

2. Background

Many models have been proposed to estimate incident delay. These models can be classified into three types based on themethods adopted: (1) methods based on queuing analysis (Morales, 1987); (2) methods based on shock wave analysis (Mes-ser et al., 1973; Wirasinghe, 1978, and Chow, 1976) and (3) methods based on freeway traffic simulation (Wickes and Lie-berman, 1980 and FRESIM, 1984). These models can also be categorized into two types based on the scales: (1) models thatfocus on total incident delay caused by incidents (Morales, 1987; Messer et al., 1973; Wirasinghe, 1978, Chow, 1976, Wickesand Lieberman, 1980 and FRESIM, 1984); and (2) models that focus on individual vehicle incident delay (Fu and Rilett, 1997).

The models referenced above have not been widely implemented and therefore present some level of difficulty for theirevaluation. These models do not capture the spatial characteristics of incident delay by not considering the characteristics oftransportation network (Kamga, 2006). They consider only the demand upstream of the incident location that arrives fromthe roadways mainline and most models are applicable only for non-congested traffic flow conditions as they utilize delayfunctions that are also based on volume to capacity ratios (v/c), which cannot exceed one.

As observed by Skabardonis et al. (1996), another major limitation of the queuing diagram method, which is often over-looked, is that it estimates delays at a specific point. Several incidents, however, may occur simultaneously along a freewaysection, and the traffic conditions at the incident of interest could be influenced by conditions upstream.

In this paper, we perform the analysis using VISTA, a DTA model tools. As other most prevalent DTA models, VISTA appliesmesoscopic traffic simulation for evaluating route choice decisions. Mesoscopic models provide a level of detail somewherebetween that provided by macroscopic and microscopic models. Typically, mesoscopic models can propagate vehicles usingmacroscopic rules, but capture microscopic detail, such as individual vehicle location and queue evolution. Mesoscopic sim-ulation provides greater computational efficiency that allows a much faster simulation or allows application to a much largernetwork (VISTA, 2002).

3. Dynamic traffic assignment

3.1. Overview

DTA models depart from the standard static assignment assumptions to address long-standing problems caused by theunrealistic assumptions of existing static planning methods. One of the main common features of DTA models is that theydeal with the dynamic nature of the network under time-varying demands. Another important feature is that DTA modelstake into account complex interactions between supply and demand in a transportation network. Such models compute thespatio-temporal path for every vehicle while accounting for real-time driver behavior. This is a great advantage over manytraditionally used models (such as CORSIM) that do not track the movement of individual vehicles but instead split traffic atintersections.

DTA models allow for modeling a variety of ITS options, a feature of great importance given the proliferation of such sys-tems in the last two decades (Peeta and Ziliaskopoulos, 2001). Existing DTA models are generally classified into two broadcategories: analytical models and simulation-based ones. The analytical models can be further categorized as mathematicalprogramming, optimal control, or variational inequality models.

Most analytical DTA models are extensions of their equivalent static formulations and tend to focus on the user equilib-rium (UE) and system optimal (SO) objectives, or on some variants of them. These models typically attempt to formulate theproblems and seek mathematical techniques to solve them. Although analytical formulations are vital to gaining insight intoDTA problems and future applications, a number of limitations raise questions regarding their applicability for real worldapplications. First, existing analytical models inevitably involve many simplifications. As a result, such models cannot ade-quately capture the true dynamics of traffic conditions such as congestion buildup and dissipation. Moreover, in realisticapplications, the network size is typically very large, and analytical models are computationally cumbersome and not prac-

C.N. Kamga et al. / Transportation Research Part C 19 (2011) 1215–1224 1217

tical for real world employment. Thus, when real world large-scale networks are considered, simulation-based DTA modelsoffer practical advantages over analytical approaches for implementation (Sisiopiku and Li, 2006).

3.2. Simulation-based DTA models

In general, simulation-based DTA models iterate between a traffic simulation module, a time-dependent shortest pathmodule, and a network-loading module. First, given a set of vehicles and their travel paths, the traffic simulation modulereplicates complex traffic flow dynamics as the vehicles are propagated through the network. The link travel times reportedby the simulator are then used to calculate the time-dependent shortest paths. Those shortest paths are then combined withall previous sets of shortest paths, and the vehicles are loaded onto the network on those paths. A new iteration then beginsas the simulator propagates vehicles through the network along the new combination of paths. The process stops when someuser-specified convergence criterion is met.

A detailed overview of the literature on DTA models, along with a discussion of current and future challenges in DTA re-search and applications, can be found in Peeta and Ziliaskopoulos (2001). A comparison of features, strengths, and limitationsof commonly used DTA simulation models is offered in Sisiopiku and Li (2006).

4. Methodology

The overall approach in this study is to use the DTA capabilities to assess the impact of designed incident scenarios on ODpairs in a network during the analysis period and to evaluate the implementation of ITS technologies on the transportationnetwork. Unlike static assignment methods, which are based on average daily traffic and fail to capture the dynamic processof an incident, DTA is particularly appropriate for studying short-term planning applications such as evaluating various inci-dent management options. The main required model capabilities include the simulation of incident conditions, the responseof individual drivers to the incidents, and the impact of the dissemination of incident information. Furthermore, the modelneeds to simulate networks that are large enough to contain not only the direct effects of incidents and the pre-plannedmanagement strategies but also the indirectly impacted areas.

The approach proposed in this study to estimate incident delay is based on the travel time difference between normal andincident conditions using DTA simulation software. This procedure avoids some of the problems in using the queuing dia-gram and directly provides the delay perceived by the motorists. The general formula for the determination of the totalnon-recurring delay (NRD) could be defined as follows:

Network NRD = network wide total travel time under incident conditions – Network wide total travel time under normal1

conditions

where,Network wide total travel time under incident conditions is the expected total travel time of all the travelers (all OD pairs)

from the occurrence time of an incident until its conclusion and until all travelers reach their destinations.Network wide total travel time under normal conditions is the total travel time of all the travelers (all OD pairs) from the

time that the incident presumably occurred until all travelers reach their destinations under normal conditions.

4.1. Model selection and main features of visual interactive system for transportation algorithms (VISTA)

A detailed review of the model approaches, capabilities, and limitations, along with the availability of models and otherresources, led to the selection of VISTA as the simulation tool for this study. VISTA is an innovative network-enabled frame-work that integrates spatio-temporal data and models for a wide range of transportation applications, including planning,engineering, and operational ones. The VISTA DTA produces the dynamic user equilibrium (DUE) paths of all vehicles fromtheir Origin to their destination-given: Infrastructure roadway geometry and topology, Traffic control data, origin/destina-tion matrix of vehicles, and a traffic simulator to propagate the traffic at the currently selected paths and allocated demandper path, it computes the spatio-temporal path for every user so that: No user can switch path and improve either his departureor his arrival time unilaterally.

A modified method of successive averages (MSA) is used to identify the paths for each OD pair using the time-dependentshortest path (TDSP) algorithm. VISTA utilizes a mesoscopic simulator called RouteSim and a DTA routine to emulate thebehavior of individual drivers and how they distribute themselves into the transportation network. RouteSim is based onan extension of Daganzo’s cell transmission model introduced by Ziliaskopoulos and Lee (1996). In this model, the road isdivided into small cells, and the cells are adjustable in length; bigger cells are used for a mid-section of a long highway seg-ment, and smaller cells are used for intersections and interchanges. Vehicles are considered to be moving from one cell toanother. Basically, traffic is moved by the simulator in platoons and not in terms of single vehicles. The simulator keepstrack of the flow in each cell and, every time step, calculates the number of vehicles that are transmitted between adjacentcells.

1 The normal state of the transportation network under prevailing incident-free conditions represents the state of the traffic during normal conditions.


Initially, the RouteSim simulator in VISTA is run with vehicles assigned to the free flow shortest paths. The link traveltimes resulting from that assignment pattern are then used to calculate a new set of shortest paths, and the simulation isrepeated with vehicles assigned to a combination of the paths in the previously calculated path set. At first, the link flowsgenerated by the free flow shortest paths vehicle assignment can be different from the link flows generated by the sim-ulation using the new set of calculated paths. However, after a certain number of iterations, the link flows will converge.Thus, iterations continue between the mesoscopic simulation and vehicle assignment until the link flows converge. Thisprocedure accounts for vehicle path choice with changes in traffic conditions (Chang and Ziliaskopoulos, 2003). Mesoscop-ic traffic simulators capture the basic traffic flow characteristics of the speed-flow density relationship from cell to cell,moving vehicles in packets. In contrast, microscopic traffic simulators model the drivers’ sub-second driving decision ofacceleration, deceleration, gap acceptance, headway, lane changing thereby providing a more realistic representation oftraffic flow propagation. For the purpose of this study a mesoscopic traffic simulator adequately propagates the trafficalong the DUE paths produced by the VISTA DTA model – each vehicle is assigned a DUE path at convergence level basedon the VISTA algorithm. For the purpose of this study the use of a mesoscopic traffic simulator is sufficient to demonstratethe impact of an incident on path travel times and delays and the presence of a VMS with rerouting capabilities. It is notedthat when more accuracy is necessary under a microscopic traffic simulator may be employed instead of a mesoscopicone. VISTA has its own microscopic traffic simulator that could be employed if necessary. However, it makes the processextremely slow due to its iterative nature and a few hours could easily turned into days until convergence therefore it isnot employed frequently.

4.2. Data needs

Although specific data needs may vary with the model selected, the VISTA DTA model typically requires the followingdata as input: (1) The topology of the network in a Geographic Information System format, (2) Roadway geometry, (3)The traffic control data: signal timing, speed limit, lane movement designation, vehicle class prohibitions, ramp meter-ing, signal preemption, (4) Bus routes, schedules and average dwelling time per bus stop, and (5) origin–destination ma-trix for each vehicle class (e.g., autos, trucks). This is also applicable to VISTA. VISTA network data are in a node-linkformat and the user can define a network with nodes and links (Ziliaskopoulos and Barrett, 2005). As an alternative,the user can directly import existing network data from other software, such as TRANPLAN and CORSIM, by use of con-version tools provided. The demand data in VISTA are in the format of origin–destination (OD) trip matrices. VISTA al-lows the user to input either dynamic demand or static demand. Dynamic OD trip matrices are departure time based,and the number of seconds after the beginning of the simulation at which the vehicle enters the network is given. Incontrast, static demand is a flat value for the total number of vehicles going from one zone/node to another for a givenperiod.

VISTA features its own GIS system and it can be interfaced with any other GIS system. Similarly it can be interfaced withany transportation planning software such as VISUM, CUBE, and TRANSCAD EMME/3, microscopic traffic simulators such asCORSIM, VISSIM, and PARAMICS, and signal optimization software such as SYNCHRO and TRANSYT.

The dynamic OD matrix is usually estimated at 15-min time intervals through the use of a static OD matrix and any avail-able traffic counts.

5. The chicago network case study

5.1. Study description

A case study was performed on an extraction of the greater Chicago network as shown in Fig. 1. This network configura-tion comprises 123 nodes (some nodes are not displayed on the Figure) and 194 OD pairs, thus allowing for alternativeroutes for OD pairs. The total demand of the network during the 10 h of simulation assignment is 162,626 vehicles. The de-mand is uniformly distributed for each of the 15 min time step of the assignment period. Two scenarios were developedusing the VISTA software. First, a base case scenario was generated to depict operational characteristics of the test networkunder normal traffic flow conditions. Then, an incident was emulated using VISTA’s Incident management module. A closureof link 13139 (two lanes) was assumed that started 180 min from the start of the simulation having an incident duration of45 min, covering a length of 2400 feet. Under the incident scenario two sub-cases were modeled:

� An incident scenario where all drivers were assumed to have no information on the incident. Under this case all driversare assumed to follow their current (or ‘‘no incident’’) paths as determined by the base case DTA. In order to emulate thesenew traffic conditions due to the roadway partial closure, the traffic simulator RouteSim is executed using the paths gen-erated by the base case DTA.� An incident scenario where all drivers are assumed to have perfect information of the incident conditions. The VISTA DTA

to estimate the new DUE is executed to emulate these conditions as the closure of two lanes have reduced link’s 13139geometry capacity.

Fig. 1. Chicago network study test bed with an incident on link #13139.


5.2. Results

Table 1 displays the results for the base case (without an incident). It shows the number of vehicles loaded onto the net-work, including a differentiation among passenger cars and commercial vehicles. The corresponding total travel time for allvehicles is provided in hours while the average travel time and standard deviation per vehicle are reported in minutes. Thetable also shows the corresponding vehicle miles traveled in the network.

Table 2 shows the corresponding simulation results for the case under incident conditions with an assumption that alltravelers are aware of the incident and are choosing paths that minimize their travel time in the network. This is a non-real-istic case since only a small fraction of travelers usually have information about the incident and its anticipated impact. InVISTA, the evaluation of the implementation of an advanced traveler information system (ATIS) can be emulated by routingvehicles to pre-determined paths. The evaluation of the traveler information system is beyond the scope of this paper. How-ever, results from scenarios of providing ATIS will be briefly examined to determine its potential impact on the transporta-tion system under incident conditions.

Table 3 depicts the travel times and vehicle miles traveled for the incident case when drivers have no information aboutthe incident. In this case, it is assumed that all travelers are following their initial (base case) routes. RouteSim, a mesoscopictraffic simulator, is executed to propagate vehicles on initial paths determined by the DTA model of the base case. No newpaths are generated from the base case. As in the case above, this situation is also not realistic. In reality, it is expected that

Table 1Simulation results for the base case.

Demand (vehicle) Total TT (h) Average TT (min/veh) STD (min/veh) VMT (miles)

All type Vehicles 162,626 20,918 7.72 2.68 973,732Passenger cars 160,249 20,542 7.69 2.68 953,486Commercial vehicles 2377 376 9.51 1.48 20,246

Table 2Incident with dynamic user equilibrium (full incident information).


All type vehicles 162,626 22,063 8.14 3.98 974,839Passenger cars 160,249 21,686 8.12 4.00 954,593Commercial Vehicles 2377 376 9.51 1.47 20,246

Table 3Incident with RouteSim (no incident information).


All Type Vehicles 162,626 27,671 10.21 9.69 973,732Passenger Cars 160,249 27,157 10.17 9.69 953,486Commercial Vehicles 2377 513 12.96 9.69 20,246

Table 4Delay for the overall network.

Complete incident information No incident information

Delay average(min/veh)

Delay STD(min/veh)

Total delay(VH)

Delay STD(VH)

Delay average(min/veh)

Delay STD(min/veh)

Total delay(VH)

Delay STD(VH)

All typevehicles

0.42 6.66 1138 18,051 2.49 12.37 6749 33,528

Passengervehicles

0.43 6.68 1148 17,841 2.48 12.37 6624 33,038

Comm.vehicles

0.00 2.95 0 117 3.45 11.17 137 443


the occurrence of the incident may result in some vehicles changing their travel plan either by changing their initial route, orchanging their departure time or destination, or canceling their trip. However, this has no major influence on the paper’sobjectives of estimating travel time of OD under incident conditions at the network level.

The incident delay is calculated according to the methods shown above and the results are illustrated in the followingtables. Table 4 shows the difference in incident delay between the case where all travelers are aware of the impact and loca-tion of the incident (full information) and the case where all drivers are following their initial base case paths (no informa-tion). These results show the importance of dissemination of traveler information systems in helping to reduce the negativeimpact of an incident on traffic operations at the network level. However, in reality, it is expected that only a small fraction oftravelers will likely have knowledge about the incident, its impact, and some alternative routes, such that they may avoid itsundesirable impacts.

Next we present the results for individual OD pairs in order to examine the impact of an incident for various travelers.Disaggregate travel times and delays for each origin–destination pair that comprise the transportation network are helpfulto understand the incident effects on vehicles upstream and downstream of the incident location. The spatial distribution ofvehicles during and after the incident is an important ‘‘parameter’’ for analyzing the impacts of an incident. The average tra-vel time distribution for some origin–destination pairs is extracted from the simulation results. The OD pairs are classified intwo groups: OD pairs with paths that include (Table 5–8) the incident location link and OD pairs with paths that do not (Ta-ble 9 and 10).

By grouping the OD pairs in these two categories, one can analyze the effect of an incident on vehicles as a direct responseto the incident impact. These tables demonstrate that the impacts of this incident are greater on vehicles with paths thatinclude the incident link compared to those with paths that do not include the incident link. The former vehicles experiencemore delay, especially if there is no information provided that may help some of them to divert away from the incident.However, it is found that providing information upstream of the incident may not always be beneficial for vehicles withpaths not traversing the incident link. A few of them experience slightly more delay resulting from the rerouting of affectedvehicles that generates an increased vehicle volume in non-congested alternate routes.

As shown in Tables 6–8, above, some vehicles are diverted away from the incident location. The diverted paths are addi-tional paths generated by the simulator on which rerouted vehicles are loaded from OD pairs. The percentages provided inthese tables are the ratio of volume using a path to the total vehicle volume of the OD pairs. For example, 71.6% of vehicles

Table 5Travel time for some OD pairs with paths that include the incident link (no provision of incident information).

OD Pairs Demand average(vehicles)

Base case averageTT (min/veh)

Base case STD(min/veh)

Incident case average TT(min/veh)

Incident caseSTD (min/veh)

102717-201929 973 10.14 0.03 15.47 0.26102090-202081 870 6.07 0.14 6.94 0.19102015-202344 1569 10.02 0.08 14.70 0.38102015-202081 639 7.00 0.17 12.49 0.51102013-202081 955 9.05 0.17 17.96 0.38101983-202344 2914 9.54 0.09 14.48 0.32101983-202081 1132 6.54 0.17 12.20 0.30101923-202344 714 11.06 0.09 19.05 0.21

Table 6Travel time for some OD pairs with paths that include the incident link (with provision of incident information).

OD pairs Path 1(vehicle)

Path 1 TT(min/veh)

Path 2(vehicle)

Path 2 TT(min/veh)

Path 3(vehicle)

Path 3 TT(min/veh)

Path 4(vehicle)

Path 4 TT(VH)

102717-201929 900 11.00 73 12.17102090-202081 834 6.39 33 6.84102015-202344 1452 10.57 107 13.29 10 11.05102015-202081 487 8.68 129 13.04 17 8.50 6 8.71102013-202081 698 11.03 57 10.32 193 11.52 7 11.71101983-202344 2869 11.70 45 10.53101983-202081 811 7.50 96 7.95 225 7.99101923-202344 662 11.92 48 15.06 4 15.11

Table 7Delay for some OD pairs with paths that include the incident link (no provision of incident information).

OD pairs % Vehicles Incident delay average (VH) OD pairs % Vehicles Incident delay average (VH)

102717-201929 100 86 101923-202207 100 65102090-202081 100 13 101923-202090 100 39102015-202344 100 122 101923-202081 100 122102015-202081 100 58 101861-202081 100 22102013-202081 100 142 101682-202344 100 1079101983-202344 100 240 101682-202090 100 204101983-202081 100 107 101682-202081 100 192101923-202344 100 95 101682-201929 100 321

Table 9Travel time for some OD pairs with paths that do not include the incident link.

OD Pairs Base case Incident – no information Incident – with information

Demand(veh)

TT(min/veh)

STD(min/veh)

Demand(veh)

TT(min/veh)

STD(min/veh)

Demand(veh)

TT(min/veh)

STD(min/veh)

115184-202344 2985 8.36 0.04 2985 8.44 0.12 2985 8.76 0.13102717-202344 2264 10.83 0.06 2264 10.91 0.11 2264 11.22 0.12102717-202339 1414 6.24 0.02 1414 6.24 0.03 1414 6.36 0.05102717-202207 870 7.45 0.04 870 7.45 0.04 870 7.45 0.04102717-202090 828 9.45 0.05 828 9.45 0.04 828 9.45 0.04102717-202081 917 11.79 0.23 917 12.66 0.20 917 12.07 0.15102344-202207 955 6.24 0.02 955 6.24 0.03 955 6.24 0.03102344-202090 955 8.24 0.03 955 8.24 0.03 955 8.24 0.03102344-202081 1362 4.68 0.03 1362 4.68 0.06 1362 4.70 0.08102339-202081 825 6.07 0.12 825 6.96 0.18 825 6.35 0.14

Table 8Delay for some OD pairs with paths that include the incident link (with provision of incident information).

OD Pairs Path 1%vehicles

Path 1 incidentdelay (VH)

Path 2%vehicles


Path 3%vehicles


Path 4%vehicles


102717-201929 92.5% 12.8 7.5% 2.5 0.0102090-202081 95.9% 4.3 4.1% 0.4102015-202344 92.5% 13.0 6.8% 6.0 0.7% 0.0102015-202081 76.2% 13.6 20.2% 13.0 2.7% 0.4 0.9% 0.2102013-202081 73.1% 23.0 6.0% 1.2 20.2% 7.9 0.7% 0.3101983-202344 98.5% 103.3 1.5% 0.7101983-202081 71.6% 13.0 8.5% 2.3 19.9% 5.3101923-202344 92.7% 9.5 6.7% 3.2 0.6% 0.3


for OD pair 101983-202081 (Table 8) remain on their initial path (Path 1) while 8.5% and 19.9% of vehicles are rerouted to asecond path (path 2) and/or a third path (path 3) respectively. The rerouting of vehicles for OD pair 101983-202081 reducesthe average delay per vehicle from 107 vehicle-hours (all vehicles in path 1) to 13, 2.3, and 5.3 vehicle-hours for path 1, path2, and path 3 respectively, after diversion. The diversion of vehicles helps reduce considerably the average travel time formost vehicles traversing the incident, therefore reducing the total delay in the network. There are some OD pairs with pathsthat do not traverse the incident link that are also affected (Tables 9 and 10). While most of these OD pairs are not impacted,

Table 10Delay for OD pairs with paths that do not include the incident link.

OD Pairs Incident – no information Incident – with information

Vehicles average Delay average (VH) STD (VH) Vehicles average Delay average (VH) STD (VH)

115184-202344 2985 4.0 7.5 2985 0.4 8.5102717-202344 2264 3.0 6.0 2264 0.4 6.8102717-202339 1414 0.0 1.0 1414 0.1 1.0102717-202207 870 0.0 1.0 870 0.0 1.2102717-202090 828 0.0 1.2 828 0.0 1.2102717-202081 917 13.3 6.6 917 0.3 5.8102344-202207 955 0.0 0.8 955 0.0 0.8102344-202090 955 0.0 1.0 955 0.0 1.0102344-202081 1362 0.0 2.0 1362 0.0 2.5102339-202081 825 12.4 4.1 825 0.3 3.6


Table 9 and Table 10 show that a few OD pairs are slightly negatively affected for both incident cases. Once again, the effectof the incident is network wide. The estimates of its impact should take into account the resulting comportment of all vehi-cles, including, as well, vehicles not necessarily traversing it. By comparing both states of the network, one can capture thetrue impact caused by the incident.

There are indications in the following figures (Figs. 2 and 3) that incidents are not just affecting vehicles on links upstreamof its location. Some vehicles originating downstream of the incident location may also experience minor delay or benefitfrom it depending on the prevailing level of demand on the roadway when it occurred. For both incident cases, the incidentimpacts on downstream vehicles are mixed. The vehicles originating upstream of the incident with paths including the af-fected link experience a longer recovery time if no traveler information is provided, compared to when it is provided. Withthe deployment of ATIS, the negative impacts of incidents may be reduced and shortened for upstream vehicles. However,vehicles originating downstream of the incident may not always benefit from the deployment of ATIS at the OD level. The

Average Delay

0

10

20

30

40

50

60

0:00:00 0:40:00 1:20:00 2:00:00 2:40:00 3:20:00 4:00:00 4:40:00 5:20:00 6:00:00 6:40:00 7:20:00 8:00:00 8:40:00 9:20:00

Vehicle Departure Time

Del

ay (

min

ute/

veh)

Information (DUE)No Information (RteSim)

Fig. 2. Temporal distribution of delay for OD (101682-201929) traversing the incident link.

Fig. 3. Temporal distribution of delay for OD (115184-202344) with origin located downstream of the incident with paths not including the incident link.


improvement of travel conditions for vehicles originating upstream may at the same time have a negative effect on thedownstream flow. Thus, the importance of considering the network wide impact is once again recommended.

6. Conclusions

In this paper, DTA principles were applied to the modeling of incident conditions in the Chicago, IL network using VISTA.More specifically, the impact of incident on travel time on OD pairs and incident delay were examined through the simula-tion of an incident with and without availability of incident-related traveler information.

The main conclusions stemming from this study are as follows:

� The impacts of incident are not just on vehicles originating upstream or traversing the incident location. Vehicles origi-nating downstream or not traversing the incident location may also negatively be impacted by the incident if one con-siders the network-wide area.� Availability of information on incident presence, along with availability of alternative routes with residual capacity that

allow rerouting of vehicles around incident congested locations, could considerably assist in improving incident manage-ment practices.� DTA could provide a powerful tool to transportation agencies to improve their existing Incident Management plans

through the evaluation of route diversion strategies and the production of more comprehensive estimates of incidentdelay.� It is proposed that DTA becomes part of the analysis tools if traffic management and information centers to support inci-

dent management and traveler information services.

The present study did not include the use of non-DUE DTA where only a partial set of users has full knowledge of trafficflow characteristics under incident conditions. This should be further explored in order to develop a more comprehensivemethodology. In addition, the model developed included only an abstraction of the Chicago network.


DTA has evolved rapidly over the past two decades; this advancement has been fueled by the needs of application do-mains ranging from real-time traffic operations to long-term planning. As a matter of fact, DTA models are a natural evolu-tion in the transportation field and are expected to become mainstream when issues related to the realism of theirassumptions and mathematical tractability are addressed in greater detail. Several simulation-based DTA models are cur-rently available for real world deployment and have already gained sophistication and significant acceptability. An opera-tional DTA model will need to be developed on a more detailed network that contains all major and minor arterials thatcould be used for rerouting of vehicles. In parallel, the DTA model should be integrated with the traffic surveillance systemof the network in order to be calibrated continuously. We note here that DTA models in their current form cannot be used forreal-time applications but as off-line tools as they are computationally inefficient.

Acknowledgements

The authors would like to thank Mr. Curtis Barrett of Vista Transport Group, Inc., for providing the Chicago network andfor assisting in using the VISTA software. We would like to thank Professor Neville A. Parker for its support.

References

Chang, E., Ziliaskopoulos, A.K., 2003. Data challenges in development of a regional assignment: simulation model to evaluate transit signal priority inChicago. Transportation Research Record 1841, 12–22.

Chow, W.M., 1976. Study of traffic performance models under an incident condition. Highway Research Record 567, 31–36.FRESIM, 1984. FRESIM user guide, Version 4.5, Office of Research And Development, Federal Highway Administration, U.S. Dept. of Transportation,

Washington, DC.Fu, L., Rilett, L.R., 1997. Real-time estimation of incident delay in dynamic and stochastic networks. Transportation Research Record 1603, 99–105.Kamga, N.C., 2006. Estimation of Network Based Incident Delay in a Transportation Network Using Dynamic Traffic Assignment, PhD Dissertation, The

Graduate School and University Center, The City University of New York, July, 2006.Messer, C.J., Dudek, C.L., Friebele, J.D., 1973. Method for predicting travel time and other operational measures in real-time during freeway incident

conditions. Highway Research Record 461.Morales, J., 1987. Analytical procedures for estimating freeway traffic congestion. ITE Journal – Institute of Transportation Engineers 57 (1), 45–49.Peeta, S., Ziliaskopoulos, A.K., 2001. Foundations of dynamic traffic assignment: the past, the present and the future. Networks and Spatial Economics 1 (3/

4), 233–265.Sisiopiku, V.P., Li, X., 2006. Overview of dynamic traffic assignment options. In: Proceedings of the 2006 Transportation Simulation Symposium (TSS 06),

April 2–6, 2006, Huntsville, Alabama.Skabardonis, A., Petty, K., Noeimi, H., Rydzewski, D., Varaiya, P., 1996. I-880 field experiment: data-base development and incident delay estimation

procedures. Transportation Research Record 1554, 204–212.VISTA, 2002. Technical Documentation: Routesim and Dynamic Traffic Assignment, Release 1.0.1., Fall.Wickes, D., Lieberman, E., 1980. Development and Testing of INTRAS: A Microscopic Freeway Simulation Model, Report FHWA RD-80, vol. 1, no. 3, pp. 106–

108.Wirasinghe, C.S., 1978. Determination of traffic delays from shock wave analysis. Transportation Research 12.Wirtz, J.J., Schofer, J.L., Schulz, D.F., 2005. Using simulation to test traffic incident management strategies: illustrating the benefits. In: Proceedings of the

84th Annual Meeting of the Transportation Research Board, January 2005, Washington, DC.Ziliaskopoulos, A.K., Lee, S., 1996. A cell transmission based assignment-simulation model for integrated freeway/surface street systems. In: Proceedings of

the 75th Transportation Research Board Annual Meeting, January 1996, Washington, DC.Ziliaskopoulos, A.K., Barrett, C., 2005. Visual Interactive System for Transportation Algorithms. <http://cluster.vistatransport.com> (accessed 16.02.06).

http://cluster.vistatransport.com

Documents

A methodology to estimate travel time using dynamic traffic assignment (DTA) under incident conditions