301

Implementing And Simulating Dynamic Traffic AssignmentWith Intelligent Transportation Systems In Cube Avenue

Peter Foytik, Mike RobinsonVMASC Old Dominion University

pfoytik@odu.edu, rmrobinson@odu.edu

Abstract. As urban populations and traffic congestion levels increase, effective use of information andcommunication tools and intelligent transportation systems as becoming increasingly important in order tomaximize the efficiency of transportation networks. The appropriate placement and employment of thesetools within a network is critical to their effectiveness. This presentation proposes and demonstrates theuse of a commercial transportation simulation tool to simulate dynamic traffic assignment and rerouting tomodel route modifications as a result of traffic information.

1. INTRODUCTION

Modeling and Simulation (M&S) of transportationis critical to developing and assessing proposedideas and technologies. Simulations of pasttransportation events allow planners to betterunderstand what really happened. By simulatingfuture changes, decision makers can greatlyimprove the roadways of tomorrow.

Alternatives are frequently proposed in differentlocations of cities for the future development ofcity and federal roadways. Proper testing ofproposed plans must be done to assure bestsolution. One area of transportation systemimprovements that has largely not benefited fromM&S testing is the installation or improvement ofIntelligent Transportation Systems (ITS).

IEEE's Intelligent Transportation Systems Societydefines ITS as systems that utilize synergistictechnologies and systems engineering conceptsto develop and improve transportation systems ofall kinds. ITS refers to efforts to add informationand communications technology to transportinfrastructure. It strives to apply advancedtechnology to resolve the problems of surfacetransportation by improving efficiency, safety, andmobility. Other objectives include reducingenergy, economic costs, and damage to theenvironment [2]. To better improve the planningof a large area such as the region of southeasternVirginia, ITS should be tested over the entirenetwork to assess the improvements in trafficflows and congestion levels. This document willdescribe efforts and research to implement ITSand vehicle driver effects from ITS in amesoscopic model using Avenue from the Cubefamily of transportation software.

2. TECHNICAL BREAKDOWN OF CUBE

2.1 Cube A Transportation Tool

Cube family of transportation tools developed byCitilabs is chosen for this projects as the tool ofchoice because it is already selected as theplanning standard by the Virginia Department ofTransportation (VDOT). Cube provides amacroscopic transportation modeling tool, amicroscopic modeling tool and a mesoscopicmodeling tool, each of these tools can integratetogether by sharing loaded networks. It alsoallows the modeler additional control with itsscripting language allowing the ability to programin vehicle reactions that the software tool was notdeveloped or intended to do through the defaultuser interface. The scripting language isproprietary, and offers flexibility to make changesto road networks and Origin Destination matrices(00).

2.1.1 Microscopic Dynasim

The Cube microscopic tool Dynasim is like othertransportation micro simulation in that the usercan simulate individual vehicle behavior creating avery accurate simulation. The problem with amicro simulation is the amount of time required todevelop and run a scenario [4]. This problemusually requires the simulated area to be reducedto a more manageable size so that the simulationcan run in a reasonable amount of time.Therefore, if the interest of the study is to see theeffects of ITS on an intersection, amicrosimulation would work quite well. This typeof simulation will show you the local effects in avery small area very well, but what if the plannersneed to see effects in a larger scale in multiplelocations at the same time? A microscopicsimulation could accomplish this but require muchmore time to set up and to run.

302

60 6060 60

74 43

29 66

36 2656 14

39 192 17

6 12

o 0

o 0

Figure 2: Shows an output from one of the roadsegments in a simulation run providing values of

Volume at each time segment.

2.1.3 Mesoscopic Avenue

Mesoscopic models are in between a micro and amacroscopic model, allowing traffic volume tochange over time through a large scale system[4). While some mesoscopic simulation toolshave more micro than macro features, Cube hasmore macro than micro features in that itsimulates volume over the road network.Visualization of the output animation appears as amicroscopic simulation where but it visualizespackets of vehicles based off of the volumecalculations instead of individual vehicles. Themesoscopic tool in Cube is very closely related tothe macroscopic tool, so close that it is actuallyjust another module added to Voyager.

Avenue is capable of reading in a list of 00matrices, one for each time segment and anetwork file. Time segments are defined timesteps that the simulation advances and also arethe defined moments when new volume can beadded to the system by a new 00 matrix. At eachtime segment the simulation will run the volumeover the network as a discrete event simulationfinding equilibrium then doing the same thing forthe next time segment. All of this is done throughAvenues Oynamicload statement. Much like thehighway module's pathload, Oynamicload uses agravity model to calculate equilibrium, but insteadof for one 00 Matrix for one time periodOynamicload will calculate equilibrium for eachtime segment using the calculated equilibriumfrom the previous time segment. The output filesthat Avenue produces are matrix files, networkfiles, data/text files, path files, packet log files anda few other types of outputs. The most importantoutput file is the network file which containsvalues on all of the road segments from the lastsimulation run. Most of the values that areoutputted on the road segments have a value foreach time segment that the simulation ran. Forexample, volume, queue length, speed, and timeare default outputs that Avenue provides, witheach segment representing each variable asvariable t where t equals the time segment itrepresents and variable representing the name ofthe variable, This allows the user to color codethe road segments over time using the Bandwidthchart display. A bandwidth chart is a display thatgives the user controls to advance time and to seehow that particular value changes.

SPEEDS6 2SPEEDS7 2

VS1 2 4

VS2 2 5\IS3 2

VS4 2VSf,?VS6 2 5VS7 2 4QlJ:I.£VS1 2Q(£lF¥S2 2

'?,o",

Figure 1: Shows an image of a Voyager outputnetwork with links color by level of service.

2.1.2 Macroscopic Voyager

The macroscopic tool within Cube, Voyager, isprobably the most used and well known tool withinthe suite. Voyager can calculate volumes of trafficover large networks. It offers a number ofmodules that users can use to simulatetransportation demand macroscopically. Eachmodule requires its own input files, and usingeither a script file generated by the module toperform a default task, or a script created by theuser to do a unique task can produce manyoutputs. Once the script runs the task the modulecan generate a number of outputs files of variousformats that can be used as inputs to othermodules or as strict outputs that visualize data.

Voyager runs the highway module whichproduces the calculated values on each segmentof the network for a period of time chosen by theuser using a gravity model [1). The highwaymodule takes as in input a daily demand matrix,then uses the command pathload to run thevolume over the network and using a gravitymodel to find an equilibrium over the network.The pathload command within highway takes afew inputs, one being the path variable. The pathvariable is used to set impedance over theroadways that are being simulated with pathload.The model developer can select differentroadways to run with different pathloadstatements allowing multiple impedances over theentire road network. A typical example would beto see the average congestion for the HamptonRoads area for an entire day. In this case, theuser could color code the road segments todisplay a range of colors representing the value ofcongestion. This type of output is useful forshowing daily traffic and is able to highlight theroadways that are being overused. UsingVoyager to model ITS is very possible and canshow the change in volume on roads due to ITS ina static sense. However if the planner wants tosimulate over time how vehicles are changingdirection and routes, then the macroscopic modelwill not completely accomplish that.

303

Packet log output is a text file containing a recordof locations where the packets have traveled foreach time segment. The user can load this fileover the network and view animations of thetransportation that was simulated. The animationis a view of packets represented as rectanglesover the loaded network. Users can control timeto advance at different speeds and as timeprogresses the animation displays packetstraveling the routes they were simulated to taketowards their destinations. Since the packet logfile is a text file, script files can be written with amatrix module to parse the log file and determinedata that can be presented in a user createdoutput file. An example of a parsing task wouldbe to locate the amount of time segments it tookeach packet to arrive at its destination, and thenaverage that value to obtain the average traveltime for the simulation.

Avenue's dynamic assignment and flexibility alongwith its informative outputs and helpfulvisualizations make it a great choice for modelingvehicle behavior from ITS. Large areas can besimulated in a reasonable amount of time and in atime stepped simulation. Mesoscopic simulationbenefits the non technical planners who need tounderstand how the simulated system is affectedby driver behaviors.

3. IMPLEMENTING A DAILY MESOSCOPICMODEL

3.1 A Hampton Roads Mesoscopic Model

Implementing a daily mesoscopic model for amajor metropolitan and especially in the HamptonRoads can be challenging. Demand must begenerated for all origins and destinations at eachtime segment. This is typically done by takingpercentages of the daily demand over each timesegment. The problem with this technique in aHampton Roads model is that the traffic patternshere resemble two peak load curve for mostroutes. One peak represents vehicles going onedirection in the morning, and another representingthe same traffic returning in the evening. The bestsolution will have demand values for the morning,lunch and evening traveling in the appropriatedirection. Because the tests being done now areprototype tests, smaller percentages of the dailydemand will work.

3.2 Mesoscopic Model For Testing

Specific tests require manually injecting traffic in atest area and applying congestion to one of theroadways in the way of this injected demand.This process is much like the process of doing amicroscopic test in that only a small test area isbeing worked on. The demand is also set upmuch like the microscopic simulation where anorigin destination matrix needs to be defined foreach time segment. Once the test area isperforming as expected the same type of driverintelligence can be applied in multiple areas for

the entire Hampton Roads network, or in one spotwith demand over the entire network to causedifferent reactions to the test area.

4. CAUSING CONGESTION

4.1 Implementing Incidents

In order to realistically model driver behaviors andthe influence of ITS, congestion must also besimulated. There are two ways that congestioncan be accurately portrayed in Avenue. Theeasiest method is to overload the system withlarge amounts of vehicle traffic volume, a morerealistic method involves injecting a simulatedincident by reducing road capacity. The mostprecise way to create congestion where needed isto create an incident. Overloading the systemwith traffic volume is effective but can beunpredictable as traffic could overly congest areasthat are not of interest.

In Cube there is not a default function to applyincidents, so to implement incidents the modelerhas to be able to use the Cube scripting language.The incidents modeled in Avenue require that theincident last as long as the time segment. Amodeler cannot request Avenue to reduce thecapacity of a roadway for one half of the timesegment because all dynamic changes andcalculations are done by time segment. Thereforeto model fifteen minute incidents Avenue wouldrequire fifteen minute time segments or a differentcapacity reduction would need to be calculated.This new capacity would equal the capacityeffects of a fifteen minute incident but at a onehour value [5]. The locations and severity of theincidents can be selected from historic data ofroad segments in Hampton Roads that are morelikely to have an incident [5], or can be assignedto specific areas that a planner would like tostudy.

The command Avenue uses to reduce capacity onroad segments is Dynamic. The Dynamiccommand only works with the variable C. InAvenue to change the capacity of the roadsegment the C variable has to be changed whichrepresents the capacity on the road segment overthe entire simulated time. When the Dynamiccommand is used with C, a value of C iscalculated for each time segment. The modelercan then alter the C value for any road segment atany defined time by using Dynamic and C. Thescript line needs to contain the A node (whichrefers to the starting point of a segment) and Bnode (which represents the end point of asegment) of the link segment, this is to assure thatthe incident occurs in the right direction and onthe right road segment. The capacity is usuallyreduced by multiplying the current C variable by areducing factor. When the time segment of theincident is over, the C variable is calculated by itsnormal calculation, capacity multiplied by lanesmultiplied by simulation length.

304

I F(li. A = 66537 and Ii. B = 66791)

DYNAMIC C[28] = Ii.C x 0.5

The equation above shows a conditionalstatement that locates the link with an A nodeequal to 66537 and a B node equal to 66791 thensets its C variable to half of its normal value attime segment 28.

5. SIMULATING DYNAMIC BEHAVIOR

5.1 Road Impedance To Control Behavior

Currently in Avenue dynamic behavior is alreadybeing simulated. The idea of this study is to bettercontrol that behavior and accurately simulate whatis really happening. As stated in thedocumentation of Avenue, impedance for all roadsegments can be defined by the user. Thisimpedance can be altered based on the road costvalue, time to traverse link values, and userdefined values. Smarter traffic can also besimulated by running multiple iterations of Avenueallowing the gravity model better equalize thenetwork reducing congestion by having vehicleschoose different routes based on the knowledgeof the previous run simulation. This is accurate toa point, if an incident is being simulated then amulti iteration run is not going to be realistic.Therefore Avenue needs to be manipulated toallow the impedance of some specific roads tochange at different times.

When the evacuation model of the HamptonRoads was developed a compliance variable wasimplemented to control the behavior of vehiclessticking to the evacuation routes [5]. This sameprincipal can be applied to this study. Earlyimplementations involved using compliancevariables over the network for each time segment.These compliance variables could then routevehicles around incident areas as if there wereinformation alerting vehicles of these areas.

The problem is that vehicles need to approach theaccident segments as if knowledge of the accidentdoesn't exist. Then once the accident knowledgecan be distributed the vehicles need to make anattempt to divert. This clearly shows that theimpedance needs to change dynamically. Theonly way to change the impedance is to changethe path variable within Dynamicload.

Dynamicload is Cube's dynamic analog of thestatic PATHLOAD which is the heart of themacroscopic simulation, and takes as an input alist of volumes for each time segment as well as apath impedance (1). The path variable can be setto COST, TIME, or a list of working link variables(LW). LW variables can contain values of linkimpedance or impedance equations and can beset, then altered after each time segment in theADJUST phase of Avenue. This provides a nicedynamic adjustment to the link impedanceproviding a more controlled environment toproduce dynamic behavior in a simulated ITSevent.

\

\\Figure 3: Shows an Avenue output network with

an incident occurring on a road segment indicatedby the large red dot.

5.1.1 Impedance set to Cost and Time

By using a mixture of Cost and Time a simple anddeterministic ITS system can be simulated. Byusing two Dynamicload statements ITS behaviorcan be achieved by using one statement for thetime segments where ITS is being simulated andthe other statement for normal times when ITS isnot active or needed. This is accomplished byusing Cost as the impedance for the Dynamicloadsimulating the ITS time segments and using Timeas the impedance for the other Dynamicloadstatement to simulate impedance without ITS.This method also requires that your demandcooperates with the Dynamicload statements. Forexample during normal road impedance theregular demand matrices will be applied to theDynamicload that has its path equal to Time, andduring those time segments the otherDynamicload's demand matrices should equalzero. When ITS effects need to be simulated thenthe Dynamicload with path equal to Cost takes theregular demand matrices and the Dynamicloadwith path equal to Time takes the zero demandmatrices. This method keeps the regular volumeflowing onto the network at all times andseamlessly simulates deterministic behavior dueto ITS.

5.1.2 Impedance set to LW variables

Deterministic behavior really isn't enough tosimulate the true behavior of ITS effects soinstead of using just Cost and Time asimpedance, LW variables are used. LW variablesfor each time segment give the ability to simulatedifferent behavior to the entire system. Tosimulate certain dynamic behaviors caused byITS, specification of road segments to roadimpedance needs to take place. New variablescan be assigned to the road network to giveweighted values that can specify which roadwayswill be effected and which roadways will stay thesame. By setting all normal roadways to a weightof 1 and the effected roadways to a value greaterthan one, a multiplicative operation to theimpedance equation will result in a greater valuefor the effected road segments and a normalvalue for the non effected segment.

305

6. FUTURE WORK AND CONCLUSION

Development and testing of these scenarios havea ways to go, but current tests show control overthe traffic in a way that can be made more realisticto mimic real driver behaviors. Using driversurvey's to obtain data that can producefrequencies of when drivers decide to abandon anormal route because of information or congestionand to reroute either to a known or an unknownroute. These frequencies can then be applied tothe LW variable equations to create realisticsimulations. Then using the data from thesurveys the model can be validated to the numberof vehicles that potentially would reroute. Moretests of manipulating the LW values to be alteredby time values per segment as well as congestionvalues are being done. Currently Avenue allowstime to dynamically alter impedance variables butthe results are inconsistent and need to beverified.

REFERENCES1. Cube Voyager Reference Guide, Version 5.0,

Citilabs inc, 2007-2008.2. W. Barfield and T. Dingus, "Human Factors in

Systems", Lawrence Erlbaum Associates.August 1997

3. A. De Palma and F. Marchal, "Real CasesApplications of the Fully Dynamic MetropolisTool-Box: An Advocacy for Large-ScaleMesoscopic Transportation Systems", KluwerAcademic Publishers, 2002.

4. M. Ben-Akiva, M. Bierlaire, H.N. Koutsopoulos,and R. Mishalani, "DynaMIT: a simulation-basedsystem for traffic prediction and guidancegeneration", Presented at Tristan III, June 1998.

5. R.M. Robinson, "Hampton Roads HurricaneEvacuation Transportation Study," Final Report,VMASC, Suffolk VA, 2008.