An agent-based approach to modelling driver route

Anagent-basedapproachtomodellingdriverroutechoicebehaviourundertheinfluenceofreal-timeinformationHusseinDiaDepartmentofCivilEngineering,TheUniversityofQueensland,Brisbane,Qld4072,AustraliaAbstractThispaperpresentsanagent-basedapproachtomodellingindividualdriverbehaviourunderthein-fluenceofreal-timetrafficinformation.ThedriverbehaviourmodelsdevelopedinthisstudyarebasedonabehaviouralsurveyofdriverswhichwasconductedonacongestedcommutingcorridorinBrisbane,Australia.Commuters
responsestotravelinformationwereanalysedandanumberofdiscretechoicemodelsweredevelopedtodeterminethefactorsinfluencingdrivers
behaviourandtheirpropensitytochangerouteandadjusttravelpatterns.Basedontheresultsobtainedfromthebehaviouralsurvey,theagentbehaviourparameterswhichdefinedrivercharacteristics,knowledgeandpreferenceswereidentifiedandtheirvaluesdetermined.Acasestudyimplementingasimpleagent-basedroutechoicedecisionmodelwithinamicroscopictrafficsimulationtoolisalsopresented.Driver-vehicleunits(DVUs)weremodelledasautonomoussoftwarecomponentsthatcaneachbeassignedasetofgoalstoachieveandadatabaseofknowledgecomprisingcertainbeliefs,intentionsandpreferencesconcerningthedrivingtask.EachDVUprovidedroutechoicedecision-makingcapabilities,basedonperceptionofitsenvironment,thatweresimilartothedescribedintentionsofthedriveritrepresented.Thecasestudyclearlydemonstratedthefeasibilityoftheapproachandthepotentialtodevelopmorecomplexdriverbehaviouraldynamicsbasedonthebelief–desire–intentionagentarchitecture.
2002ElsevierScienceLtd.Allrightsreserved.Keywords:Autonomousagents;Intelligenttransportationsystems;Travellerinformationsystems;Travelbehaviour;Microscopictrafficsimulation1.IntroductionTheprovisionofreal-timetravelinformationisincreasinglybeingrecognisedasapotentialstrategyforinfluencingdriverbehaviouronroutechoice,tripmaking,timesoftravelandmodeTransportationResearchPartC10(2002)331–349www.elsevier.com/locate/trcE-mailaddress:[e-mail](H.Dia).0968-090X/02/$-seefrontmatter
2002ElsevierScienceLtd.Allrightsreserved.PII:S0968-090X(02)00025-6choice.Understandingtravellers
responsetothisinformationisthereforecriticaltothedesignandimplementationofeffectiveintelligenttransportsystemsstrategiessuchasmobileorfixedad-vancedtravellerinformationsystems(ATIS).Thesesystemsprovidedriverswithreal-timeinfor-mationabouttrafficconditions,accidentdelays,roadworkandrouteguidancefromorigintodestination.Someofthemethodsusedforprovidingdriverswiththisinformationincludetrafficinformationbroadcasting,pre-tripelectronicrouteplanning,on-boardnavigationsystems,elec-tronicrouteguidancesystemsandstrategicallylocatedvariablemessagesigns(VMS).Theprin-cipalaimofthesesystemsistoinfluencedrivers
behaviouronroutechoiceanddeparturetimedecisionsinordertoimprovemobilityandreducetrafficcongestion.Despitetheobviousneedforassessinguseracceptanceandthepotentialimpactsofthesesystemsintermsofimprovingtrafficconditionsforindividualdriversandtheoveralltransportationsystem,therehasbeenalackofmodelstoevaluatetheirfullimpacts(Ben-Akivaetal.,1997).Typically,ATIShavebeenevaluatedthroughoperationaltests,travelsurveysortrafficsimulators(BonsallandParry,1991).Althoughthesetestsandexperimentsareveryuseful,theyareveryexpensivetoconductanddonotallowforeffectiveevaluationofdifferentalternatives.Computersimulationmodels,ontheotherhand,allowfortestingalternativesystemdesignsbeforeconductingoperationaltests,thusresultinginmoreeffectiveoperationaltestsandimplementation.Thesesimulationmodelstypicallyconsistoftwomaincomponents:adynamicdriverbehaviourmodelandatrafficsimulationmodel.Thisstudyaimstodemonstratethefeasibilityofusingautonomousagentsformodellingdy-namicdriverbehaviourandanalysingtheeffectsofATISontheperformanceofacongestedcommutingcorridorinBrisbane,Australia.Theoverallmodellingframeworkwillconsistofadecisioncomponentthatdeterminesdrivers
responsestothesuppliedinformationandatrafficsimulationcomponent.Theagent-baseddriverdecisioncomponentwillbeusedtodetermineindividualdrivers
preferencesandrouteswitchingdecisionsasafunctionofthesuppliedinfor-mation.UnlikepreviousstudieswhichmodelleddriverresponsetoATISbasedondatacollectedfromeithertravelsimulatorsortravelsurveys,thisresearchisbasedonabehaviouralsurveyofcongestiononareal-worldtrafficcommutingcorridor.Theresultsfromthebehaviouralsurveywereusedtodeterminethefactorsinfluencingroutechoicedecisionsandformedthebasisoftheagent-baseddriverbehaviourmodel.Thispaperwillfirstpresentanoverviewofagent-basedapplicationsintransportationengi-neering.Thiswillbefollowedbyanoverviewofexistingdynamicdriverbehaviourmodelsandsomeoftheirlimitations.Abehaviouralsurveyofdrivers,whichwasconductedonacongestedcommutingcorridorinBrisbane,willthenbediscussedandtheresultsfromanumberofdiscretechoicemodelsusedfordeterminingthefactorsinfluencingroutechoicewillbepresented.Theagent-basedframeworkformodellingdriverbehaviourandresponsetoinformationisthendis-cussedinthecontextofacasestudyonthesamecommutingcorridorwherethetravelbehavioursurveywasconducted.Anumberofmodelapplicationareasarethenidentifiedanddetailsofongoingresearcheff