Research Collection

Conference Paper

Dynamic ride sharing implementation and analysis in MATSim

Author(s): Wang, Biyu; Liang, Hong; Hörl, Sebastian; Ciari, Francesco

Publication Date: 2017-09-12

Permanent Link: https://doi.org/10.3929/ethz-b-000183727

Rights / License: In Copyright - Non-Commercial Use Permitted

This page was generated automatically upon download from the ETH Zurich Research Collection. For moreinformation please consult the Terms of use.

ETH Library

Dynamic Ride Sharing Implementation and Analysis in MATSim

Biyu Wang Institut für Verkehrsplanung und Transportsysteme (IVT) Bahnhaldenstrasse 9, 8052 Zurich wangb@student.ethz.ch

Hong Liang Institut für Verkehrsplanung und Transportsysteme (IVT) Paul-Feyerabend-Hof 5 8049 Zurich hliang@student.ethz.ch

Sebastian Hörl Institut für Verkehrsplanung und Transportsysteme (IVT) HIL F 33.3 Stefano-Franscini-Platz 5 8093 Zurich sebastian.hoerl@ivt.baug.ethz.ch

Dr. Francesco Ciari Institut für Verkehrsplanung und Transportsysteme (IVT) HIL F 33.2 Stefano-Franscini-Platz 5 8093 Zurich ciari@ivt.baug.ethz.ch

ABSTRACT

Introduction

In many urban areas, low occupancy rate of private cars leads to the aggravation of trafficexternalities, such as severe traffic congestion and pollutant emissions. To solve these problems,approaches such as traditional ridesharing (also referred as carpooling) were proposed as analternativeformoftravel,whichwouldmovemorepeoplewhileusingtheexistinginfrastructureandwouldusevehiclesmoreefficiently.[1]IntheUnitedStates,HOV(high-occupancyvehicle)laneshavealreadybeenconstructedtoencourageridesharingsince1991,culminatingwiththepassageoftheIntermodalSurfaceTransportationEfficiencyAct,which favoredhigh-occupancyvehicle (HOV) laneconstruction[2]. However, carpooling requires long- term commitments among people and impliesthemhavingfixedschedulesandoriginanddestinationpoints,whichisnotsuitableforthefastpaceof life in today’s cities.Thus, carpoolingusage rate forwork tripsdecreasedsignificantly in theUSduringtheperiodofthe1970s–2000s,[3]peakingin1970withacommutemodeshareof20.4%,by2011itwasdownto9.7%.[4]

Nowmobileinternettechnologyubiquityhasrenewedtheinterestaroundridesharingasapossibleway tomitigate trafficexternalities. In the last fewyearsdynamic ridesharing, alsoknownas real-time ridesharing, gained traction. Compared with traditional carpooling, a dynamic ridesharingsystemcanprovidematchesbetweendriversandpassengersonvery shortnotice,whichmeansaride-sharecanbeestablishedonlyafewminutesbeforedeparturetime.Additionally, inadynamicride sharing system, each trip is considered individually, and can accommodate by design tripsfrom/toanypointatanytime[3],whileadrivercangetamatchatanypointalonghisride.

Thedevelopmentofalgorithms foroptimalmatching in real-timeand for fastdetour computationhavebeentackledrecentlybyseveralresearchers.[5-8]However,asforthefurtherstudyofitseffectonnetwork trafficperformanceandon the individual travelbehavior, it is stillnearlyblank. In thefew attempts documented in the literature, given the difficulties in dealing with such problemsanalytically,simulationshavebeenusedtotacklethem.

This paper reports on a research effort were the agent-based simulation MATSim is used and itaimedat

a) ImprovingthealreadyexistingcapabilityofMATSimofdealingwiththeDynamicVehicleroutingproblem;

b)Assessingtheimpactofaride-sharingserviceonthetransportationsystemaswellastoprovideatheoreticalbasisforrelevantpoliciesandmeasurements.

Methodology

TheMulti-AgentTransportSimulation(MATSim)isaplatform[9],canperformnetworkloadingswithmillions of persons or vehicles (represented through the agent paradigm) and trace each agentthroughoutthewholeday.

InMATSim’saDynamicVehicleRoutingProblem(DVRP)extensionwasalreadydevelopedandusedto simulate and analyze regular and shared taxi services. [10-14]That contribution creates a dynamicvehicle routing system, using the concept of dynamic agent for the taxi vehicles. Unlike regularagents, this kind of agents, are notified of any new relevant event during the simulation, and re-routedtakingintoaccountthecurrentsituation[15].However,theDVRPisespeciallydesignedforthetaximode,whichsharessimilaritiesbutalsohavesomefundamentaldifferenceswithride-sharing.Fromasimulationstandpoint,taxisalsohavedynamicactivitychains,pickupanddropoffactivitiesandneedtoberouteddynamically.However,taxiagentsdonotbelongtothepopulationofagentsandthereforetheirplansdonotneedtogetascoreortobereplanned.Thisisclearlydifferentforridesharingdriversandpassengers.Therefore,thecurrentDVRPframeworkhadtobeupdated,anda specificRideShareAgentwascreated to represent ridesharingsystems.Onceanagentchoosesride sharedriveras its tripmode, itwill act asRideShareAgent,whichhas theattributesofbothdynamicagentandnormalagent.RideShareAgentwillswitchtoDynamicAgentwhenitisonthelegof ridesharedrivermodeandwill switchbacktopopulationagentwhen itdecidestoexecutethenextactivity.Theconcreteimplementationisasfollows:

Figure1ImplementationofRideShareAgent

Request allocation is another core problem of the ride-share dynamic system, which should behandled in both temporal and spatial aspect. This problem has already attracted attentions fromsomescholars.Geisbergerdevelopedanalgorithmtosolvethefastdetourprobleminridesharing.[7]AndShuoMa’salgorithmdealtwith theride-share taxi searchingandscheduling.[8]Basedontheseresults,therequestallocationsystemwasdevelopedasfollows:

Figure2RequestAllocationSystem

Likewise, a reasonable scoring configuration is also a decisive issue in the new model, whichdeterminestherelationshipbetweensupply,theshareofridesharedriver,anddemand,theshareofridesharepassenger.

Casestudyandpreliminaryresults

Some preliminary tests, executed on scenario of Sioux Falls (SD, USA)[16], allowed testing thefunctionalities introduced. Sioux Falls is a small city and themodel has 24 zones, 76 links and 24nodes[17]. It was chosen to test the new ride-sharing implementation because, while being acompleteandtoalargeextentrealisticscenario,itisneverthelesssimpleandsmallenoughthatthecomputation time isnot toohigh. To further simplify theproblem, themaximumcapacityof eachvehicleis2whichmeanseachdrivercanonlyhaveonepassengerinthecar.Otherconfigurationsofvehicleareexactlysameasvehiclesofcarmode.

In thissimulationcasestudy,driveragentsplansareevaluatedwithaspecialscoringsystem.Theygetpositiveutilitywhentheyareonalinkwithpassengers,butgetnegativeutilityasthenormalcarmode when driving alone. (See more details in Table 1) The parameters of rideshare driver andridesharepassengerarebasicallydeterminedbytheprice.Forexample,passengerneedtopaydriver30 dollars per hour. By convertingmonetary cost to utility, passengerwill lose 30 * 0.062 = 1.86

utilityperhour,whichissetastravellingRideSharePassengerintheconfiguration.Atthesametime,afterexcludingthenormalcarconsumption,ridesharedriverwillget1.86–0.992=0.868utilityperhour,which issetastravellingRideShareDriver intheconfiguration.Thesimulationwill runfor100iterations, in the first 80 rounds agent will have the possibility to changemode, using a randomimitator,andinthelast20roundsagentwillchoosethemodewithhighestscore.

Table1:Configuration

Parameter value unit

writeExperiencedPlans TURE

BrainExpBeta 1

constantPt -0.124 utils

constantCar -0.562 utils

onstantWalk 0 utils

constantRideSharePassenger 0 utils

constantRideShareDiver -0.562 utils

earlyDeparture -1.5 utils/hr

lateArrival -2 utils/hr

learningRate 0.4

marginalUtilityOfMoney 0.062 utils/unit_of_money

marginalUtlOfDistanceWalk 0 utils/m

monetaryDistanceRateCar 0 unit_of_money/m

monetaryDistanceRatePt 0 unit_of_money/m

monetaryDistanceRateRideSharePassenger 0 unit_of_money/m

monetaryDistanceRateRideShareDriver 0 unit_of_money/m

performing 0.96 utils/hr

traveling -0.992 utils/hr

travelingPt -0.18 utils/hr

travelingWalk -1.14 utils/hr

travelingRideShareDriver 0.868 utils/hr

travelingRideSharePassenger -1.86 utils/hr

utilityOfLineSwitch 0 utils

waitingPt -0.18 utils/hr

Figure3ModeShareofBasicScenario

Figure4ModeShareafterRideShareImplementation

Asthepresentresultsshows,thenewtripmodeofridesharingindeedaffectstrippatternsofagentscomparedtothebasescenario.Currentlythefarerateforridesharemodeiszeroandchangingmoderandomlyisencouragedforeachiteration,whichcouldexplaintherelativelyevenlydistributedmodeshare.Aproperestimationoftheparameters,basedonrevealedorstatedpreferences,aswellassettingarealisticpricewouldobviouslyinfluencenotonlythebehaviorofridesharepassengersanddrivers,butalsothebalanceofthedemand-supplyrelationship.Thiswouldbeneededinordertohaveresultswhichcouldbeusedforpolicymaking.

25%

33%

42%

car pt walk

16%

21%

21%

22%

20%

car pt walk ridesharedriver ridesharepassenger

Figure5ExecutedScoreofdifferenttripmodesinbasicscenario

Figure6ExecutedScoreofdifferenttripmodeswithridesharing

Thefurtherexpectedresultwillbetrafficperformanceofthenewridesharemodecomparedwithothertrafficmodes,aswellastherelationshipbetweennewmodeshareandvariouspricingstrategies.Theworkwillbethereforeextendedtofindoptimalpricingstrategiesinordertoreachdifferenttargets(VKMminimization,welfaremaximization,etc.).

1. RobertoBaldacci,V.M.,AristideMingozzi,AnExactMethodfortheCarPoolingProblemBasedonLagrangeanColumnGeneration.OperationsResearch,2004.52.3:p.422-439.

2. JianLingLi,P.E.,StephenMattingly,KavehSadabadi,IsaradattaRasmidatta,MarkBurris,WhoChoosestoCarpoolandWhy?:examinationofTexascarpoolers.TransportationResearchRecord:JournaloftheTransportationResearchBoard,2007.2021:p.110-117.

3. AmberLevofsky,A.G.,Organizeddynamicridesharing:Thepotentialenvironmentalbenefitsandtheopportunityforadvancingtheconcept.TransportationResearchBoard2001AnnualMeeting,2001.

4. NelsonD.Chan,S.A.S.,Ridesharinginnorthamerica:Past,present,andfuture.TransportReview,2012.32(1):p.93-112.

5. NielsA.H.Agatz,A.L.E.,MartinW.P.Savelsbergh,XingWang,SustainablePassengerTransportation:DynamicRide-Sharing.ReportSeriesResearchinManagement,2010:p.30.

6. NielsA.H.Agatz,A.L.E.,MartinW.P.Savelsbergh,XingWang,Dynamicride-saring:AsimulationstudyinmetroAtlanta.TransportationResearchPartB,2011.45:p.1450-1464.

7. RobertGeisberger,D.L.,SabineNeubauer,PerterSander,LarsVolker,FastDetourComputationforRideSharing.2009.

8. ShuoMa,Y.Z.,OuriWolfson,T-Share:Alarge-scaledynamictaxiridesharingservice.DataEngineering(ICDE),2013IEEE29thInternationalConference,2013:p.410-421.

9. Horni,A.,Nagel,K.andAxhausen,K.W.(eds.)TheMulti-AgentTransportSimulationMATSim.2016:London:UbiquityPress..DOI:http://dx.doi.org/10.5334/baw.License:CC-BY4.0.

10. JoschkaBischoff,M.M.,Agent-basedsimulationofelectrictaxicabfleets.TransportationResearchProcedia,2014.4:p.191-198.

11. MichalMaxiejewski,K.N.,SimulationanddynamicoptimizationoftaxiservicesinMATSim.TransportSystensPlanningandTrasportTlematics,2013.

12. MichalMaxiejewski,K.N.,TowardsMulti-AgentSimulationoftheDynamicVehicleRoutingProbleminMATSim.nternationalConferenceonParallelProcessingandAppliedMathematics,2012:p.551-560.

13. YeqianLin,W.L.,FengQiu,HeXu,ResearchonOptimizationofVehicleRoutingProblemforRide-sharingTaxi.Procedia-SocialandBehacioralSciences,2012.43:p.494-502.

14. MichalMaxiejewski,J.B.,KaiNagel,AnAssignment-BasedApproachtoEfficientReal-TImeCity-ScaleTaxiDispatching.IEEEIntelligentSystems,2016.31:p.68-77.

15. MichalMaxiejewski,M.,KaiNagel,Towardsmulti-agentsimulationofthedynamicvehicleroutingprobleminMATSim,inInternationalCoferenceonParallelProcessingandAppliedMathematics.2011:SpringerBerlinHeidelberg.

16. SebastianHörl,A.E.,K.Axhausen,Implementationofanautonomoustaxiserviceinamulti-modaltrafficsimulationusingmatsim.2016,ChalmersUniversityofTechnology,:Gotheburg.

17.Github.[online]https://github.com/bstabler/TransportationNetworks.[AccessedNov2016]