研究解説 179 AnalysisofTraficDataColectionSystemsfor ImprovingSensorPlacements TianJIANG *,MarcMISKA **,ShinjiTANAKA *** andmasaokuwahara *** Abstract Inthispaper,weanalyzehighwaydetectionsystem underdiferenttraficscenarios.traficdatacolectiondependsonnot onlysensortechnologies,butalsowheredetectorsareplaced.datavalidityhighlyvariesondiferenttraficconditionsand detectionlocations.toasesthedetectioneficiency,agenericdatasupplyfunctionisdefinedtodescribedetectionsystem performance.inaddition,speeddatacolectionisselectedasanexampletodetermineaspecificdatasupplyfunctionwhich depictstherelationshipbetweenspeeddetectionvariationandtraficstate.thesupplyfunctionisvalidatedbyacasestudy basedontokyometropolitanexpresway. Introduction Themobilitydemandofoursocietyincreasedrapidlyinthe lastdecades.toensureaccesibilitythetraficinfrastructure needstobeusedmoreeficiently.onewaytoachievethisis dynamic trafic management(dtm) such as traveltime estimation,rampmetering,speed limits,accidentdetection, routeguidance,hardshoulderutilizationandetc.themore inteligence those systems have,the more they rely on information accuracy and reliability.ifsuch information is unavailable,thealgorithmscannotperform totheexpectations, andmighteventurnouttobecounterproductive.therefore, vitalforthetraficmanagementcycleistraficdatacolection toenablethefeedbacklooptotraficoperations. Roadsidedetectionisnowadaysthemostcommonwayof traficdatacolection.itisnecesarytohaveadetectionplan, whichiscosteficientanddesignedtoservethepolicyofthe road authority. The sensor technology and the network geometryaregeneralytakenintoaccountwhenroadauthorities planthedatacolectionsystem.howeverdatavaliditystrongly dependsontraficconditionswhicharenothighlyconcernedin theplanningstage. To analyze the detection eficiency ofa network,the relationshipbetweentraficdatadetectionvariationandthe traficstatehastobeconsidered.forinstance,ifadetectoris * DepartmentofCivilEngineering,TheUniversityofTokyo ** SmartTransportCenter,QueenslandUniversityofTechnology *** Advanced Mobility Research Center(ITS Center),Instituteof IndustrialScience,TheUniversityofTokyo placedinafreeflowroadlink,speeddatacouldrepresentthe vehiclespeedfortheentirelinkwhichmeansonedetectorcan providefulspeedknowledgeofthelink.ontheotherhand, whenthecongestionisoccured,moredetectorsneedtobe instaled to colect information in diferent trafic state segments.therefore,theguidanceforexploringdatadetection variation isvery importantforhighway detection system planning. Inafirststep,wereviewsensortechnologiesforspeeddata colectionandidentifylocationsofphysicalbotlenecksbased on network geometry.secondly,according to the trafic simulationresults,wedefinethedatasupplyfunctiontodepict detectionvariationofspeeddata.thenc1routeoftokyo MetropolitanExpreswayisselectedasacasestudytovalidate thesupplyfunction.attheendofthispaper,thesuitable detectorlocationisdiscusedforthepotentialimprovementsof thedetectionsystem. SensorTechnologies Trafic detection highly relies on sensor technologies. Publicationsdealingwiththediferentsensortechnologies,their operationandcostcanbefoundinklein(2001),fhwa(2006), Mimbelaetal.(2007),Marcetal.(2009).Wereviewroadside detectionforspeeddatacolectionandclasifyitintotwo groupsasshownintable1. Firstgroupofdetectorcanmeasuretheindividualvehicle speeddirectlywhilethevehiclepasingthedetectorlocation. Averagespeedindesigneddataaggregationtimeiscalculated fortraficmanagement.theothergroupofdetectorisnotable 63
180 63 巻 2 号 (2011) Table1.Fixeddetectorsforspeeddatacolection ramp),speed changes even in the free flow condition. Therefore,botleneckdetectionisnecesarytodeterminethe esentialareafortraficstatechanges.we summarizethe physicalbotleneckinformationintable2. Toidentifyotherbotleneckswhichdependontraficdemand andtoexplorehowspeeddatadetectionvarieswithtraficstate, we utilize dynamic trafic asignment (DTA) in trafic simulationtocolectspeeddataindiferenttraficscenarios. TraficsimulationbasedonDTA tomeasurespeeddirectly.instead,doubledetectorsarerequired tobeplacedneareachotherandvehiclespeediscalculated based on vehicle pasing time and distance between two detectors.therefore,detectorlocationisesentialfactorfor speeddatacolection. Botleneckdetection Botleneckisvitalforspeeddatadetection.Therearespecific pointsthatarenotoriousforcausingcongestiononadailybasis. Thistypeofbotleneckisrelatedtoitsphysicaldesign,we detectthesebotleneckbasedonhighwaystructureinformation and network analysis. For details of network analysis methodologywerefertojianget.al.(2010). Somebotlenecksareonlyactivewhentraficdemandisover roadcapacity.forinstance,iftraficdemandisinarelatively low level,speeddoesnotchangeathighwaymergingarea. However,when demand increases and congestion occurs, mergingareabotleneckisactiveandqueuestartstogrowtothe upstream.ontheotherhand,whenvehicleenteralow speed segmentfrom ahighspeedsegment(ex.from mainroadtoof SimulationbasedonDTAisusedforinvestigatingbotleneck activity and queue lengths.the properties ofdta have importantimplicationsonitsabilitytoportraytheactualtravel behavior,andhenceonthefidelityandaccuracyofthemodel results.fordetailsofdtatheory,werefertokuwaharaet.al. (2001)andTianet.al.(2010). A sample network is established by trafic simulation softwarevissim asshowninfigure1.inmanyrealisticurban highwaynetworks,fixeddetectorsareevenlyplacedincertain distance from severalhundredsofmetersto a couple of kilometers.asaresult,itisverydificulttodeterminethe traficstatebetweentwodetectors.however,wecouldaddas manydetectorsasneededinthesimulatednetworktoget detailedtraficinformationanddescribetraficstateaccurately. The network includes three origins (1,2,3) and three destinations (4,5,6). Data colection points are evenly distributedinevery50meterstorecordspeeddata.byutilizing DTA,WeadjustOD volumeasinputsofthesimulationto colectspeeddataunderdiferenttraficscenarios. Table2.Physicalbotleneckinformation Figure1.Samplenetworkfortraficsimulation Based on simulation results,we provide vehicle speed varyingwithtimenearatypicalbotlenecklocationinfigure2. Accordingtothespeeddistribution,itclearlyshowsthetrafic state,thebotlenecklocationandthequeuetail.wecould observeadecelerationsegmentbeforethequeuesegmentand anaccelerationsegmentafterit.thespeedintheupstream and 64
181 Figure2.Speeddistributionbasedonsimulationdata downstream freeflow segmentsisquiteconstantwithhigh speed.thespeedinqueuesegmentisalsoconstantbutwithlow speed.thisinformationwilbeusedtodeterminethespeed datasupplyfunction. Speeddatasupplyfunctionforfixeddetectors According to detectors placed in diferenttrafic state segments,two speed datasupply functionsaredefined as folowing: (1)Detectorplacedattheconstantspeedsegment Basedonstatisticanalysis,wedefinedetectioninfluencearea asthesegmentwhichthedetectorisplacedinanditsneighbor segments.basicaly,accordingtotraficstate,therearetwo typesofsegmentforspeed measures,the constantspeed segmentandthevariablespeedsegment.alconstantspeed suchasfreeflowandqueuesegmentsareconnectedbyvariable speedlikedeclarationandaccelerationsegments.basedonthe boundaryoftwodiferenttraficstatesegmentsandspeed information,wedevelopthespeeddatasupplyfunctionfor detectorplacedinconstantspeedsegmentasshowinfolowing equations: Baseonthesupplyfunction,wecreateestimatedsupply curveinfigure3andfigure4.detectorplacedineitherfree flow segmentorqueuing segmentcould provide accurate informationintheplacedsegment.datasupplydecreasesin variablesegments.howmuchinformationitcouldprovidefor itsneighborconstantspeedsegmentsdependsonthediference betweentheirtraficstates.aslongasthedetectorplacedin sameconstantspeedsegment,datasupplydoesnotreplyonthe detectorlocation. Figure3.Speeddatasupplyfordetectorplacedinfreeflowsegment Figure4.Speeddatasupplyfordetectorplacedinqueuingsegment (2)Detectorplacedatthevariablespeedsegment Comparedtodetectorplacedinconstantspeedsegment,data supplyissensitivetodetectorlocationwhenitisplacedin variable speed segment. The related supply function is introducedasfolowing: 65
182 63 巻 2 号 (2011) AsshowninFigure5,thedetectoronlyprovidestheful informationatdetectorlocationandcouldnotsupplyaccurate andstablespeedinformationfortheentirelink.therefore,to enhance the detection eficiency,we should avoid placing detectorsinthevariablespeedsegment Figure5.Speed datasupply fordetectorplaced in variablespeed segment Figure6.MEXC1Routenetworkinthesimulationbasedondynamic traficasignment Thereaderisreminded,thatbasedonlocalmeasurements, speed diference between two neighbor areas should be determined with localknowledge.the variables leftare boundarylocationssuchasthepositionofthebotleneckand thetailofthequeue.whilethebotlenecklocationusualyis static,thetailofthequeuekeepsmoving,dependingonthe demand.tobeabletogenerateaknowledgemapofanetwork, oritslinks,dtaisusedtogetanestimationofthequeuetail positionovertime. CaseStudy ByRunningDTAmodelsinVisim,botleneckactivitiesand queuelengthscanbeinvestigated.tovalidatethespeedsupply functionunderrealworldconditions,wehavechosenc1route oftokyometropolitanexpresway(mex)forthecasestudy. MostvehicleusingMEX areinstaledetcdevices,andetc dataincludeindividualentranceandexittimeoftheseequipped vehicle,thereforebyaggregatingtotalnumberofequippedcars withsameod pair,realod volumeofc1networkcanbe created.weselectmorningpeakfrom 6am to9am asstudy period,andprepare6continuousodmatricesasdtainputsto simulatetraficstatevariationovertime.thec1networkis created based on map information with the actualscale visualizedinfigure6 Accordingtothesimulationresults,traficstateofaten kilometerlinkwithseveralbotlenecksisvisualizedinfigure7, atimespacediagram.forinstance,aphysicalbotleneckis Figure7.TraficstateoftheC1networkinthemorningpeak locatedaround5000m locationandthequeuetailpositioncan bedeterminedbyspeeddiference. Withthisinformationonecancalculateaknowledgemap dependingonthetimeasshowninfigure8,whenadetectoris placedaround4000m location.theredareameansthedetector providefulknowledgeinthosepositionsovertime,whilethe blueareashowsnospeedinformationisavailable.atthe beginningofthesimulationperiod,sincemostpartsofthelink isinfreeflow condition,thedetectorcanprovideaccurate speeddatauptotheendofthelink.withthebotleneckactive 66
183 Figure9:Rankingofdetectorlocationswiththeircoverageareasbased onlinkknowledgemap Conclusion Figure8.Fulknowledgemapofthelinkprovidedbythedetectorat 4000location in thedownstream,theknowledgeareadecreasesand the detectorcanonlyprovidetheinformationuptothenearest queuetail.forthelast40minutes,whenthequeueisbackupto thedetectorlocation,thedetectorstartstocolectthespeed information in the queue.since the vehicle speed in the congestionareaisnotasconstantasinthefree-flowarea,the informationdetectorcanprovideinthequeueisnotsuch accurate. Takingthisfulknowledgemapasabasis,itisposibleto asesmonitoringinformationfrom diferentdetectorsetings, andtooptimizethedetectorplacement. SuitabledetectorlocationsbasedonDTA Takingnetworkinformationfrom historicaldataandthe knowledgemap,itisposibletodetermineprefereddetector location.withaknownbotlenecklocation,andahighpriority tomonitorifabotleneckisactive,theprimelocationfora detectoristhebotleneckitself,eventhoughtheinformation spanduringpeekconditionsisrelativelysmal.thislocationis folowedbytheupstream location,thatholdsinformationfor queueestimationalgorithmsandthedetectionofthestream behindthebotleneck,withoutanyadditionalinformation.the rankingisshowninfigure9. Inthispaperwehavedefineddatasupplyfunctionthat explorestherelationshipbetweentraficconditionsanddata colection.by utilizing DTA theory in trafic simulation, botleneckactivitiesandqueuetaillocationsareinvestigated, thatstronglyafectshighwaytraficdetection.basedonthis information,itisposibletoreachpotentialimprovementsof detectorplacementswhichwilcontributetohighwaydetection system asesmentandoptimization. (Manuscriptreceived.February2,2011) Reference 1) Klein,L.A.,DataRequirementsandSensorTechnologiesfor ITS,Norwood,MA,ArtechHouse,2001 2) U.S. Department of Transportation, Federal Highway AdministrationReportNo:FHWA-HRT-06-139,TraficDetector Handbook:ThirdEdition-Volume I,October2006 3) Mimbela,L.& Klein,L.A.,SummaryofVehicleDetectionand Surveilance Technologies used in InteligentTransportation Systems,2007 4) MarcMiska,TianJiangandMasaoKuwahara:TowardsCost Eficient Trafic Data Colection, Vol.40, Proceedings of Infrastructure Planning,Japan Society of CivilEngineers, 2009.11 5) Tian Jiang,Marc Miska and Masao Kuwahara:Highway NetworkAnalysisandDetectionAsesmentFramework,17th WorldCongresonInteligentTransportationSystems,Busan, Korea,2010.10 6) MasaoKuwahara,andTakashiAkamatsu:Dynamicuseroptimal asignmentwithphysicalqueuesforamany-to-manyodpatern, TransportationResearchPartB:Methodological,Volume 35, Isue5,June2001,Pages461-479 7) Tian Jiang,Marc Miska and Masao Kuwahara: Detector PlacementOptimizationBasedonDTAandEmpiricalData,The ThirdInternationalSymposium ondynamictraficasignment, Takayama,Japan,2010.07 67