交通灯外文文献

原文出处AustralasianTransportResearchForumAdelaideFUZZYLOGICTRAFFICSIGNALCONTROLZEESHANRAZAABDYPREPAREDFORDRNEDALT.RATROUTINTRODUCTIONSignalcontrolisanecessarymeasuretomaintainthequalityandsafetyoftrafficcirculation.Furtherdevelopmentofpresentsignalcontrolhasgreatpotentialtoreducetraveltimes,vehicleandaccidentcosts,andvehicleemissions.Thedevelopmentofdetectionandcomputertechnologyhaschangedtrafficsignalcontrolfromfixed-timeopen-loopregulationtoadaptivefeedbackcontrol.Presentadaptivecontrolmethods,liketheBritishMOVA,SwedishSOS(isolatedsignals)andBritishSCOOT(area-widecontrol),usemathematicaloptimizationandsimulationtechniquestoadjustthesignaltimingtotheobservedfluctuationsoftrafficflowinrealtime.Theoptimizationisdonebychangingthegreentimeandcyclelengthsofthesignals.Inarea-widecontroltheoffsetsbetweenintersectionsarealsochanged.Severalmethodshavebeendevelopedfordeterminingtheoptimalcyclelengthandtheminimumdelayatanintersectionbut,basedonuncertaintyandrigidnatureoftrafficsignalcontrol,theglobaloptimumisnotpossibletofindout.Asaresultofgrowingpublicawarenessoftheenvironmentalimpactofroadtrafficmanyauthoritiesarenowpursuingpoliciesto:−managedemandandcongestion;−influencemodeandroutechoice;−improvepriorityforbuses,tramsandotherpublicservicevehicles;−providebetterandsaferfacilitiesforpedestrians,cyclistsandothervulnerableroadusers;−reducevehicleemissions,noiseandvisualintrusion;and−improvesafetyforallroadusergroups.Inadaptivetrafficsignalcontroltheincreaseinflexibilityincreasesthenumberofoverlappinggreenphasesinthecycle,thusmakingthemathematicaloptimizationverycomplicatedanddifficult.Forthatreason,theadaptivesignalcontrolinmostcasesisnotbasedonpreciseoptimizationbutonthegreenextensionprinciple.Inpractice,uniformityistheprinciplefollowedinsignalcontrolfortrafficsafetyreasons.Thissetslimitationstothecycletimeandphasearrangements.Hence,trafficsignalcontrolinpracticearebasedontailor-madesolutionsandadjustmentsmadebythetrafficplanners.Themodernprogrammablesignalcontrollerswithagreatnumberofadjustableparametersarewellsuitedtothisprocess.Forgoodresults,anexperiencedplannerandfine-tuninginthefieldisneeded.Fuzzycontrolhasproventobesuccessfulinproblemswhereexactmathematicalmodellingishardorimpossiblebutanexperiencedhumancancontroltheprocessoperator.Thus,trafficsignalcontrolinparticularisasuitabletaskforfuzzycontrol.Indeed,oneoftheoldestexamplesofthepotentialsoffuzzycontrolisasimulationoftrafficsignalcontrolinaninter-sectionoftwoone-waystreets.Eveninthisverysimplecasethefuzzycontrolwasatleastasgoodasthetraditionaladaptivecontrol.Ingeneral,fuzzycontrolisfoundtobesuperiorincomplexproblemswithmultiobjectivedecisions.Intrafficsignalcontrolseveraltrafficflowscompetefromthesametimeandspace,anddifferentprioritiesareoftensettodifferenttrafficflowsorvehiclegroups.Inaddition,theoptimizationincludesseveralsimultaneouscriteria,liketheaverageandmaximumvehicleandpedestriandelays,maximumqueuelengthsandpercentageofstoppedvehicles.So,itisverylikelythatfuzzycontrolisverycompetitiveincomplicatedrealintersectionswheretheuseoftraditionaloptimizationmethodsisproblematic.BenefitsanddisadvantagesoffuzzysystemsFuzzylogichasbeenintroducedandsuccessfullyappliedtoawiderangeofautomaticcontroltasks.Themainbenefitoffuzzylogicistheopportunitytomodeltheambiguityandtheuncertaintyofdecision-making.Moreover,fuzzylogichastheabilitytocomprehendlinguisticinstructionsandtogeneratecontrolstrategiesbasedonprioricommunication.Thepointinutilizingfuzzylogicincontroltheoryistomodelcontrolbasedonhumanexpertknowledge,ratherthantomodeltheprocessitself.Indeed,fuzzycontrolhasproventobesuccessfulinproblemswhereexactmathematicalmodellingishardorimpossiblebutanexperiencedhumanoperatorcancontrolprocess.Ingeneral,fuzzycontrolisfoundtobesuperiorincomplexproblemswithmulti-objectivedecisions.Atpresent,thereisamultitudeofinferencesystemsbasedonfuzzytechnique.Mostofthem,however,sufferill-definedfoundations;eveniftheyaremostlyperformingbetterthatclassicalmathematicalmethod,theystillcontainblackboxes,e.g.defuzzification,whichareverydifficulttojustifymathematicallyorlogically.Forexample,fuzzyIF-THENrules,whichareinthecoreoffuzzyinferencesystems,areoftenreportedtobegeneralizationsofclassicalModusPonensruleofinference,butliterallythisnotthecase;therelationbetweentheserulesandanyknownmany-valuedlogiciscomplicatedandartificial.Moreover,theperformanceofanexpertsystemshouldbeequivalenttothatofhumanexpert:itshouldgivethesameresultsthattheexpertgives,butwarnwhenthecontrolsituationissovaguethatanexpertisnotsureabouttherightaction.Theexistingfuzzyexpertsystemsveryseldomfulfilthislattercondition.Manyresearchesobserve,however,thatfuzzyinferenceisbasedonsimilarity.Kosko,forexample,writes'Fuzzymembership...representssimilaritiesofobjectstoimpreciselydefinedproperties'.Takingthisremarkseriously,westudysystematicallymany-valuedequivalence,i.e.fuzzysimilarity.Itturnsoutthat,startingfromtheLukasiewiczwell-definedmany-valuedlogic,weareabletoconstructamethodperformingfuzzyreasoningsuchthattheinferencereliesonlyo