交通信号智能控制系统外文文献及翻译doc.docx
《交通信号智能控制系统外文文献及翻译doc.docx》由会员分享,可在线阅读,更多相关《交通信号智能控制系统外文文献及翻译doc.docx(14页珍藏版)》请在冰豆网上搜索。
交通信号智能控制系统外文文献及翻译doc
Agentcontrolledtrafficlights
Author:
DankoA.Roozemond,JanL.H.Rogier
Provenance:
DelftUniversityofTechnology
Introduction
Thequalityof(urban)trafficcontrolsystemsisdeterminedbythematchbetweenthecontrolschemaandtheactualtrafficpatterns.Iftrafficpatternschange,whattheyusuallydo,theeffectivenessisdeterminedbythewayinwhichthesystemadaptstothesechanges.Whenthisabilitytoadaptbecomesanintegralpartofthetrafficcontrolunititcanreactbettertochangesintrafficconditions.Adjustingatrafficcontrolunitisacostlyandtimelyaffairifitinvolveshumanattention.Thehypothesisisthatitmightofferadditionalbenefitusingself-evaluatingandself-adjustingtrafficcontrolsystems.Thereisalreadyamarketforanurbantrafficcontrolsystemthatisabletoreactiftheenvironmentchanges;thesocalledadaptivesystems."Real"adaptivesystemswillneedpro-activecalculatedtrafficinformationandcycleplans-basedonthesecalculatedtrafficconditions-tobeupdatedfrequently.
Ourresearchoftheusabilityofagenttechnologywithintrafficcontrolcanbesplitintotwoparts.Firstthereisatheoreticalpartintegratingagenttechnologyandtrafficcontrol.Thefinalstageofthisresearchfocusesonpracticalissueslikeimplementationandperformance.Herewepresenttheconceptsofagenttechnologyappliedtodynamictrafficcontrol.Currentlywearedesigningalayeredmodelofanagentbasedurbantrafficcontrolsystem.Wewillelaborateonthatinthelastchapters.
Adaptiveurbantrafficcontrol
Adaptivesignalcontrolsystemsmusthaveacapabilitytooptimisethetrafficflowbyadjustingthetrafficsignalsbasedoncurrenttraffic.Allusedtrafficsignalcontrolmethodsarebasedonfeed-backalgorithmsusingtrafficdemanddata-varyingfromyearstoacoupleofminutes-inthepast.Currentadaptivesystemsoftenoperateonthebasisofadaptivegreenphasesandflexibleco-ordinationin(sub)networksbasedonmeasuredtrafficconditions(e.g.,UTOPIA-spot,SCOOT).Thesemethodsarestillnotoptimalwheretrafficdemandchangesrapidlywithinashorttimeinterval.Thebasicpremiseisthatexistingsignalplangenerationtoolsmakerationaldecisionsaboutsignalplansundervaryingconditions;butalmostnoneofthecurrentavailabletoolsbehavepro-activelyorhavemeta-rulesthatmaychangebehaviourofthecontrollerincorporatedintothesystem.Thenextlogicalstepfortrafficcontrolistheinclusionofthesemeta-rulesandproactiveandgoal-orientedbehaviour.Thekeyaspectsofimprovedcontrol,forwhichcontributionsfromartificialintelligenceandartificialintelligentagentscanbeexpected,includethecapabilityofdealingwithconflictingobjectives;thecapabilityofmakingpro-activedecisionsonthebasisoftemporalanalysis;theabilityofmanaging,learning,selfadjustingandrespondingtonon-recurrentandunexpectedevents(Ambrosinoetal..,1994).
Whatareintelligentagents
Agenttechnologyisanewconceptwithintheartificialintelligence(AI).TheagentparadigminAIisbaseduponthenotionofreactive,autonomous,internally-motivatedentitiesthatinhabitdynamic,notnecessarilyfullypredictableenvironments(Weiss,1999).Autonomyistheabilitytofunctionasanindependentunitoveranextendedperiodoftime,performingavarietyofactionsnecessarytoachievepre-designatedobjectiveswhilerespondingtostimuliproducedbyintegrallycontainedsensors(Ziegler,1990).Multi-AgentSystemscanbecharacterisedbytheinteractionofmanyagentstryingtosolveavarietyofproblemsinaco-operativefashion.BesidesAI,intelligentagentsshouldhavesomeadditionalattributestosolveproblemsbyitselfinreal-time;understandinformation;havegoalsandintentions;drawdistinctionsbetweensituations;generalise;synthesisenewconceptsand/orideas;modeltheworldtheyoperateinandplanandpredictconsequencesofactionsandevaluatealternatives.Theproblemsolvingcomponentofanintelligentagentcanbearule-basedsystembutcanalsobeaneuralnetworkorafuzzyexpertsystem.Itmaybeobviousthatfindingafeasiblesolutionisanecessityforanagent.Oftenlocaloptimaindecentralisedsystems,arenottheglobaloptimum.Thisproblemisnoteasilysolved.Thesolutionhastobefoundbytailoringtheinteractionmechanismortohaveasupervisingagentco-ordinatingtheoptimisationprocessoftheotheragents.
IntelligentagentsinUTC,ahelpfulparadigm
AgenttechnologyisapplicableindifferentfieldswithinUTC.Theonesmostimportantmentioningare:
informationagents,agentsfortrafficsimulationandtrafficcontrol.Currently,mostapplicationsofintelligentagentsareinformationagents.Theycollectinformationviaanetwork.Withspecialdesignedagentsuserspecificinformationcanbeprovided.Inurbantraffictheseintelligentagentsareuseableindeliveringinformationaboutweather,trafficjams,publictransport,routeclosures,bestroutes,etc.totheuserviaaPersonalTravelAssistant.Agenttechnologycanalsobeusedforaggregatingdataforfurtherdistribution.Agentsandmultiagentsystemsarecapableofsimulatingcomplexsystemsfortrafficsimulation.Thesesystemsoftenuseoneagentforeverytrafficparticipant(inasimilarwayasobjectorientedprogramsoftenuseobjects).Theapplicationofagentsin(Urban)TrafficControlistheonethathasourprimeinterest.Hereweultimatelywanttouseagentsforpro-activetrafficlightcontrolwithon-lineoptimisation.Signalplansthenwillbedeterminedbasedonpredictedandmeasureddetectordataandwillbetunedwithadjoiningagents.Themostpromisingaspectsofagenttechnology,theflexibilityandpro-activebehaviour,giveUTCthepossibilityofbetteranticipationoftraffic.CurrentUTCisnotthatflexible,itisunabletoadjustitselfifsituationschangeandcan'thandleun-programmedsituations.Agenttechnologycanalsobeimplementedonseveraldifferentcontrollayers.ThisgivestheadvantageofbeingclosetocurrentUTCwhileleavingconsiderablefreedomatthelower(intersection)level.
Designingagentbasedurbantrafficcontrolsystems
Theidealsystemthatwestriveforisatrafficcontrolsystemthatisbasedonactuatedtrafficcontrollersandisabletoproactivelyhandletrafficsituationsandhandlingthedifferent,sometimesconflicting,aimsoftrafficcontrollers.Theproposeduseoftheconceptofagentsinthisresearchisexperimental.
Assumptionsandconsiderationsonagentbasedurbantrafficcontrol
Therearethreeaspectswhereagentbasedtrafficcontroland-managementcanimprovecurrentstateoftheartUTCsystems:
-Adaptability.Intelligentagentsareabletoadaptitsbehaviourandcanlearnfromearliersituations.
-Communication.Communicationmakesitpossibleforagentstoco-operateandtunesignalplans.
-Pro-activebehaviour.Duetotheproactivebehaviourtrafficcontrolsystemsareabletoplanahead.
Tobeacceptableasreplacementunitforcurrenttrafficcontrolunits,thesystemshouldperformthesameorbetterthancurrentsystems.TheagentbasedUTCwillrequireon-lineandpro-activereactiononchangingtrafficpatterns.AnagentbasedUTCshouldbedemandresponsiveaswellasadaptiveduringallstagesandtimes.Newmethodsfortrafficcontrolandtrafficpredictionshouldbedevelopedascurrentonesdonotsufficeandcannotbeusedinagenttechnology.Theadaptabilitycanalsobedividedinseveraldifferenttimescaleswherethesystemmayneedtohandleinadifferentway(Rogier,1999):
-gradualchangesduetochangingtrafficvolumesoveralongerperiodoftime,
-abruptchangesduetochangingtrafficvolumesoveralongerperiodoftime,
-abrupt,temporal,changesduetochangingtrafficvolumesoverashortperiodoftime,
-abrupt,temporal,changesduetoprioritisedtrafficoverashortperiodoftime
Onewayofhandlingthebalancebetweenperformanceandcomplexityistheuseofahierarchicalsystemlayout.Weproposeahierarchyofagentswhereeveryagentisresponsibleforitsownoptimalsolution,butmaynotonlybeinfluencedbyadjoiningagentsbutalsoviahigherlevelagents.Theseagentshavethetaskofsolvingconflictsbetweenlowerlevelagentsthattheycan'tsolve.Thisrepresentscurrenttrafficcontrolimplementationsandidea's.Onefinalaspecttobementionedistherobustnessofagentbasedsystems(ifallcommunicationfailstheagentrunson,iftheagentfailsafixedprogramcanbeexecuted.
Tobeabletokeepourfirsturbantrafficcontrolmodelassimpleaspossiblewehavemadethefollowingassumptions:
welimitourselvestoinnercitytrafficcontrol(roadsegments,intersections,corridors),wehandleonlycontrolledintersectionswithdetectors(intensityandspeed)atallroadsegments,weonlyhandlecarsandweusesimplerulebasesforknowledgerepresentation.
Typesofagentsinurbanintersectioncontrol
Aswedividethesysteminseveral,recognisable,partswedefinethefollowing4typesofagents:
-Roadsarerepresentedbyspecialroadsegmentagents(RSA),
-Controlledintersectionsarerepresentedbyintersectionagents(ITSA),
-Forspecific,defined,areasthereisanareaagent(higherlevel),
-Forspecificroutestherecanberouteagents,thatspansseveraladjoiningroadsegments(higherlevel).
Wehavenotchosenforoneagentpersignal.Thismayresultinamoresimplesolutionbutavailabletrafficcontrolprogramsdonotfitinthatkindofagent.Wedeliberatelychooseamorecomplexagenttobeabletousestandardtrafficcontroldesignalgorithmsandprograms.Theideastillistheoptimisationonalocallevel(intersection),butwithlocalandglob