ABSTRACT
ABSTRACT
Thisdissertationfocusedonsomeissuesonapproximatedynamicprogramming(ADP)anditsapplicationsintransportation.Themainworkandkeycontributionswere
assummarized
1.Basedthefollowing:Steffensen’Smethodandformerweights’values,anewoptimizedon
acceleratedalgorithmwaspresentedforneuralnetwork.Onthebasisofthisnewoptimizedacceleratedalgorithm,anewADHDP(Action.DependentHeurigicDynamicProgramming)method
proposed.Abasedonformerweights’valuewasmadedata(ADHDP(D)forshort)wasADHDPandADHDP(D).detaileQanalysisbetweenthe
Simulationresultsshowedthegoodperformanceofthisnewalgorithm.Furthermore,aninitialweightscompositionalmethodwasproposed.Simulationresultsdemonstratedtheconvergencepropertyisimprovedeffectivelyincomparisonwithinitialweightsrandomlypreassigned.
2.Consideringonthattrafficsystemis
onastrongnonlinearanduncertainsystem,anidentificationmethodbasedADPwasdevelopedtoestimatetheparametersofthe
generaldiscrete—timenonlineartrafficflowsystem.Withrigorousanalysis,itwasshownthattheproposedidentificationscheme,independentoftheprecisetrafficflow
tomodel,canguaranteetheconvergence.Anumberofsimulationresultswereprovided
theefficacyoftheproposedapproach.
3.Anadaptivesingle—exponentsmoothingbasedonADPwasputforwardtoselectthesmoothingcoefficientdynamically.Withrigorousanalysis,itwasshownthattheproposedpredictionschemecanguaranteetheconvergence.Thesimulationresultsverifidtheeffectivenessoftheproposedalgorithm.
4.Theoptimal
onsignaltimingproblemwasinvestigatedforanurbanintersection.BasedthetechniqueADP,theoptimalsignaltimingcontrollerswere
proposedforthetwo—phaseintersectionandthree—phaseintersection,respetively.ThesimulationexamplesshowedthatthecontrolalgorithmbasedonADPcanallocategreentimerationallyandachievetheequilibriumofqueuelength.Furthermore,takingtheoptimalsignaltimingofthree—phaseintersectionforexample,simulationsverifidthattheinitialweightsaffecttheconvergenceresult.
5.BasedonthetechniqueofADRtheproblemsoflocalrampandcoordinatedramp
toavoidthemeteringwerediscussed.TheADPcontrollerwasdesigneddifficultyof
trafficflowmodeling.SimulationresultsdemonstratedthenewcontrollershavebetterV