1、大連理工大學(xué)碩士學(xué)位論文城市路網(wǎng)實(shí)時(shí)動(dòng)態(tài)交通信息預(yù)測(cè)方法的研究姓名：史慧敏申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：計(jì)算機(jī)應(yīng)用技術(shù)指導(dǎo)教師：譚國(guó)真20091217城市路網(wǎng)實(shí)時(shí)動(dòng)態(tài)交通信息預(yù)測(cè)方法的研究StudyonReal—TimeDynamicTrafficInformationPredictioninUrbanRoadNetworksAbstractIntelligentTransportationSystems(ITS)istheintematio

2、nallyrecognizedbestwaytelieveurbantra]fjficcongestionRoadtrafficinforrnationisanindispensableprerequisiteandmaincontentforallITSprojectsT11eproblemthathowtogetthesemessagesinashorttimeandhowtOusetheseinforrnationstodeter

3、minequicklytheoptimaldrivingpathhasbecomeacuttingedgeissuesinthefieldofITS，trafficnetworkshorttimetrafficinformationpredictiontheory，theprosandconsofmodelandalgorithmdirectlyaffectthewholeITSimplementationThetrafficflowa

4、ndtraveltimethroughsectionisthemostbasicparametersoftrafficinformationItnotonlyrepresentsthephysicalpropertiesandtransportpropertiesofsomesection，butalsoisthemostdirectinterestconcernedbyusersShorttermtrafficinformationp

5、redictioncallproviderealtimetrafficinformationthroughurbantrafficinformationreleaseplatformforthetrip，alsohelpthemtothepathselectionandinduction乃eartificialneuralnetworkisanewmethodofmathsmodelingBecauseofbetteradaptabil

6、ity，ithasbeenwidelyappliedintrafficforecastingBackpropagation(BP)algorithmisusuallyusedtotraintheneuralnetwork，butitconvergesslowlyandeasilygetsintolocalminimumTospeeduplearningofneuralnetwork，manyparallellearningalgorit

7、hmsareproposednisthesisusesakindofparallelnonlinearoptimizationtechniquetotrainnetworks，realizingparallelismbasedonlearningalgorithmBPalgorithmisimprovedbyparallelselfscalingquasi—Newton(SSPQN)algorithmIneachiteration，as

8、etofsearchdirectionsisgeneratedEachsubtaskcarriesoutinexactlinearsearchalongeachdirectiontofindtheoptimalpointThisthesistestsandanalysestheabovealgorithmintheMPICHparallelenvironmentTheexperimentalresultshowsthatwiththes

9、ametrainingprecisiontheSSPQNparallelalgorithmeffectivelyimprovestheconvergencespeed，hasbetterforecasteffectsthantraditionalBPneuralnetwork，andmeetstherequirementoftrafficflowforecastingInordertoachieveapredictionoftraffi

10、cvolumeinarbitrarysectionofthenetwork，thethesisusemultivariatestatisticalanalysisofthemultidimensionalscalingofurbanroadnetworkcorrelationanalysisfornondetectorsectionitachievesthedivisionoftheentirenetwork，andprediction