广义神经网络的研究及其在交通流预测中的应用

《广义神经网络的研究及其在交通流预测中的应用》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、大连理工大学硕士学位论文广义神经网络的研究及其在交通流预测中的应用姓名：苑文江申请学位级别：硕士专业：计算机应用技术指导教师：谭国真20050316广义神经网络的研究及其在交通流预测中的应用ResearchofGeneralizedNeuralNetworkandItsApplicationinTrafficFlowForecastingAbstractAsanimportantaspectofIntelligemTransportationSystemsors)，trigflowguidanceisconsideredasanopti

2、mumwaytoimprovetrRf并Cefficiencyandmobility．TheessentialoftheTrafficFlowGuidanceSystems(TFGS)aresupplingreal-timeexact蛐cinformation．Trafficflowisimportantinformationinurbantraffic，SOtrafficflowforecastinghasimportantsignificance．Therearemanyfactorsthatcarlinfluencethetraf

3、iCicflow，alloftheseresultsinthedifficultiesofreal-timewafficflowforecasting．Owingtothegoodadaptability，neuralnetworkhasbecomeacommonmodelforinformationforecasting．Basedontraditionalneuralnetwork,thispaperpresentsainteUigentneuronmodel，whichiscomposedoflinearlyindependemf

4、unctionsandSigmoidfunctionwithadjustableparameters．Itisprovedthattheinformationstorageabilityofthisintefiigemneuronisgreatlyimpmvedcomparedwithtraditionalones，consequentlygreatlyimprovestheinformationprocessingabilityofthewholeneuralnetwork．Meanwhile，inordertoreducethesi

5、zeoftheneuralnetwork’Sinput,thispaperUSeSthecorrelationtheorytoanalyzethecorrelationbetweenneighborroadsectiOIlS，andchoosethetra伍cflowofdifferentroadsections，whichhasstrongcorrelationwiththebeingforecastingoneasneuralnetwork’Sinputs，andestablishthetrafficflowmodelbasedon

6、generalizedneuralnetwork．Experimentresultsshowthat,thegeneralizedneuralnetworkconvergesfasterthantraditionalBPneuralnetwork,andmeetpracticalrequirementswell．Inordertogreatlyimprovedtheconvergespeedofgeneralizeneuralnetwork，thispaperdesignsaparalleltrairfingalgorithm，whic

7、hisbasedontrainingsetdecomposition．Thisparalleltrainingalgorithmusesanewcommunicationprofile．Thisprofilegreatlyreducesthecommunicationcostoftheparallelalgorithm．Experimentresultsshowthat，thisparalleltrainingalgorithmiseffectiveforreducingthetrainingtimeofgeneralizedneura

8、lnetwork．KeyWords：IntelligentNeuron；GeneralizedNeuralNetwork；TrafficFlowForecasting；ParallelComputing；G