多因素融合下基于AGC-LSTM的短时交通速度预测

打开文本图片集

引用格式：，，，等.多因素融合下基于AGC-LSTM的短时交通速度预测[J].现代电子技术，2025，48（18）：9-16.

关键词：智能交通；短时交通速度预测；特征融合；组合深度学习；图卷积网络；长短期记忆网络；注意力机制中图分类号：TN711-34；U491.14 文献标识码：A 文章编号：1004-373X（2025）18-0009-08

Short-time traffic speed prediction based on AGC-LSTM under multi-factor fusion

CHENYujia1，GAOMingxial，XIANGWanli1，MOJunwen² （1.SchoolofTraficand Transportation，Lanzhou JiaotongUniversity，Lanzhou73oo7O，China; 2.School of Economics and Management，Lanzhou JiaotongUniversity，Lanzhou 73oo7o，China）

Abstract：Inorder to more accuratelypredictthe dynamic changes of short-term traffc speed in rapidlychanging urban environments，anexteralfctorsfusecomponent（EF-Component）isonstructedbyconsideringhistoricaldataweatheractors， andsurounding pointsof interest （POIs），andfurther incorporatingroadconditionfactors.Basedontheexisting deplearning model，theurbanshort-termtraffcspeedpredictionmodelintegratinggraph-convolutionalnetwork（GCN），lng-short-rm memory （LSTM）network，andatentionmechanismundertheeffectof multiplefactors，EF-AGC-LSTM（atention-enhancedGCN embeddedLSTMwithmulti-factorfusion）isstudied.TheEF-Componentisusedtosynthesizemulti-influencingfactors，spatio temporal featuresof traffcspeedare extracted bymeansofGCNandLSTM，whichembeds theGCNintothe LSTM gating computation tosynchronouslyobtainspatio-temporalfeaturesofdata，andthenatention mechanismisusedtoautomatically identifyandenhancetheinfluencefeaturesofkeyexternalfactors，soastoimprovetheperformanceofthemodel.Theexample validationonavelocity datasetfrom Luohu District，Shenzhen，Chinawasconducted.Theresultsshowthatincomparison with thebaselinemodel，thepredictioneffctofEF-AGC-LSTMisimprovedgreatly.IncomparisonwiththetraditionalGCN-LSTM model，the meanabsoluteeror（MAE）androot meansquare error （RMSE）of predictionarereduced by4.3%and 3.3%， respectively，and theaccuracyisimprovedby1.4%.The predictedMAEandRMSEarereducedby1.22%and 0.87% respectively，afterintroducingtheroadconditionfactors.Thecomprehensiveconsiderationofmultipleinfluencingfactorscan makethe predictionofshort-termtraficspeedbeimprovedfurther，andtheEF-AGC-LSTMmodelcanwellrealizetheprediction of short-term trafic speed under the fusion of multiple factors to meet the needs of urban management.

Keywords：intellgent transportation;short-term traffcspeed prediction;feature fusion;combinatorialdp learning;graph convolutional network;longand short-termmemorynetwork;attentionmechanism

0 引言

随着城市化进程的加快，城市交通问题日益突出，短时交通速度预测成为智能交通系统（ITS）中的一个关键环节。（剩余11470字）