一种时空融合自适应低通多图卷积神经网络

打开文本图片集

关键词：交通流量预测；图卷积神经网络；深度学习；智慧城市；性能提升中图分类号： 文献标志码：A 文章编号：2095-2945（2025）30-0050-05

Abstract：Thispaperproposesaspatiotemporal fusionadaptivelow-passmulti-graphconvolutionalnetwork （STFALMGCN） fortraffcflowprediction.First，weproposeacausalgating linearunitmoduletoextractthetemporalcharacteristicsof trffic flow，usecausalconvolutiontogreatlyreducethetrainingparameters，andthegatinglinearunitspedsupthetrainingspeed. Secondly，adynamicallylearmedadaptiveadjacencymatrixisconstructedusingmultipleinformationandbasedonthespatial informationgivenbythegraphstructuretobuildaspatialgraph，andglobalinformationandhideninformationarefulyutilized tofindtheoptimalcorelationgraph.Finalltemporalandspatialfeauresareintegratedtofurtherimprovepredictionaccuracy. Experimentswereconductedontworeal-worlddatasets，andtheexperimentalresultsshowedthatourmodelperformedbeter prediction accuracy.

Keywords：traffcflow prediction；graph convolutional neural network;dep learning;smart city;performance improvement

近年来，交通网络通常表现为非欧几里得空间，而卷积操作在欧几里得空间中取得了显著效果，因此，针对非欧几里得空间的卷积技术应运而生，图卷积神经网络便是其中之一。（剩余5402字）