摘要
针对当前路网交通流量预测方法中存在的和长距离空间依赖性能力不足等问题,设计了一种基于时空图注意力神经网络的智能交通流预测方法(STMGAN)。通过时间注意力模块采集各节点动态时间参数,建立时空融合模块完成时空特征的数据融合,通过门控融合的方法自主融合空间与时间特征。研究结果表明:本次设计的STMGAN各项性能指标均明显优于其他基准方法,该方法采用屏蔽机制以获得全局相似度,通过提升关注度获取时间和空间的动态关联,获得更高精度。STMGAN在长期预测过程中保持准确性,通过连续变化实现注意力在邻近节点与远距离功能相似区集中分布,能够识别全局功能相似区,捕捉长距离依赖性,有效提升预测准确性。该研究有助于提高智能交通效率,响应目前高速科技发展的需求。
Aiming at the existing problems of road network traffic flow prediction methods and the lack of long distance spatial dependence ability,an intelligent traffic flow prediction method(STMGAN)based on spatiotemporal graph attention neural network is designed.By collecting the dynamic time parameters of each node through the time attention module,a spatio-temporal fusion module is established to complete the data fusion of spatio-temporal features,and the spatial and temporal features are automatically fused through the gated fusion method.The research results show that the performance indicators of STMGAN in this design are significantly better than other benchmark methods.STMGAN model adopts a masking mechanism to obtain global similarity,and achieves higher accuracy by improving the dynamic correlation between time and space of attention.STMGAN maintains accuracy in the process of long-term prediction,and realizes concentrated distribution of attention in nearby nodes and distant functional similar regions through continuous changes,which can identify global functional similar regions,capture long-distance dependencies,and effectively improve prediction accuracy.The research will help to improve the efficiency of intelligent transportation and meet the needs of the current rapid technological development.
作者
侯同娣
李超
陈杰
HOU Tongdi;LI Chao;CHEN Jie(School of Economic and Trade Management,Yancheng Polytechnic College,Yancheng 224005,Jiangsu,China;School of Computer Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,Jiangsu,China;College of Information Security,Yancheng Polytechnic,Yancheng 224005,Jiangsu,China)
出处
《中国工程机械学报》
北大核心
2025年第5期851-855,共5页
Chinese Journal of Construction Machinery
基金
江苏省产学研合作资助项目(BY2022078)。
关键词
智能交通
图注意力神经网络
图掩码机制
特征融合
intelligent transportation
graph attention neural network
picture mask mechanism
feature fusion
