摘要
目前大多数基于时空关系构建预测交通流量的模型分别使用不同类型的深度学习方法获取交通数据的时间相关性和空间相关性,割裂了交通流量的时空同步相关性。使用局部时空图和注意力机制,构造图卷积网络模型同时捕获地铁拥挤度数据的时空同步动态相关性。实验对比分析显示,提出的基于注意力机制时空同步图卷积神经网络的地铁拥挤度预测方法比没有使用局部时空图模型预测方法的平均绝对误差减小了0.07%、均方根误差减小了0.16%,比没有使用注意力机制模型预测方法的平均绝对误差减小了1.72%、均方根误差减小了3.24%。
Currently,most models for predicting traffic flow based on spatiotemporal relationships use different types of deep learning methods to capture the temporal and spatial dependencies of traffic data separately,thereby neglecting the synchronous spatiotemporal correlations of traffic flow.By using local spatiotemporal graphs and attention mechanisms,a graph convolutional network model is constructed to simultaneously capture the synchronous dynamic spatiotemporal correlations of subway crowding data.Experimental comparative analysis shows that the proposed subway crowding prediction method based on the attention mechanism-based synchronous spatiotemporal graph convolutional neural network reduces the mean absolute error by 0.07%and the root mean square error by 0.16%compared with the prediction method that does not use the local spatiotemporal graph model.It also reduces the mean absolute error by 1.72%and the root mean square error by 3.24% compared with the prediction method that does not usethe attention mechanism model.
作者
宋莹
戴政君
王博
SONG Ying;DAI Zheng-jun;WANG Bo(Beijing Key Laboratory of Internet Culture and Digital Dissemination Research,Beijing Information Science and Technology University,Beijing 100101,China;Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190,China;Software Engineering College,Zhengzhou University of Light Industry,Zhengzhou Henan 450002,China)
出处
《计算机仿真》
2025年第11期236-241,共6页
Computer Simulation
基金
国家自然科学基金面上项目(61872043)
计算机体系结构国家重点实验室开放课题(CARCHA202103)。
