摘要
提出了一种新的深度学习框架--图注意力网络(GAT)与门控循环单元(GRU)组合的时空交通流量预测模型。图注意力网络用来学习复杂的拓扑结构来捕获空间依赖,门控循环单元学习交通数据的动态变化来捕获时间依赖。利用加利福尼亚高速公路数据进行模型验证,实验结果表明,该模型相对于其他主流预测模型进一步降低了预测误差,在交通流预测问题中适用性更强。
A new deep learning framework,a spatiotemporal traffic flow prediction model combined with Graph Attention Network(GAT)and Gated Recurrent Unit(GRU),is proposed.Graph attention networks are used to learn complex topologies to capture spatial dependencies,and gated recurrent units are used to learn the dynamics of traffic data to capture temporal dependencies.The model is verified by using California highway data.The experimental results show that the model further reduces the prediction error compared with other mainstream prediction models,and is more applicable to traffic flow prediction problems.
作者
赵静
李昕
ZHAO Jing;LI Xin(School of Electronics and Information Engineering,Liaoning University of Technology,Jinzhou 121001,China)
出处
《辽宁工业大学学报(自然科学版)》
2022年第3期170-176,共7页
Journal of Liaoning University of Technology(Natural Science Edition)
基金
辽宁省教育厅高校科研基金项目(LJKZ0625)。
关键词
交通流预测
深度学习
图注意力网络
门控循环单元
时空相关性
traffic flow prediction
deep learning
graph attention network
gated recurrent unit
spatial-temporal correlation