基于自适应动态图卷积循环网络的交通流预测
Traffic Flow Prediction Based on Adaptive Dynamic Graph Convolutional Recurrent Network
唐晨嘉 1曾伟 1赵振兴1
作者信息
- 1. 华中科技大学人工智能与自动化学院 武汉 430074
- 折叠
摘要
针对实际交通状况中节点之间存在的动态变化关系,提出一种自适应动态图时空预测模型TAGGRU,基于编码器-解码器网络结构对交通数据动态时空特征融合建模.将节点嵌入与时间编码结合为时空编码,并以此构建动态邻接图,用以表示节点关系的时间演化.将交通流数据与动态邻接矩阵共同输入编码器,通过自适应门控循环单元进行特征提取.编码器和解码器之间添加交互注意力模块,将历史特征进行转换,以生成未来特征表示,通过特征维度变换得到最终输出.结果表明:该模型有较优的预测性能.
Abstract
Aiming at the dynamic relationship between nodes in actual traffic conditions,an adaptive dynamic graph spatio-temporal prediction model TAGGRU was proposed.Based on the encoder-de-coder network structure,the dynamic spatio-temporal characteristics of traffic data were fused and modeled.Combining node embedding and time coding into space-time coding,a dynamic adjacency graph was constructed to represent the time evolution of node relationship.The traffic flow data and the dynamic adjacency matrix were input into the encoder,and the features were extracted by the a-daptive gating cycle unit.An interactive attention module was added between the encoder and the de-coder to transform historical features to generate future feature representations,and the final output was obtained through feature dimension transformation.The results show that the model has better prediction performance.
关键词
交通流预测/时空编码/自适应动态图/门控循环单元Key words
traffic flow forecasting/spatial-temporal embedding/adaptive dynamic graphs/gated re-current unit引用本文复制引用
出版年
2024
NETL
NSTL
万方数据