Spatio-Temporal Residual Graph Convolutional Network for Short-Term Traffic Flow Prediction
Qingyong Zhang, Meifang Tan, Changwu Li, Huiwen Xia, Wanfeng Chang, Minglong Li
Abstract
Accurate spatio-temporal traffic flow prediction is a significant research direction in the intelligent transportation system. Current prediction methods have limitations in spatio-temporal feature extraction, and the prediction results have poor performance. In this paper, a short-term traffic flow prediction model based on a Spatio-Temporal Residual Graph Convolutional Network (STRGCN) is proposed to solve the problem of poor accuracy in extracting the spatial and temporal correlation in the short-term traffic flow prediction task. Firstly, a Deep Full Residual Graph Convolutional Network (DFRGCN) module is used to learn the spatial correlation. Secondly, a Bidirectional Gated Recurrent Unit based on the Attention mechanism (ABi-GRU) is used to accurately obtain the temporal dependence of traffic flow data. Finally, the experimental results show that the STRGCN model achieves better prediction performance and stability on three publicly available datasets compared to the baseline methods.