Dynamic Spatial-Temporal Graph Convolutional GRU Network for Traffic Forecasting

Accurate and real-time traffic state forecasting has become an important and challenging task. The correlations of temporal and spatial are important to the accuracy of traffic forecasting. We propose a novel deep learning framework, spatial-temporal GCN-GRU(ST-GGRU), to capture the spatial and temporal dependence. The proposed ST-GGRU is combined with Graph Convolutional Network and GRU. Graph Convolutional Network is used to capture the complex spatial dependence and the GRU is used to capture the traffic data temporal dependence. Experiments show that our ST-GGRU network can capture the spatial-temporal correlations throughout traffic data and the outperform state-of-art baselines algorithms on real-world traffic datasets.