A Method of Traffic Flow Forecasting Using Spatio-Temporal Graph Convolutional Networks

Traffic flow prediction is important to boost traffic congestion management and urban planning. However, while the graph neural network-based methods captured spatial features, the temporal characteristics were dropped in the previous study. This paper proposes a traffic flow prediction method based on spatio-temporal convolutional networks (ST-GCN) which can learn spatial and temporal features simultaneously. A main component of ST-GCN is spatio-temporal convolutional blocks, which were designed to process time series data of node features and an adjacency matrix of road networks. Experimental results with cross-sectional traffic volume data showed the impact of the proposed method on predicting traffic flow accurately.