Traffic Network Socialization: An Adaptive Spatio-Temporal Graph Convolutional Network for Traffic Prediction

Accurate traffic prediction is important for developing intelligent transportation system (ITS). We take inspiration from the graph convolutional network (GCN) technology of link prediction in social networks. Traffic and social networks are similar in the link prediction structure. Link prediction in social networks is related to user information and topology information; moreover, the future traffic flow of nodes is related to neighbor nodes and historical traffic flow. This study proposes an adaptive spatio-temporal GCN for traffic prediction based on similarities in the link prediction structure. First, considering the traffic flow data socialization problem, the road network nodes are compared to users in social networks, and the relationship between users is mapped to spatial correlation in traffic flow data. Furthermore, because of the hidden spatial dependence between road network nodes, an enhancing GCN based on an adaptive adjacency matrix is developed to enhance system robustness. Second, aiming at the dynamic spatio-temporal correlation of traffic data, the dynamic spatio-temporal graph module (DST-graph module) is proposed, which is based on the modeling ability of the transformer for long time series. The module captures the dynamic spatio-temporal correlation and the long-term temporal dependence. Finally, a gate fusion module is designed to effectively integrate the learned temporal-spatial features of traffic flow to improve system robustness and prediction accuracy. Multiple experiments have been performed on four real-world datasets. The results show that, compared with other baseline methods, the proposed model achieves additional accuracy for long-term traffic flow under complex traffic conditions.