Short-Term Vehicle Speed Prediction Based on Spatio-Temporal Traffic Graph Networks

Predictive information is an important research direction in vehicle energy management. As the most intuitive item among numerous predictive information, the accurate and real-time prediction of vehicle speed is significant for vehicle motion control and energy management optimization. However, traditional vehicle speed prediction methods generally predict future speed changes based on the experience of historical vehicle speed changes and current vehicle speed information, which belongs to prediction in the pure time dimension. The spatial information of traffic scenes that directly affects vehicle behaviors has not been widely applied. Aiming at the complex spatio-temporal correlation in the vehicle motion process, this paper proposes a spatio-temporal graph convolutional network (GCN-LSTM). This network combines graph convolutional network (GCN) and long short-term memory network (LSTM) and is used for vehicle speed prediction. Among them, the GCN can handle the complex topological structures in graph data to extract the spatial features of traffic scenes, and the LSTM can handle the time features of the dynamic changes in traffic scenes. Training and testing are carried out on two datasets of traffic congestion and smooth traffic. The simulation results show that compared with other benchmark models, the proposed model has the smallest prediction error. Meanwhile, based on the datasets, the impact of the expansion of the amount of spatial information on the prediction accuracy of the model in this paper is verified. The results indicate that abundant spatial dimension information can further improve the prediction accuracy.