GCN-SATO: A Graph Convolutional Network with Self-Attention based Car-Following Model

Car following is critical for the overall safety, efficiency, and smooth operation of autonomous vehicles in traffic. However, the existing car-following model primarily focuses on local feature extraction, overlooking the spatial-temporal relationships between vehicles and key information within the time series. This limitation negatively impacts prediction accuracy and generalization capabilities. To tickle these problems, this paper proposes a novel car-following method based on a Graph Convolutional Network (GCN) and a self-attention mechanism. Firstly, the GCN network is utilized to construct the spatial-temporal structure of the car-following model, which can effectively obtain topological structure relationships among vehicles. Secondly, the self-attention module is designed to assign different attention weights to different neighbors of a node in the car-following model, which can facilitate the capture of various aspects of node relationships simultaneously and enhance its expressive power. Finally, the Multi-Layer Perceptron (MLP) layer is employed to predict the future behaviors of following vehicles. The proposed car-following model (CF) was trained and evaluated using five real-world datasets: HighD, SPMD, Waymo, NGSIM, and Lyft. The results indicate that our method outperforms other existing methods in Mean Squared Error (MSE) of spacing and achieves a zero collision rate on all datasets. These findings indicate that the GCN-SATO proposed in this paper enhances safety and smooth operation compared to other existing models.