ST-GAN: Spatial-Temporal Graph Attention Network for Traffic Prediction

Accurate traffic flow prediction is crucial for Intelligent Transportation Systems (ITS) to optimize traffic management, enhance safety, and improve efficiency. However, effectively capturing the complex spatiotemporal dynamics of traffic data, including periodicity, spatial heterogeneity, and long-term dependencies, presents a significant challenge. This paper proposes the Spatial-Temporal Graph Attention Network (ST-GAN) for traffic flow prediction. ST-GAN integrates a temporal self-attention module to model temporal dynamics and a dynamic graph self-attention module to capture evolving spatial dependencies. Experiments from four real datasets demonstrate that ST-GAN outperforms benchmark methods performance, especially the MAE value of the PeMS08 dataset experiment, which is improved by 8.62 %. This method highlights its potential and practical value in traffic flow prediction.