Spatio-Temporal Multi-Embedded Attention Aware Transformer for Traffic Flow Prediction

Traffic flow prediction is a crucial component of intelligent transportation systems, where accurate prediction facilitates traffic management and improves efficiency. Currently, there are numerous challenges in predicting traffic flow, such as heterogeneity and sudden fluctuations. With the advancement of research, more methods have been applied to traffic flow prediction, such as graph learning models and Transformer models. In particular, models based on the Transformer architecture have garnered widespread attention due to their unique attention mechanisms and outstanding performance across various domains. This paper proposes a Transformer model that performs temporal encoding at different time granularities, followed by temporal attention computation to capture time feature dependencies across different time spans. Additionally, a spatial attention layer is introduced to capture dependencies between road nodes. The proposed method is validated using real-world traffic datasets, and experimental results demonstrate that our model achieves superior performance.