Spatio-temporal Fusion of Transformer and Global Feature Mining for Traffic Flow Prediction

The primary challenge in traffic flow prediction centers on effectively capturing the spatio-temporal dependencies within traffic data. To address these challenges, we propose a Spatio-Temporal Feature Fusion Model based on Transformer and a Global Feature Mining Module. The aim is to overcome the high resource consumption issue of the Transformer model when processing large-scale traffic data, as well as its potential shortcomings in capturing subtle spatio-temporal dynamics. The model is capable of precisely capturing the spatio-temporal characteristics of traffic data, achieving seamless integration of temporal and spatial correlations, and revealing the interconnections between global and local features. Through extensive experiments on five real-world traffic datasets, the research results demonstrate a significant improvement in prediction accuracy of our proposed method compared to existing models.