Research on Tunnel Traffic Flow Prediction Model Based on Graph Neural Networks

To address the complex spatiotemporal dependencies and dynamically evolving spatial relationships in tunnel traffic flow prediction, a macro–micro collaborative two-stage prediction method is proposed.The Grey Wolf Optimizer (GWO) is first employed to optimize the GRU model for predicting incoming traffic flow at the tunnel entrance, providing reliable macro-level input for subsequent modeling.Based on this, a spatiotemporal graph structure is constructed, and an FSE-ST-GCN model integrating an adaptive adjacency matrix with spatial and channel attention mechanisms is developed to capture dynamic spatial dependencies and enhance key feature representation.Experiments are conducted using real-world traffic flow data collected from the Shizuizi Tunnel on the Jilin–Caoshi Expressway. The results show that the proposed method outperforms baseline models in terms of MAE, RMSE, and MAPE, achieving superior prediction accuracy and stability. This work provides effective technical support for refined tunnel traffic management and lighting control.