Optimizing Urban Traffic Flow Prediction: Integrating Spatial–Temporal Analysis with a Hybrid GNN and Gated-Attention GRU Model

In this paper, a cutting-edge hybrid model that synergizes graph neural networks (GNNs) with gated-attention gated recurrent units (GRUs) to predict traffic flow is introduced. This innovative approach uniquely combines the GNN's capability to unravel complex spatial relationships within traffic networks with the GRU's proficiency in capturing temporal dynamics, significantly enhanced by an attention mechanism. The model excels in understanding both the intricate connectivity of urban traffic systems and their fluctuations over time, leading to a notable improvement in prediction accuracy. Comparative analyses reveal that our model outperforms traditional LSTM, standalone GRU, GNN, and other deep learning models in traffic flow forecasting. The superior performance stems from its adept handling of spatial–temporal data, making it a promising tool for urban planners and traffic management, potentially revolutionizing traffic flow predictions and contributing to more efficient and well-informed urban traffic control.