Attention-Based Spatio-Temporal Recurrent Neural Network Prediction Model

Accrate urban traffic flow prediction is a fundamental information processing task crucial for the efficiency of Intelligent Transportation Systems (ITS) and smart city infrastructure, directly impacting traffic management, congestion mitigation, and logistics optimization. Addressing the core challenge of modeling complex spatiotemporal dependencies in traffic data, this paper proposes a novel deep learning algorithm named the Spatio-Temporal Gated Recurrent Unit with Attention Mechanism (ST-GRUA). Leveraging the power of Recurrent Neural Networks (RNN), ST-GRUA integrates Gated Recurrent Units (GRU) to capture long-term temporal patterns and introduces a dedicated spatial attention mechanism to dynamically model the intricate spatial correlations within road networks. Crucially, a multi-head attention fusion layer intelligently combines the extracted temporal and spatial features. Experimental validation on a simulated urban traffic dataset demonstrates that ST-GRUA significantly outperforms traditional statistical models (e.g., ARIMA) and state-of-the-art deep learning baselines across key metrics (MAE, RMSE, MAPE). The enhanced accuracy and robustness of ST-GRUA position it as a powerful computational tool for intelligent traffic control systems and automated decision-making within the broader context of AI- driven urban informatics and automation.