Integrated spatial-temporal feature alignment with graph convolutional and gated recurrent networks for traffic flow prediction

Abstract

Accurate traffic flow prediction is essential for Intelligent Transportation Systems (ITS), yet capturing the complex spatiotemporal relationships within traffic data remains challenging due to dynamic traffic patterns and the non-Euclidean structure of road networks. Existing models struggle to adapt in real time, limiting their prediction accuracy and reliability. This study introduces the Spatiotemporal Feature Alignment with Graph Convolutional and Gated Recurrent Unit (STF-GGRU) model to address these limitations. By integrating a novel Integrated Spatiotemporal Feature Alignment (ISTFA) module, which combines Dynamic K-Nearest Neighbor (D-KNN) and Centered Kernel Alignment (CKA), the model dynamically captures critical spatial and temporal interactions. The STF-GGRU model achieves superior prediction accuracy, with RMSE values of 27.18 and 11.1 on the PeMSD4 and PeMSD8 datasets, outperforming traditional methods such as ARIMA, GRU, LSTM, and advanced neural models. These results demonstrate STF-GGRU’s potential for robust, real-time traffic predictions, marking a significant advancement in ITS capabilities.