An Advanced Attention-Based Neural Model for Traffic Flow Forecasting with Integrated Temporal Correlation Analysis

Reliable traffic flow predictions play a key role in supporting both drivers and traffic control personnel in making well-informed choices. Achieving high-precision traffic forecasting requires a thorough consideration of both spatial-temporal dependencies and the intricate temporal relationships present in the data. However, many current methodologies primarily emphasize temporal continuity while often overlooking broader temporal interactions. This research presents a novel multi-modal attention-based neural network aimed at capturing both long- and short-term sequence dependencies (LSTSC) for traffic forecasting. By leveraging attention mechanisms, the model effectively discerns spatiotemporal relationships within sequential data. The proposed approach has been rigorously evaluated on two benchmark datasets, PeMS08 and PeMSD7(M), where it has demonstrated superior accuracy and reliability, particularly in long-range predictions.