Design and Implementation of Intelligent Traffic State Estimation by GraphMasked X-TempFormer

The escalating complexity of urban transportation systems has established traffic prediction as a critical enabler for intelligent mobility solutions. While accurate traffic forecasting underpins route optimization and congestion management, conventional methodologies—such as autoregressive integrated moving average (ARIMA) and graph neural networks (GNNs)—are hindered by their inability to concurrently model spatiotemporal dependencies. Existing approaches often isolate temporal trends or spatial correlations, neglecting the dynamic interplay between these factors inherent to traffic flow. To address this limitation, we propose a novel framework—GraphMasked X-TempFormer, that integrates parallelized attention mechanisms to independently yet synergistically decode temporal sequences and spatial graph structures. Leveraging historical speed data from defined observation windows, our model generates high-fidelity predictions for subsequent intervals by explicitly disentangling temporal evolution and spatial interactions. Comprehensive benchmarking against three baseline models (ARIMA, LSTM, GCN) under identical data conditions demonstrates superior performance. The architecture further incorporates end-to-end trainable components and adaptive noise suppression, ensuring robustness in diverse urban traffic scenarios. This work advances spatio-temporal modeling for intelligent transportation systems, offering a scalable solution for real-world deployment.