Time-Aware Spatio-Temporal Graph Neural Networks for Traffic Forecasting
2026-99-2415
11/20/2026
- Content
- Accurately modeling intricate geographical interconnections and dynamic temporal patterns is still a significant difficulty in traffic forecasting, playing a crucial role in intelligent transportation systems. The majority of contemporary spatiotemporal graph neural networks are limited in their capacity to capture long-range temporal dependencies and time-varying spatial correlations because they rely on pre-defined or self-learning graph topologies in conjunction with recurrent-based temporal modeling. This study suggests a Time-Aware Spatio-Temporal Graph Neural Network (TASTGNN) for traffic forecasting in order to overcome these drawbacks. In order to enable the learned graph structures to dynamically adjust to changing traffic correlations over time, we first provide a time-aware graph learning technique that explicitly includes temporal context into spatial dependency modeling. Second, by using self-attention processes along both spatial and temporal dimensions, we develop an attention-enhanced spatio-temporal modeling framework that simultaneously captures long-range temporal dependencies and dynamic interactions among road network nodes. The efficacy of the suggested TASTGNN is shown by extensive tests carried out on two sample real-world traffic benchmark datasets.
- Citation
- Yin, X., Qian, Y., Zhou, W., and Xu, W., "Time-Aware Spatio-Temporal Graph Neural Networks for Traffic Forecasting," 2026 2nd International Conference on Intelligent Transportation and Future Mobility (ITFM2026), Xi'An, China, March 27, 2026, .