Deep learning-based short-term origin-destination demand prediction in urban rail transit systems during holidays
Document Type
Journal Article
Publication Date
2025
Subject Area
place - urban, mode - rail, ridership - demand, ridership - modelling
Keywords
Urban rail transit, deep learning, short-term OD demand prediction, holidays
Abstract
Short-term origin-destination (OD) demand prediction during holidays is both critical and challenging for ensuring efficient Urban Rail Transit (URT) operations, particularly due to sharp demand fluctuations and issues such as data sparsity. This study proposes a short-term holiday OD demand prediction model for URT systems, called dynamic multi-graph convolution gated recurrent unit network (DM-GCGRU). Specifically, the model integrates multi-source heterogeneous data fusion, graph construction, and a dynamic graph convolution gated recurrent unit to capture complex spatiotemporal patterns. A pre-training strategy is adopted to enhance feature learning and mitigate data sparsity, while a physics-informed loss guides training by incorporating the relationship between inflow and OD demand. The proposed model is evaluated on two real-world holiday OD datasets, demonstrating superior predictive accuracy compared to state-of-the-art baselines. Ablation studies further confirm the contribution of each model component.
Rights
Permission to publish the abstract has been given by Taylor&Francis, copyright remains with them.
Recommended Citation
Zhang, S., Zhang, J., Deng, C., Yang, Y., Chen, X., & Yang, L. (2025). Deep learning-based short-term origin-destination demand prediction in urban rail transit systems during holidays. Transportmetrica B: Transport Dynamics, 13(1), 2521366.
