Short-Term Passenger Flow Forecasting for Urban Rail Transit Based on Multi-View Spatial Graph Fusion
2026-99-2608
11/20/2026
- Content
- Accurate short-term passenger flow forecasting is important for the operational management of urban rail transit systems. However, most existing graph-based forecasting methods rely primarily on physical network topology and may fail to capture latent correlations between non-adjacent stations. To address this limitation, this study proposes a Multi-View Spatio-Temporal Graph Neural Network, termed MV-STGNN. Four complementary spatial graphs are constructed to represent physical topology, OD flow correlation, functional similarity, and geospatial proximity. A learnable fusion mechanism is then introduced to integrate the four graph views, and a graph convolutional gated recurrent module is employed to jointly extract spatial and temporal features. The proposed method is evaluated using 184 days of AFC data collected from nine consecutive stations on an urban metro line. Experimental results show that MV-STGNN achieves lower MAE, RMSE, and MAPE values than the selected baseline models. Ablation experiments further demonstrate the complementary contribution of the multi-view graph structures and the adaptive fusion mechanism. These results confirm the effectiveness of the proposed method for short-term passenger flow forecasting in the investigated metro corridor.
- Citation
- Ding, Y. and Xu, Y., "Short-Term Passenger Flow Forecasting for Urban Rail Transit Based on Multi-View Spatial Graph Fusion," 2026 2nd International Conference on Transportation Infrastructure and Engineering (ICTIE2026), Shandong, China, April 17, 2026, https://doi.org/10.4271/2026-99-2608.