Learning dynamic interaction for multimodal transportation systems: A joint passenger flow prediction approach based on spatiotemporal hypergraph convolution networks
Document Type
Journal Article
Publication Date
2025
Subject Area
mode - bus, mode - subway/metro, place - urban, place - asia, planning - integration, planning - methods, ridership - forecasting, ridership - modelling
Keywords
Multimodal transportation systems, Joint passenger flow prediction, Spatiotemporal characteristic, K-means clustering, Temporal convolution, Hypergraph convolution
Abstract
In metropolitan areas, multimodal transportation systems form a complex passenger flow network with intricate spatiotemporal dependencies across different transit modes. Existing approaches often treat metro and bus systems as independent, station-based modes, neglecting their direct flow transfer and mutual influence. To address this gap, this study proposes a joint passenger flow prediction framework that explicitly models inter-modal passenger flow transfer between metro and bus systems. Firstly, A hypergraph game K-means clustering algorithm is introduced to capture latent structural correlations between the two modes. Then, a multi-layer stacked Spatiotemporal Hypergraph Convolutional Network (STHCN) is proposed for the first time, integrating temporal gated convolutions with spatial hypergraph convolutional networks and systematically incorporating a hypergraph self-attention mechanism to effectively capture both intra- and inter-modal spatiotemporal dependencies of the two public transportation systems. Finally, real-world large-scale multimodal datasets covering metro and bus passenger flows in Beijing are employed to validate the effectiveness and robustness of the proposed approach. Extensive experiments indicate that the proposed STHCN outperforms several classic and state-of-the-art models across different forecasting horizons and spatial resolutions, achieving a 6.79% and 3.38% overall improvement over the best-performing baseline model at metro and bus passenger flow predictions, respectively. Ablation studies further confirm the effectiveness of each module in the proposed architecture and highlight the model’s ability to learn dynamic interdependencies within multimodal passenger flow networks robustly.
Rights
Permission to publish the abstract has been given by Elsevier, copyright remains with them.
Recommended Citation
Luo, D., Wang, J., Lu, W., Ding, W., Yan, X., & Yang, H. (2025). Learning dynamic interaction for multimodal transportation systems: A joint passenger flow prediction approach based on spatiotemporal hypergraph convolution networks. Transportation Research Part C: Emerging Technologies, 178, 105257.
Comments
Transportation Research Part C Home Page:
http://www.sciencedirect.com/science/journal/0968090X