Predicting network-wide metro passenger flow under large-scale disruptions in a mega-scale city

Accurately predicting passenger flow distribution within a metro network during large-scale disruptions is crucial for maintaining the resilience of metro systems in megacities. Such disruptions not only risk causing severe station congestion but also exacerbate operational delays across the entire network, thereby affecting the overall efficiency of urban transportation systems. Thus, investigating passenger flow distribution under disruption conditions is essential for alleviating congestion and optimizing emergency responses. This study proposes a predictive framework based on a dynamic Bayesian network (DBN) path choice model and a spatio-temporal graph convolutional network (ST-GCN) to accurately forecast network-wide passenger flow distribution under disruption conditions in megacities. The framework leverages ST-GCN to capture the spatiotemporal dependencies and complex relationships among stations and integrates a DBN to simulate passengers’ path adjustment behavior, facilitating real-time updates of optimal travel paths during disruptions. Simulation results using real-world disruption data from the Shanghai Metro demonstrate that the proposed model outperforms benchmark models such as long and short-term time-series network (LSTNet), gated convolutional neural network (CNN), diffusion convolutional recurrent neural network (DCRNN), multi-task multi-graph neural network (MTMGNN), and convolutional long short-term memory (ConvLSTM) in terms of prediction accuracy across two major disruption scenarios. The findings indicate that the proposed framework significantly enhances the accuracy of passenger flow prediction under disruption conditions, providing operators with reliable early warnings and effective decision-making support, thereby improving the disaster resilience and operational robustness of metro systems.