Against the backdrop of China’s “dual-carbon” goals, accurate analysis and prediction of subway passenger flows are crucial for optimizing operational efficiency and advancing low-carbon urban transportation. Beijing’s subway network exhibits pronounced spatiotemporal heterogeneity across workdays, weekends, and holidays, yet existing studies often rely on static networks or single-scale temporal analyses, failing to capture dynamic flow evolution. To address this gap, this study develops a dynamic time-varying network framework with a 15 min temporal granularity, integrating sliding time-window analysis, node strength evaluation, and betweenness centrality for bottleneck identification. A Temporal–Spatial Fusion Gated Recurrent Unit (TSF-GRU) model is proposed to fuse temporal dependencies, spatial correlations, and network topology for short-term passenger flow forecasting. Results show distinct flow patterns: workdays feature a “concentrated commuting” dual peak, holidays a “steady continuous” leisure pattern, and weekends an “extended flexible” hybrid pattern. Station functions and bottleneck evolution vary dynamically across date types, with transportation hubs central on holidays/weekends and business nodes dominating workday peaks. The TSF-GRU model achieves a test-set MAPE of 7.62% and bottleneck prediction accuracy of 92.3%, outperforming traditional methods. This study provides a feasible pathway for refined, low-carbon subway operations in megacities and methodological support for achieving dual-carbon goals.
Zhang et al. (2026) studied this question.