Evacuation Performance of Passenger Spaces and Facilities in Underground Subway Stations for Heterogeneous Ambulatory Passengers

Urban subway systems accommodate heterogeneous passenger flows, including elderly, disabled, and luggage-laden passengers, whose dense peak-hour movements aggravate underground evacuation difficulties. Current design codes adopt oversimplified computational formulas, while existing studies mainly focus on local bottlenecks and post-event crowd management, neglecting systematic evacuation performance assessment along full passenger routes. This study investigated 128 single-line underground stations in Shenzhen using 27 spatial and facility parameters, integrating pedestrian walking experiments, agent-based simulation, K-means clustering, and interpretable machine learning to evaluate evacuation performance of heterogeneous ambulatory passengers and identify critical controlling factors. Results show: (1) all stations were categorized into five configurations based on platform type, spatial scale, and stair–escalator layout; (2) population-weighted mean walking speed is 1.267 m/s—crutch users slowest, blind individuals most variable; (3) 27.34% of stations exceed the adopted 6 min evacuation-time criterion under peak-hour conditions; Cluster 5 (side platforms with sequential stairs and escalators arrays) shows the longest times and highest risk (83.33% exceed 6 min), while Cluster 1 (island platforms with central T-shaped stair) has the shortest times; (4) concourse length/width, stair–escalator spacing, and mid-platform stair width are critical, exhibiting nonlinear effects with diminishing returns. Tailored retrofits validated via individual station cases reduced evacuation times by 5.35–29.04%, all within 6 min. This framework establishes the role of spatial design in evacuation efficiency, offering a risk-screening and preventive optimization approach for operational underground subway stations.