Modeling Drone-Assisted Subway Fire Evacuation Based on Digital Twin and Human–Fire-Station Interaction

Subway fire evacuations face severe challenges due to high passenger density, spatial unfamiliarity, and toxic smoke, which often render traditional static signage ineffective and cause chaotic congestion. To address this, we propose a drone-assisted evacuation strategy evaluated via a high-fidelity digital twin platform coupling PyroSim and AnyLogic. The developed human–fire-station interaction model integrates CO-induced physiological degradation and a binary logit model to capture bounded-rational decision-making and herding behaviors. Using the Suzhou Olympic Sports Centre Station as a case study, simulations reveal that spatial familiarity’s impact on spontaneous evacuation exhibits strict diminishing marginal returns. Under a realistic low-familiarity scenario (25%), unguided evacuation requires 312 s with severe bottlenecking. Drone guidance actively intercepts this chaos, slashing clearance time to 237 s—a 24.0% improvement in efficiency. Furthermore, aerial directives successfully transform panicked clusters into structured platoons, mitigating stampede risks and ensuring the safe egress of vulnerable demographics. This study provides robust quantitative evidence for integrating unmanned aerial systems into smart transit emergency management, serving as a high-fidelity virtual testbed for future evacuation drills.