We present a practical digital-twin simulation for mass evacuation when an electric bus ignites at a fixed site (charging station or bus depot), releasing toxic gases in dense urban space. The system helps planners anticipate time-evolving conditions and choose faster, safer actions. The architecture comprises four parts. (1) A gas dispersion model that uses Aerial Location of Hazardous Atmosphere (ALOHA) software to generate hazard-segmented threat zones (e.g., PAC levels), projecting these zones onto the facility and nearby road network as dynamic high-risk and medium-risk areas. (2) A site-space evacuation system that models movement within the facility (e.g., charging station/bus terminal/vehicle interiors) with rules and potential-field which considered local layout, obstacles, and crowding. (3) A roadway evacuation system for evacuees and vehicles on surrounding roads, with behaviors distinct from the site layer (e.g., transition rules dependent on distance and density, density-dependent speeds, rerouting around toxic masks) and guidance toward assembly points. (4) A strategy system presently scoped to two levers: allocating priority of usable links proximate to the fire for fire engines and ambulances, and selection of perimeter assembly points. The system is being instantiated for Kaohsiung, Taiwan, using open road-network and pedestrian-flow data with scenarios over incident time/location, wind, and gas release magnitude. Key performance indicators are total evacuation time and total toxic-exposure time. We employ a heuristic algorithm to select assembly points and assign usable links priority, reducing both metrics. By linking site-level and road-level evacuation, the model better matches real operations and gives clearer guidance on where to block roads, route people, and speed emergency response.

Keywords

OR in disaster relief, Evacuation, Digital twin, Agent-based simulation, Toxic gas incident