摘要
In the event of a large-scale toxic gas incident in an urbanized area, the mass evacuation of people located in the affected area is a highly challenging yet essential measure. We formulate a two-phase agent-based evacuation simulation model informed by real data from the National Science and Technology Center for Disaster Reduction (NCDR) in Taiwan and Aerial Location of Hazardous Atmosphere (ALOHA) software which can be leveraged by disaster management decision makers to bolster preparation and response mechanisms in the context of a toxic gas incident. Both phases of the evacuation simulation framework utilize agent-based rules tailored to manage the risk related to a toxic gas incident. Phase 1 is designed to realistically simulate the movement of pedestrian agents through a cell-based road network from their location in the hot zone or warm zone to a pre-determined set of assembly points to which high capacity public vehicles are dispatched. In Phase 2, the public vehicles also follow natural agent-based rules to select a shelter destination located outside the affected area and then to traverse through the cellular road network from assembly point to shelter. To pre-select an efficacious set of assembly points to be utilized in the event of an actual evacuation, a heuristic based on the k-means clustering algorithm is designed and implemented. We applied our two-phase agent-based evacuation simulation framework to carry out an empirical study based on a hypothetical massive toxic gas incident in an industrial zone in a large Taiwanese city. The empirical study looks at the effect of several model parameters on the evacuation simulation process, including the choice of assembly point arrangement. Compared to the current government policy, the proposed assembly point arrangement is shown to respectively decrease the average Phase 1 and Phase 2 evacuation times by 22% and 13%, which can be critical in reducing or preventing exposure to harmful levels of toxic gas.