Distributionally robust optimization for fire station location under uncertainties

Emergency fire service (EFS) systems provide rescue operations for emergencies and accidents. If properly designed, they can decrease property loss and mortality. This paper proposes a distributionally robust model (DRM) for optimizing the location of fire stations, the number of fire trucks, and demand assignment for long term planning in an EFS system. This is achieved by minimizing the worst-case expected total cost, including fire station construction cost, purchase cost for fire trucks, transportation cost, and penalty cost for not providing adequate service. The ambiguity in demands and travel durations distributions are captured through moment information and mean absolute deviation. A cutting plane method is used to solve the problem. Due to fact that it is computationally intensive for larger problems, two approximate methods are introduced; one that uses linear decision rules (LDRs), and another that adopts three-point approximations of the distributions. The results show that the heuristic method is especially useful for solving large instances of DRM. Extensive numerical experiments are conducted to analyze the model's performance with respect to different parameters. Finally, data obtained from Hefei (China) demonstrates the practical applicability and value of the model in designing an EFS system in a large metropolitan setting.