Hedging against disruptions with ripple effects in location analysis

Supply systems are subject to disruptions whose impact may not remain confined, but might actually propagate across the network. We consider the problem of optimally protecting a capacitated median system with a limited amount of protective resources subject to disruptions. Specifically, the type of disruption studied is characterized by correlation effects between the facilities, and may result in partial or complete disruption of the facilities involved. The model optimizes protection plans in the face of large area disruptions; i.e., disruptions that affect regions rather than single elements of the system. Examples may be earthquakes, storms, floods, fires, hurricanes, droughts, the spread of diseases, the spread of chemical agents, and cascading failures. The model is also a general framework for the family of fortification problems in the context of locationanalysis, as it includes uncapacitated facilities and single-target disruptions as special cases. We provide a tri-level formulation of the problem, and we propose an exact solution algorithm which makes use of a tree-search procedure to identify which facilities to protect. The procedure is enhanced by a dual-based pruning rule. The underlying disruption problem is reformulated as a single-level mixed-integer program. The algorithm has been tested on a dataset based on the 2009 L’Aquila earthquake. We verify empirically the efficiency of the pruning rule, and we provide an evaluation of the importance of considering propagation effects in the disruptions.