Evolving Robust Facility Placements

Robustness is a desired quality in many real-world engineered systems. The p-median problem is an optimization process in which a set of facilities must be found to minimize the average distance between each individual in a given population and the nearest facility. By introducing perturbations drawn from a specified distribution during evolution, we simulate the effect of natural disasters or other catastrophes on the placement of facilities. We use a non-dominated multi-objective procedure to select facilities with high fitness and robustness. We show that facilities evolved in this way are similarly fit and more robust than optimal solutions evolved without perturbation. Importantly, evolved facility layouts are much less susceptible to large catastrophic failures that are of the greatest concern in the placement of public infrastructure.