SR-1: a simulation-based algorithm for the capacitated vehicle routing problem

In this paper we present SR-1, a simulation-based heuristic algorithm for the Capacitated Vehicle Routing Problem (CVRP). Given a CVRP instance, SR-1 uses an initial "good solution", such as the one provided by the classical Clarke and Wright heuristic, in order to obtain observations for the variable "distance between two consecutive nodes in a route". These observations are then fitted by a statistical distribution, which characterizes the inter-node distances in good solutions. Then, the fitted distribution is employed to generate a large number of new random solutions with similar edge-size distribution. Thus, a random but oriented local search of the space of solutions is performed, and a list of "best solutions" is obtained. This list allows considering several properties per solution, not only aprioristic costs, which can be practically used when making multiple-criteria decisions. Several tests have been performed to discuss the effectiveness of this approach.