A GRASP×ELS approach for real-life Location Routing Problems
Abstract
This paper addresses capacitated location-routing problems (CLRP). Those distribution network design problems involve depot (or hub) location, fleet assignment and routing decisions. The distribution networks under investigation are characterized by several depots, by a capacitated homogeneous vehicle fleet and by a set of customer nodes to be serviced with demands. The objective is to assign the serviced nodes to depots and to design the vehicle routes. The optimal solution minimizes both the depot cost and the total route distance in such a way that the total customer demand assigned to one depot is upper bounded by its capacity. The proposed framework is a greedy randomized adaptive search procedure (GRASP) combined with an evolutionary local search (ELS). Classical instances used in benchmarks are based on complete symmetric weighted graphs. Moreover, most of them deal with homogenous depots. We propose a new set of instances based on asymmetric sparse graphs and strongly non homogenous set of depots. The distances between pairs of nodes are non Euclidean.