Innovative Hybrid Approaches for Vehicle Routing Problems

This thesis deals with the efficient resolution of Vehicle Routing Problems (VRPs). The first chapter faces the archetype of all VRPs: the Capacitated Vehicle Routing Problem (CVRP). Despite having being introduced more than 60 years ago, it still remains an extremely challenging problem. In this chapter I design a Fast Iterated-Local-Search Localized Optimization algorithm for the CVRP, shortened to FILO. The simplicity of the CVRP definition allowed me to experiment with advanced local search acceleration and pruning techniques that have eventually became the core optimization engine of FILO. FILO experimentally shown to be extremely scalable and able to solve very large scale instances of the CVRP in a fraction of the computing time compared to existing state-of-the-art methods, still obtaining competitive solutions in terms of their quality. The second chapter deals with an extension of the CVRP called the Extended Single Truck and Trailer Vehicle Routing Problem, or simply XSTTRP. The XSTTRP models a broad class of VRPs in which a single vehicle, composed of a truck and a detachable trailer, has to serve a set of customers with accessibility constraints making some of them not reachable by using the entire vehicle. This problem moves towards VRPs including more realistic constraints and it models scenarios such as parcel deliveries in crowded city centers or rural areas, where maneuvering a large vehicle is forbidden or dangerous. The XSTTRP generalizes several well known VRPs such as the Multiple Depot VRP and the Location Routing Problem. For its solution I developed an hybrid metaheuristic which combines a fast heuristic optimization with a polishing phase based on the resolution of a limited set partitioning problem. Finally, the thesis includes a final chapter aimed at guiding the computational evaluation of new approaches to VRPs proposed by the machine learning community