Hybrid adaptive large neighborhood search for vehicle routing problems with depot location decisions

Abstract

This article considers three variants of the vehicle routing problem (VRP). These variants determine the respective depot locations from which customers are supplied, i.e., the two-echelon VRP (2E-VRP), the location routing problem (LRP), and the multi-depot VRP (MDVRP). Both the LRP and the MDVRP can be formulated as special cases of the 2E-VRP, so that all three problem classes can be readily solved via a single solution approach. We develop such a unified solution approach for all three problem classes based on the recently proposed hybrid adaptive large neighborhood search (HALNS). The HALNS uses a population of solutions generated by an efficient ALNS. Individuals of this population are subject to a crossover and selection phase, using elements of genetic algorithms resulting in a hybrid heuristic. Computational experiments on several sets of instances from literature demonstrate the competitive performance of the HALNS. The HALNS outperforms all approaches for solving the 2E-VRP and is on par with heuristics that are dedicated either to the LRP or the MDVRP. Furthermore, the HALNS shows superior robustness, i.e., the variance of results from several runs is comparatively low. The HALNS especially outperforms all existing pure ALNS implementations on these problem classes, demonstrating the value of hybridization. Additionally, the HALNS finds three new best-known solutions for LRP instances.