Experimental Evaluation of Meta-Heuristics for Multi-Objective Capacitated Multiple Allocation Hub Location Problem

Abstract

Multi-objective capacitated multiple allocation hub location problem (MOCMAHLP) is a variation of classic hub location problem, which deals with network design, considering both the number and the location of the hubs and the connections between hubs and spokes, as well as routing of flow on the network. In this study, we offer two meta-heuristic approaches based on the non-dominated sorting genetic algorithm (NSGA-II) and archived multi-objective simulated annealing method (AMOSA) to solve MOCMAHLP. We attuned AMOSA based approach to obtain feasible solutions for the problem and developed five different neighborhood operators in this approach. Moreover, for NSGA-II based approach, we developed two novel problem-specific mutation operators. To statistically analyze the behavior of both algorithms, we conducted experiments on two well-known data sets, namely Turkish and Australian Post (AP). Hypervolume indicator is used as the performance metric to measure the effectiveness of both approaches on the given data sets. In the experimental study, thorough tests are conducted to fine-tune the proposed mutation types for NSGA-II and proposed neighborhood operators for AMOSA. Fine-tuning tests reveal that for NSGA-II, mutation probability does not have a real effect on Turkish data set, whereas lower mutation probabilities are slightly better for AP data set. Moreover, among the AMOSA based neighborhood operators, the one which adds/removes a specific number of links according to temperature (NS-5) performs better than the others for both data sets. After analyzing different operators for both algorithms, a comparison between our NSGA-II based and AMOSA based approaches is performed with the best settings. As a result, we conclude that both of our algorithms are able to find feasible solutions of the problem. Moreover, NSGA-II performs better for larger, whereas AMOSA performs better for smaller size networks.