Metaheuristics based on genetic algorithm and tabu search for vehicle routing problem with stochastic demands

Abstract

This study considers a Vehicle Routing Problem with Stochastic Demands (VRPSD) where the demands are unknown when the route plan is designed. The VRPSD objective is to find an a priori route under preventive restocking that minimizes the total expected cost, subject to the routing constraints, under the stochastic demands setting. Various metaheuristics based on Genetic Algorithm (GA) and Tabu Search (TS) were proposed to solve VRPSD. This study began with investigating the effect of static and dynamic tabu list size in TS. The results showed the advantage of dynamic tabu list size in significantly reducing the probability of cycling. Further, Reactive Tabu Search (RTS) which has never been used in VRPSD was introduced. This study showed that RTS give significant improvement to the solution quality of TS. This study then explored the enhancement of GA for VRPSD by proposing Adaptive GA (AGA), Breeder GA (BGA) and two types of Hybrid GA with Tabu Search (HGATS). Solutions generated using AGA were better than solutions from fixed parameter setting, and the use of AGA reduce the amount of time required in finding the appropriate mutation probability values of GA. The BGA also gave an improvement to the solution quality of GA. Different schemes of incorporating TS to GA lead to a significantly different performance of the HGATS algorithms. Next, comparative studies between metaheuristics implemented in this study were carried out including the comparison with previous research on GA for VRPSD. The HGATS showed superiority in terms of solution quality compared to other metaheuristics, followed by BGA and RTS in the second and third best performance respectively. Furthermore, the proposed bi-objective Pareto BGA gave better solution qualities compared to Pareto GA. Finally, the use of metaheuristics in a case study of solid waste collection reduced significantly the company current operation cost.