A multi-stage heuristic algorithm based on task grouping for vehicle routing problem with energy constraint in disasters

Abstract

In recent years, serious disasters have happened frequently, causing significant loss of lives and substantial economic consequences. Considering that rescue vehicles are mainly powered by electricity, they cannot compete with the conventional ones in terms of the cruising range. Therefore, it is of great significance to study how rescue vehicles can execute multiple rescue tasks quickly and efficiently with energy constraint. In this paper, a multi-stage vehicle routing algorithm based on task grouping (MSVR-TG) is proposed for the rescue vehicle routing problem with energy constraint (VRPEC) in disasters. In the task grouping stage, a novel K-means algorithm based on angular density (K-means-ad) is suggested to address the instability of the result of K-means caused by the random selection of initial cluster centers. It can balance the travel distance of each vehicle as much as possible, which is critical when vehicles are energy constrained. In the sequence planning stage, a problem-specific genetic algorithm (PSGA) with multiple population initialization rules and the route improvement strategy is introduced to balance the convergence speed and population diversity, achieving a promising planning result. In the route adjusting stage, various removal and insertion heuristics in the large neighborhood search (LNS) are designed to adjust the routes, thereby further improving the planning result. The routes participating in adjustment are obtained via the developed route selection rule. The validity of each stage in MSVR-TG has been verified in a series of scenarios. A comparison about the proposed algorithm against several advanced algorithms is constructed, and results reveal that MSVR-TG is superior over compared algorithms for VRPEC.