A two-stage hybrid heuristic approach combining genetic algorithm and variable neighborhood descent for the clustered electric vehicle routing problem

Abstract

This paper considers a new variant of the Electric Vehicle Routing Problem (EVRP), termed the Clustered Electric Vehicle Routing Problem (CluEVRP). In CluEVRP, all customers are pre-divided into clusters, and each charging station is either located within a cluster or independent of any cluster. Each electric vehicle must complete service for all customers within the current cluster before proceeding to the next cluster or returning to the depot. Electric vehicles can charge at any available charging station while serving a cluster, but incur a penalty cost upon entering each cluster. The objective is to minimize the total logistics cost, comprising vehicle startup costs, cluster entry penalty costs, and energy consumption costs. To solve CluEVRP, a two-stage hybrid heuristic combining a Genetic Algorithm (GA) and Variable Neighborhood Descent (VND) is proposed (HGA-VND), where GA ensures population diversity and VND enhances local search capability. To evaluate the algorithm’s performance, 75 test instances are adapted from classic Clustered Vehicle Routing Problem (CluVRP) dataset, incorporating electric vehicle characteristics. Computational results demonstrate that HGA-VND consistently obtains high-quality solutions within reasonable time for both CluVRP and CluEVRP instances, exhibiting good performance. Furthermore, sensitivity analysis indicates that moderately increasing vehicle capacity, optimizing battery configuration, and adopting lightweight designs can significantly reduce total operating costs. This study extends traditional EVRP research by introducing clustered customer distribution, enriching solutions for routing problems in practical logistics networks, particularly for “milk run” models in industrial parks, and providing significant managerial insights.