Cluster-Centric Local Search Strategies for Enhanced Multi-Objective Logistics Optimization

Solving multi-objective vehicle routing problem with time windows (VRPTW) is a challenge in last-mile logistics. While traditional methods have predominantly focused on minimizing travel distances and vehicle numbers, recent innovations acknowledge the need for a more refined balance across various objectives. These advancements incorporate a range of heuristic algorithms, blending hybrid techniques that merge genetic algorithms with local search strategies, and employing advanced methodologies such as particle swarm optimization and multi-objective evolutionary algorithms based on decomposition (MOEA/D). Such developments signify a shift towards more effective solutions for VRPTW. In response to these challenges, this paper introduces a novel approach that integrates MOEA/D with K-means clustering for generating efficient solutions. In our approach, we refine the tri-objective VRPTW efficiency by establishing a criterion that prioritizes customer sequences based on their spatial proximity and angular alignment relative to cluster centers and depot, utilizing an integrated metric of distance and cosine angles, which can enhance the local search process and effectively balance the three objectives in VRPTW. Experimental results on Solomon's 56 VRPTW 100-customer dataset instances demonstrate the superiority of our approach.