Hybrid knowledge-guided reinforcement learning with adaptive variable neighborhood search for dynamic multi-depot electric vehicle routing problems

The growing demand for environmentally friendly and economical transportation has driven development in self-driving cars and optimized travel routes. To meet these goals, efficient route planning is essential for cutting costs and minimizing harm to the environment. This paper addresses the complex challenge of optimizing routes for autonomous electric vehicles in a dynamic and uncertain environment. We focus on the Dynamic Multi-Depot Electric Vehicle Routing Problem with Time Windows (DMDEVRPTW), considering multiple depots, fluctuating demands, and time constraints. To tackle this problem, we propose a novel hybrid framework combining Deep Reinforcement Learning (DRL), Graph Attention (GAT) mechanisms, and an adaptive metaheuristic optimization algorithm (AVNS). Our approach leverages knowledge-guided techniques to improve solution quality. Extensive experiments demonstrate the superiority of our method in terms of total distance traveled and computational efficiency compared to state-of-the-art alternatives. This research contributes to the advancement of sustainable and efficient transportation systems by providing effective solutions for electric vehicle routing optimization.