QMOEA: A Q-learning-based multiobjective evolutionary algorithm for solving time-dependent green vehicle routing problems with time windows

Abstract

The vehicle routing problem with time windows (VRPTW) is critical in the fields of operations research and combinatorial optimization. To promote research on the multiobjective VRPTW, a time-dependent green VRPTW (TDGVRPTW) is introduced in this study. Subsequently, a Q-learning-based multiobjective evolutionary algorithm (QMOEA) is proposed to solve the TDGVRPTW, where three objectives are simultaneously considered: total duration of vehicles, energy consumption, and customer satisfaction. In QMOEA, a hybrid initial method is devised that includes four problem-specific heuristics, to generate initial solutions with a high level of quality and diversity. Then, considering the problem features, two Pareto-front-based crossover strategies are designed to learn from the approximate Pareto front to explore the search space and accelerate the convergence process. Moreover, five local search operators are selected by a Q-learning agent at the decision point, to enhance local search abilities. Finally, a set of instances based on a realistic logistic system is presented to verify the effectiveness and superiority of QMOEA.