Solving the green reverse logistics problem in e-commerce using a reinforcement learning based genetic algorithm

This study explores the two-phase green reverse logistics problem with time windows and a focus on perishable items that pose a significant challenge in the management of returned goods in e-commerce. We proposed a mixed integer programming model that considers carbon emissions, fuel consumption costs, facility establishment and operating costs, among other factors.

We incorporated reinforcement learning concepts to adjust parameters in traditional genetic algorithms, which often have inflexible parameter settings, thereby enhancing both the efficiency and quality of the solutions. The Q-learning algorithm was adopted as the learning method, and various action combinations of reinforcement learning were explored and compared. We further evaluated the performance of different genetic algorithm variations. The results indicate that the proposed algorithm provides high-quality solutions, and that effective parameter configuration significantly impacts the algorithm’s overall performance.