Deep reinforcement learning-based optimisation of reverse logistics transport in closed-loop supply chain

Traditional optimisation methods often struggle to address the stochastic and dynamic characteristics of reverse logistics. Deep reinforcement learning (DRL), with its ability to learn adaptive policies under uncertainty, is well suited to this challenge. This study investigates the optimisation of green transportation strategies in reverse logistics systems through a DRL framework. By formulating the problem as a Markov decision process, this study integrates economic profitability, carbon dioxide emissions reduction and transportation efficiency into a dynamic reward function. The experimental results of the basic model and the extended model show that the superior performance of the DRL-based approach, which significantly surpasses traditional methods such as genetic algorithms and random strategies in adaptability and optimisation effectiveness. These findings provide theoretical foundations and practical insights for advancing sustainable logistics practices in complex systems.