This article proposes a model predictive control (MPC) method for complex dynamic e-commerce closed-loop supply chains, considering time delay and dual channel mixed recycling. It was found that MPC achieved high accuracy (MAE 0.21, RMSE 0.27, MAPE 1.12%) for sine function requirements with obvious fluctuation patterns, demonstrating stability and low error. In the scenario of random pulse demand, using the “rolling optimization + dynamic learning” mechanism, compared with reinforcement learning and collaborative planning prediction and replenishment methods, MAE decreased by 28.6% and 35.1%, and RMSE decreased by 31.2% and 38.7%, respectively, verifying the adaptive advantage of MPC under dynamic demand. This study elucidates how MPC optimizes inventory management, logistics scheduling, and recycling strategies in e-commerce supply chains, providing theoretical insights and practical guidance to improve operational efficiency and overall supply chain performance.
Jun Hang (Fri,) studied this question.