Deep Reinforcement Learning Approach for Capacitated Supply Chain optimization under Demand Uncertainty

• Main Authors: Peng, Zedong, Zhang, Yi, Feng, Yiping, Zhang, Tuchao, Wu, Zhengguang, Su, Hongye
• Format: Conference Proceeding
• Language: English
• Subjects: deep reinforcement learning; demand uncertainty; Dynamic scheduling; Machine learning; Mathematical model; Optimization; supply chain optimization; Supply chains; Uncertainty; vanilla policy gradient
• ISSN: 2688-0938
• DOI: 10.1109/CAC48633.2019.8997498

With the global trade competition becoming further intensified, Supply Chain Management (SCM) technology has become critical to maintain competitive advantages for enterprises. However, the economic integration and increased market uncertainty have brought great challenges to SCM. In this paper, two Deep Reinforcement Learning (DRL) based methods are proposed to solve multi-period capacitated supply chain optimization problem under demand uncertainty. The capacity constraints are satisfied from both modelling perspective and DRL algorithm perspective. Both continuous action space and discrete action space are considered. The performance of the methods is analyzed through the simulation of three different cases. Compared to the baseline of (r, Q) policy, the proposed methods show promising results for the supply chain optimization problem.