A Deep Q-Learning Approach for Continuous Review Policies with Uncertain Lead Time Demand Patterns

Setting safety stock levels and expediting replenishment lead time are tactical policies to hedge against demand fluctuations and supply uncertainties in supply chains. However, the large variable space of planning possibilities makes it difficult to obtain an optimal trade-off policy in a dynamic supply chain environment. In this paper, a deep Q-learning approach for dynamic reorder point decisions of a continuous review system in two-stage supply chains is proposed to explore the proactive decisions of combining uncertain lead time demand patterns and lead time expediting activities. The simulation outcomes show that the deep Q-learning model improves decision performances evidently and identifies the lead time selection tactics depending on expediting cost.