Reinforcement Learning for Inventory Management

Abstract

The decision of “how much to order” at each stage of the supply chain is a major task to minimize inventory costs. Managers tend to follow particular ordering policy seeking individual benefit which hampers the overall performance of the supply chain. Major findings from the literature show that, with the advent of machine learning and artificial intelligence, the trend in this area has been heading from simple base stock policy to intelligence-based learning algorithms to gain near-optimal solution. This paper initially focuses on formulating a multi-agent four-stage serial supply chain as reinforcement learning (RL) model for ordering management problem. In the final step, RL model for a single-agent supply chain is optimized using Q-learning algorithm. The results from the simulations show that the RL model with Q-learning algorithm is found to be better than Order-Up-To policy and 1–1 policy.