Reinforcement Learning for Inventory Management

Abstract

A comparison between four common reinforcement learning algorithms, namely deep Q network (DQN), double deep Q network (DDQN), prioritized experience reply (DQN + PER) and double DQN + PER; and discussion on the methodology with the limitations and advantages of each algorithm are included in this paper. In order to provide these insights, OpenAI environments that demonstrate the working of these algorithms was used. Mountain car environment was used to generalize our results and prove the consistency of our insights. Insights were derived by evaluating basic parameters like, episode length, minimum rewards, maximum rewards and average rewards. This study discusses strategies for including reinforcement learning in supply chain management by using it for inventory management.