Optimizing Supply Chain Management with Approximate SARSA and Reinforcement Learning

Supply chain optimization and inventory management have become increasingly critical issues in the era of globalization and the post-pandemic world. These topics have garnered significant attention across various sectors. However, due to the high-dimensional state and action spaces, the diverse and dynamic environmental variables, the need for continuous decision-making over time, and the data-dependence and lack of flexibility of traditional methods, we model the problem as a Markov Decision Process (MDP). To address it, we employ Approximate SARSA, a method using Q-value approximation, alongside RL methods based on different feature processing techniques. Our experiments evaluate and extend the applicability of the model, while also attempting to provide interpretability.