In recent years, research into decision-making problems using reinforcement learning in uncertain environments has been actively conducted. However, in the real world, when making decisions using reinforcement learning, not only is excellent performance required, but the basis for the decision is also required to be explained. The Markov decision process is a framework for expressing an environmental model for reinforcement learning. In this study, we apply dynamic programming to an inventory management problem formulated as a Markov decision process to derive an ordering policy. To explain the ordering policy, we use a local explanation method for the pair of state and action, and calculate the influence of each inventory amount. We generated an explanation of the ordering policy in natural language using a large language model. We also selected a kernel width that affects the results of the explanation method. The results showed that the explanation method can effectively present the influence of each inventory amount in the inventory management problem, and the results of the explanation method could be generated in natural language using a large language model. Additionally, the accuracy of the explanation method can be improved by selecting an appropriate kernel width. This study proposes a method to make it easier for people without specialized knowledge of inventory management to understand the ordering policy. It contributes to improving the explainability of decision-making problems using reinforcement learning.