Reinforcement learning approaches for specifying ordering policies of perishable inventory systems

- Developed RL policies perform better than the other algorithms.
- RL learn better when the age of the products is used in state representation.
- Age and demand variance are important for perishable inventory management.
- The value of the age becomes critical when the lifetime of the product decreases.

In this study, we deal with the inventory management system of perishable products under the random demand and deterministic lead time in order to minimize the total cost of a retailer. We investigate two different ordering policies to emphasize the importance of the age information in the perishable inventory systems using Reinforcement Learning (RL). Stock-based policy replenishes stocks according to the stock quantities, and Age-based policy considers both inventory level and the age of the items in stock. The problem considered in this article has been modeled using Reinforcement Learning and the policies are optimized using Q-learning and Sarsa algorithms. The performance of the proposed policies compared with similar policies from the literature. The experiments demonstrate that the ordering policy which takes into account the age information appears to be an acceptable policy and learning with RL provides better results when demand has high variance and products has short lifetimes.