Using the proximal policy optimisation algorithm for solving the stochastic capacitated lot sizing problem

This paper studies the multi-item stochastic capacitated lot-sizing problem with stationary demand to minimise set-up, holding, and backorder costs. is a common in industry, concerning both inventory management production planning. We study applicability of Proximal Policy Optimisation (PPO) algorithm this problem, which type Deep Reinforcement Learning (DRL). The modelled as Markov Decision Process (MDP), can be solved optimality small instances by using Dynamic Programming. In these settings, we show that performance PPO approaches optimal solution. For larger an increasing number products, solving intractable, demonstrate solution outperforms benchmark Several adjustments standard are implemented make it more scalable instances. linear growth computation time for algorithm, present method explaining outcomes algorithm. suggest future research directions could improve scalability explainability