A Policy Gradient Based Reinforcement Learning Method for Supply Chain Management

Technological advances of the recent decades have significantly affected the business world, the retail business in particular. Retailers need to innovate to maintain a competitive edge over competitors and ensure business sustainability. Therefore Research and Development is a crucial component of businesses growth. Intuition based approaches are replaced by supply chain computerized solutions such as inventory management, warehousing, allocation and replenishment. This paper aims at building a reinforcement learning agent capable of placing optimal orders for the sake of constructing a replenishment plan for next period. The goal is to develop a novel method of inventory replenishment. We base the developed module on the recent breakthroughs of reinforcement learning research in using deep neural networks for control. We compare the classical RL methods to the recently introduced Proximal policy optimization algorithm. As far as we know, this is the first time PPO is used in Supply Chain Management.