Reinforcement learning approach to product allocation and storage

Abstract

In this thesis I investigated a reinforcement learning (RL) approach to address effective space utilization for warehouse management. RL in the domain of machine intelligence, it is an approach that learns to achieve a given goal by trial and error iterations with its environment. In this research I explored a solution framework for a warehouse management problem faced by a local distributor in Massachusetts. The distributor is challenged by an increase in inventory levels, and by warehouse management decisions in order to handle the high volume of inventory. Although most distributors utilize warehouse management systems (WMS), in some events it leads to inaccurate and ineffective recommendations from the WMS, and therefore resulting in suboptimal warehouse operations and management. These events include: the dynamic nature of the environment (i.e., fluctuating demand for inventory, high inventory levels), inefficiencies on the floor (i.e., slow rate in replenishing the inventory), and in other events selecting the inappropriate WMS for a certain warehouse needs.