Can Deep Reinforcement Learning Improve Inventory Management? Performance and Implementation of DRL At a Global Consumer Goods Shipper
The popularity of reinforcement learning is growing but is it effective in operations? We provide proof of concept that deep reinforcement learning (DRL) can be applied to classic, yet intractable operations problems such as dual-sourcing or dual-mode inventory replenishment problems. Step-by-step guidance on how to apply DRL at a consumer goods shipper is offered together with a careful discussion of its performance, strengths and weaknesses.
- DRL is a promising concept to improve operations
- Sufficient data and a realistic model of the environment are necessities
- Companies should invest in know-how as the current techniques are more technical compared to simple heuristics
Joren Gijsbrechts is a PhD student at the Operations Management department of the Faculty of Economics and Business, KU Leuven. He obtained his Master’s degree in Business Engineering, majoring in transportation and logistics, and worked two years in the consumer goods industry prior to starting his PhD. Joren develops applied models and tools to optimize operations management problems such as inventory control, transport mode choice decisions, reducing variability in the supply chain and offshoring. His main methodological focus consists of optimization methods combining machine learning (reinforcement learning or approximate dynamic programming), simulation, (mixed) integer programming and stochastic programming. In addition to research, Joren provides lectures and courses on the impact of the latest machine learning algorithms on the field of operations management.