Optimizing production planning and control: A potential analysis of reinforcement learning

Abstract Producing companies are increasingly challenged by the complex and volatile environment they have to deal with. The collapse of supply chains due to the COVID-19 pandemic and the chip crisis are two examples that emphasize the relevance of handling global disruptions. Production planning and control (PPC) especially suffers from an uncertain environment, leading to frequent replanning and rescheduling activities. Several publications highlight the importance of innovative approaches from Reinforcement Learning (RL) to manage the challenges in PPC. The underlying structure of the PPC task essentially determines the contribution RL can make. While some selected tasks already show excellent potential regarding the applicability of RL (e.g., order dispatching), others are rarely supported by RL yet (e.g., order management). However, it is unclear how research and industry can identify tasks and problems in PPC that are well-suited for applying RL. To meet this research demand, this publication provides a methodology to identify the potential of RL for PPC tasks. The methodology was developed through literature research and expert workshops. It was applied to a real-world use case, indicating the applicability and success potential of the suggested methodology.