Modelling Framework for Reinforcement Learning Based Scheduling Applications

Over the last years, reinforcement learning has been extensively applied to schedule complex and dynamic systems. There are multitudes of simulation environments and algorithms, which hinder standardization and impede testing the suitability of reinforcement learning for specific scheduling applications and their easy implementation. This article proposes a framework to model production systems easily and transform them into standard industry simulation software to solve this issue. This framework contains major elements of classic production systems and references them adequately to allow effortless modelling. Furthermore, the domain models’ adjacent systems and their respective functionalities are described to facilitate reinforcement learning-based scheduling. This study demonstrates the framework's applicability using an existing dynamic scheduling problem. The experiences during modelling and training of the reinforcement learning subsequently are discussed.