A Standardised Environment for the Application of AI in Production Scheduling Research

Abstract

Production scheduling contains a wide variety of nondeterministic polynomial time hard problems that are differentiated by machine setups, constraints and optimisation targets. Traditionally, scheduling is run overnight for the next day and cannot adapt to a dynamic situation. Newer research methods such as deep reinforcement learning aim to address this problem, but require robust environments which can be generalised to these different setups in order to be examined properly. Currently, researchers must create their own environments when examining production scheduling problems due to the often proprietary nature of the research undertaken. This process both takes time and means that comparisons between methods are difficult. This work introduces a suitable environment
which can be applied to different setups in order to reduce this wasted time and allow easier comparisons between current research methods.