Dynamic scheduling in a job-shop production system with reinforcement learning
Csaba Kardos, Catherine Laflamme, Viola Gallina, Wilfried Sihn
Abstract
Fluctuating customer demands, expected short delivery times and the need for quick order confirmation creates a fast-paced scheduling environment for modern production systems. In this turbulent scene, using the data provided by intelligent elements of cyber-physical production systems opens up new possibilities for dynamic scheduling. The paper introduces a reinforcement learning approach, in particular Q-Learning, to reduce the average lead-time of production orders in a job-shop production system. The intelligent product agents are able to choose a machine for every production step based on real-time information. A performance comparison against standard dispatching rules is given, which shows that in the presented dynamic scheduling use-cases the application of RL reduces the average lead-time.