The integration of digitalization in production systems is now a crucial requirement for both new as well as existing systems. By utilizing digital twins, which are virtual simulations of actual production systems, we can streamline analysis processes, improve our understanding of the systems, and unlock numerous optimization opportunities. In tandem, integrating machine-learning methods enables the processing of digital data, unveiling novel production control strategies. This paper aims to develop a control method for a production system operating in a dynamic environment using reinforcement learning. The designed reinforcement learning agent's role is to minimize production costs in terms of inventory and backordered demands and the dynamic behavior of the system is achieved through discrete event simulation.