Due to the fourth industrial revolution, the central element of product manufacturing - production planning and control – has been subject to various new challenges as well as opportunities. The literature review shows that topics like complexity, dynamics, and new organizational forms are already being focused on. Still, aspects such as derivation of actions and the transfer of knowledge in unknown situations are among the greatest challenges for existing methods. The method developed in this thesis addresses these challenges and investigates possible solution strategies. Additionally, the new method is evaluated over a variety of scenarios and compared to other methods. Different characteristics and complexity levels of actions, the observation space, and the amounts of data required for successful training are analyzed. Finally, it is shown that the new method can fulfill the requirements of production planning and control and perform well in unknown scenarios. Additionally, the method is capable of human-like performance and can be used in a real-world scenario.