This paper presents a method to combine a discrete-event simulation model of a logistics network with a deep reinforcement learning agent. The agent applies different actions to the logistics network in order to learn a strategy to improve the cost structure and the performance of the logistics network. The return for the agent is derived from the costs and the performance of the logistics network which is measured in terms of logistics costs and β service level. Possible actions comprise the relocation of inventory and the adjustment of transport relations. The authors developed a method to represent the state of the simulation model with an image generated from the simulation input data and used a well-known deep Q-learning algorithm with a convolutional neural network to train the agent.