Deep Reinforcement Learning (DRL) can optimize the operating strategies of industrial energy supply systems (IESS), enhancing energy efficiency and flexibility while reducing costs. This work conducts and evaluates a real-world application of Proximal Policy Optimization (PPO), a widely used DRL algorithm, on the complex ETA Research Factory’s supply system. Our approach involves five steps: system boundary definition and data accumulation, modelling and validation, DRL implementation, simulation-based evaluation and finally, application of the DRL-based controller on the real system. In simulation experiments, the controller demonstrates a 43% reduction in operational costs with respect to the conventional controller while maintaining operational stability within safety limits, thus proving to be a cost-effective and reliable solution. During real-world application, the DRL-based controller upheld safety and robustness. Future research should aim at establishing comparability under various environmental conditions and closing the gap between simulation and reality.