Demand response for cleaning machines: A comparative study of deep reinforcement learning and model predictive control in application

Demand response strategies offer significant potential for optimisation of electrical energy consumption in industrial manufacturing. This paper presents a comparative analysis of two demand response algorithms, using model predictive control (MPC) and deep reinforcement learning (DRL). While existing algorithms for demand response implementation often rely on simulation for validation, this study contributes by implementing MPC and DRL on a real aqueous parts cleaning machine, providing a quantitative and qualitative comparison. In the experiments, the MPC algorithm achieves an energy cost reduction of 58 %, the DRL of 57 %, in comparison to standard operation. The qualitative comparison underscores the importance of domain knowledge in implementing both algorithms effectively. While DRL shows promise in operational efficiency, its black-box nature and implementation challenges warrant careful consideration in industrial settings. This study not only provides insights into the advantages and disadvantages of both methodologies in practical scenarios, but also sets a benchmark for future research in industrial energy optimisation.