Deep Learning Optimal Control for a Complex Hybrid Energy Storage System

Deep Reinforcement Learning (DRL) proved to be successful for solving complex control problems and has become a hot topic in the field of energy systems control, but particular case thermal storage (TES) systems, only few studies have been reported, all them with complexity degree TES system far below one this study. In paper, we step forward through DRL architecture able deal an innovative hybrid system, devising appropriate high-level operations (or policies) over its subsystems that result optimal from or monetary point view. The results show policy can reduce operating costs by more than 50%, as compared rule-based (RBC) policy, cooling supply reference residential building Mediterranean climate during period 18 days. Moreover, robustness analysis was carried out, which showed that, even large errors parameters simulation models corresponding error multiplying factors up 2, average cost obtained original model deviates optimum value less 3%, demonstrating solution wide range errors.