Controlling distributed energy resources via deep reinforcement learning for load flexibility and energy efficiency

Behind-the-meter distributed energy resources (DERs), including building solar photovoltaic (PV) technology and electric battery storage, are increasingly being considered as solutions to support carbon reduction goals increase grid reliability resiliency. However, dynamic control of these in concert with traditional loads, effect efficiency demand flexibility, is not yet commonplace commercial products. Traditional rule-based algorithms do offer integrated closed-loop optimize across systems, most often, PV systems operated for arbitrage charge management, the provision services. More advanced approaches, such MPC have been widely adopted industry because they require significant expertise develop deploy. Recent advances deep reinforcement learning (DRL) a promising option operation DER loads reduced setup effort. there limited studies that evaluate efficacy methods multiple subsystems simultaneously. Additionally, research has conducted simulated environments opposed real buildings. This paper proposes DRL approach uses deterministic policy gradient algorithm HVAC storage presence on-site generation. The algorithm, trained on synthetic data, was deployed physical test evaluated against baseline current best-in-class strategies. Performance delivering efficiency, load shift, shed tested using price-based signals. results showed DRL-based controller can produce cost savings up 39.6% compared controller, while maintaining similar thermal comfort building. project team also simulation components developed during this work an OpenAIGym environment made it publicly available so prospective researchers leverage alternate algorithms.