Data-Driven Design of Control Strategies for Distributed Energy Systems

The flexibility afforded by distributed energy resources in terms of energy generation and storage has the potential to disrupt the way we currently access and manage electricity. But as the energy grid moves to fully embrace this technology, grid designers and operators are having to come to terms with managing its adverse effects, exhibited through electricity price volatility, caused in part by the intermittency of renewable energy. With this concern however comes interest in exploiting this price volatility using arbitrage-the buying and selling of electricity to profit from a price imbalance-for energy cost savings for consumers. To this end, this paper aims to maximize arbitrage value through the data-driven design of optimal operational strategies for distributed energy resources (DERs). Formulated as an arbitrage maximization problem using design optimization principles and solved using reinforcement learning, the proposed approach is applied toward shared DERs within multi-building residential clusters. We demonstrate its feasibility across three unique building clusters, observing notable energy cost reductions over baseline values. This highlights a capability for generalized learning across multiple building clusters and the ability to design efficient arbitrage policies for energy cost minimization. The scalability of this approach is studied using two test cases, with results demonstrating an ability to scale with relatively minimal additional computational cost, and an ability to leverage system flexibility toward cost savings.