Community Battery Management using Reinforcement Learning in a residential scenario
The increasing interest in the residential renewable energy sources leveraged the development of the Battery Energy Storage Systems (BESS) ecosystems that allow the losses mitigation caused by the lack of intelligent storage of unused energy. BESS's have reduced user's costs by storing the overproduced energy by the renewable sources and using it when the demand was bigger than the production. Despite that, cost minimizing can yet be improved by applying a Energy Management System (EMS) which takes in consideration several factors and draws a policy to reduce costs. Model-based approaches have been thought as a solution but the increasing complexity of the models turned the researchers into a model-free direction using Reinforcement Learning. This paper creates an innovative agent, trained by Deep Reinforcement Learning algorithms, capable of reducing the costs in a building. A literature review of other approaches, a solution formulation to solve the problem, and evaluation using synthetic data and data supplied by an energy company are contained. Furthermore, evaluation scenarios and techniques that can lead to agent' instability are mentioned. Results indicate that the agent has better performance in terms of revenue when compared with the same BESS operation without the agent.
This work was supported by national funds through Fundação para a Ciência e a Tecnologia (FCT) with reference UIDB/50021/2020 (INESC-ID).
Email Address of Submitting Authorsergio.email@example.com
ORCID of Submitting Author0000-0002-8627-3338
Submitting Author's InstitutionINESC-ID, Instituto Superior Técnico, Universidade de Lisboa
Submitting Author's Country