Reinforcement Learning-Based Battery Energy Management in a Solar Microgrid

The intermittent nature of Renewable Energy Sources (RES) leads to a mismatch between electricity supply and demand, thus, there is a need for energy storage and load management. This paper presents a framework using Reinforcement Learning (RL) to control the operation of a battery storage device in a microgrid. Here, the microgrid considered consists of a photovoltaic (PV) system, inverters, residential consumer and battery storage. The optimal operation of the battery is formulated as a Markov decision process. A deterministic setting is considered with the weather forecast for PV production and electricity consumption known in advance. The agent learns an optimal energy management policy by using its past experiences. The developed solution was tested in a Belgian case with local load and PV production profiles of residential consumers. Index terms microgrid, reinforcement learning, battery, PV, storage.