Efficient use of PV in a Microgrid using Reinforcement Learning

Abstract

Artificial Intelligence is a new concept to optimize or schedule the energy storage system of the Microgrid. The reinforcement learning (RL) method can be used in the effective scheduling of the battery connected to the microgrid. The proposed strategy aims to reduce energy costs while prioritizing both energy balance and user comfort within the microgrid. The key innovation lies in developing an optimal policy for battery actions (charging, discharging, idle) using a model-free stochastic approach. One significant aspect that sets this work apart from others is its acknowledgment of the non-deterministic nature of the state of charge (SOC) of the battery. Unlike systems that solely rely on grid charging, our approach takes into account the unpredictability of renewable energy sources, particularly solar power, which heavily depends on varying time instances and weather conditions throughout the day. Consequently, the SOC of the battery exhibits non-deterministic behavior due to the uncertainty in the availability of excess renewable energy for charging. The RL-based policy presented in this research capitalizes on the effective utilization of photovoltaic sources, optimizing the battery’s discharge and idle states. By intelligently adapting to the dynamic energy supply from renewable sources, the proposed approach ensures that the battery is charged only when surplus energy is available beyond fulfilling the overall system load demand.