Reinforcement Learning-Based Energy Management Algorithms Effect on Microgrid Physical Properties.

This paper studies the application of Artificial Intelligence (AI), precisely Reinforcement Learning (RL), to minimize energy costs of a grid-connected community Microgrid (MG). This MG is composed of photovoltaic (PV) panels, dynamic and static loads, and a battery storage system (BSS). RL algorithms learn through interacting with a MG simulator to make decisions in real time that minimizes energy costs without prior knowledge of uncertainties related to PV production, load consumption and electricity costs. In this article, two RL algorithms, Q-learning and State-Action-Reward-State-Action (SARSA), are studied. These algorithms were benchmarked with two baselines in order to test its effectiveness. The first baseline involves no storage, and the second baseline involves storage management provided by a linear programming (LP) algorithm. The main contribution of this work is the investigation of the RL-based energy management approaches on the MG voltage magnitude at the Point of Common Coupling (PCC). In this sense, an electrical model of the studied MG is adopted. Simulation results shows effective results of the data-driven methods (based on Q-learning and SARSA) where no physical constraints violating have been recorded.