Reinforcement Learning-based Energy Management System for Microgrids with High Renewable Energy Penetration

This paper presents a microgrid energy management system (M-EMS) integrating artificial intelligence (AI) and the Internet of Things (IoT), designed for high renewable energy penetration microgrids. The system comprises three control stages. First, the day-ahead planning considers the solar power generation forecasts, load consumption forecasts, and electric vehicle (EV) charging demand. A linear programming optimization algorithm generates optimal scheduling and power transaction recommendations. Second, the daily economic dispatch optimization occurs in five-minute time windows, including energy storage system (ESS) and EV charger scheduling, and outputs power adjustment strategies for solar smart inverters. Lastly, the real-time control utilizes an AI-based reinforcement learning (RL) model within the five-minute time window at one-minute intervals, swiftly adapting microgrid ESS and EV charging power to tackle uncertainties in renewable energy generation and load consumption. The proposed M-EMS has been tested in a commercial building microgrid. With the proposed system, power utilization efficiency improves by 4%, power savings increase by 121%, power losses reduce by 61%, power transaction aggregation surges by 135%, and peak demand reduction reaches 17%. This approach aims to address uncertainties in renewable energy generation and load electricity consumption, enhancing the overall efficiency and reliability of microgrids.