Autonomous Micro-Grids: A Reinforcement Learning-Based Energy Management Model in Smart Cities

The growing electricity consumption of communities has raised concerns about the environmental impact of traditional energy sources. To mitigate these concerns, governments are promoting to utilize renewable energy sources. However, the intermittent nature of renewable energy poses significant challenges for gird stability. Micro-grids have emerged as promising solutions to address these challenges. A micro-grid is an independent electric system that can generate, distribute, and manage electricity within a small area. It offers many advantages such as; peak load reduction, minimized load variability, and enhanced power quality. Energy management systems (EMS) within micro-grids play a significant role in overcoming operational challenges. They are designed to control micro-grid systems with the goal of flattening, smoothing, and reducing the curve of electrical demand. This helps to reduce the operational costs of electricity generation, transmission and distribution. Reinforcement Learning (RL) has been an important research area for EMS systems. By leveraging historical and real-time data, RL enables effective control of EMS systems within micro-grids. However, despite the advancements in this area, many of these research is challenging to reproduce. In this work, we use SAC and PPO RL agents in a micro-grid architecture. We make use of Citylearn framework to test our agents. We compare our agents with the Rule Based Controller (RBC). Our test results show that our solution is able to improve the micro-grid performance by effectively smoothing the electricity consumption.