Optimization of multi agent based energy management systems using reinforcement learning for microgrids

Abstract

People's concerns about environmentally friendly power generation have given rise to a new concept known as distributed generation. The integration of renewable-based distributed generation resources into the main power grid, on the other hand, is difficult due to their constant monitoring and control requirements. As a result, microgrids have been identified as appropriate platforms for integrating distributed generation resources and loads. Instead of using traditional centralized control architecture, these microgrids use distributed control systems like multi-agent-based systems as a novel approach. Social, reactive, proactive, and autonomous are the common features of these control agents. These agents can be improved by using machine learning-based technologies to introduce intelligence. As a result, the focus of this research is on using Reinforcement Learning as an experimental machine learning method to optimize the energy generation function of a grid-connected microgrid so that the microgrid's grid dependency is minimized as the agent learns. To determine the best technique for system optimization, the proposed microgrid's performance is tested using single and multi-agent reinforcement learning models in Python. A hardware testbed is developed for the selected high-performance model to demonstrate the practical applicability of reinforcement learning.