Reinforcement Learning Approach in Microgrid Energy Management System Based on Firefly Moving Regression for Optimization

With the increasing demand for electricity and the occurrence of energy crises, the development of microgrids (MGs) has become more common. The utilization of Renewable Energy Sources (RES) in Microgrids is more advantageous. This study examines the integration of Distributed Generation (DG) and Solar Photovoltaic (PV) systems into a Microgrid. The primary aim of the objective function is to minimize the accumulation of power differences between the generation from renewable energy sources (RES), such as solar photovoltaic, and the demand within the microgrid system. The ability to produce electricity or receive power from the utility grid (UG) is dependent upon the energy capacity of the Microgrid. In this context, the Firefly Algorithm (FA), Moving Regression (MR) combination between Moving Average (MA) and Linear Regression (LR), and Reinforcement learning approach is Deep Q Network (DQN) strategy are employed as effective tools for addressing the optimization problem in discussion. The proposed methodology is being applied in the domain of Microgrid, and the outcomes of implementation of the Firefly algorithm and Reinforcement learning are being compared with the original dispatch methods. During the process of dispatching, a certain amount of power is allocated as backup power to mitigate the impact of power outages, hence improving the overall supply reliability of the microgrids. In this work, the viability of the proposed method in power dispatch strategy is visualized in Python Environment. A program-enabled industrial computer has been placed in Microgrid 1 at National Central University to assess the viability of the algorithm and the optimization outcome of the scheduling approach in terms of system supply reliability.