Model Free Control of Islanded Microgrids

Abstract

Micro-grids have been a major means of electricity using renewable energy sources. Since a decade the researchers are working on different control and management techniques for the power and load management in micro-grids working in islanded mode. When a micro-grid operates in islanded mode it has to fulfill the power demand all by itself. As renewable energy resources have an intermittent nature so frequency control is a major issue while dealing with them. Frequency deviation indicates the power deficiency in micro-grid. Considering the behavior of Renewable energy sources and load variations this research proposes a model free control of micro-grid using Reinforcement learning for secondary frequency regulation of a micro-grid. Deep reinforcement learning is employed on micro-grid environment con sisting a massive number of distributed energy resources for reducing the frequency deviations produced in islanded mode. The system is developed in MATLAB 2022a using both value based and policy based method of deep reinforcement learning and validated in terms of frequency deviations and overall cost minimization. The proposed model free control is compared with distributed model predictive controller, adaptive Linear Quadratic Regulator and PID. The results show a 95 percent reduction in frequency deviation as compared to distributed model predictive control and adaptive Linear Quadratic Regulator respectively.