Reinforcement learning-driven dynamic Model Predictive Control for adaptive real-time multi-agent management of microgrids
Darioush Razmi, Oluleke Babayomi, Zhenbin Zhang
Abstract
Nowadays, renewable energy sources (RESs) are widely used to enhance the performance of existing energy distribution systems . The emergence of microgrids and the integration of these resources have created new opportunities, while also presenting significant challenges. These challenges involve fluctuations in power generation and the critical need to ensure power quality, necessitating innovative management strategies for effective solutions. Multi-agent systems can effectively address these issues by facilitating decentralized control and coordination among diverse energy sources. To effectively deal with the combined challenges of distributed control and real-time power regulation under dynamic operating conditions, this paper proposes an integrated approach that combines model predictive control (MPC) with a multi-agent system (MAS). This combination enables adaptive, scalable, and coordinated control in the presence of renewable energy variability. This paper firstly presents an improved Model Predictive Control (MPC) strategy aimed at improving system performance . Then, reinforcement learning is effectively applied within the multi-agent system framework to optimize the control coefficients, enhancing the adaptability and overall performance of the control approach. The performance of the system has been evaluated using MATLAB Simulink, ensuring accurate and realistic assessments through comprehensive analysis.