Reinforcement Learning for Decentralized Robust Optimal Voltage Control of Uncertain Islanded DC Microgrid Under ZIP Load

This paper delves into the application of robust optimal control theory for voltage regulation in DC microgrids with uncertain ZIP loads. The primary challenge in DC microgrids with local ZIP loads is addressed through a two-phase approach encompassing classical robust control and data-driven control methodologies. Initially, the robust control problem for voltage regulation is tackled using an undiscounted optimal approach. Subsequently, the classical structure of the proposed robust optimal control scheme is converted into a data-driven control strategy employing a reinforcement learning (RL) algorithm. Given the system's unmatched uncertainties, a virtual control input is necessary during the robust control problem-solving process, preventing the extension to a model-free control strategy. By converting the unmatched uncertainties into matched ones in the first phase, a data-driven robust control strategy is achieved using the RL-based algorithm in the second phase. The simulation results which are obtained using MATLAB/SimPowerSystems toolbox showcase the effectiveness of the data-driven approach in achieving stability and adaptability in uncertain DC microgrid environments.