A Generalized Dynamic Nonsmooth Control Design for DC Microgrids based on Deep Reinforcement Learning

A novel generalized dynamic nonsmooth control framework, integrated with the deep reinforcement learning (DRL) algorithm, is developed in this paper for the primary layer control of DC microgrids. Aiming to strike a balance between the adaptiveness and robustness in traditional nonsmooth control, first, a benchmark nonsmooth controller is designed with reference to the existing literature. Second, the synthesis dynamic nonsmooth control strategy is presented, in which a soft actor-critic (SAC) algorithm is utilized to realize parameter self-configuration according to the current operating condition. The contributions of the proposed control framework can be summarized as the following twofold: the robustness and adaptiveness of the closed-loop system can be feasibly endowed. Besides, the large-signal stability of the DC/DC converter system can be guaranteed. The efficacy of the investigated approach is verified by simulation studies based on MATLAB/Simulation platform.