A Comparative Study of Reinforcement Learning and Classic Nonlinear Methods for Stabilization of DC Microgrids Supplying Constant Power Loads

This paper provides a review of stabilization methods for constant power loads (CPLs) in DC microgrids. As a result of their non-linear characteristics, CPLs may lead to voltage and frequency fluctuations, resulting in instability issues. Specifically, the focus of this study is on source-side stabilization methods. DC microgrids face various obstacles, including the unpredictable nature of CPLs and renewable energy sources, unexpected changes in energy demand, and parallel operation of sources. For these issues to be addressed, an adaptable control and stabilization method needs to be developed without requiring a precise model of the system. A review of non-linear control techniques in the literature is presented in this paper along with their advantages and disadvantages. A DC microgrid may benefit from data-driven control methods, such as Reinforcement Learning (RL), because classic control methods require a detailed model of the system. The RL is a model-free machine learning technique that is capable of learning from interactions with the environment. This paper provides an overview of the application of RL for CPLs in DC microgrids and discusses future research directions in this field.