Distributed Stability-Guaranteed Reinforcement Learning-Based Control for Microgrid Voltage Regulation

Microgrids and interconnected power networks are inherently complex and nonlinear, and their operation faces growing challenges with the increasing integration of renewable energy sources (RES). The unpredictable and variable characteristics of RES create several difficulties in controlling crucial variables such as voltage and frequency. Reinforcement learning (RL) has demonstrated strong potential in tackling these situations, as it performs well in uncertain, nonlinear environments. However, guaranteeing the safety and stability of RL-based controllers in power systems is still a major concern. This work introduces a distributed control method for voltage regulation using inverter-based distributed energy resources. The proposed strategy employs a neural network, whose parameters are tuned through an RL-driven optimization process. To ensure stability, independent Lipschitz-like constraints are applied to each local controller, guaranteeing exponential convergence of the voltage mismatch within the microgrid. These Lipschitz constraints are derived by solving a constrained optimization problem designed to maximize the feasible set for the neural network parameters. The correctness and effectiveness of the proposed approach are further assessed in simulation.