Robust Stability Control of Droop-Controlled Microgrids Under Worst-Case Parameter Uncertainty via Adversarial Deep Reinforcement Learning

Microgrids based on droop-controlled inverters can achieve small-signal stability control by adjusting the droop coefficients. However, due to discrepancies between modeling parameters and actual system parameters in inverter-based microgrids, the droop settings determined from the model often fail to deliver the expected control performance in real-world applications. To address this critical issue, this paper proposes a droop coefficient tuning strategy based on adversarial reinforcement learning, which exhibits strong robustness against system parameter uncertainties. The uncertain parameter behavior and the robust tuning of droop coefficients are respectively modeled as a Markov Decision Process (MDP) and an Adversarial Markov Decision Process (AMDP). The twin delayed deep deterministic policy gradient (TD3) algorithm is employed to solve these processes, yielding the worst-case uncertainty policy and the corresponding robust droop control strategy. Finally, case studies are conducted on a microgrid system with four inverter-based resources (IBRs) and a modified IEEE 33-bus distribution system to validate the effective ness and scalability of the proposed method.