"Model-Less Forecast-Proactive Intelligent Control for Inverter in Islanded Microgrid"

"Islanded inverter-dominated microgrids with high penetration of renewable energy sources face significant voltage regulation challenges under unbalanced and time-varying load conditions. Conventional voltage controllers, such as proportional\u2013integral\u2013derivative (PID) schemes, rely on accurate system modeling and careful parameter tuning, which limits their adaptability under uncertain operating conditions. This paper proposes a model-less forecast-proactive Reinforcement Learning (RL) voltage control framework for islanded AC microgrids supplied through a single DC source and grid-forming inverter. The proposed controller directly adjusts the inverter voltage reference using only local measurements, without requiring an explicit network model. A simple non-learning forecast feature based on the next-hour load schedule is incorporated to improve anticipatory control actions during peak demand periods.The effectiveness of the proposed approach is validated on a four-bus islanded AC microgrid with unbalanced and dynamic loads. Comparative studies against a model-based PID controller and an LSTM-based baseline demonstrate that the proposed RL controller achieves voltage regulation performance comparable to optimally tuned PID control while maintaining strict voltage limits of 0.98-1.02 pu. The results confirm that the proposed model-less framework provides a practical and robust alternative for intelligent voltage control in islanded microgrids."