Adaptive Fuzzy Attention Inference to Control a Microgrid Under Extreme Fault on Grid Bus

Abstract

Power quality of a microgrid falls due to large penetration in renewable sources with rapid fluctuations. This article develops a double Q Learning (DQL) based adaptive fuzzy attention inference (FAI) to control a microgrid connected to the grid. Detailed dynamic modeling is presented with renewable sources such as photovoltaic (PV), wind, fuel, and microalternators with respective power electronics interfaces. The inference scheme controls the phase angle of the static compensator (STATCOM) coupled with capacitive energy storage device in the microgrid. STATCOM provides reactive power, while the capacitive storage corrects real power imbalances. Numerical results showed the developed control scheme restores stability rapidly in the event of severe symmetrical three-phase to ground fault on the grid bus. Rule base and corresponding membership function of the inference dynamically adapts based on the DQL mechanics. Prioritized experience and shallow neural net reward model are integrated where importance sampling of the temporal difference based rewards, contributes to overcome microgrid transients. The proposed control outperform conventional optimized proportional-integral-derivative (PID), model predictive, and sliding mode control schemes to restore normal operation of the converters across different levels of electromechanical transients.