Computationally Efficient Predictive Control of Inverters in AC Microgrids Using Random Forests

Accurate real-time control of three-phase inverters in AC microgrids is challenged by the need to effectively manage multi-variable control while meeting tight sampling instant time limit. Model predictive control (MPC) offers efficient constraint handling, but it often suffers from computational complexity. To address this, this paper implements a random forest (RF)based controller that is trained offline with extensive historical data extracted from MPC. Thereby reducing computational demands and improving adaptability to varying loading conditions. The proposed RF-based controller is assessed under various operating scenarios of the AC microgrid, including resistive, inductive, and capacitive loading conditions, demonstrating its ability to maintain high-quality sinusoidal output voltage with low total harmonic distortion (THD). The findings prove that this approach not only preserves the advantages of MPC but also significantly enhances computational efficiency, offering a promising solution for modern power electronics applications.