Design of a robust neural network-based controller for frequency stability in microgrids

The utilization of renewable energy sources (RESs), including solar and wind, in microgrids (MGs) present a critical challenge for maintaining system stability, mostly because of the elimination of mechanical inertia traditionally provided by synchronous generators. This study presents a controlling technique to guard disturbances in the islanded MG. To address this challenge, a multi-layer feedforward neural network (MLFFNN) is used to enhance the frequency stability of an islanded MG. The MLFFNN controller is compared to traditional controllers such as proportional-integral-derivative-acceleration (PIDA), proportional-integral-derivative (PID) and virtual inertia (VI) to evaluate its performance and effectiveness. Three scenarios are studied: load variations, RES fluctuations, and a combined case including both load variations and RES variability. A comparative study between VI, PID, PIDA, and MLFFNN controllers has been carried out, and shows that the MLFFNN combined with the VI controller is better than VI, PID, PIDA and MLFFNN controllers in all cases. Compared with the uncontrolled system, the MLFFNN with VI reduced the maximum frequency deviation from 4.92 to 3.86× \:10⁻⁵ Hz in the third case. Finally, the MLFFNN combined with the VI controller provided better performance for frequency stability.