Critical Features in AdaBoost Algorithm for Islanding Detection: A Review

Islanding detection in microgrids with high renewable energy integration is a critical challenge due to the variability of inverter-based distributed generation (DG) and reduced fault current contributions. Traditional islanding detection methods, including passive, active, and hybrid techniques, face limitations such as non-detection zone (NDZ), power quality degradation, and implementation complexity. Machine learning (ML) approaches, particularly the AdaBoost algorithm, have emerged as promising solutions for enhancing detection accuracy and reliability. This review focuses on the role of feature selection in AdaBoost-based islanding detection, highlighting its impact on classification accuracy, computational efficiency, and resilience to noise. Key electrical features such as voltage deviation, frequency deviation, rate of change of frequency (ROCOF), total harmonic distortion (THD), and impedance variations are examined for their effectiveness in differentiating islanding from grid disturbances. Advanced feature engineering techniques, including wavelet transforms (WTs), recursive feature elimination, and correlation analysis, are explored to optimize the selection of critical parameters. Despite its advantages, the AdaBoost algorithm encounters challenges in high-renewable grids, such as feature redundancy, noisy data, and evolving grid dynamics. Addressing these issues requires adaptive feature selection methods, real-time signal processing, and hybrid ML frameworks. Future research should focus on dynamic AdaBoost models, standardized datasets, and integration with deep learning techniques to improve islanding detection performance. By refining feature selection strategies, AdaBoost can be further optimized for accurate, efficient, and real-time islanding detection in modern power networks.