Islanding Detection in Power Distribution Systems Using a Hybrid Machine Learning Approach

The increasing integration of distributed generation units, especially synchronous generators, into modern power systems emphasises the critical need for reliable islanding detection methods. This paper proposes a hybrid islanding detection approach that combines passive monitoring of multiple electrical parameters—such as negative sequence voltage, generator field voltage, and the rate of change of positive sequence impedance—with intelligent machine learning techniques. To enhance detection accuracy, the approach incorporates feature standardisation, Adaptive Synthetic Sampling for data augmentation, Recursive Feature Elimination for optimal feature selection, and a stacking ensemble integrating XGBoost and Random Forest models. The method is evaluated under various network conditions, including load switching, capacitor bank operations, faults, and islanding events, using a modified IEEE 34‐bus distribution system. Simulation results demonstrate that the proposed hybrid method achieves high accuracy, robustness, and generalisability across diverse operating conditions, making it a promising solution for reliable islanding detection in distribution networks.