High-impedance fault detection in resonant grounded distribution networks using multi-fractal detrended fluctuation analysis and an improved whale-optimized support vector machine

Detecting High-Impedance Faults (HIFs) in resonant grounding distribution networks is challenging due to weak fault currents, dynamic transition resistance and strong background disturbances. This paper proposes a hybrid protection scheme that combines Multi-Fractal Detrended Fluctuation Analysis (MFDFA) with an Improved Whale Optimization Algorithm (IWOA) tuned Multi-level Support Vector Machine (MSVM), denoted IWOA-MSVM. The transient responses of low-impedance faults and HIFs are analytically characterised and grouped into three regimes: low-resistance, high-impedance overdamped and high-impedance underdamped states. MFDFA is applied to zero-sequence currents to extract a compact set of multifractal features, which form the input vector to the IWOA-MSVM classifier. The IWOA incorporates inverse learning, nonlinear convergence control and adaptive restart to tune the key Support Vector Machine (SVM) hyperparameters, thereby improving both classification accuracy and convergence reliability. Simulation studies on a 10 kV resonant grounding distribution network show that the proposed method attains 96.92 % overall recognition accuracy, surpassing other metaheuristic-optimized SVM schemes and maintaining correct identification of the most critical high-impedance underdamped faults. Additional tests under additive measurement noise and feeder reconfiguration demonstrate strong noise immunity and structural adaptability. The results indicate that integrating multifractal analysis with optimized small-sample machine learning provides a practical and robust solution for high-impedance fault detection in modern distribution networks. • Forms an analytical model of high-impedance faults in resonant grounded feeders. • Uses MFDFA on zero-sequence currents to derive compact multifractal fault features. • Classifies low and high impedance fault regimes with an IWOA tuned SVM framework. • Achieves 96.92 % fault recognition accuracy on a 10 kV resonant grounded network. • Maintains robust performance under measurement noise and feeder reconfiguration.