Detection of High-Impedance Fault in Distribution Networks Using Frequency-Band Energy Curve

Detecting high-impedance faults (HIFs) in distribution networks poses a significant challenge for conventional relay devices due to low fault currents and various characteristics such as weaker fault features, distortion offset, and background noise interference. This article introduces a novel and streamlined method for HIF detection, which ingeniously integrates frequency-band energy curve (FBEC) analysis and Gaussian smoothing to extract trend changes, thus enhancing the precision and effectiveness of HIF detection. The proposed method utilizes continuous wavelet transform (CWT) to extract the time-frequency spectrum from the zero-sequence current. By analyzing the feature-band energy, the FBEC is computed. To mitigate noise interference and enhance the periodic change pattern, a Gaussian filter is applied for smoothing. Distinguishing between HIF and normal operations, including low impedance fault (LIF), capacitor switching (CS), inrush current (IC), and ferromagnetic resonance (FR), is achieved by analyzing the peak points of FBEC. The proposed method’s performance was extensively validated through simulations and field data. The performance of the proposed method was extensively validated through a series of simulations and detailed analysis of real-world field data. The results demonstrated an excellent detection performance on field data, with an impressive accuracy rate of 86.5% and an
$\text{F}_{{1}}$
-score of 0.87. Moreover, we examined the method’s resilience against noise using data with a signal-to-noise ratio (SNR) of 20 dB, resulting in a detection accuracy of 77.3% and an
$\text{F}_{{1}}$
-score of 0.79. These findings underscore the method’s clear physical meaning, strong interpretability, and versatility, establishing its effectiveness and practicality for real-world applications.