•  
  •  
 

Keywords

distribution network; high-resistance ground fault; panoramic waveform of traveling wave; RMT; multi-head self-attention mechanism; wavelet packet Shannon entropy

Abstract

When a high-resistance grounding fault occurs in a distribution network, the fault signal is weak, and traditional methods based on single-feature extraction are prone to causing protection maloperation. In response, this paper introduces an effective identification method for high-resistance ground faults in distribution networks based on differential features of traveling wave voltage signals and a "retentive network meets vision transformer" (RMT) model. Initially, the traveling wave voltage signal is extracted to implement wavelet packet time-frequency features, and the differences in the time-frequency domain responses between high-resistance ground faults and normal disturbance conditions are visualized. Then, a composite RMT model integrating retentive network (RetNet) and vision transformer (ViT) is constructed, utilizing RetNet's memory mechanism and incorporating temporal priors into ViT to enhance the memory and attention to rare events and thereby address the issues of limited sample data and their imbalanced category distribution in distribution networks. Subsequently, the panoramic waveform of traveling wave is input into the RMT model, and hyperparameters are optimized to improve the model's discriminative performance, so as to achieve precise identification of high-resistance ground faults. Finally, a simulation test is conducted to verify the feasibility and effectiveness of the proposed method. Simulation results confirm that this approach overcomes the limitations of traditional methods that rely on a single feature quantity, maintaining high classification accuracy even with small, imbalanced datasets, and accurately identifying high-resistance ground faults under various fault conditions.

DOI

10.19781/j.issn.1673-9140.2026.02.011

First Page

117

Last Page

126

Download

Share

COinS