•  
  •  
 

Keywords

tree-line grounding fault, Bayesian optimization, XGBoost, multi-feature fusion

Abstract

High-resistance grounding faults induced by tree-line contact are characterized by weak signal features and difficult fault localization, and are one of the significant hidden risks in power grid operations. Existing studies mostly focus on the feature extraction, monitoring, and early warning of faults, but there are still deficiencies in the precise identification and classification of tree-line grounding faults. To address this issue, a tree-line grounding fault identification method combining Bayesian optimization-extreme gradient boosting (Bayes-XGBoost) was proposed. First, a tree-line grounding fault simulation model was established to simulate the zero-sequence current characteristics of different tree species under high-resistance grounding faults, and the signal envelope was extracted combining the Hilbert transform. Second, by analyzing the feature differences of tree-line grounding faults, multi-feature parameters such as shape entropy, waveform smoothness, and root mean square variation rate were designed to comprehensively characterize the global and local characteristics of fault signals. Finally, the Bayes-XGBoost model was adopted for multi-feature fusion classification, and Bayesian optimization was utilized to automatically adjust the model hyperparameters. Experimental results indicate that this model performs excellently in the task of distinguishing tree-line grounding faults from other single-phase high-resistance grounding faults.

DOI

10.19781/j.issn.1673-9140.2026.03.010

First Page

99

Last Page

108

Download

Share

COinS