•  
  •  
 

Keywords

transformer; fault diagnosis; over-sampling; extreme gradient boosting algorithm; northern goshawk optimization algorithm

Abstract

To enhance the diagnostic accuracy of transformer fault, raise the efficiency of model recognition, and mitigate the adverse effects of imbalanced samples on transformer fault recognition models, a transformer fault pattern recognition model with double-layer extreme gradient boosting (XGBoost) based on borderline synthetic minority over-sampling technique (BSMOTE) and northern goshawk optimization (NGO) algorithms is proposed. Firstly, BSMOTE is used to expand minority class samples, and a balanced dataset is obtained. Secondly, the noncoding ratio method is used to establish multi-dimensional feature quantities, and XGBoost is used to determine the optimal feature subset. Then, an NGO algorithm is used to optimize the XGBoost parameters, and a transformer fault diagnosis model is obtained, achieving accurate recognition of transformer faults. Finally, practical cases are adopted to make a simulation analysis of the proposed method. The diagnostic accuracy of the proposed method is 2.88%, 4.03%, 4.44%, and 7.47% higher than that of the recursive feature elimination (RFE), random forest (RF) feature screening, categorical boosting (CatBoost) feature extraction, and the 19-dimensional feature, respectively. The results show that the method proposed in this article has higher fault recognition accuracy, lower misjudgment rate, and stable performance.

DOI

10.19781/j.issn.1673-9140.2025.06.004

First Page

32

Last Page

42

Download

Share

COinS