Transformer winding looseness diagnosis method based on multiple feature extraction and sparrow search algorithm optimized XGBoost

In order to solve the problem of overlap and insufficient anti-interference ability under different load conditions in diagnosing transformer winding looseness using a single feature quantity, a vibration signal diagnosis method for transformer winding looseness based on kernel principal component analysis (KPCA) and extreme gradient boosting (XGBoost) optimized by improved sparrow search algorithm (SSA) was proposed. Firstly, feature quantities in vibration signals were extracted from three dimensions: time domain, frequency domain, and entropy; Then, the feature quantity was dimensionally reduced through grid search optimized KPCA; Finally, a fault diagnosis model based on XGBoost was constructed and sparrow search algorithm was used to optimize the parameters for achieving accurate identification of transformer winding looseness faults under different currents. The experimental verification was conducted on a 110 kV transformer. The diagnosis results show that the extracted feature quantities can accurately reflect the fault characteristics, have stronger anti-interference ability, and the diagnostic accuracy rate of the diagnostic model is 99. 00% . Compared with other diagnostic algorithms, the accuracy and stability are higher, and have good recognition effects under different load conditions. © 2024 Editorial Department of Electric Machines and Control. All rights reserved.