Distribution Transformer Winding Fault Detection Based on Hybrid Wavelet-CNN

Abstract

This paper presents a new decision logic approach for protecting and distinguishing internal faults in power transformers. This method uses the feature extraction technique based on wavelet transform and artificial neural network. The proposed method is designed based on the difference between the energies of the wavelet transform coefficients produced by short circuit fault currents in a certain frequency band. First, the operation of the transformer under phase-to-ground and phase-to-phase short circuit faults was examined. Then, the resulting secondary winding current was transferred to the discrete wavelet transform. This method is used to analyze components of the signal at different scales. Based on the results, it has been shown that due to the good temporal and frequency characteristics of the wavelet transform, the features extracted by the wavelet transform have more distinct features than those extracted by the fast Fourier transform. As a result, the fault detection process was improved by integrating the obtained wavelet transform values into artificial intelligence models. This step allows the analysis process to be automated and faults to be identified more quickly and accurately. The proposed method is more efficient and faster and achieves a higher success rate than the traditional method.