Sound Diagnosis Method of Power Transformer Discharge Fault Based on CEEMDAN-SVM

Abstract

The sound signal characteristics of the power transformer discharge fault is the key to reliable identification of the power transformer discharge fault. However, the sound signal of the power transformer discharge fault is non-stationary and susceptible to environmental interference. Therefore, this paper proposed a sound diagnosis method for the power transformer discharge fault based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and support vector machine (SVM). First, the CEEMDAN was used to decompose the power transformer's sound signal, thus a series of intrinsic mode functions (IMF) that reflect the sound signal's local properties could be obtained. The kurtosis of each IMF was solved to select the suitable IMF components for signal reconstruction and denoising. Secondly, the reconstructed signal is decomposed by CEEMDAN, and the singular spectral entropy and marginal spectral entropy are extracted to form the eigenvector. Finally, the discharge fault of the power transformer is classified and identified by SVM. The simulation results demonstrate that the proposed method can obtain a recognition rate of more than 80% of discharge fault when the power transformer noise interference is taken into account and that it may be utilized to identify and diagnose the power transformer discharge defects.