Study On Variational Mode Decomposition And VPMCD Method For Pattern Recognition Of Partial Discharge In Power Transformers

A power transformer is a kind of pivotal equipment in an electrical network.Its serviceability has a significant influence on the safe operation of electrical network.The condition of its electrical insulation system determines the reliability of the power transformer.Partial discharge is one of the important signs and manifestation of insulation degradation in a power transformer.In order to effectively analyze partial discharge signals,time-frequency analysis method,feature extraction method and pattern recognition method for partial discharge have been conducted in this paper.The main contents of this research are as follows.A novel time-frequency analysis method based on variational mode decomposition(VMD)and Wigner-Ville distribution(VMD-WVD)is proposed.Firstly,the known partial discharge signals are decomposed by VMD and several band-limited intrinsic mode functions(BLIMFs)are extracted.Then these Wigner-Ville distributions of BLIMFs are computed and the Wigner-Ville distribution of each BLIMF is added linearly to reconstruct the Wigner-Ville distribution of the original signal.The VMD-WVD method can effectively restrain the cross terms of Wigner-Ville distribution,and guarantee a high energy aggregation and good time-frequency resolution.The analyses of simulation and experimental signals show that the proposed method can effectively analyze the time frequency characteristics of partial discharge signals.A novel method based on VMD and Hilbert transform(Hilbert-VMD)is proposed for feature extraction of partial discharge signals.Firstly,the known partial discharge signals are decomposed by VMD and several BLIMFs are extracted.Then each BLIMF is processed by Hilbert transform and the marginal spectrum of each BLIMF is calculated.Finally,the features of partial discharge signals in the frequency domain are extracted based on the marginal spectrum of each BLIMF.This method is used to extract the features of partial discharge signals in laboratory environment.The experimental results show that the proposed method can effectively extract the frequency-domain characteristics of partial discharge signals and the features extracted by the proposed method have a higher correct recognition rate.An improved VPMCD based on kernel partial least squares(KPLS)regression,which is called as KPLS-VPMCD,is proposed for pattern recognition of partial discharge signals.The kernel function is introduced in KPLS-VPMCD method,and the KPLS regression is used to replace the ordinary least squares(LS)regression in the original VPMCD method.It can effectively tackle non-linear relations without using any given mathematical model.Moreover,KPLS-VPMCD is still useful when the high multi-collinearity exists among the variables of samples or the number of training samples is small.Experiments and analyses have been done using both UCI standard datasets and the extracted features of partial discharge signals.The experimental results validate the effectiveness of the proposed method for pattern recognition.An unknown partial discharge signal recognition method based on sample-weighted FCM clustering is proposed to realize the identification of unknown samples in partial discharge signal to be identified.Firstly,the existing known partial discharge signals are clustered by FCM to determine the clustering center of the known classes.Then,the sample weights of known partial discharge signals and partial discharge signals to be identified are calculated respectively.Finally,the weights of the partial discharge signals to be identified are compared with the threshold to determine whether they belong to the known class.In this method,an adaptive threshold of sample weight is determined by using Otsu criterion.The effectiveness of the proposed method is validated by case analysis.In addition,the proposed method also provides a new analysis method for the identification of unknown partial discharge signals.