Transformer is a key equipment in power system

and partial discharge (PD) is not only the main cause of insulation degradation of transformer

but also an important manifestation of insulation degradation. As for the complex noise information in high frequency PD signal

a kind of improved noise reduction method based on mutual information and kurtosis value of the parameterless empirical wavelet transform is proposed. The method is based on the empirical wavelet transform (EWT) decomposition

the Gaussian kernel function is constructed to smooth the spectrum to generate the scale space

the k-means clustering is used to screen the boundary points of frequency band

then

the mutual information and kurtosis values of the mode components decomposed by EWT of the non-parametric scale space for merging and optimization

the redundant spectrum dividing points and repartitioning the spectrum are eliminated. Moreover

the valuable component which is obtained by the division of new spectrum boundary. The noise components are removed from the valuable components and the remaining components are subjected to generic threshold noise reduction and then the signal is reconstructed to achieve the PD signal noise reduction effect. Personnal involvement in the method of this paper is not needed and the mode aliasing problem is solved. The simulations

laboratory and field tests show that this method

compared with existing wavelet transform noise reduction and empirical modal decomposition based noise reduction methods

is more effective in suppressing the noise information in PD signals.