Partial Discharge Feature Extraction Based On Synchronous Extrusion Fourier Transform And Multiscale Dispersion Entropy

Partial discharge is a concomitant phenomenon of insulation faults in electrical equipment,and contains various types of information on the types of insulation faults.Since different partial discharge signals correspond to different insulation faults of the equipment,the identification of partial discharge types has important guiding significance for accurately grasping the insulation state of the transformer and rationally arranging maintenance.The key to the identification of discharge type is the extraction of discharge characteristics.In view of the current poor stability of partial discharge signature recognition and low recognition rate,this paper studies the feature extraction method of partial discharge signal based on multi-scale analysis and entropy information.The main contents are as follows:A feature extraction method based on synchronous squeeze window Fourier transform(WFSST)and multi-scale spread entropy(MDE)is proposed(WFSST-MDE).Firstly,the synchronous discharge window Fourier transform algorithm is used to decompose the partial discharge signals of four typical insulation faults collected under laboratory conditions,and the intrinsic mode-like function components(IMTFs)of the partial discharge signals are obtained.Then,for each type of partial discharge signal,the number K of modal components suitable for type identification is selected by the center frequency method,and the multi-scale scatter entropies of the K modal components are respectively calculated and combined into original feature quantities.On this basis,the original feature quantity is optimally reduced by the maximum correlation minimum redundancy criterion,and finally the classification is implemented by the support vector machine(SVM)classifier.The experimental results show that the WFSST-MDE method can extract the characteristics of the partial discharge signal well regardless of noise interference.The extracted features can accurately describe the fluctuation warning and complexity difference of different types of discharge signals,and the MDE method.Compared with the empirical mode decomposition(EMD)MDE method(EMD-MDE),the WFSST-MDE method has higher recognition rate and better noise robustness.In order to further improve the spectrum analysis ability of the synchronous extrusion window Fourier transform on the partial discharge signal and reduce thespectral aliasing and spectrum leakage during modal component extraction,a WFSST method based on Kaiser self-convolution window(KCWFSST)is proposed.And combined with multi-scale dispersion entropy,a feature extraction method based on KCWFSST-MDE for partial discharge signal is constructed.The Kaiser self-convoluting window has excellent sidelobe performance,which can effectively suppress the influence of spectral leakage and modal aliasing on modal component extraction,so that the MDE of the modal component can better reflect the fluctuation characteristics and time-frequency complexity of the PD signal.The experimental results show that the recognition accuracy of KCWFSST-MDE algorithm based on Kaiser self-convolution window is higher than WFSST-MDE algorithm.