The Pattern Recognition And Transmission Characteristic Of Power Transformer Partial Discharge Signals

Partial discharge is not only an important inducing factor of insulation fault for power transformers,but also a highly sensitive form of insulation fault.Therefore,the researches of PD signal detection,PD fault mode recognition and fault location have been paid much attention.The experimental models of four typical PD faults which are air gap discharge,corona discharge in air,surface discharge in oil and surface discharge in air are made,building an experimental platform,extracting original signals of typical PD faults and studying the pattern recognition of PD faults.Firstly,because partial discharge signals are non-stationary,nonlinear and easy to be interfered,a new method of partial discharge threshold denoising based on complete empirical mode decomposition and ensemble empirical mode decomposition is proposed with combining empirical mode decomposition.Making use of the new method and the conventional wavelet threshold denoising algorithm to denoise for partial discharge signals,the results show that the new method achieves better denoising effects,which verifies effectiveness of the new method and lays a foundation for signal feature extraction and pattern recognition.Secondly,partial discharge types of different insulation faults are recognized based on autoregressive coefficient features.According to PRPD,the 36 dimensional statistical feature parameters are extracted from two dimensional spectrums,three-dimensional spectrums and grayscales.And then making use of principal component factor method,the 36 dimensional feature parameters are reduced to 8 dimensional feature parameters which have physical meanings.In order to ensure feature information's primitiveness,this paper attempts to extract feature parameters by using autoregressive model algorithm.Comparing recognition results of two kinds of features by using BP neural network and SVM,it is found that the recognition rates of autoregressive coefficient features are higher than that of statistical features,which indicates that the autoregressive coefficient features are superior to the statistical features.Thirdly,partial discharge types of different insulation faults are recognized through hypersphere support vector machine.The recognition rates of BP and SVM are lower as a result of complex PD signals,various fault types,limited partial discharge samples and nonlinear feature parameters.In this paper,based on autoregressive coefficient features,the partial discharge types are recognized and analysed through making use of hypersphere support vector machine which is optimized by particle swarm algorithm.The recognition rate of optimized hypersphere support vector machine is higher,which has guiding significances to improve pattern recognition rates of partial discharge.Finally,the XFDTD which is based on finite difference time domain method is used to study the propagation characteristic of electromagnetic waves in transformer windings.In this paper,the transformer windings are constructed by XFDTD and the PD signals are simulated by pulse signal source,setting the imaginary monitoring points to measure the electromagnetic wave's attenuation.The results show that the amplitude attenuation rates are above 65% when the electromagnetic waves pass through the windings;The amplitude attenuation degree of electromagnetic waves is more serious and the time delay differences of electromagnetic wave propagation become larger when the widths of transformer windings become larger.