| Power transformer is an important equipment in power system.Its reliable operation directly affects the security of power system.However,due to various reasons,the internal insulation of transformer will be damaged and partial discharge will occur.In order to ensure the safe operation of transformer,it is particularly important to identify and evaluate the partial discharge in transformer through certain technology.At present,there are many methods to detect partial discharge.On the basis of ultrasonic detection technology,the time-frequency analysis of partial discharge ultrasonic signal is carried out,and the selection,processing and extraction of characteristic parameters of ultrasonic signal are studied.Firstly,three kinds of partial discharge models are designed according to the common defects in transformer,which are suspended discharge model,surface discharge model and needle-plate discharge model.Then,an experimental platform is built in the partial discharge Laboratory of Shanghai Academy of Electrical Sciences.Finally,a partial discharge experiment is carried out.The traditional double exponential oscillation simulation function of partial discharge signal is improved to make the simulated ultrasonic signal closer to the actual signal.Then the denoising effect of several commonly used wavelet functions is compared,and the wavelet function with good denoising effect is selected to denoise the ultrasonic signal.In the aspect of extracting characteristic parameters of ultrasonic signal waveform,the method of improving statistical accuracy of discharge times is studied.Starting with two parameters,parameter interval(HDT)and ringing count,the influence of these two parameters on discharge times statistics is studied.The accuracy of discharge statistics is improved by setting these two parameters,so as to improve the validity.of parameters extraction.Fast Fourier Transform(FFT)is used to analyze the differences of bandwidth and peak spectrum of three kinds of discharge.In the aspect of envelope analysis of ultrasonic signal,several different methods of envelope drawing are studied.By comparing and analyzing the effects of various methods,the extremum method with time window is put forward to get envelope based on extremum method,and the appropriate width of time window is put forward to make the envelope drawing smoother.Then the influence of different interpolation methods on envelope is studied.The advantages and disadvantages of pchip interpolation and spline interpolation are mainly compared.According to the characteristics of both methods,we use spline interpolation method to analyze the amplitude-frequency characteristics of envelope and use pchip interpolation method to extracting the characteristic parameters of envelope.Finally,based on the envelope analysis method,the non-linear time-varying high-frequency ultrasonic signal is transformed into smooth low-frequency signal.The amplitude-frequency characteristics of the three discharge envelopes are analyzed by FFT,compared with the spectrum of the original signal before,the envelope spectrum of the three discharge types can be distinguished better.Based on empirical mode decomposition(EMD),three different discharge ultrasonic signals are decomposed.By analyzing the decomposed IMF components,the characteristics of three different discharges in decomposition layers,IMF amplitude and energy are studied.According to the energy distribution,energy entropy is introduced to characterize the energy distribution.Finally,five features,i.e.decomposition layer n,maximum amplitude of IMF Amax,layer of maximum amplitude of IMF mA,layer of maximum energy of IMF mE and energy entropy Hen are proposed as the characteristic parameters of partial discharge pattern recognition.Finally,five feature parameters are extracted to form feature vectors,and pattern recognition is carried out by BP neural network.The recognition results prove the validity of the five feature parameters selection. |
PDF Full Text Request