Analysis And Recognition Of Vibration Signal Of Power Transformer

Power transformer is an important equipment in the power system,and plays an important role in electrical energy transmission.Therefore,it is necessary to monitor the condition of the power transformer.Among the various state monitoring methods,the vibration method can perceive the rich state information of internal transformer components when electrified.Besides,the collection equipment is simple,safe and economical.This thesis mainly focused on the analysis and identification methods of power transformer vibration signals.Considering the correlation between the vibration signal and the operating condition data,the Lempel-Ziv algorithm was used to calculate the complexity of the transformer vibration signal and quantitatively analyze the observed vibration signals of the transformer under different operating conditions and different years.Combining the vibration mechanism of the transformer,the changing trend of the transformer state was predicted.To deal with the redundant dimensions in the transformer vibration signal,a vibration signal processing method based on the variational modal decomposition algorithm was proposed.The transformer vibration signal was decomposed into several components with a central frequency from low to high.Combining with the Pearson correlation coefficient method,the frequency distribution of the transformer vibration signal was analyzed.The threshold method was used to select highly correlated components to reconstruct signal.Thus,the weakly correlated components were removed while retaining important information in the reconstructed signal,and the stability of the data was enhanced.In the classification and recognition of vibration signals,to overcome the shortcomings of traditional methods,a method combining continuous wavelet transform and convolutional neural network was proposed.The two-dimensional time-frequency map obtained by wavelet transform of vibration signal was taken as the input of the convolutional neural network,which is the state-of-the-art method in image recognition tasks.Moreover,it was verified on the bearing standard data set to confirm the good effect of this method in the classification of vibration data.Based on the Py Charm integrated development environment,a visualization software for transformer vibration signal identification was designed and developed,which can be used to identify the timefrequency map of transformer vibration.