The main source of transformer vibration is the core and winding.The failure of the core and the winding can change the vibration signal of the transformer,and the fault type can cause different signal changes.Based on the self-designed transformer vibration signal acquisition system,the transformer vibration signal was collected.Due to much of transformer vibration signal noise,matching pursuit based on genetic optimization algorithm was used to get rid of the noise,which is called signal sparse representation.The different fault types after the noise signal was analysized and the fault diagnosis method was finally determined.Researches show that,the base frequency of the vibration of the box can reflect the fault of the core,and the proportion of the high frequency component can reflect the fault of the winding.Once the core has some trouble,the fundamental frequency components can change obviously,and the energy of the main frequency amplitude can be calculated by using the FFT.When the winding doesn't work well a large amount of high frequency component willproduce.Using the wavelet-high and low frequency energy algorithm to calculate the spectral envelope of high and low frequency,we can compare the ratio of the low frequency component with total energy to determine winding fault block off.