The analysis of GIS vibration signal can timely and effectively find faults existed in the equipment. Feature extraction is the most important step in the fault diagnosis of GIS equipment. In view of the shortcoming that the feature obtained by single kind of feature extraction method is likely lack of specificity in this paper

the feature extraction method of GIS vibration signal based on composite features is proposed. First

such two kinds of time frequency methods as singular value decomposition and rotation-invariant technique based on total least square criterion are combined to extract the characteristic quantity with definite physical meaning in the vibration signal. The characteristic quantity without definite physical meaning is extracted by variational modal decomposition and multifraction form of the data sequence. Then

the new feature quantity is obtained by combining those two feature quantities and the classification effect of this method is compared with that of the single feature extraction method. The method proposed in this paper has higher recognition rate and more superiority.