Study On Denoising And Feature Extraction For Transformer Partial Discharge Signal

Power transformer is one of the main equipments in power system, and its safe and stable operation is very important to the power system. Local defects in transformer insulation structure will lead to partial discharge(PD) when the applied voltage reaches a certain level, and PD will cause further deterioration of the insulation and thus creating a vicious cycle. As PD signal can sensitively reflect the damage extent of the transformer insulation and at the same time there are differences between PD signal characteristics caused by different types of insulation defects, so effective detection and analysis of PD signal is of great significance for the safe operation and targeted repair and maintenance of the transformer.The strong on-site noise suppression and the efficient feature extraction of detection signals for pattern recognition are key parts for the analysis of PD signal. Based on the analysis of the characteristics of transformer PD, the denoising method and the feature extraction method of PD signal are studied. The main work of this paper is as follows:In order to effectively remove electromagnetic interferences in transformer PD signals, this paper proposes applying a new signal decomposition algorithm——variational mode decomposition(VMD) and wavelet to transformer PD signal to suppress the periodical narrow-band interferences and white noise because of the problems of empirical mode decomposition such as modal aliasing. Firstly, VMD can decompose noisy signal into several modes, which are compact around a center pulsation respectively, and the white noise is automatically filtered out in the decomposition process; then, extract modes which contain PD information for reconstruction and finally, remove residual periodical narrow-band noises by wavelet denoising so as to realize interference suppression. The analysis results of simulations and measured signals show that the proposed method can eliminate two kinds of interferences well and retain the PD characteristics, verify that the approach which uses VMD and wavelet to remove the interferences is efficient and thus it is fit for PD signal denoising.In allusion to the defect of traditional statistical spectrum feature extraction of transformer PD pattern recognition such as high dimension and low recognition accuracy, a novel method to extract the feature of PD grayscale image based on gray level co-occurrence matrix(GLCM) and local binary pattern(LBP) is proposed. According to the proposed method, grayscale image is transformed to GLCM to obtain 8 features of GLCM from a macro perspective and relative grayscale response of neighbor pixels is calculated based on LBP to obtain 10 features of LBP from a micro perspective. PD signals of four experimental models are collected by using pulse current method, combining two kinds of features, support vector machine is used as the classifier to recognize four PD types, and one traditional feature extraction method is used for comparison. The results show that the proposed method can overcome the defect of high dimension and also have a high recognition accuracy, effectively identify the four PD models, and verify that the proposed method is effective.