Study On Aging Stage Recognition Methods Of Oil-paper Insulation Based On Cylinder-plate Model Discharge Features

Power transformers are the most important equipment in the field of transmissionand distribution network, there is no doubt that the security and reliability of power gridoperation will be significantly improved if the insulation aging status of transformerscan be exactly evaluated. Partial discharge (PD) is an inevitable phenomenon occurredinside transformers due to various kinds of latent defects, it has been proved thatcharacteristic parameters associated with PD can be used as effective evaluationindicators for the aging condition assessment of transformer insulation. Furthermore,PD monitoring technology has the advantages of non-destructive and on-linemeasurement, which make its application prospect broad. However, up to now, researchon aging condition assessment of oil-impregnated transformers based on PD analysis isstill relatively infrequent and along with some shortage, such as the oneness of defecttypes, model sizes, test voltages and assessment methods, which precisely is the focusof this paper.In this paper, a130℃accelerated thermal aging test of oil-paper insulation underlaboratory conditions is carried out, and PD signals of surface discharge models intransformer oil with two kinds of cylinder electrode diameters at different aging stagesand test voltages are collected. After that, the influence of aging degree, electrodediameters and test voltages on surface discharge characteristics is analyzed, the agingstages of oil-paper insulation are recognized by PD grayscale images and (2D)2PCAalgorithm. Finally, a two-order feature selection strategy combining MRMR algorithmwith SVM-RFE algorithm (MRMR+SVM-RFE) is proposed to further improve theaging recognition performance. The main achievements made in the paper are asfollows:①The variation trend of aging degree, electrode diameters and test voltages onsurface discharge characteristics is found. As the aging degree of oil-paper insulationbecomes deeper, discharge phase moves towards zero-crossing points of the appliedvoltage firstly, then shrinks and moves back apparently; The maximum PD charge doesnot show a monotonous growth trend. On the other hand, the impact of electrodediameters and test voltages on PD characteristics is less than that of aging degree.②A new method for distinguishing the aging stages of oil-paper insulation basedon PD grayscale images and (2D)2PCA algorithm is proposed, and a highest recognition accuracy of79.53%is achieved. Under the same parameter settings of classifiers, thehighest recognition accuracy based on27-dimensional statistical characteristics andPCFA algorithm is75.58%, which proves the effectiveness of the proposed aging stagerecognition method.③A “MRMR+SVM-RFE” combined feature selection strategy is put forward toimprove the aging recognition performance of oil-paper insulation. Comparativeanalysis of aging recognition performance under four cases, namely without featureselection, feature selection by MRMR, feature selection by SVM-RFE and featureselection by “MRMR+SVM-RFE”, is implemented. Classification results show that theaging stage recognition of oil-paper insulation based on “MRMR+SVM-RFE” achievesthe highest recognition accuracy of85.58%, which verifies the effectiveness of theproposed feature selection method.