Study On Condition Assessment Of Oil-paper Insulation In Power Transformer

Power transformer is one of the key equipment in the whole power system, undertaking the task of the voltage conversion. The safe and reliable operation or not of transformer will directly affect the safety and reliability of the power system. Oil-paper insulation is the main form inside the power transformer, it will age and be failure under the condition of electrical and thermal stress, and may lead to the failure of the transformer. The interruption of the entire grid may occur by the single point failure of the transformer. Therefore, study on the aging process and the health condition assessment of oil-paper insulation is of great significance.The characteristic parameters of oil-paper insulation are analyzed, and the characteristic parameters that can reflect the operation of state comprehensively and accurately are chosen to assess the condition of oil-paper insulation in working power transformer in this paper. Tests of DGA and oil quality are analyzed. Also the accuracies of different types of DGA method in oil paper insulation fault diagnosis are analyzed.Health index (HI) is proposed to evaluate oil-paper insulation condition. The HI1, HI2and HI3of oil-paper insulation are calculated based on the industry standards, utility experts’ judgments and the Duval triangle method. In order to deal with the influence of noise data and error, the fuzzy c-means(FCM) algorithm is used to calculate HI4.These four sets of health indices are combined into one final health index(HIf).Finally, training database is formed with the HIfand test results of DGA and oil quality.In terms of the current main problems condition assessment of oil-paper insulation in power transformer, support vector machine (SVM) and fuzzy support vector machine (FSVM) are introduced on the basis of oil-paper insulation condition assessment. The oil-paper insulation condition assessment models are established based on the LIBSVM. The models are tested by the verification sample and the classification accuracies are analyzed. The class imbalance problem in training dataset is dealt by over-sampling and under-sampling techniques, the classification accuracies of SVM、FSVM are compared before and after implementing processing techniques. In order to choose characteristic parameters that can reflect the condition of the oil-paper insulation best as input of SVM. characteristic parameters are selected by calculating con-elation index. The influence of the number of characteristic parameters on the classification accuracy of SVM is analyzed. At last the BP neural network is introduced to condition assessment of oil-paper insulation, and the accuracies of SVM、FSVM and BP neural network are compared.