Research On Transformer Faults Diagnosis Based On Clustering And Fuzzy Support Vector Machine

As the very expensive hub equipment in power system, a transformer plays an important role in the safety and reliability of power supply. And how to detect and identify transformer fault has been an important topic of electrical sector. Up to now, Oil Dissolved Gas Analysis (DGA) is an effective method to discover and identify transformer fault. The essence of DGA aims at finding an objective and accurate description of the relationship between the types of transformer faults and the feature information of dissolved gas. In this paper, based on the theories of support vector machines and fuzzy support vector machines, a hierarchical incremental algorithm of clustering fuzzy support vector machines that can help transformer faults diagnosis has been proposed, and its effectiveness has also been proved by experimental results.The main work has been done as follows:i. According to the relationship between dissolved gases in transformer oil and transformer fault, a transformer fault diagnosis model and its solution procedure are proposed derived from multi-class SVM theory.Based on the concept of feature extraction in pattern recognition, a hierarchical structure is employed to extract the input features closely related to the model of classification, and it has effectively suppressed the interference of redundant information. By comparison of the diagnosis results, the best extracting model is selected. ii. The adaptive parameter optimization algorithm has both increased the flexibility of parameter selection for SVM and enhanced the convergence speed.iii. On the basis of supporting the SVM transformer fault diagnosis model, the FSVM transformer fault diagnosis model has been established, with a fuzzy membership degree determined according to the Euclidean Distance between the sample data and its corresponding cluster center. This method has depressed the influence of noises and outlier.iv. With the incremental learning algorithm, sample information has been rationally and effectively added into the FSVM diagnosis model, so as to effectively make up for the shortcomings of off-line learning of the FSVM model.