Research Of Transformer Remote Faults Diagnosis Based On Fuzzy Support Vector Machine

Large power transformer, as the key equipment of power system, plays influential role to ensure the safe operation of power system. Real-time detecting the insulation state of transformer, accurately predicting the fault and avoiding possible trouble are important measures to ensure the safe operation of the power system, to improve equipment utilization and reduce costs of equipment maintenance. In this thesis, an advanced algorithm-fuzzy support vector machine (FSVM) is used to study the fault diagnosis of transformer which is based on Dissolved Gas Analysis (DGA) method and the exploration and research of related content of fuzzy support vector machine algorithm is made, so the recognition rate of transformer fault diagnosis is improved. A remote transformer fault diagnosis system which was based on B/S model was also developed, in which the technology of the fault diagnosis, Internet, Matlab Web Server was utilized.The main contents and conclusions of this article as follows:(1) The common faults of power transformers are analyzed in this thesis, the common methods of transformer fault diagnosis are described.Though many solutions have been applied to the diagnosis of transfomers'failures,they still have many flaws. A fault diagnosis method based on fuzzy support vector machine is proposed.(2) The study of a multi-classification based on fuzzy support vector machineThe Multi-classifier is formed by constituting the multiple binary classifiers for support vector machines which is evolved from the binary classifier. It focus on several multi-classification which is commonly used.And their advantages and disadvantages were analyzed in detail.(3) The study of determining the fuzzy membershipThe performance of fuzzy vector machines is significantly influenced by fuzzy membership, so it focus on the methods of calculating a more reasonable sample fuzzy membership which leads a better classification performance of the fuzzy support vector machines.(4) The study of parameters optimization algorithm for fuzzy support vector machineThe classfication performance of an FSVM model is largely dependent on the selection of the model parameters. Two algorithms of parameters optimization-the grid search and the genetic algorithm-was compared, and the appropriate parameters optimization method for fuzzy support vector machine model was selected.(5) According to DGA methods, a transformer fault diagnosis model is proposed derived form multi-classification fuzzy support vector machine theory.MATLAB and other tools were used to constructe multi-classification model based on fuzzy support vector machine which been introduced into the transformer fault diagnosis. The simulation experiments show that the FSVM classifier constructed by this thesis have higher testing accuracy than the traditional SVM multi-class algorithms and back-propagation neural network.(6) The web interface of Internet-based transformer remote fault diagnosis system was compiled.The overall constitution and function of transformer fault diagnosis system are studied.The project of applying Matlab web Serve to resolve the problem of is put forward.Transformer fault diagnosis system has been developed based on Matlab Web Serve.Such functions as remote datatransmitting,data analyzing, remote diagnosing,result exporting on web are realized in this system.The experiment of transformer remote fault diagnosis system was carried out, the results showed that transformer fault diagnosis system based on FSVM is feasible and accurate.