Research On Fault Diagnosis Of Power Transformer Based On Dissolved Gases Analysis And Support Vector Machine

Power transformer is the key large equipment in the power transmission process.It plays an important role in the whole electric power system, and especially theoperation of oil-immersed power transformer is related to the power system security,reliability and economy. While the performance of the transformer insulation is thekey factors affecting its working life most of time, so the transformer’s repair workhas become an important task in power sector. All the time since, whether after therepair or prophylactic repair, these can not meet the repair requirements according totheir own fault characteristics. State maintenance based on online conditionmonitoring and fault diagnosis can solve problems of" lack of maintenance " and"excessive maintenance" well according to the characteristics of transformer faults.In the insulation fault diagnosis of transformer, dissolved gas analysis (DGA)has been recognized as one of the most effective methods. Whether foreign IEC threeratio methods or Chinese highly modified three ratio methods analyze and diagnosetransformer are based on dissolved gas in oil of transformer as the research object.Therefore, this paper research transformer uses the data of dissolved gas in oil as theoriginal sample of the model.In view of the traditional ratio coding method has problem of ratio of encodedcombinatorial is deficiency and the coding boundary is fuzzy, this paper uses supportvector machine (SVM) to classify transformer fault. Support vector machine is akind of machine learning method based on statistics theory, and it can maketwo-dimensional space linearly non-separable samples map into the high dimensionspace into a linearly separable through the introduce of kernel function. This paperuses above idea for the diagnosis of internal insulation fault of oil-immersedtransformer. According to the classification problem, this paper combined the data ofdissolved gas in oil with support vector machine and uses "one to one" algorithm toestablish the transformer fault diagnosis model based on DGA-SVM. This paper usesLibsvm toolbox to build the simulation model which is based on MATLAB, and theresults show that use the support vector machine into oil immersed transformer faultdiagnosis has good generalization ability and application prospect.In order to improve the fault diagnosis accuracy of classifier of support vectormachine further, this paper uses the K-CV method to find SVM penalty parameter C and parameter of kernel function g for optimal according to the idea of crossvalidation. This paper also uses the optimal parameters which are found tooptimization the model. The MATLAB simulation results show that the modeloptimized through optimization parameter reaches the intended purpose that improvethe accuracy of fault diagnosis effectively.