Transformer Fault Diagnosis Based On Support Vector Machine Optimized By Simulated Annealing

Power transformer is one of the most important equipment in power system, its runningstate directly affects the security and stability of power system. Because of the complexity ofits internal structure in the long-term operation, fault is inevitable. With enhancement ofsociety and economy development for the requirements of good reliability and high securityof power system. Fault diagnosis technology research of power transformer is very importantto improve the operation reliability and scientific management level.Based on discussion of the present situation of fault diagnosis technology with dissolvedgas analysis (DGA) of transformer, the traditional three-ratio method has disadvantages offuzzy ratio boundary, so intelligent fault diagnosis technology has become the trend ofresearch. Therefore, this thesis proposes a method which combined simulated annealing (SA)algorithm and support vector machine (SVM), SA algorithm is used to optimize theparameters of SVM and acquire simulated annealing support vector machine model, denotedas SA-SVM model.Firstly, this thesis studies the transformer fault diagnosis technology based on DGA, anddiscusses the advantages and disadvantages about early ratio methods. The necessity ofartificial intelligent fault diagnosis for transformer has been presented. In order to reflect therelationship fully between the internal fault of transformer and the characteristic of gas, fivekinds of characteristics of gas concentration ratio data amounted to fifteen groups as pre-inputhave been proposed and gene selection recursive feature elimination (RFE) have been used toselect fifteen features which act as the final fault diagnosis model input. In the establishmentof the SVM classifier, the utilization of multi-classifiers method, the choice of kernel functionand the optimization of SVM parameters have been discussed. Aiming at the optimization ofthe most influential parameter to classification effect in the classifiers of SVM, the SAalgorithm is introduced to parameter optimization and the optimization process has beenrealized. Finally, the transformer fault diagnosis method based on RFE-SA-SVM is putforward with the selected feature subsets using RFE as input and the type of transformer faultdiagnosis as output. In order to avoid unreadable programming function handier in Matlab,the GUI interface is finished. Compared with single model, the RFE-SA-SVM model issuperior. Using this model to analyze examples, the effective RFE-SA-SVM model is verified,and this model has the application value.