Research Of Power Transformer Fault Diagnosis System Based On Rough Set And Genetic Support Vector Machine

Power transformer is the key equipment in power system,whose normal operation guarantees that the whole power system operates securitily,stably,reliably.With the rise of the voltage level and the increase of the capacity, the probability of transformer fault is becoming increasingly high.Therefore, there is a very important practical significance to carry out the study about diagnosis of the transformer fault.It helps detect early transformer latent fault and repair timely.Thus it improves power system supply reliability rate.Dissolved gas analysis (DGA) is an effective method of discovery and diagnosis of transformer fault. Method of discovery and diagnosis of transformer fault based on DGA also transits from the traditional ratio method to the intelligent diagnosis method.But most of the intelligent diagnosis method is difficult to get ideal diagnostic results because of the small samples and incomplete information of samples.Aiming at this problem,this paper will be able to deal effectively with incomplete data and reduction of data based on Rough Set theory(RS).Also it explores a new kind of intelligent transformer fault diagnosis method research which has excellent performance in combining the Support Vector Mechanism theory (SVM) with application to fault diagnosis of transformer and introduces genetic algorithm to parameter optimization of support vector machine.Firstly this paper puts forward the necessary of artificial intelligent fault diagnosis,which based on the study of transformer fault diagnosis with the analysis of dissolved gas in oil and analysis of traditional ratio method’s advantages and disadvantages.Then it studies rough set theory and related concepts, rough set of discrete methods for continuous data and the decision table attribute reduction method.This paper equally frequency discretes and Genetic algorithm reduces the ratio property of original fault sample with the Rosetta software according to rough set data processing steps,then gets the final decision table which has been attributely reduced and rule merged.The decision table acts as the new input of the fault diagnosis model in this paper. On the establishment of support vector machine classifier, this paper carry on an in-depth study about related support vector machine classification method, support vector machine kernel function selection and support vector machine parameter optimization and so on. Genetic algorithm is introduced for parameter optimization on the problem of parameter optimization of the greatest impact on the support vector machine classifiers,then genetic support vector machine parameter optimization process is obtained.Finally,this paper choose the final decision table which has been dealed with the rough set sample as input,and the type of transformer fault diagnosis as output,use one-on-one multi-classification method, the RBF kernel function and genetic algorithm optimization parameters,finally programs the model of transformer fault diagnosis which based on rough set and genetic support vector machine with the use of LIBSVM toolbox in MATLAB2011. The chosen algorithm superiority has been verified after many experiments and analysis. Compared to the conventional modified three ratio method and BP neural,the chosen algorithm’s superiority can be verified.The method of fault diagnosis proved to be effective when the model is used in examples. So it has certain application value.