Research On Fault Diagnosis Of Power Transformer Based On PCA-SVM

As a power conversion and electromagnetic switching equipment, power transformer is very important in the power transmission and allocation. Since the complexity and non-linear characteristics of fault types and fault characteristics of power transformer, the mathematical model of fault is difficult to establish. With the characteristic of self-learning and non-linear mapping, a new way is provided to solve the problem of power transformer fault classification and decision by support vector machine. Therefore, combining the dissolved gas analysis (DGA) technology,the principal component analysis (PCA) and support vector machine (SVM) are regarded as the key to study the fault diagnosis of power transformer.Firstly, the relationship between characteristic gas and fault category of power transformer is analyzed in this thesis. In order to further increase the fault information contained in the 6-dimensional data obtained by DGA technology, the 13-dimensional concentration ratio data of characteristic gas is incorporated.Secondly, a complete binary tree is constructed by fuzzy C-means clustering to classify the fault categories. The fuzzy clustering is used to get classified labels for the problem that fuzzy clustering is large amount of calculation and difficult to be classified online, and four SVM sub-classifiers are obtained based on the complete binary tree classification structure. Finally, for the problem that the 19-dimensional data dimension is too high, the principal component analysis is used to reduce the dimensions of the data and generates new comprehensive variables. The number of principal component is determined by a method, which is combined accumulated variance contribution rate with multiple correlation coefficient. The improved grid search (GS) algorithm is used to optimize the penalty coefficient C and kernel parameter g of RBF kernel function for support vector machine. The training speed and classification accuracy of SVM are further improved while preventing the parameter from falling into the local optimum.Through simulation and case diagnosis, it is shown that on the basis of 93.03%retention rate, data mapping from 19-dimensional to 5-dimensional is realized by the principal component analysis method. The strong capability of fault diagnosis above 90% is achieved by PCA-SVM algorithm, which has been arrived the requirement of power transformer fault diagnosis.