A new model of faults classification in power transformers based on data optimization method

The current paper aims to present a hybrid model for classifying faults in power transformers. The innovation of this paper can be shown by introducing a new method of data optimization based on the combination (subsets) of the set (C-set) method and the unsupervised fuzzy C-means (FCM) clustering algorithm. The new method aims to solve the quandaries of unbalanced data, outliers, and boundary ratios that exist in conventional and artificial methods. Considering the compactness of dissolved gas analysis (DGA) data, the C-set method is employed for dividing the data samples according to their fault types into multiple groups without repetition. The FCM algorithm is then adopted for the samples pre-selection process by obtaining the cluster centers of each C-set class. The obtained cluster centers are combined to form the labeled expert training data (LETD) set, which will train the 1-vs-1 multiclass support vector machine (MCSVM) with a linear kernel. The proposed model diagnosis obtained accuracy is 88.9%. Our proposed model has been compared with other soft computing and traditional models. The experimental results of this model revealed high performance in classifying transformer faults and improving the fault identification accuracy.