Neural networks and damage pattern recognition in power transformer diagnostics

Abstract

The results of detecting deformations and damage of power transformer windings using the transformer Frequency Response Analysis (SFRA) are presented taking into account RG CIGRE A2.26, Standard IEC 60076-18, Standard IEEE C57.149. The technology of pattern recognition by signal images for diagnostics of winding damage, type of defect, its localization is implemented. Neural networks are used - where in a multilayer perceptron with backpropagation of error - the work of neurons in a hierarchical network is imitated. Parametric methods - such as the Naive Bayes classifier - a probabilistic classifier based on the Bayes formula with the assumption of independence of features among themselves for a given class, which greatly simplifies the classification task due to the assessment of one-dimensional probability densities instead of one multidimensional. An algorithm for recognizing patterns of defects and damage to power transformers using the results of diagnostics by the FRA method has been developed.