Bayesian classifier is the tool of increasing the efficiency of defects recognition in power transformers

Abstract

The article considers the method of forming a statistical Bayesian classifier in relation to the problems of operational diagnostics and rapid evaluation of the technical condition of transformer equipment. It is proposed to use the classifier as a regular means to improve the reliability of defect recognition in power oil-filled transformers based on the analysis of dissolved gases in oil. A stochastic approach to the formation of the classifier in a conditions linearly realized dichotomy of technical status classes is developed. As a distinguishing feature, a nonlinear function of the primary parameters of state is used. This simultaneously achieves both a reduction in the dimension of the feature space and an improvement in the characteristics of the random distribution. The proposed approach allows to form a decisive rule that minimizes the total error of decision-making regardless of the impact on the object of random operational factors. The results of the study of stochastic properties of the distributions of the distinguishing feature for each of the selected classes of states are obtained. The algorithm to perform statistical calculations and procedures for recognizing the current state of the transformer using the generated decision rule is designed. The results of the study illustrate the possibility of practical application of the developed approach in the real exploitation of power transformers.