Bayesian Network and Compact Genetic Algorithm Approach for Classifying Partial Discharges in Power Transformers.

This paper presents a statistical learning method capable of classifying the incidence level of partial discharges in power transformers. By using the results from acoustic emission measurements, it is possible to detect the presence of partial discharges inside the equipment, allowing the qualitative health monitoring of the transformer’s insulation. Therefore, the use of a Bayesian Network is proposed, combined with a Compact Genetic Algorithm tailored for solving mixed integer programming problems, for discretization of the continuous metrics extracted from acoustic emission measurement. Comparing the results with Multilayer Perceptron Neural Network and Decision Tree and after a suitable amount of runs of the algorithm, it was verified that the Bayesian Networks presented superior results. [ABSTRACT FROM AUTHOR]