Diagnosis of Power Transformer Oil Using KNN and Nave Bayes Classifiers

摘要

In this paper, the K nearest neighbor (KNN) and Naïve Bayes classifiers are used to diagnose the power transformer oil insulation utilizing dissolved gas analysis (DGA). The diagnosis accuracy remains an important factor. Indeed, to obtain the maximum accuracy rate, different vectors are used as inputs for both classifiers. These vectors consist in DGA in ppm, DGA in percentage, Duval triangle reports, Rogers ratios and Dornrenberg ratios. Otherwise, PD, DI, D2, TI&T2 and T3 are the classes of faults adopted as output vector of the classifiers. A related database and the IEC TC 10 one have been used to verify the accuracy rate of both classifiers. Using Duval triangle reports, the KNN algorithm provides the highest accuracy rate than the Naïve Bayes algorithm one.