Machine Learning-based Regression and Classification Models for Oil Assessment of Power Transformers

Abstract

Expensive and widely used power and distribution transformers need to be monitored to ensure the reliability of the power grid. Evaluating the transformer oil different parameters is vital to determine the transformer insulation health conditions. In this paper, both regression and classification models based on machine learning are used to test the correlation between the interfacial tension values (IFT) of the transformer oil with other oil test results, namely, breakdown voltage, acidity, color, dissipation factor and water content. Experimental results with oil samples obtained for 730 units indicate that both acidity and color have the highest correlation with IFT. Nevertheless, other parameters like breakdown voltage and dielectric dissipation factor contributes marginally in increasing the classifier output accuracy when added to the acidity and color. 2020 IEEE.