Assessment of the health condition of oil-immersed transformers using logistic regression and poisson distribution

In this paper, a method is proposed to assess the health condition of oil-immersed transformers using a logistic regression analysis and Poisson distribution. The parameters of the regression model are estimated by using a maximum-likelihood cost function developed based on Poisson distribution for modeling transformer health condition that include oil breakdown voltage (BDV), acidity of oil, 2-Furaldehyde content, water content in oil, dissolved combustible gas (DCG), and dissipation factor (DF). The minimization of the criterion function is carried out by the Gradient Descent technique. The values of the health indices of 30 working power transformers below 110 kV are obtained from the regression model. Using preset ranges of health indices, the transformers are classified into four different health conditions. It is shown that the results are close to expert evaluations and on par with recently reported ones in the literature, thus signifying a reliable and effective assessment of transformer health.