Supervised machine learning model for predictive analysis of dielectric response of insulating liquids

Abstract

The present study deals with application of a supervised machine learning (ML) technique to predict and explain the trends in the dielectric properties of the oil samples with change in temperature. The insulation system of the transformer mainly consists of the conventional mineral oil (MO) and the solid insulation like kraft paper and pressboards. High temperature, ageing and oxidation of the oil reduce the lifetime of the insulation. Therefore, it is of utmost importance to carry out periodic monitoring of the insulation system to avert any untoward failures in the power system network. Nanofluid (NF) is evolving as a dielectric liquid for usage in various high voltage apparatus for the purpose of insulation and heat transfer because of its certain advantages. For formulating the NF, semiconductive Titanium oxide (TiO2) nanoparticle (NP) is dispersed into the MO in a specific volume percentage. The study in this work aims at predictive analysis of the dielectric properties like permittivity and dielectric losses of MO and MO-NF considering its dielectric response using the frequency domain spectroscopy (FDS). For the predictive study, the supervised ML model known as decision tree regression is used as it is one of the most powerful tool for prediction. The model is developed using a dataset of 355 experimentally measured values of the dielectric properties with temperature range of 30oC to 90oC. These results indicate the variation in the dielectric properties of both MO and MO-NFs and help to comprehend the changes in the oil properties at a wide range of frequencies.