Analysis of Electric Field and Temperature Distributions of Non-Uniformly Contaminated Silicone Composite Insulators Using Deep Learning

Abstract

This study examines the insulation performance of a silicone composite insulator under various contamination conditions. The non-uniform pollution of a silicone composite insulator, operating in a 34.5 kV, 50 Hz power grid under atmospheric conditions, was analyzed using COMSOL Multiphysics. Considering that the conductivity of the contamination layer on the insulator surface affects leakage currents, surface temperature, and electric field distribution, five critical regions of the insulator surface, significant in terms of contamination, were identified, and distinct conductivity levels were assigned to each. Various combinations of these conductivity levels and voltage values were used to calculate the electric field (kV/cm) in the frequency domain and temperature (degrees C) in the time domain. Data sets were generated for all possible combinations at 29 critical points along the leakage distance. The aim was to identify critical conditions for the electric field and temperature, thus providing a closer approximation to actual operating conditions. Using the obtained data, a Deep Neural Network (DNN)-based model was developed to predict the insulator's response under varying contamination, current density, and voltage conditions. The model demonstrated consistent predictions for electric field and temperature values under nonuniform pollution conditions. The predictive performance of the proposed model was validated through comparative analysis with established machine learning techniques, including Support Vector Machine (SVM) and Random Forest (RF). The model demonstrated consistent predictions for electric field and temperature values under nonuniform pollution conditions.