Comparative Evaluation of Artificial Neural Networks and Monte Carlo Simulation for Transformer Insulating Oil Lifetime Prediction

Abstract

Transformer insulating oil is an important factor for the reliability and service life of power transformers. It provides electrical insulation and heat dissipation. Accurate lifetime prediction is essential for asset management and condition-based maintenance. In this study, the comparison of Artificial Neural Network (ANN) and Monte Carlo Simulation (MCS) techniques is presented to predict transformer insulating oil lifetime based on three physicochemical parameters such as acid content, moisture content and breakdown voltage. The model was developed and validated on an experimental dataset of 18 transformer insulating oil samples. The ANN model was based on a multilayer perceptron architecture with three hidden layers (80-80-40 neurons). The MCS model was run for 3000 simulation iterations to include the input uncertainty. The model performance was assessed using mean Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE) and coefficient of determination (R²). The ANN model produced better results with a MAPE of 8.20%, an RMSE of 4.15 months and an R² of 0.98, surpassing the MCS model, which achieved a MAPE of 11.50%, an RMSE of 9.40 months and an R² of 0.89. The results presented show that ANN is a more accurate and reliable methodology for the prediction of transformer insulating oil lifetime that allows efficient condition-based maintenance and transformer asset management.