Abstract
Accurately estimating the degree of polymerization (DP) of distribution transformers is important for assessing the condition and performance of these important electrical devices. This study presents a comprehensive analysis of various predictive models used to estimate DP, including the proposed regression model and developed artificial neural network (ANN), compared with other models and a set and actual value of measured DP of distribution transformer. This assessment uses key performance metrics including Mean Absolute Deviation (MAD), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). The proposed regression model emerged as a robust predictor, with a MAD of 0.176, MSE of 0.051, RMSE of 0.225, and MAPE of 2.78% with MAD of 0.176, MSE of 0.051, RMSE of 0.225 and MAPE of 2.78%. Notably, this model closely matches the accuracy of the "DP Measure" benchmark, which has a MAD of 0.193, an MSE of 0.052, an RMSE of 0.227, and a MAPE of 2.94%. Although the ANN model performed quite well, it showed a slightly higher degree of error, with a MAD of 0.348, an MSE of 0.176, an RMSE of 0.419 and a MAPE of 5.53%. This higher error degree is caused by aging factors of distribution transformer.
Comparative analysis with existing models shows that the regression model proposed outperforms these methods, thus demonstrating its effectiveness. Effective in estimating DP and consequently the remnant lifespan of the transformer. The outcomes show that the suggested regression model and actual measurements provide valuable tools for transformer maintenance and condition monitoring, helping to prevent costly failures and prolonging time. Further, this study highlights the importance of accurate PD estimation in distribution transformers and provides the basis for the practical application of predictive models, providing insight into performance and reliability of them. Furthermore, it encourages future studies to explore the applicability of these models for different types of transformers and operating scenarios, thereby helping to improve reliability and durability of the electrical system.
Keywords: Degree of Polymerization, Distribution Transformers, Predictive Models, Regression Analysis, Artificial Neural Network, Statistical Tools.