Application of Machine Learning for Predictive Maintenance in Power Transformer Health Assessment:

A Comparative Study of Support Vector Machine, Artificial Neural Network, and Random Forest

Abstract

Electricity plays an irreplaceable role in human daily life. To meet the increasing demand for electrical energy, a reliable electrical system, such as a power transformer, is required. Power transformers hold a crucial role in the electrical power system, where the long-term reliability of the transformer is closely related to the safety and stability of the power system. Therefore, transformer maintenance must be carried out to anticipate sudden failures and ensure the overall reliability of the electrical power system. These assessments can be performed in various ways, including the Health Index and Dissolved Gas Analysis. The Duval Pentagon Method (DPM) and Duval Triangle Method (DTM) are used in Dissolved Gas Analysis to ascertain the condition of transformers. In this development, a comparison of three machine learning models—SVM, ANN, and Random Forest—was made using the DPM and DTM datasets to obtain the model with the highest accuracy. The confusion matrix was applied to each DTM and DPM method with several split ratios for training and testing sets. The splits included 90:10, 80:20, 75:25, and 60:40. The model with the highest accuracy will be implemented in a transformer maintenance information system to determine the transformer's condition. The results of the Health Index and Dissolved Gas Analysis calculations can determine the appropriate recommendations for power transformer actions.