Abstract

Oil-paper insulation is considered the most popular method for insulating the windings within power transformers because of its ability to withstand high electrical and thermal stress. During the life time of power transformers, oil-paper insulation is aged, and hydrocarbon gases are released. Dissolved gas analysis (DGA) is globally recommended as a reliable diagnostic technique used to identify the fault inside these transformers. Several methods have been previously presented to recognize these faults depending on the data obtained from DGA. This paper proposes a new model that converts Duval pentagon diagnostic graph that defines fault zones using a set of linear inequalities to a numerical table, enabling a fully numerical, automated, and simplified fault identification with 100 % accuracy of traditional Duval pentagon 1, 2 and combined. A comprehensive dataset was generated using the proposed model, covering a wide range of possible gas concentration scenarios. Based on this dataset, a smart diagnostic model employing a Gaussian Support Vector Machine (SVM) was developed to predict transformer faults directly from dissolved gas ratios without the need for manual centroid calculations with up to 98.1 % accuracy of Duval Pentagon according to 20 % test dataset. The proposed model is validated through real case studies, achieving identical results of Duval Pentagon. Comparative analysis against conventional DGA methods and other AI-based techniques demonstrates the superior performance of the presented model in terms of accuracy, simplicity, and practicality.

Key words: — Dissolved Gas Analysis (DGA), Duval Pentagon Modification, Gaussian Support Vector Machine (SVM), Machine Learning Algorithms (MLA), and Smart Diagnostic.