Classification and Prediction of Incipient Faults in Transformer Oil by Supervised Machine Learning using Decision Tree
Raymon Antony Raj, D Sarathkumar, Sampath Kumar Venkatachary, Leo John Baptist Andrews
Abstract
Transformer condition evaluation is a crucial technique for displaying a transformer’s health index. Key gases (KG), which dissolve in oil and become inseparable, are developed because of the different nascent flaws formed in the transformer enclosure. The exact defect that caused the gases to evolve may be found using the composition of KG present in the transformer. Five KG, including hydrogen H <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> , methane (CH <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">4</inf> ), ethane (C <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> H <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> ), ethylene C <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> H <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">4</inf> ), and acetylene (C <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> H <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> ), are employed in this study as an input to anticipate the real incipient defect. In the inquiry that follows, classification is carried out using a Decision Tree (DT) and the data is trained using a Supervised Machine Learning (SML). The dataset is made up of 180 different input KG compositions and defects that are trained, identified, and predicted using a predictive model. 16 more datasets with KG composition are utilized for prediction. TPR versus FNR, PPV vs FDR, AUC, parallel coordinates plot, confusion matrix, and lowest classification error are used to assess the statistical correctness. The accuracy of the projected data using DT and optimized DT is 62.9%, however the accuracy of the optimized DT with principal component analysis (PCA) is only 55%. Therefore, the suggested DT fault classification approach may be utilized to accurately forecast the onset of faults.