Classification of Fault and Stray Gassing in Transformer Oil using SVM, NB and KNN Algorithms

Power transformer is one of the most crucial components in the power system network. A major fault on the transformer can disrupt the power supply, thus causing substantial losses. The dissolved gas analysis (DGA) is used to detect incipient fault based on the transformer oil. However, stray gassing of oil could give false indication to the result. This paper aims to develop a model for considering the results obtained from DGA to investigate transformer oil fault condition. Machine learning (ML) algorithms which are Naïve Bayes (NB), support vector machine (SVM) and K-nearest neighbour (KNN) are trained to classify the DGA data into three categories; not determined (N/D), fault, and stray gassing. The algorithms achieved an accuracy of 93.0%, 95.4% and 97.7% respectively. Overall, the algorithms’ performance was tested and verified using various user-input data, where correct classification was achieved successfully.