Transformer Monitoring using Supervised Learning based Classifiers

Dissolved gas analysis (DGA) functions as a diagnostic technology which helps utilities distinguish transformers running properly from those needing preventive upkeep through analysing formed gases. The main function of DGA enables fault detection through gas analysis of transformer systems. DGA practitioners widely utilize the key gas method (KGM) for transformer health index evaluation through the detection of certain gases and their development patterns. DGA technique alone capable to indicate existing cases. However, the technique is less effective in failure prediction unless combined with prediction techniques. The transformer health index classification process requires multiple supervised prediction models including K-nearest neighbours (KNN), linear discriminant analysis (LDA), quadratic discriminant analysis (QDA) and decision tree algorithms. The evaluation of models takes place through accuracy measurement and precision and recall calculation. The KNN model delivered the best results among the bench of analytics methods with 93.28% accuracy and 94.12% precision and recall at 92.44%.