Machine Learning Model for Transformer Health Monitoring and Fault Detection

Health monitoring and predictive maintenance of electrical equipment may help in the replacement of the equipment before its failure and maintain the continuity of the power to the consumers. Amongst several electrical equipments, transformers are widely used in various transmissions as well as distribution networks. The system operators conduct several tests on a periodic basis to have a real-time track of the working condition of the power transformer. However, these tests rely on human involvement, resulting in increased maintenance time and potential errors caused by variations in the operator’s experience and understanding. Thus various continuous and non-invasive techniques to monitor the health of the equipment are needed. The work undertaken provides insight into the current health condition of the power transformer and foresees the likelihood of various types of faults using machine learning. It also alarms the user about the type of fault that is probably to occur in the near future. The type of fault depends on the ratio of the dissolved gases. Dissolved Gas Analysis along with Rogers Ratio is taken as a predictor. The work undertaken evaluates the health of the oil-immersed transformer and provides a comparative analysis of various machine learning-based classification models namely the logistic regression (log R), k-nearest neighbors (k-NN), XGBoost, Random Forest (RF), and Decision Tree (DT) in predicting the same. The outcomes of the various models confirm the superiority of the Random Forest model in predicting the state of the oil-immersed power transformer with an accuracy of 97.17%.