Identification and Classification of Transformer Faults Using a Feature Engineering Based Machine Learning Model

Power transformers are vital pieces of equipment in the power grid. Repairing or replacing power transformers is expensive. Due to this, monitoring of transformers is crucial to track and observe the quantities of interest in a transformer. The Dissolved Gas Analysis method analyzes the gases that are produced based on the decomposition of oil during the use of power transformers. By measuring the levels of gases dissolved in the oil, one can monitor the health of the transformer. This paper presents a Muli-Layer Perceptron model that utilizes backpropagation for Condition Monitoring of power transformers using gas data. Furthermore, Feature Engineering is implemented along with a Synthetic Minority Oversampling Technique to create an algorithm that enables accurate classification with unbalanced data between classes