Studying Intelligent Techniques Acting in Large Power Transformer Monitoring

Abstract

The presented development is an intelligent diagnostic system for transformers that studied machine learning techniques to determine the operational status of these transformers. The study of these techniques is initiated by observing the quantities that define the operational behavior of large transformers, aiming to identify anomalies in their operation from data from sensors that equipment it in the functioning environment. This large power transformer has a theoretical service life of above 20 years and a low failure rate. Thus, obtaining failure values, which have their evolution monitored for large transformers, is almost nil. Therefore, a supervised machine training methodology to diagnose these cases is practically unfeasible. The study carried out with several traditional intelligent techniques can verify this. Several supervised methods (Closest Neighbor K-th Neighbor, Support Vector Machine, Radial Base Function, Decision Trees, Random Forest, Neural Network, AdaBoost, Gaussian Naive Bayes, and Quadratic Discriminant Analysis) were studied.