Análises de desempenho de transformadores de potência imersos em óleo mineral isolante com aplicações de inteligência artificial

Abstract

The monitoring of the performance index of power system electrical equipment, such as power transformers, is essential to ensure that these assets operate with the expected reliability, thereby guaranteeing the continuous and adequate supply of electric energy to end-users. However, this monitoring process presents considerable challenges, including the scarcity of reliable data, the high variability of operational conditions, and the complexity of the physical degradation phenomena associated with failure occurrence. In this context, the objective, in this work, is to develop and evaluate Artificial Intelligence (AI)–based tools for predicting incipient faults in these assets, enabling optimized maintenance planning. In the methodology, the following developments were carried out: implementation of the Local Outlier Factor (LOF) algorithm for outlier detection across different performance datasets from predictive techniques; particle filters for forecasting the performance index and classifying the operational state of power transformers; and a tool composed of binary classifiers for transformer performance based on dissolved gas analysis (DGA) parameters and physicochemical tests. The results show that these models achieve satisfactory performance in classifying the operational state of the equipment under several evaluated scenarios. Therefore, the proposed approaches represent viable and effective alternatives for monitoring the performance index of high-voltage equipment, contributing to the early identification of faults and to the optimization of maintenance interventions.