Resumen

FREIRE, A.S.; ASTUDILLO, J.C.; QUINATOA, C.I.  y  ARIAS, F.R.. Interpretation of Gases Dissolved in Dielectric Oil Using Random Forests for the Detection of Anomalies in Power Transformers. Revista Técnica energía [online]. 2023, vol.19, n.2, pp.90-98. ISSN 2602-8492.  https://doi.org/10.37116/revistaenergia.v19.n2.2023.544.

The following paper presents a machine learning tool for the interpretation of anomalies in power transformers using the random forest method. Using the results of gas chromatography tests on dielectric oil from several published papers, the data set delivered by the dissolved gas analysis (DGA) in quantities of parts per million (ppm), the amount of hydrocarbon gases such as hydrogen (H2), methane (CH4), ethane (C2H6), ethylene (C2H4) and acetylene (C2H2) that serve to diagnose the internal state of the transformer is used. Due to the reduced number of collected data, there is a disadvantage to apply artificial neural networks, support vector machine, among others that need large amounts of data for each variable, but satisfactorily they are solved using random forests, because this methodology classifies better the data of smaller amount. The learning obtained by training is validated with the states obtained by the test data under IEC 60599 and IEEE C57-104, which encompass 4 diagnostics such as high energy discharge, low energy discharge, normal state and overheating, resulting in a final corroborative validation criterion for the algorithm by comparing the diagnostic results with the random forests.

        · resumen en Español     · texto en Español     · Español ( pdf )