Interpretación de gases disueltos en aceite dieléctrico mediante redes neuronales para la detección de anomalías en transformadores de potencia de la subestación Novacero

The following document presents an automatic learning tool for the interpretation of dissolved gases in power transformers of the Novacero substation, using application algorithms such as neural networks and random forests with Python programming language. Through the results of gas chromatography tests in dielectric oil from various published articles, the data set delivered by the Analysis of Dissolved Gases (AGD) is used in quantities of parts per million (ppm), the amount of hydrocarbon gases as hydrogen (H2), methane (CH4), ethane (C2H6), ethylene (C2H4) and acetylene (C2H2) that serve for learning and diagnosis of failure results. The algorithm implementation process is carried out with 128 training data and 64 test data to verify the proposed learning. The result obtained by the training through the use of automatic learning is validated with the states obtained by the test data and AGD reports provided by the substation, under the IEEE C57.104-2019 standard, the results are analyzed applying the triangle method. of Duval showing four state diagnoses such as high energy discharge, low energy discharge, normal state and overheating, obtaining as a result a corroborative and applicable final validation criterion to interpret the gases dissolved in dielectric oil.

• Main Author: Freire Freire, Armando Salvador (author)
• Format: masterThesis
• Language: spa
• Published: 2023
• Subjects: ANÁLISIS DE GASES DISUELTOS; TRANSFORMADORES DE POTENCIA; APRENDIZAJE AUTOMÁTICO; REDES NEURONALES; ELECTRICIDAD
• Online Access: http://repositorio.utc.edu.ec/handle/27000/10014