Redução de Dimensionalidade para Diagnóstico de Falhas em Transformadores de Potência

Power substations face high operating costs for implementing fault monitoring in power transformers, which requires monitoring equipment installation and signal processing. This work investigates two approaches to alleviate such costs in the context of gas-in-oil-based fault diagnosis: (i) modern dimensionality reduction techniques or (ii) brute-force reduction using fewer gas sensors. Both approaches incorporate machine learning architectures and are explored here for a minimum (unity) output dimension. In addition to being less costly, a brute-force solution with the exclusive use of hydrogen gas proves superior for most performance metrics.