An Intelligent System for Furfural Estimation in the Power Transformers

Power transformers are supposed to be an expensive and critical component of a power system and so its schedule maintenance is an important aspect near the utilities. The cellulose paper used as the solid insulating material of the transformer deteriorates regularly due to progressive aging. As a result, it produces several degradation by-products of cellulose insulation into the transformer oil. Furfurals are among the major by-product of cellulose and are exploited to estimate the physical state of the transformer’s dielectric and the electrical insulation directly and noninvasively. In the present work, an intelligent system is proposed and developed that predicts the level of furfural in the transformer oil. The system makes predictions using easily quantifiable parameters, enabling utilities to avoid suffering financial losses. The proposed system employs the Adaptive Neuro Fuzzy Inference System (ANFIS) technique with temperature and moisture as the input and 2-Furfuraldehyde (2-FAL) as the output. An optimum arrangement of the developed model to find accurate outputs has been achieved by the suitable selection of number and types of the membership functions of the network. Also, an exhaustive study is made to optimize the training data points to achieve the accurate predictions at minimum iterations. The proposed system is validated with nine test conditions of temperature and moisture, and the output thus obtained is verified through existing life models in the literature.