Insulation Life Loss Prediction of an Oil-Filled Power Transformer Using Adaptive Neuro-Fuzzy Inference System

Abstract

In this work, the performance accuracy of the Adaptive Neuro-Fuzzy Inference System (ANFIS) in short-term prediction of insulation life loss is evaluated. A 50 Hz, Dyn11, 1000 kVA 11/0.4 kV oil-filled indoor power transformer, feeding an essential facility with mostly nonlinear and seasonally changing loads, is used. The 1735 Fluke power logger unit and the Fluke 59 mini-infrared thermometer are respectively used for total load current and ambient temperature recordings. The ANFIS, such as implemented in MatLab R2019b software package, is invoked to perform 24-hour computation and subsequently predict the status of insulation life for 7 consecutive days based on 24-hour measurement of the load current, ambient temperature and the hottest-spot temperature. Results show a MAPE of 6.51% for this technique in short-term prediction of insulation life loss of an oil-filled power transformer.