Metodologia para Estimar a Vida Útil de Para-raios de Óxido de Zinco Baseada na Aplicação do Sistema Neuro-Fuzzy Adaptativo
Keywords: useful life estimation, artificial intelligence, forecasting models, time series, metal oxide surge arresters
AbstractThis work presents a methodology to be used to estimate the Metal Oxide Surge Arresters’ (MOSAs’) useful life, based on the prediction model based on the Adaptive Neuro-Fuzzy Inference System (ANFIS). For this purpose, a database consisting of MOSA’s total leakage current was used. Thus, it was possible to build a time series composed of values of the third harmonic component of the leakage current. This component is one of the indicators of the MOSAs’ degradation level most used in the surge arresters monitoring. Subsequently, the forecasting models based on artificial intelligence (ANFIS and Support Vector Regression - SVR) were implemented, evaluated, and compared. During implementations, the ANFIS model was tested with three distinct membership functions: Gaussian, Generalized Bell-Shaped, and Pi-Shaped; and the SVR model was tested with three different kernel functions: Gaussian, Linear, and Polynomial. The performance of each implemented model was evaluated using the determination coefficient in the training phase of the models and the mean absolute percentage error, in the validation phase. Considering the results obtained by the ANFIS e SVR models, it was found that the forecasts made using the ANFIS model with Gaussian membership function were the most accurate forecasts. Thus, the proposed methodology, which applies the ANFIS model with Gaussian function, was used to estimate the MOSA’s useful life.