Adaptive Neuro-Fuzzy Inference System Application of Flashover Voltage of High-Voltage Polluted Insulator

The paper proposes an adaptive neuro-fuzzy inference system called ANFIS for predicting the flashover voltage of external insulators. High voltage insulators were the subject of actual testing, which produced a database for the application of artificial intelligence ideas. The experiments were conducted using different concentrations of synthetic pollution (distilled brine), with each concentration denoting the amount of contamination per milliliter of area. The database offered flashover voltage values for various pollution levels and electrical conductivity levels in each isolation zone. Adaptive neuro-fuzzy inference employed a hybrid learning algorithm to determine suitable membership functions, minimizing the root mean square error as the performance criterion. The primary parameters affecting flashover voltage were identified: applied high voltage, conductivity of the artificial impurity, and amount of impurity in the insulation. For both training and test data, precise predictions were obtained by using membership functions in the shape of a triangle with three fuzzy sets. During testing, the technique demonstrated a low mean absolute percentage error (0.027011) and a high coefficient of determination (0.999997049). Comparison with practical tests yielded a root mean square error of 0.0128623, confirming the effectiveness of the Adaptive Neuro-Fuzzy Inference System in estimating the critical flashover voltage for newly designed insulators under different operating conditions.