AC Surface Flashover Voltage Prediction of Epoxy Resin by Tree-Based Model

AC surface flashover is a prevalent insulation fault in gas-insulated equipment. Given the complex dependency of this fault on various factors and the limitations of destructive experimental methods, this paper proposes the use of tree-based machine learning models as an economical and efficient alternative to obtain the flashover voltage. The analysis concentrates on salient features impacting AC surface flashover, such as electroluminescence (EL), water contact angle (WCA), and percentage of silicon atoms (PSA). The extra trees algorithm was employed and optimized by feature selection. The model, utilizing six features—EL, WCA, PSA, Polished status, Ra, and surface resistivity—yielded excellent performance with MAE, MSE, MAPE, and R2 scores of 0.50, 0.37, 1.47%, and 0.93, respectively. The study highlights the potential of tree-based models in providing insights and prediction of surface flashover voltage for the insulation in gas-insulated equipment.