A Prediction Method for the Surface Arc Inception Voltage of Epoxy Resin Based on an Electric Field Feature Set and GS-SVR

To address the critical challenges posed by the complex coastal climate on the external insulation of electrical equipment, research into the prediction of the surface arc inception voltage of epoxy resin under multiple conditions is of great significance for preventing failures and guiding operations and maintenance. In this regard, we propose a prediction method for surface arc inception voltage based on grid search-optimized support vector regression (GS-SVR). Using a 21-dimensional electric field feature set along the shortest inter-electrode path as model input, high-accuracy prediction of surface arc inception voltage under complex conditions is achieved. The results demonstrate that the model accurately predicts surface arc inception voltage with limited samples, achieving a mean absolute percentage error (MAPE) of 6.24%. Furthermore, the non-uniform coefficient-based dataset partitioning method improves prediction accuracy compared to random partitioning, with the lowest MAPE of only 2.39%. The findings provide theoretical and technical support for improving the anti-pollution flashover and anti-condensation performance of epoxy resin insulating materials.