Prediction of discharge voltage for rod-plane gap based on regularized IRM-NN and its generalization analysis.

The discharge characteristics of long air gaps, as an important basis for the design of external insulation in ultra-high voltage transmission and transformation projects, have always been of great concern to researchers. The discharge characteristics of the rod-plane air gap, as a typical problem, are also the research topic of many researchers. At present, although there are many methods for calculating and predicting the discharge voltage of rod-plane gap, a considerable number of methods cannot guarantee their generalization ability when facing large-scale temperature and humidity changes. Therefore, in order to achieve accurate calculation of discharge voltage under extreme temperature and humidity conditions, this paper proposes a method based on regularization invariant risk minimization neural network (IRM-NN) to predict the discharge voltage of rod-plane gap. The influencing factors of rod-plane gap discharge were systematically analyzed, and key feature quantities were extracted as input for training the model. The average absolute percentage error of the model on the test set is only 2.59%, which verifies that the model can effectively extrapolate to application scenarios outside of the experimental conditions. The proposed method for calculating the 50% discharge voltage between the rod-plane gap can adapt to a wide range of temperature, humidity, and certain electrode size changes, providing reference for the study of discharge characteristics in long air gap.