Dependence Mapping of the Static Electric Field Distribution Characteristics to Dielectric Strength of Long Air Gap Based on Random Forest

To evaluate the dielectric strength of a long air gap under a positive 250/ $2500~\mu $ s impulse waveform, various parameters should be taken into account, among which the electric field (EF) is the most important. This article proposes a random forest (RF) model to construct the dependence mapping from the static EF distribution characteristics of long air gaps to their dielectric strengths. Forty interelectrode features were adopted to describe the EF distribution of an air gap and their correlation degrees with the 50% discharge voltages were analyzed. A high-quality feature subset was input to train the RF model with some limited discharge test data. The dielectric strengths of large-size sphere-plane long air gaps, which are used in valve hall, were forecast by the RF model, and the results show a mean absolute percentage error (MAPE) of only 2.0% for 17 samples. The model was also extrapolated to the grading ring and cap to ground gaps, and the predicted dielectric strengths have similar variation trend with the experimental values. This study can provide references for insulation distance selection of shield fittings in converter station or substations.