典型电极短空气间隙的击穿电压混合预测

Hybrid Prediction of the Breakdown Voltages of Short Air Gaps with Typical Electrodes

空气间隙的击穿电压与放电起始之前的电场特征存在多维非线性关系。采用支持向量分类机（support vector classifier,SVC）建立了预测模型,以28个电场特征量作为SVC的输入,以间隙在加载电压下耐受（-1）或击穿（1）作为SVC的输出,基于正交设计原理选取9个样本对SVC模型进行训练,成功实现了球隙、棒–板和棒–棒短空气间隙的工频击穿电压混合预测,72个测试样本预测结果的平均绝对百分比误差仅为2.39%;对球–板、棒–球和大球–小球间隙的击穿电压分别进行预测,预测值与试验值基本相符,平均绝对百分比误差分别为1.75%、2.78%和4.57%,验证了该预测方法的有效性和泛化性能。

There is a multidimensional nonlinear relationship between the breakdown voltage of air gap and the electric field features before discharge inception. We established a prediction model by support vector classifier（SVC）. Taking 28 electric field features as the input of the SVC, and the output parameters as-1 and 1, respectively,we characterized whether the gap would withstand or breakdown under an applied voltage.Selecting 9 samples by the orthogonal design principle,we used the samples to train the SVC model.The hybrid prediction of the power frequency breakdown voltages of short air gaps including sphere-sphere, rod-plane and rod-rod geometries was successfully realized. The mean absolute percentage error（MAPE） of the predicted results of 72 test samples is only 2.39%. The power frequency breakdown voltages of sphere-plane, rod-sphere and big sphere-small sphere gaps were predicted. The predicted results are basically in accordance with the experimental values and the MAPEs are respectively 1.75%, 2.78%, and 4.57%, which verifies the validity and generalization performance of the prediction method.