摘要 本文通过GIS实验装置平台设置的4种缺陷类型的放电情况,用特高频（UHF）检测法获取的PD信号,分别以时域、频域等17组统计参数作为PD类型的特征量,通过局部线性嵌入（locally linear embedding,LLE）方法将17组特征量进行降维处理,得到9组有效特征参数。并用多分类相关向量机（M-RVM）作为识别方法,在110kV电压下获取的实验检测数据作为训练和预测样本,结果取得86%的理想识别率,验证了LLE与M-RVM结合的识别系统的有效性。
Abstract：In this paper, the PD signal obtained by UHF detection method is used as the characteristic quantity of PD defect types, through the four kinds of defect types arranged by the GIS experiment platform, 17 groups of feature quantities are selected by time domain and frequency domain, etc. Then, useingLocally linear embedding (LLE) method to reduce the dimensionality of 17 sets of feature groups, and nine effective characteristic parameters are obtained. (M-RVM) was used as the recognition method. The experimental data obtained at 110kV voltage were used as training and prediction samples, The results show that above 86% recognition rate is achieved, the validity of the identification system combined with LLE and M-RVM was verified.
2017, Vol. 18 
