基于主动学习SVM的变压器局部放电模式识别  

Partial discharge pattern recognition in power transformer based on active learning SVM

作　　者：尚海昆[1] 徐扬[1] 苑津莎[1] 李洪强[2] 

机构地区：[1]华北电力大学电气与电子工程学院,河北保定071003 [2]中国电子科技集团公司第五十四所,河北石家庄050081

出　　处：《华北电力大学学报（自然科学版）》2013年第4期27-31,106,共6页Journal of North China Electric Power University：Natural Science Edition

基　　金：国家自然科学基金资助项目(61204027);中央高校基本科研业务费专项资金资助(13XS26)

摘　　要：对局部放电类型进行有效识别是准确评估变压器绝缘状态的重要前提。然而放电类型识别过程中往往需要大量训练样本的学习才能达到较高的识别精度。针对这种情况,提出一种基于主动学习支持向量机的局部放电模式识别新方法。该方法在选择训练样本的过程中采用主动学习的方法,挑选出对分类器最有价值的放电样本进行训练。研究结果表明,该方法在保障识别精度的前提下有效减少了放电信号需要的训练样本个数,缩短了训练时间,提高了样本的学习效率。Effective discrimination between different partial discharge （PD） types is an important premise for correct condition evaluation of insulation in power transformers. In the process of PD recognition, a large number of training samples are required for attaining higher recognition accuracy. A novel method based on active learning Support Vector Machine is proposed for PD recognition aiming at the current problem. The most valuable PD samples for SVM are ex- tracted with active learning method. Results demonstrate that, the proposed method reduced the number of PD training samples and shortened the training time, which improved the learning efficiency of PD samples.

关 键 词：SVM 主动学习 变压器 局部放电 识别 

分 类 号：TM83[电气工程—高电压与绝缘技术]