摘要
在交联聚乙烯(XLPE)电力电缆的局部放电模式识别的研究中,为解决标注样本数量过少而导致识别率低下的问题,引入基于半监督学习的方法进行电缆局部放电模式识别研究。制作了XLPE电缆的四类典型绝缘缺陷,从局部放电信号中提取20种统计特征参数,对基于半监督学习的一致性模型方法与基于有监督学习的J48,k近邻,BP神经网络等方法进行了对比,并采用主成分分析进行优化。研究结果表明半监督学习能充分利用已标注样本的特征信息和未标注样本的分布信息,增强分类器的性能,提高局部放电模式识别的准确率。而通过主成分分析的方法能降低样本特征维数,有效提高半监督学习算法速度。
In the research of pattern recognition on partial discharge （PD） in XLPE cable,insufficient labelled data may lead to low recognition rate.To solve the problem consistency model （CM） based on semi supervised learning（SSL）theory is introduced.Twenty types of statistical characteristics are extracted from PD pulse sequence from four typical models of insulation defects in a XLPE power cable.A comparison between CM based on SSL and supervised methods（J48,k Nearest Neighbor and BP Neural Network）is conducted,and CM is optimized using principal component analysis.The comparison result shows that CM method takes full advantage of both diversified characteristic information from manually labelled data and distribution information from unlabelled data to enhance the performance of the classifier and improve the recognition rate efficiently.Principal component analysis method can reduce the characteristic dimension of samples and speed up the algorithm of semi supervised learning.
出处
《电力系统保护与控制》
EI
CSCD
北大核心
2011年第14期40-46,共7页
Power System Protection and Control
关键词
XLPE电缆
局部放电
半监督学习
模式识别
主成分分析
XLPE cable
partial discharge
semi supervised learning
pattern recognition
principal component analysis
