摘要
局部放电模式识别的输入特征量选择是非常关键的步骤。针对油纸绝缘中5种典型局部放电类型,从其相间局部放电(PRPD)谱图中提取出31个统计算子。分别运用K-W检验、类内类间距离比、顺序前进法以及遗传算法等4种方法对这些算子进行了选择优化。分别用这些选取的特征量组合作为输入向量,通过BP神经网络这个统一的模式识别技术来比较研究这4种特征选择方法,结果表明,顺序前进法和遗传算法由于考虑了特征量之间的相关性,所选择的特征量优于另外2种方法。
Feature selection is a key step in partial discharge （PD） pattern recognition of. In this paper, five PD defect models are established according to the common PD defects in oil immersed transformer. PD signals of the models are collected under differen! experiment conditions and 31 statistical operators are extracted from PRPD pattern. K-W test, ratio between distance in class and out of class, sequential forward selection and genetic algorithm are used for feature selection. Selected features are used as the input vector for BP neural network, and the four feature selection methods are compared by the recognition results. The result shows that the feature＇s selected by sequential fnrward selection and genetic algorithm which consider the relevance among features are better than those selected by the other two methods.
出处
《陕西电力》
2011年第11期1-4,9,共5页
Shanxi Electric Power
基金
国家自然科学基金资助项目(50877064)
关键词
局部放电
模式识别
特征选择
BP神经网络
partial discharge
pattern recognition
feature selection
BP neural network
