中国电机工程学报
中國電機工程學報
중국전궤공정학보
ZHONGGUO DIANJI GONGCHENG XUEBAO
2014年
34期
6228-6237
,共10页
陈奎%韦晓广%陈景波%牛俊萍
陳奎%韋曉廣%陳景波%牛俊萍
진규%위효엄%진경파%우준평
小电流接地故障%样本数据%经验模态分解%故障测度%信息增益度%主成分分析法%SMOTE%ADABOOST
小電流接地故障%樣本數據%經驗模態分解%故障測度%信息增益度%主成分分析法%SMOTE%ADABOOST
소전류접지고장%양본수거%경험모태분해%고장측도%신식증익도%주성분분석법%SMOTE%ADABOOST
small current grounding%sampled data%empirical mode decomposition (EMD)%fault measure%information gain degree%personal computer assistant (PCA)%SMOTE%ADABOOST
针对零序暂态分量的特点以及现有的信息融合技术在小电流接地故障选线中具有样本数据不均衡、维数灾难和经验风险高的缺陷,分析选线样本的特性,提出基于样本数据处理和ADABOOST法的小电流接地故障选线的新方法。首先,通过经验模态分解和快速傅里叶变换对零序信号进行故障特征提取,然后利用故障特征建立线路故障测度和利用信息增益度建立方法故障测度,进一步通过主成分分析法对故障特征样本进行降维处理以及利用 SMOTE 采样法处理样本的不均衡性,最后将处理后的数据运用ADABOOST进行综合选线。通过系统模型仿真,验证了主成分分析法和SMOTE 采样法对样本数据处理的合理性以及利用ADABOOST选线的有效性,结果表明所提方法应用于选线具有较高的准确率和灵敏度。
針對零序暫態分量的特點以及現有的信息融閤技術在小電流接地故障選線中具有樣本數據不均衡、維數災難和經驗風險高的缺陷,分析選線樣本的特性,提齣基于樣本數據處理和ADABOOST法的小電流接地故障選線的新方法。首先,通過經驗模態分解和快速傅裏葉變換對零序信號進行故障特徵提取,然後利用故障特徵建立線路故障測度和利用信息增益度建立方法故障測度,進一步通過主成分分析法對故障特徵樣本進行降維處理以及利用 SMOTE 採樣法處理樣本的不均衡性,最後將處理後的數據運用ADABOOST進行綜閤選線。通過繫統模型倣真,驗證瞭主成分分析法和SMOTE 採樣法對樣本數據處理的閤理性以及利用ADABOOST選線的有效性,結果錶明所提方法應用于選線具有較高的準確率和靈敏度。
침대령서잠태분량적특점이급현유적신식융합기술재소전류접지고장선선중구유양본수거불균형、유수재난화경험풍험고적결함,분석선선양본적특성,제출기우양본수거처리화ADABOOST법적소전류접지고장선선적신방법。수선,통과경험모태분해화쾌속부리협변환대령서신호진행고장특정제취,연후이용고장특정건립선로고장측도화이용신식증익도건립방법고장측도,진일보통과주성분분석법대고장특정양본진행강유처리이급이용 SMOTE 채양법처리양본적불균형성,최후장처리후적수거운용ADABOOST진행종합선선。통과계통모형방진,험증료주성분분석법화SMOTE 채양법대양본수거처리적합이성이급이용ADABOOST선선적유효성,결과표명소제방법응용우선선구유교고적준학솔화령민도。
According to the characteristics of the zero sequence transient component and the defects that existing information fusion technology had the unbalanced sampled data, dimension disaster and great empirical risk applied to fault line detection for small current grounding system, the characteristics of the sampled data at fault line selection were analyzed and the method based on sampled data processing and ADABOOST was proposed. The fault feature extracted from the zero sequence component by using empirical mode decomposition and fast Fourier transform at first. And then the line fault measure was established by the fault feature and the method fault measure was established by the information gain degree. Thirdly, personal computer assistant was used for the sampled data of the fault feature to reduce dimension and SMOTE treated the unbalanced sampled data. Finally, ADABOOST was applied to global diagnosis by data of dimension reduction. Personal computer assistant and SMOTE to treat the sampled data is reasonable and ADABOOST to global diagnosis is effective through simulation system. The results showed that the method that applied to fault line detection is accurate and sensitive.