基于深度置信网络的电压暂降特征提取及源辨识方法  

Deep Belief Network Based Method for Feature Extraction and Source Identification of Voltage Sag

作　　者：李丹奇 梅飞[2] 张宸宇 沙浩源 郑建勇[1] 李陶然 LI Danqi;MEI Fei;ZHANG Chenyu;SHA Haoyuan;ZHENG Jianyong;LI Taoran(School of Electrical Engineering,Southeast University,Nanjing 210096,China;College of Energy and Electrical Engineering,Hohai University,Nanjing 211100,China;Electric Power Research Institute of State Grid Jiangsu Electric Power Co.,Ltd.,Nanjing 211103,China)

机构地区：[1]东南大学电气工程学院,江苏省南京市210096 [2]河海大学能源与电气学院,江苏省南京市211100 [3]国网江苏省电力有限公司电力科学研究院,江苏省南京市211103

出　　处：《电力系统自动化》2020年第4期150-158,共9页Automation of Electric Power Systems

基　　金：江苏省重点研发计划资助项目（BE2017030）~~

摘　　要：提出一种基于深度置信网络(DBN)的电压暂降特征提取与暂降源辨识方法,利用DBN的特征提取能力对实测波形数据进行特征自提取,解决了人工提取特征过度依赖专家经验,受未知特征影响较大不具备一般性的问题。采用多隐层结构网络学习特征最终实现暂降源辨识。该模型集特征提取器与分类器于一体,优化了模型结构框架,提高了暂降源辨识效率。对模型最优参数进行选择,建立适用于电压暂降实测数据类型的DBN模型,对电网实测暂降数据进行特征提取与暂降源辨识,通过对比验证了DBN方法在特征提取与暂降源识别上的优越性,适用于实际工程。A method for voltage sag feature extraction and sag source identification based on deep belief network(DBN)is proposed.The feature extraction ability of DBN is used to extract the feature data from the measured waveform data,which solves the problem that the artificial features rely too much on expert experiences and are short of generality.The multi-hidden layer structure network learning feature is used to realize the identification of the sag source.The model integrates feature extractor and classifier together to optimize the model structure and improve the efficiency of sag source identification.The optimal parameters of the model are selected,and DBN model for the measured voltage sag data is established.The feature extraction and sag source identification of the real-time measured sag data of power grid are carried out.The superiority of DBN method in feature extraction and sag source identification is verified,which is available for practical engineering.

关 键 词：电压暂降 深度学习 深度置信网络 特征提取 暂降源辨识 

分 类 号：TM714.2[电气工程—电力系统及自动化]