作　　者： ; ; ; ; ; ;

机构地区： 重庆大学电气工程学院输配电装备及系统安全与新技术国家重点实验室

出　　处： 《电网技术》 2012年第8期31-37,共7页

摘　　要： 雷电及操作过电压的识别,对改进和提高电力系统绝缘配合水平具有重要意义。提出了一种基于时频矩阵奇异值分解(singular value decomposition,SVD)和多级支持向量机(support vector machine,SVM)的雷电及操作过电压识别方法,通过对过电压信号的小波分解,构建多尺度时频矩阵,利用SVD对该矩阵进行奇异值分解,将信号分解到不同的时频特征子空间,然后获取过电压信号的奇异谱,并计算奇异谱的特征量,将这些特征量作为多级SVM的输入,实现雷电及操作过电压的辨识。对变电站实测5种过电压信号的计算表明:提取的特征量维数低,对过电压信号的电磁干扰具有相对稳定性;采用的识别方法训练次数少,识别率高,能够实现雷电及操作过电压的准确分类。 It is significant for improving and enhancing insulation coordination level of power system to identify lightning and switching surges. Based on singular value decomposition （SVD） of time-frequency matrix and support vector machine （SVM）, a method to identify lightning and switching surges is proposed. Through applying wavelet decomposition to voltage signals a multi-scale time-frequency matrix is constructed, then this time-frequency matrix is decomposed by SVD and the signal components are decomposed into different time-frequency characteristic subspaces to attain singular spectrum of voltage signal, and characteristic quantities of singular spectrum are calculated and taken them as the input vectors of multilevel SVM to identify lightning and switching surges. Calculation results of five overvoltage signals measured in a certain substation show that the dimensionality of the extracted characteristic quantities is low and extracted characteristic quantities possess relative stability to electromagnetic disturbances in overvoltage signals; less training samples are needed for the proposed identification method that exists high recognition rate. Thus, using the proposed method, lightning and switching surges can be classified accurately.

关 键 词： 雷电过电压 操作过电压 特征提取 奇异值分解 过电压识别 多级支持向量机

领　　域： [电气工程]