基于奇异值分解的局部放电模式识别方法

Pattern Recognition for Partial Discharging Using Singular Value Decomposition

提出了一种基于奇异值分解的变压器局部放电模式识别方法。通过搭建人工缺陷实验环境并采集样本数据,计算每个样本的统计特征参数,构成实验数据的样本矩阵。对样本矩阵进行奇异值分解,判断保留矩阵的特征是否明显,确定最佳保留矩阵的阶数,从而得到降维后的类型特征空间描述矩阵和类中心描述向量组。对现场采集的样本数据进行计算得到待分类的样本向量,并用类型特征空间描述矩阵进行线性变换,然后计算变换后的向量与类中心向量组中每个向量的距离,从而得到分类的判断结果。该算法简单而且高效,能够实现局部放电检测中各种放电信号的有效区分,局部放电模式识别召回率约为91.3%。

A pattern recognition method based on singular value decomposition（SVD） for partial discharge in transformers is proposed. By setting up an experimental environment with artificial defects and calculating the statistical parameters from the data obtained from each sample, the sample matrix is constructed. SVD is then carried out for the sample matrix. After dimensional reduction by decomposing the matrix, the best order for the remained matrix is judged by the singular value. Then, the low-dimensional description matrix of feature space and the class-center vectors are obtained. The classified sample vector which is acquired on Site is formulated by linear transforms of the description matrix. The result of classification is gotten by calculating the distances between the transformed vector and the class-center vector. The proposed method is simple and efficient. It has the ability to recognize effectively various signals of partial discharge. The experiments show that the recall rate of partial discharge is about 91.3%.