EMD-SVD及粒子群优化的SVM变压器局部放电模式识别

The Pattern Recognition of Partial Discharge Based on EMD-SVD and PSO Optimizing Parameters of SVM

为了对变压器的局部放电信号进行特征量提取以及模式识别,在分析EMD和SVD理论的基础上,提出了EMD-SVD和PSO-SVM相结合的方法。将选取的四种去噪后的局部放电信号(空气中电晕、沿面、气隙,油中气隙)经EMD分解为由高到低的固有模态函数,再利用SVD对其进行数据压缩,提取出14个反应PD信号本质的特征量,并将其输入到经粒子群优化的支持向量机进行模式识别。仿真结果表明,此方法能够较好地识别出四种局部放电信号,与未经优化的SVM、GA-SVM、GRID-SVM相比,经粒子群优化的支持向量机分类准确率较高、速度较快。

In order to extract and distinguish the pattern for the partial discharge signals of transformer, after analyzing thoretical of EMD and SVD, the method of combining between EMD-SVD and PSO-SVM is proposed. The four selected denoising partial discharge signals（ corona,surface,cavity in air, cavity in oil）decomposes into intrinsic mode functions from high to low through EMD, using its SVD realizes data compression, thereby extracting the feature quantity of fourteen reactive nature of the PD signal. At the same time,pattern discrimination by entering into the PSO optimizing parameters of SVM. The result show that, this method can identify the four kinds of partial discharge signals preferably, and compared with non-optimized SVM, GA-SVM, GRID-SVM, it has higher classification accuracy rate and faster by PSO optimizing parameters of SVM.