PCA和KICA特征提取的变压器故障诊断模型

Transformer Fault Diagnosis Model Based on PCA and KICA Feature Extraction

为了充分利用主元分析(PCA)和核独立主元分析(KICA)特征提取的互补性,提高变压器故障分类正确率,提出了基于PCA和KICA特征提取的变压器故障诊断模型。该模型中,首先,将油中溶解气体分析(DGA)测试样本投影到PCA空间中进行特征提取,采用多核支持向量机(MKSVM)作为分类器进行预分类,采用核密度估计方法估计阈值将测试样本预分类为易识别或难识别样本;对难分类样本则再次投影到KICA空间,采用另一MKSVM作为分类器进行分类识别,实现PCA和KICA双空间特征提取算法;最后,根据故障特征,建立变压器故障诊断模型。实验结果表明,所提出的双空间算法对变压器故障的识别率达到88.61%,比单空间算法和IEC3比值法的识别率分别高10%和24%。

In order to take full advantage of mutual complementary between principal component analysis （PCA） and kernel independent component analysis （KICA） to enhance the accuracy of classifying transformer faults, we proposed a novel diagnosis model of transformer faults bas.ed on PCA and KICA dual-space feature extraction algorithm. In this model, firstly, a dissolved gas analysis （DGA） test sample is projected to PCA subspace, while a multiple-kernel support vector machine （MKSVM） is adopted as classifier to predict the sample＇s class label. The test sample is pre-classified into a difficult one or an easy one through comparing the prediction result with a threshold obtained by kernel density estima- tion method. If it is a difficult sample, it would be re-projected to KICA subspace where another MKSVM is used to identify the sample＇s class label. Hence, a feature extraction algorithm in PCA and KICA dual-space is achieved. Finally, a multilayer diagnosis model is set up according to the fault characteristics of transformer. Experimental results show that the proposed method has accuracy in classification of 88.61%, which is 10% and 24% higher than that of single subspace algorithm and IEC three-ratio method, respectively.