不对称样本下基于支持向量机的变压器故障诊断

为解决基于支持向量机(SVM)的变压器故障诊断中因样本不对称导致诊断准确率降低的问题,提出了一种改进的向上采样策略和SVM结合的方法。首先通过K-近邻算法提取少数类样本数据中的边界数据集并生成新的少数类随机样本,在此基础上向少数类样本中添加人工生成的随机新样本使得两类样本数量达到基本均衡。对比均衡样本和不对称样本下的SVM分类模型的性能,结果表明:该方法能够有效降低SVM分类平面的偏移程度,进一步提高了SVM变压器故障诊断的准确率。

Transformers Fault Diagnosis Based on SVM for Unbalanced Data

In the process of power transformer fault diagnosis based on support vector machine(SVM),unbalanced data makes the classifying hyper plane of SVM shift to the minority samples,which decreases the diagnostic accuracy.This paper proposed a method combining over sampling strategy and SVM to solve the problem.K-nearest neighbor algorithm is used to extract the boundary data sets from the minority samples.On this basis,the new random samples are generated and added into the minority samples to make the two kinds of samples balanced consequently.Comparing the performance of SVM classification model with balanced data sets and unbalanced data sets,the experiment results show that the proposed method can reduce the deviation of SVM classifying hyper plane effectively.