基于SVM SMOTE的电力变压器故障样本均衡化方法

Equalization Method of Power Transformer Fault Sample Based on SVM SMOTE

在变压器故障诊断领域,数据集不平衡性带来的极端值、噪声等问题严重影响了分类算法的故障识别能力。为此,提出了一种基于支持向量机(supportvectormachine,SVM)合成少数类过采样(syntheticminority over-samplingtechnique,SMOTE)算法的电力变压器故障样本均衡化方法,并结合机器学习进行故障诊断,以解决不平衡数据集下变压器故障诊断整体精度低的问题。首先,从原理、特点、应用等方面对传统SMOTE算法和SVM SMOTE算法进行了大量研究和分析;然后,以变压器油中溶解气体为样本集,构建了基于故障样本均衡化的变压器故障诊断模型;最后,对所提方法进行了算例仿真。结果显示:相较于传统变压器故障诊断算法,采用SVMSMOTE算法对故障样本进行均衡化后,各评价指标均有大幅提升,其中整体分类准确度αmacro-F1提升了18.9%。研究结果证明所提方法可以有效解决不平衡数据集下变压器故障样本漏判率高的问题,并在其分类上具有更高的精度。

In the field of transformer fault diagnosis,the extreme values and noise caused by the imbalance of a data set have seriously affected the fault recognition ability of the classification algorithm.Therefore,a method of power transformer fault sample equalization based on the support vector machine(SVM)synthetic minority over-sampling technique(SMOTE)algorithm is proposed,and machine learning is combined for fault diagnosis to effectively solve the problem of low overall-precision of transformer fault diagnosis under the unbalanced data set.Firstly,the traditional SMOTE algorithm and SVM SMOTE algorithm from the aspects of principle,characteristics,and application were researched and analyzed in detail Then,the gas concentration dissolved in transformer oil was taken as a sample set,and the transformer fault diagnosis model based on fault sample equalization was constructed.Finally,simulation study was carried out.The result shows that,compared with traditional transformer fault diagnosis,the evaluation indexes have been greatly improved after fault sample equalization with SVM SMOTE,and theαmacro-F1,namely,overall accuracy,has increased by nearly 18.9%.The research results prove that the proposed method can be adopted to effectively solve the problem of high missing report ratio of transformer fault sample under unbalanced data set,and outperform traditional classification algorithms with respect to accuracy.