基于特征变换的DGA诊断范例推理方法

Pearson相关系数是一种衡量变量间线性关系的方法,广泛用于变压器中油中气体故障诊断(DGA)的范例推理匹配算法。但是,现有方法存在偏袒数据区间较大的特征以及认为所有特征对相关系数判定的贡献相同这两个问题。因此,在深入分析DGA色谱数据的基础上,提出采用对数特征变换方法缩小特征值域来解决偏袒大数据区间特征的问题,采用均方差特征赋权区分特征贡献度的方法进一步提高DGA故障检测效果,并构造了基于特征变换和特征权重的Pearson相关系数DGA诊断(FTW_Pearson)算法。实验结果表明,FTW_Pearson算法的DGA诊断正确率优于业界普遍使用的大卫三角形法、未考虑特征变换和权重的Pearson相关系数法以及贝叶斯算法和神经网络算法。

DGA Fault Diagnosis Based on CBR Method with Feature Transformation

Pearson correlation coefficient is a way to measure the linear relationship between two variables,which is widely used as CBR matching algorithm for DGA fault diagnosis.However,the traditional application has two problems:discriminating in favor of the features which have larger data range and regarding equally the contributions of all features.To address these issues,the paper proposed the log-function feature transforming method to narrow the data range to solve the discrimination problem and proposed the mean square deviation feature weighting method to distinguish the contribution levels to improve the accuracy of DGA fault diagnosis.Experimental results show that the proposed FTW_Pearson algorithm is superior to David triangle method which is popularly used in real applications,the traditional Pearson algorithm without feature transforming/feature weighting,the Bayes algorithm and the BPNN algorithm.