基于特征表征度与多变量预测模式识别的变压器故障诊断

Transformer Fault Diagnosis Based on Feature Representation Degree and Multi-Variable Predictive Pattern Recognition

针对现有的基于油中溶解气体分析的诊断方法忽略了特征对故障类型的表征力度差异,限制了故障区分度的问题,提出了一种新的变压器故障诊断方法。该方法首先将主成分分析与灰色关联度分析相结合,用于度量特征量表征各故障类型的重要程度,并确定各故障类型下各特征量的权重;其次,构建一种多变量预测模式识别方法,用于电力变压器故障分类。其基本思想为:基于训练样本数据建立各类别下特征量之间相互表达的数学模型,即各类别的预测模型;将待测样本特征量输入至已建立好的预测模型中,并输出对应的特征量预测值;基于各特征量的权重信息,以预测值与实际值的加权误差平方和最小为判据,确定样本所属类别。最后通过与神经网络、支持向量机等方法进行对比,验证了所提方法的有效性。

The existing diagnostic methods based on the analysis of dissolved gas in oil ignore the difference in the strength of feature to fault type,and thus limits fault discrimination.In order to solve the problem,this paper proposes a new transformer fault diagnosis method.Firstly,principal component analysis is combined with grey correlation degree analysis to characterize the importance degree of each feature and determine the weight of each feature under each fault type.Secondly,a multivariate predictive pattern recognition method is constructed for power transformer fault classification.This method is based on the training samples to establish a mathematical model of feature quantities’mutual representation of each category,i.e.models for all types of prediction;the features of the samples to be tested are input into the established prediction model,and the predicted value of the corresponding characteristic quantity is output.Then,based on the weight information of each feature,the category of the sample is determined by taking the minimum square sum of the weighted error between the predicted value and the actual value as the criterion.Finally,the effectiveness of the proposed method is verified by comparing with neural network and support vector machine.