一种基于改进BP神经网络的变压器故障诊断方法

A Method of Transformer Fault Diagnosis Based on Improved BP Neural Network

变压器内部故障与油中溶解不同组分、不同含量的气体之间的关系是目前进行变压器故障诊断的主要研究内容。文章提出一种基于BP神经网络的变压器故障诊断方法,其在传统反向传播神经网络(BPNN)基础上,引入ResNet残差网络模块思想,并在第Ⅳ和第Ⅴ残差模块中嵌入支持向量机(SVM)分类器,从权重角度筛选对诊断结果准确率更具影响的特征向量;构造多个子数据集并对在分布式平台进行并行化训练后获得的不同性能的子分类器进行投票决策。结果表明,相比于基于传统BPNN的诊断方法,本文提出的方法对变压器故障诊断的平均准确率提升了8.56%,验证了该方法能提高变压器故障类型诊断的性能。

Relationship between transformer internal faults and gases dissolved in oil with different components and different contents is the main research content of transformer fault diagnosis.Based on the traditional BPNN,this paper presents a transformer fault diagnosis method which introduces the idea of ResNet module,embeds the SVM classifier in the fourth and fifth residual modules,and selects the feature vectors that have more influences on the accuracy of the diagnosis result from the perspective of weight.Subclassifier with different performances is obtained by constructing several sub data-sets and parallellizing training on distributed platform to make voting decisions.Results show that compared with the traditional BPNN-based diagnosis method,average accuracy of the proposed method for transformer fault diagnosis is improved by 8.56%,which verifies that the proposed method can improve the performance of transformer fault diagnosis.