非均衡数据集下基于孪生卷积网络的变压器绕组变形故障识别方法

Deformation Fault Identification Method for Transformer Windings Based on Twin Convolutional Network Under Unbalanced Data Set

变压器绕组是变压器中最常发生故障的部分,故障类型多且常见程度不同。目前已有学者将机器学习应用于变压器绕组变形故障识别,但存在数据集不均衡时预测准确率低、运算时间长、所需样本量大等问题。为了解决上述的问题,文中提出了一种非均衡数据集下基于孪生卷积网络的变压器绕组变形故障识别方法,收集了变压器故障样本并搭建多种故障诊断模型进行对比以验证所提出方法的有效性。经过模型训练和验证,使用孪生卷积网络在非均衡数据集下进行变压器绕组变形故障识别正确率达到90%左右,高于卷积网络(CNN)、支持向量机(SVM)等其他方法的正确率。

The winding is the most frequently faulted part of transformers,with multiple types and different probability of occurrence.At present,some researchers apply the machine learning to the deformation fault identification of transformer winding which,however,has such problems as low prediction accuracy under data skew,long operation time and large number of samples.For solving the problems mentioned above,a fault identification method for transformer winding using twin convolutional network under unbalanced data set is proposed in this paper.The fault sample of the transformer is collected and multiple types of fault models are set up for comparison to verify the effectiveness of the proposed method.Throughout model training and verification,the accuracy rate of deformation fault identification of transformer winding under the unbalanced data set reaches around 90%,which is higher than the accuracy rate of other methods such as convolutional neural networks(CNN)and support vector machine(SVM).