基于格拉姆角差场和CNN-BiGRU的变压器故障识别法  

Transformer Fault Identification Method Based on Gramian Angle Difference Field and CNN-BiGRU

作　　者：许耀博 杨信强 徐广超 杨诗豪 段国勇 XU Yaobo;YANG Xinqiang;XU Guangchao;YANG Shihao;DUAN Guoyong(College of Electrical Engineering and New Energy,China Three Gorges University,Yichang 443000,China)

机构地区：[1]三峡大学电气与新能源学院,湖北宜昌443000

出　　处：《电子科技》2025年第4期73-79,共7页Electronic Science and Technology

基　　金：国家自然科学基金(U2034203);强电磁工程与新技术国家重点实验室开放课题(AEET2022KF005)。

摘　　要：针对变压器绕组故障特征难以提取、诊断精度较低等问题,文中在频响曲线的基础上提出了一种基于格拉姆角差场(Gramian Angular Difference Field,GADF)和双向门控循环卷积神经网络(Convolutional Neural Network-Bidirectional Gated Recurrent Unit,CNN-BiGRU)的变压器故障识别方法。针对原始特征对不同故障类型区分度小的问题,提出了一种移动窗计算法对样本片段进行处理。结合格拉姆角差场变换得到谱特征,将一维数据映射成为三维图像数据。文中分析了不同故障类型在谱特征上的分布特性,将所得谱特征作为输入,通过循环卷积神经网络对故障片段数据进行分类得到识别结果。相较于传统方法,所提方法在特征差异上更明显,准确率得到进一步提高,其对切片分类精度达到了96.2%,验证了该方法的可行性。In view of the problems such as the difficulty in extracting the fault characteristics of transformer windings and the relatively low diagnostic accuracy,this study proposes a transformer fault identification method based on the GADF(Gramian Angular Difference Field)and the CNN-BiGRU(Convolutional Neural Network-Bidirectional Gated Recurrent Unit)on the basis of the frequency response curve.In response to the problem that the original features have a small discriminative ability for different fault types,a moving window calculation method is proposed to process the sample segments.By combining with the Gram angular difference field transformation,the spectral features are obtained,realizing the mapping of one-dimensional data into three-dimensional image data.The distribution characteristics of different fault types in the spectral features are analyzed.Taking the obtained spectral features as the input,the fault segment data are classified through the recurrent convolutional neural network to obtain the identification results.Compared with the traditional methods,the proposed method has more obvious feature differences,and the accuracy is further improved.The simulation results show that the classification accuracy of the slices reaches 96.2%,and the high accuracy of the diagnostic results verifies the feasibility of this method.

关 键 词：变压器 故障诊断 格拉姆角差场 谱特征 深度学习 循环卷积神经网络 高维空间特征 

分 类 号：TN99[电子电信—信号与信息处理] TP183[电子电信—信息与通信工程]