基于改进ACGAN样本增强的变压器故障诊断技术  

Power Transformer Fault Diagnosis Technology Based on Improved ACGAN Data Augmentation

作　　者：卢锦玲[1] 朱晨菲 LU Jinling;ZHU Chenfei(School of Electrical and Electronic Engineering,North China Electric Power University,Baoding 071003,China)

机构地区：[1]华北电力大学电气与电子工程学院,河北保定071003

出　　处：《电力科学与工程》2021年第11期42-51,共10页Electric Power Science and Engineering

基　　金：国家科技支撑计划(2015BAA06B03)。

摘　　要：基于深度学习算法的故障诊断需要足量的样本作为训练数据集。变压器故障数据匮乏将导致故障诊断准确率较低。对此,提出了一种基于辅助分类GAN的故障诊断方法。该方法引入自我注意机制,提取故障样本的全局特性,以提高生成样本的质量;并加入梯度惩罚,以提高模型收敛速度和训练稳定性。运用该方法对失衡样本进行增强扩充,并在变压器振动试验数据集上进行验证。仿真结果表明,该方法能够有效改善数据不平衡带来的影响,增强扩充后的故障诊断准确率提高了3.4%。The fault diagnosis algorithm based on the deep learning needs enough samples as its training data set.The lack of fault data usually leads to low accuracy of fault diagnosis.To solve this problem,this paper proposed a transformer fault data augmentation method based on auxiliary classification GAN.This method introduces a self-attention mechanism to extract the global characteristics of fault samples to improve their quality,and adds the gradient penalty to improve the convergence speed and stability of the training process.This paper uses this method to enhance the quality of samples.The effectiveness of the method is verified by using the transformer vibration test data set.The simulation results show that this method can improve the accuracy of fault diagnosis by 3.4%after the data augmentation.

关 键 词：电力变压器 故障诊断 数据增强 自注意力机制 辅助分类生成对抗网络 

分 类 号：TM73[电气工程—电力系统及自动化]