样本不均衡下基于CGAN-CNN的逆变器故障诊断方法  

Inverter Fault Diagnosis Method Based on CGAN-CNN under Sample Imbalance

作　　者：孙权[1] 彭飞 李宏胜[1] 于翔海 孙国栋 SUN Quan;PENG Fei;LI Hongsheng;YU Xianghai;SUN Guodong(School of Automation,Nanjing Institute of Technology,Nanjing 211167,China;College of Automation Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China)

机构地区：[1]南京工程学院自动化学院,南京211167 [2]南京航空航天大学自动化学院,南京211106

出　　处：《电源学报》2024年第6期318-326,共9页Journal of Power Supply

基　　金：国家自然科学基金资助项目(61901212);江苏省高等学校自然科学研究重大资助项目(20KJA510007);江苏省配电网智能技术与装备协同创新中心开放基金资助项目(XTCX201909)。

摘　　要：三相逆变器是电动汽车电机驱动系统的重要部件,当出现故障时因发生时间较短导致故障样本规模受限,进而造成样本不均衡。为解决上述问题,提出1种融合条件生成对抗网络CGAN(conditional generative adversarial network)与卷积神经网络CNN(convolutional neural network)的逆变器故障诊断方法。首先将相电流作为故障敏感信号,经快速傅里叶变换FFT(fast Fourier transform)得到其频域特征,并进行归一化预处理;然后将各样本添加标签后输入CGAN模型进行对抗训练,生成各故障模式下的新样本。最后,采用CNN模型实现逆变器各类故障模式判别。实验结果表明,基于CGAN-CNN的故障诊断正确率可达98%以上,说明所提样本生成方法优于传统合成少数类过采样技术SMOTE(synthetic minority over-sampling technique)方法和生成对抗网络GAN(generative adversarial network)方法,可为新能源电动汽车智能运维提供理论支撑。The three-phase inverter is an important part of the motor drive system in an electric vehicle(EV).When a fault occurs,the fault sample size will be limited due to the short occurrence time,resulting in sample imbalance.To solve this problem,an inverter fault diagnosis method combining conditional generative adversarial network(CGAN)and convolutional neural network(CNN)is proposed in this paper.First,the phase current is taken as a fault sensitive signal,its frequency-domain characteristics are obtained by fast Fourier transform,and normalized preprocessing is carried out.Then,each sample is labeled and input into the CGAN model for countermeasure training to generate new samples in each fault mode.Finally,the CNN model is used to distinguish various fault modes of inverter.Through experimental research,it is found that the fault diagnosis accuracy based on CGAN-CNN can reach more than 98%,indicating that the proposed sample generation method is better than the traditional Smote and GAN methods.The results in this paper provide theoretical support for the intelligent operation and maintenance of new energy EVs.

关 键 词：故障诊断 样本不均衡 样本生成 条件生成对抗网络 卷积神经网络 

分 类 号：TM464[电气工程—电器]