基于ACGAN的功率变换器参数性故障诊断方法  

Method of Diagnosing Parametric Fault of Power Converter Based on Auxiliary Classifier Generative Adversarial Nets

作　　者：傅宏辉 王友仁[1] 孙灿飞[1] 孙权[1] FU Honghui;WANG Youren;SUN Canfei;SUN Quan(College of Automation Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China)

机构地区：[1]南京航空航天大学自动化学院

出　　处：《机械制造与自动化》2019年第6期159-163,共5页Machine Building & Automation

基　　金：南京航空航天大学研究生创新基地(实验室)开放基金资助(kfjj20170323);中央高校基本科研业务费专项资金资助

摘　　要：提出了一种基于辅助分类生成对抗网络的功率变换器参数性故障智能诊断方法。首先采集功率变换器的测点电压与支路电流信号,提取信号的时域特征,构成故障特征向量。采用对抗学习机制训练生成器和判别器,由ACGAN中生成器构造与真实故障特征分布近似的伪数据,从而将伪数据与真实数据同时用于训练判别器,判别器通过判别真伪数据来训练生成器。以Buck变换器为例,验证了所提出的故障诊断方法的可行性,结果表明ACGAN故障诊断方法相对于传统神经网络具有更高的故障诊断率与更优的泛化性能。This paper proposes an intelligent method of diagnosing parametric fault based on auxiliary classifier generative adversarial nets(ACGAN).The voltage and current signals of the power converter are collected first,and then the time domain characteristics of the signal are extracted,which is used to constitute the fault feature vectors.The generator and discriminator are trained by the adversarial learning mechanism,and the pseudo-data similar to the real fault features are constructed by the generator in ACGAN,then the pseudo-data and the real data are simultaneously used to train the discriminator,and the generator is trained by discriminator discriminating the real or fake data.Buck converter is taken as an example,the feasibility of this method is verified by the simulation.The results show that ACGAN fault diagnosis method has higher fault diagnosis rate and better generalization performance than the traditional neural network.

关 键 词：故障诊断 对抗学习机制 功率变换器 深度学习 ACGAN 样本生成 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置]