华中科技大学学报（自然科学版）2023，Vol.51Issue(12)：29-36,8.DOI:10.13245/j.hust.231205
双重对抗自编码数据扩张方法及应用
Data expansion method and application of couple adversarial auto-encoder
摘要
Abstract
To solve the problems of insufficient data and low quality in deep learning,combining the training stability of the auto-encoder network with the adversarial nature of the generation adversarial network,a data expansion method of couple adversarial auto-encoder(CAAE)was proposed.The method was applied to the fault diagnosis of permanent magnet synchronous motor(PMSM).In the process of model training,the network model was used to learn the distribution of the original data.The compressed variables and the decoded reconstructed data were input to the corresponding discriminators respectively to form a couple confrontation to ensure the quality of the generated data.After training,the random variables that satisfied a specific distribution were input into the trained network model with category labels to generate data of the corresponding category.Taking the motor inter-turn short circuit fault as an example,on the self-built motor data set,the fault data was expanded through couple adversarial auto-encoder.The validity of the model was verified according to the improvement of various indicators.Results show that,compared with traditional data expansion methods,couple adversarial auto-encoder can generate higher quality data which can improve the diagnostic accuracy effectively.关键词
双重对抗自编码(CAAE)/数据扩张/生成对抗网络/自编码网络/永磁同步电机Key words
couple adversarial auto-encoder(CAAE)/data expansion/generation adversarial network/auto-encoder network/permanent magnet synchronous motor(PMSM)分类
信息技术与安全科学引用本文复制引用
许小伟,敖金艳,刘光华,王亚玮..双重对抗自编码数据扩张方法及应用[J].华中科技大学学报（自然科学版）,2023,51(12):29-36,8.基金项目
国家重点研发计划资助项目(2022YFE0125200) （2022YFE0125200）
国家自然科学基金资助项目(51975426) （51975426）
湖北省重点研发计划资助项目(2021BAA018,2022BAA062). （2021BAA018,2022BAA062）
