基于全局优化GAN的不平衡数据故障诊断

Fault Diagnosis of Unbalanced Databased on Global Optimization GAN

为解决不平衡数据深度学习特征提取不准确,导致误分类率高的问题,提出了一种基于全局优化生成对抗网络的不平衡数据故障诊断方法。首先利用自动编码器解码网络和深度神经网络故障诊断结果指导生成器的训练,有效地避免了模型崩溃和梯度消失的问题。然后设计了一种两级判别器,通过增加深度神经网络故障诊断模型作为附加判别器,同时,采用传统的判别器对不合格的故障样本进行滤波。通过生成器和两个分级鉴别器交替优化,同时提高了生成器的生成能力以及鉴别器的识别能力。实验结果表明提出的方法能够有效地提升不平衡样本故障诊断精度。

In order to solve the problem of high misclassification rate caused by inaccurate deep learning feature extraction of un-balanced data,a fault diagnosis method based on global optimization generation countermeasure network was proposed.Firstly,the fault diagnosis results of automatic encoder decoding network and deep neural network were used to guide the training of gen-erator,which effectively avoided the problems of model collapse and gradient disappearance.Then,a two-stage discriminator is designed,in which a deep neural network fault diagnosis model was added as an additional discriminator.At the same time,the traditional discriminator was used to filter the unqualified fault samples.The generator and two discriminators were optimized al-ternately to improve the generating ability of the generator and the recognition ability of the discriminator.The experimental re-sults show that the proposed method can effectively improve the fault diagnosis accuracy of unbalanced samples.