摘要
为提高化工过程的故障诊断效果,将深度可分离卷积(DSC)和长短期记忆网络(LSTM)相结合,提出基于DSC-LSTM的集合型故障诊断方法,采用时空结合的方式从两个角度提取特征进行故障诊断。首先对数据进行归一化处理后将其送入DSC网络,通过DSC提取空域特征,同时对数据进行降维;再将DSC的输出作为LSTM的输入,通过LSTM提取时域特征,然后通过全连接层(FC)进行故障诊断;最后在田纳西-伊斯曼(TE)化工过程上对该方法进行验证。结果表明,DSC-LSTM集合方法可有效地提高故障诊断的准确率指标。
In order to improve the fault diagnosis of chemical processes, this paper combines Depthwise Separable Convolution(DSC)and Long Short-Term Memory network(LSTM),and proposes an ensemble fault diagnosis method based on DSC-LSTM,which uses the combination of space-time to extract features from two perspectives for fault diagnosis. Firstly, the data is normalized and fed into the DSC network, and the spatial domain features are extracted through DSC while the data is processed with dimensionality reduction.And then the output of DSC is used as the input of LSTM,and the time domain features are extracted through LSTM,and then the fault diagnosis is made through the fully connected layer(FC).Finally, the method is validated on the Tennessee-Eastman(TE)chemical process.The results show that the DSC-LSTM ensemble method can effectively improve the accuracy index of fault diagnosis.
作者
于桂仙
杨青
刘彦俏
YU Guixian;YANG Qing;LIU Yanqiao(Shenyang Ligong University,Shenyang 110159,China)
出处
《沈阳理工大学学报》
CAS
2021年第4期6-10,共5页
Journal of Shenyang Ligong University
基金
辽宁省教育厅科学研究项目计划(LG201917)。
