摘要
智能的故障诊断技术是处理工业大数据的一种有效方法,该技术能够快速、高效地处理原始数据并提供准确的诊断结果。为更好提取出数据的有效信息,针对化工数据的时序性和高维非线性的特点,本文提出一种基于扩展长短期记忆网络(LSTM)的集合型故障诊断方法(ELSTM)。先用LSTM处理数据得到包含原始数据时空信息的隐层输出,然后利用卷积神经网络(CNN)从多维数据中提取特征的能力,先用1D卷积提取每一个时间序列内部的局部特征,再用2D卷积提取相邻时间序列之间存在的相互依赖特征,最终将提取的特征数据经过全连接层,得到分类结果。将ELSTM用于复杂系统的田纳西-伊士曼过程(TE),实验结果表明,所用方法比标准的LSTM网络、自编码(autoencoder)具有更高的识别准确率。
Intelligent fault diagnosis technology is an effective method for processing industrial big data.It can process raw data quickly and efficiently and provide accurate diagnosis results.In order to better extract the effective information of the data,aiming at the time series and high-dimensional nonlinear characteristics of chemical data,this paper proposes an ensemble fault diagnosis method based on extended long short-term memory network(LSTM),referred to as ELSTM.First,LSTM is used to process the data to obtain the hidden layer output containing the spatiotemporal information of the original data,and then CNN is used to have a strong ability to extract features from multi-dimensional data.1 D convolution is used to extract local features within each time series,and then 2 D Convolution is used to extract the interdependent features between adjacent time series,and finally the extracted feature data passes through the fully connected layer to obtain the classification result.ELSTM is used in the Tennessee Eastman process(Tennessee Eastman,TE)of complex systems.Experimental results show that the method used has higher recognition accuracy than the standard LSTM network and autoencoder.
作者
王丹丹
陈刚
杨青
WANG Dandan;CHEN Gang;YANG Qing(Shenyang Ligong University,Shenyang 110159,China)
出处
《沈阳理工大学学报》
CAS
2020年第4期70-75,共6页
Journal of Shenyang Ligong University
基金
辽宁省教育厅科学研究项目计划(LG201917)
辽宁省自然科学基金指导计划(20180550801)
