文献类型：期刊文献

英文题名：Process monitoring of nonlinear uncertain systems based on Part Interval Stacked Autoencoder and Support Vector Data Description

作者：Wu, Qiqi[1];Lu, Weipeng[1];Yan, Xuefeng[1,2]

机构：[1]East China Univ Sci & Technol, Key Lab Smart Mfg Energy Chem Proc, Minist Educ, Shanghai 200237, Peoples R China;[2]POB 293, MeiLong Rd 130, Shanghai 200237, Peoples R China

年份：2022

卷号：129

外文期刊名：APPLIED SOFT COMPUTING

收录：;WOS:【SCI-EXPANDED(收录号:WOS:000863068900004)】；

基金：The authors are grateful for the support of National key re- search and development program of China (2021YFC2101100) , and National Natural Science Foundation of China (21878081) .

语种：英文

外文关键词：Stacked autoencoder; Process monitoring; Tennessee Eastman process; Industrial wastewater treatment plant; Support Vector Data Description

摘要：Chemical process measurement may cause inaccuracy and uncertainty in some parts of the measure-ment data due to equipment aging and environmental conditions, such as temperature, humidity, gas pressure, gas flow, and other related factors. Uncertain data of this kind commonly fluctuate in an interval centered on the true value. Given the uncertain characteristics of some parts of the measurement data, the certain of some parts, and the nonlinearity of the industrial process, the present paper proposes a novel algorithm of process monitoring and fault diagnosis called Part Interval Stacked Autoencoder and Support Vector Data Description (PISAE-SVDD). The loss function of Stacked Autoencoder (SAE) is improved in this algorithm. In certain measurement data, the reconstruction error value is the mean square error of the original input data and output data. By contrast, uncertain measurement data provide a certain allowable range for reconstruction error value. The error value of measurement data is considered as zero within this range. The reconstruction error value of measurement data beyond the allowable range is calculated in the same way as for certain measurement data. At the same time, the characteristic information of the industrial process is extracted through the strong nonlinear characterization ability of SAE. Support Vector Data Description is then used to obtain the control limit of the fluctuation range of the normal working condition based on the SAE feature information data extracted from normal samples. This approach is adopted to realize the process monitoring of the industrial process with partial uncertain measurement. To detect fault, this algorithm was applied to the numerical simulation, the Tennessee Eastman process, and the process in industrial wastewater treatment plants, and then compared with other advanced algorithms. The results indicate the excellence and high efficiency of the PISAE-SVDD algorithm in the field of process monitoring. (C) 2022 Elsevier B.V. All rights reserved.