【摘要】：工業(yè)過程數(shù)據(jù)具有高斯和非高斯混合分布的特點。獨立因子分析(IFA)采用一維高斯混合模型擬合任意的因子分布,因此可以處理高斯和非高斯混合的問題。雖然在給定因子數(shù)的前提下變分IFA算法可以有效地縮短建模時間,但是獨立因子數(shù)的選擇仍然需要較長的計算時間。此外,若IFA的因子數(shù)選擇不當,會造成部分因子的信息遺留在觀察變量的殘差中,導致GSPE監(jiān)控指標的監(jiān)控性能變差。為了解決IFA在實際應用中存在的問題,本文結(jié)合了IFA和FA方法。首先使用FA輔助IFA選取獨立因子數(shù),以進一步減小IFA建模時間;其次使用FA對IFA的殘差進行再處理,以解決由于獨立因子數(shù)選擇不當造成的問題。最后將該方法應用于田納西-伊斯曼(TE)過程和乙烯裂解爐過程的監(jiān)控中,實驗結(jié)果驗證了該聯(lián)合方法的有效性。
[Abstract]:Industrial process data have the characteristics of mixed distribution of Gao Si and non-Gao Si. Independent factor analysis (IFA) uses a one-dimensional Gao Si mixed model to fit arbitrary factor distributions, so it can deal with the mixed problem of Gao Si and non-Gao Si. Although the variable IFA algorithm can effectively shorten the modeling time under the condition of given factor number, the selection of independent factor number still needs a long calculation time. In addition, if the factor number of IFA is not selected properly, the information of some factors will be left in the residual of observation variables, which will lead to the deterioration of monitoring performance of GSPE monitoring index. In order to solve the problem of IFA in practical application, this paper combines IFA and FA methods. In order to further reduce the modeling time of IFA, the FA aided IFA is used to select the independent factor number, and the FA is used to reprocess the IFA residuals to solve the problem caused by improper selection of independent factor numbers. Finally, the method is applied to the monitoring of Tennessee Eastman (TE) process and ethylene cracking furnace process. The experimental results show the effectiveness of the combined method.
【作者單位】： 華東理工大學化工過程先進控制和優(yōu)化技術(shù)教育部重點實驗室;
【基金】：國家自然科學基金(61134007,21376077,21303102) 上海市研發(fā)平臺建設項目(13DZ2295300) 上海市自然科學基金(16ZR1407300)
【分類號】：TP277;TQ06

【相似文獻】

相關期刊論文 前3條

1 黃蕾;;IFA秘書長支招中國化肥業(yè)[J];中國科技財富;2006年04期

2 ;第68屆IFA年會引發(fā)的思考[J];農(nóng)資科技;2000年04期

3 Changkyoo Yoo;Minhan Kim;Sunjin Hwang;Yongmin Jo;Jongmin Oh;;非高斯建模方法的間歇過程的在線預測監(jiān)控和預測模型(英文)[J];Chinese Journal of Chemical Engineering;2008年01期

，

本文編號：2263746