CIESC Journal ›› 2017, Vol. 68 ›› Issue (5): 1977-1986.DOI: 10.11949/j.issn.0438-1157.20161395

Previous Articles     Next Articles

Atmospheric tower soft sensor based on regression and mutual information of kernel slow features

JIANG Xinyi1, DU Hongbin1,2, LI Shaojun1   

  1. 1 Key Laboratory of Advanced Control and Optimization for Chemical Processes (Ministry of Education), East China University of Science and Technology, Shanghai 200237, China;
    2 Research Institute of Petro China Dushanzi Petrochemical Company, Karamay 833699, Xinjiang, China
  • Received:2016-09-29 Revised:2017-01-22 Online:2017-05-05 Published:2017-05-05
  • Supported by:

    supported by the National Natural Science Foundation of China (21676086,21406064).

基于核慢特征回归与互信息的常压塔软测量建模

蒋昕祎1, 杜红彬1,2, 李绍军1   

  1. 1 华东理工大学化工过程先进控制与优化技术教育部重点实验室, 上海 200237;
    2 中国石油天然气股份有限公司独山子石化研究院, 新疆 克拉玛依 833699
  • 通讯作者: 李绍军
  • 基金资助:

    国家自然科学基金项目(21676086,21406064)。

Abstract:

A novel soft sensor method based on slow feature regression (SFR) was proposed for industrial process with nonlinear and dynamic characteristics. First, a dynamic dataset was built by adding time-delay data and information redundancy was reduced by selecting variables according to mutual information maximization criteria. Then, kernel function was introduced into slow feature analysis(SFA)to improve capability of processing nonlinear data and the kernel slow features were used for regression. Through analysis of sample variation, kernel slow feature analysis(KSFA)could extract components with slowly varying dynamics, characterize trend of industrial process effectively, and improve precision of regression modelling. Finally, effectiveness and feasibility of the proposed method were verified by soft sensor model of constant top oil dry point and constant first line dropping point in atmospheric tower.

Key words: slow feature analysis, mutual information, dynamic modeling, atmospheric tower, petroleum, prediction

摘要:

针对工业过程的非线性及动态特性,提出了一种新的慢特征回归软测量方法。该方法首先通过添加时延数据构造动态数据集,利用互信息最大化准则筛选变量从而减少信息冗余的影响。同时该方法在慢特征分析的基础上引入核函数扩展,加强模型处理非线性数据的能力,并将获得的核慢特征用于回归建模。核慢特征分析通过分析样本的变化,提取具有缓慢变化特征的成分,可以有效地刻画工业过程的变化趋势,提升回归模型精度。最后该方法的有效性在常压塔常顶油干点与常一线初馏点的软测量模型中得到了验证。

关键词: 慢特征分析, 互信息, 动态建模, 常压塔, 石油, 预测

CLC Number: