化工学报 ›› 2021, Vol. 72 ›› Issue (3): 1480-1486.DOI: 10.11949/0438-1157.20201674

• 过程系统工程 • 上一篇    下一篇

基于MIC的支持向量回归及其在化工过程中的应用

顾俊发1(),许明阳2,马方圆1,林治宇2,纪成1,王璟德1(),孙巍1   

  1. 1.北京化工大学化学工程学院,北京 100029
    2.中化泉州石化有限公司,福建 泉州 362103
  • 收稿日期:2020-12-02 修回日期:2020-12-10 出版日期:2021-03-05 发布日期:2021-03-05
  • 通讯作者: 王璟德
  • 作者简介:顾俊发(1992—),男,硕士研究生,gjf1139612668@163.com

Support vector regression based on maximal information coefficient and its application in chemical industrial processes

GU Junfa1(),XU Mingyang2,MA Fangyuan1,LIN Zhiyu2,JI Cheng1,WANG Jingde1(),SUN Wei1   

  1. 1.College of Chemical Engineering, Beijing University of Chemical Technology, Beijing 100029, China
    2.Sinochem Quanzhou Petrochemical Co. , Ltd. , Quanzhou 362103, Fujian, China
  • Received:2020-12-02 Revised:2020-12-10 Online:2021-03-05 Published:2021-03-05
  • Contact: WANG Jingde

摘要:

在化工生产中,软测量方法可以有效解决某些关键变量由于仪表故障而无法实时获取数据的问题。在建立软测量模型时,变量及回归方法的选取会直接影响模型的准确率。特别是在现代化工中,过程变量众多且变量间存在着冗余且复杂的非线性关系。对此,本文提出了一种基于最大信息系数的支持向量回归算法,利用最大信息系数在非线性相关性度量的优势,选择合适的辅助变量,避免了全部变量作为输入所造成的数据冗余。在此基础上,利用支持向量回归方法建立软测量模型,实现对软测量目标的预测。该方法被应用于存在仪表故障的某催化重整装置进料换热器热端压降的软测量中,结果表明该方法可以有效地实现对压降的软测量,实现了对仪表故障时的数据校正。

关键词: 算法, 预测, 过程系统, 数据校正, 最大信息系数, 变量筛选

Abstract:

In chemical production, soft-sensing methods can effectively solve the problem that some key variables cannot be obtained in real time due to instrument failure. When building a soft-sensing measurement model, the accuracy of the model will be directly affected by the selection of variables and regression methods, especially in modern chemical industry, there are a large number of process variables with complex nonlinear relationships among them. Therefore, it is important to select proper variables and regression methods. In this paper, a support vector regression (SVR) algorithm based on maximum information coefficient (MIC) is proposed for soft-sensing measurement. Benefiting from the advantages of MIC in nonlinear correlation measurement between the process variables and target variable, the data redundancy can be avoided by selecting the appropriate modeling variables. On this basis, the SVR method is further applied to extract the relationship between the modeling variables and the target variable, by which a soft sensing model is established to predict the target variable. This method is applied to the soft-sensing measurement of the hot end pressure drop of a heat exchanger in a catalytic reforming unit. The results show that this method can effectively realize the soft measurement of pressure drop and the data correction when the sensor failure occurs.

Key words: algorithm, prediction, process systems, data correction, MIC, variable selection

中图分类号: