华南理工大学学报(自然科学版) ›› 2009, Vol. 37 ›› Issue (8): 56-60.

• 机械工程 • 上一篇    下一篇

基于INLR-PPLS的非线性多传感耦合信息建模方法

洪晓斌 刘桂雄 叶挺东 黄国健 陈铁群   

  1. 华南理工大学 机械与汽车工程学院, 广东 广州 510640
  • 收稿日期:2008-08-03 修回日期:2008-09-12 出版日期:2009-08-25 发布日期:2009-08-25
  • 通信作者: 洪晓斌(1979-),男,博士后,主要从事新型智能传感技术、网络化测控的研究. E-mail:mexbhong@scut.edu.cn
  • 作者简介:洪晓斌(1979-),男,博士后,主要从事新型智能传感技术、网络化测控的研究.
  • 基金资助:

    广东省自然科学基金资助项目(7000815);中国博士后基金资助项目(20070420779);广东省教育部产学研结合项目(2007A090302039);华南理工大学博士后创新基金资助项目(20080215)

Approach to Modeling Nonlinear Multi-Sensor Coupling Information Based on INLR-PPLS

Hong Xiao-bin  Liu Gui-xiong  Ye Ting-dong  Huang Guo-jian  Chen Tie-qun   

  1. School of Mechanical and Automotive Engineering, South China University of Technology, Guangzhou 510640, Guangdong, China
  • Received:2008-08-03 Revised:2008-09-12 Online:2009-08-25 Published:2009-08-25
  • Contact: 洪晓斌(1979-),男,博士后,主要从事新型智能传感技术、网络化测控的研究. E-mail:mexbhong@scut.edu.cn
  • About author:洪晓斌(1979-),男,博士后,主要从事新型智能传感技术、网络化测控的研究.
  • Supported by:

    广东省自然科学基金资助项目(7000815);中国博士后基金资助项目(20070420779);广东省教育部产学研结合项目(2007A090302039);华南理工大学博士后创新基金资助项目(20080215)

摘要: 针对线性PLSR(偏最小二乘回归)在多传感信息回归建模中存在的不足,提出了一种基于INLR(Implicit Nonlinear Latent Variable Regression)-PPLS(Polynomial Partial Least Squares)的非线性多传感耦合信息建模方法.该方法通过线性PLSR对多传感信息进行预处理,达到降维的目的;基于INLR建立外模型非线性样本矩阵变换方程并线性化,进而采用PPLS进行内模型非线性映射,并对多传感非线性回归模型实行反求解.最后,将该方法应用于液态乙醇浓度测控系统,结果表明该方法较线性PLSR预测准确度提高21%.

关键词: 多传感信息, 偏最小二乘回归, INLR PPLS

Abstract:

:In order to remedy the shortcomings of linear PLSR ( Partial Least Squares Regression) in multi-sensor information regression modeling, a novel modeling approach based on INLR (Implicit Nonlinear Latent Variable Regression)-PPLS (Polynomial Partial Least Squares) is put forward. In this method, multi-sensor information is preprocessed by means of linear PLSR to reduce the dimension, and a nonlinear sample-matrix transform formula of the outer model is established and linearized based on INLR. Then, the nonlinear mapping of the inner model is performed via PPLS and the reverse regression model is obtained. The proposed method is finally applied to the measurement and control system of liquid alcohol concentration. It is found that the prediction accuracy of the pro- posed approach is 21% higher than that of linear PLSR.

Key words: muhi-sensor information, partial least squares regression, INLR, PPLS