化工学报 ›› 2019, Vol. 70 ›› Issue (12): 4741-4748.DOI: 10.11949/0438-1157.20190606

• 过程系统工程 • 上一篇    下一篇

基于k-近邻互信息的发酵过程高斯过程回归建模

赵荣荣(),赵忠盖(),刘飞   

  1. 江南大学轻工过程先进控制教育部重点实验室,江苏 无锡 214122
  • 收稿日期:2019-05-31 修回日期:2019-08-30 出版日期:2019-12-05 发布日期:2019-12-05
  • 通讯作者: 赵忠盖
  • 作者简介:赵荣荣(1993—),女,硕士研究生,1252082423@qq.com
  • 基金资助:
    国家自然科学基金项目(61573169)

Gaussian process regression modeling of fermentation process based on k-nearest neighbor mutual information

Rongrong ZHAO(),Zhonggai ZHAO(),Fei LIU   

  1. Key Laboratory of Advanced Control for Light Industry Process, Ministry of Education, Jiangnan University, Wuxi 214122, Jiangsu, China
  • Received:2019-05-31 Revised:2019-08-30 Online:2019-12-05 Published:2019-12-05
  • Contact: Zhonggai ZHAO

摘要:

发酵过程中基质浓度往往无法在线测量,采用高斯过程回归(GPR)建立基质浓度的估计模型,实现了其软测量。不同于传统软测量方法对基质浓度的估计,该方法不仅可以得到估计值,还能够得到其估计方差。考虑到发酵过程中各变量之间的非线性、相关性,为了提高模型的预测性能,在模型建立之前首先用k-近邻互信息(k-MI)辅助变量选择方法对模型的输入变量进行选择。从青霉素发酵过程的应用结果来看,采用kMI-GPR方法取得了较好的估计效果。

关键词: 发酵过程, 高斯过程回归, k-近邻互信息, 软测量

Abstract:

The concentration of the substrate during fermentation is often not measured online. In this paper, Gaussian process regression (GPR) is used to establish an estimation model of substrate concentration, and its soft measurement is realized. Different from traditional regression models, the GPR model can not only predict the quality value, but also provide the estimation variance. In order to improve the prediction performance of the model in the nonlinear fermentation process with correlated variables, the input variables of the model are selected by the k-nearest neighbor mutual information (k-MI) method before the model development. The application results of penicillin fermentation process show the ideal prediction performance based on the kMI-GPR model.

Key words: fermentation process, Gaussian process regression, k-nearest neighbor mutual information, soft sensor

中图分类号: