基于多模型的发酵过程高斯过程回归软测量建模研究（英文）  

A multi-model soft sensor development using Gaussian process regression for fermentation processes

作　　者：梅从立[1] 尹梁[1] 杨铭[1] 廖志凌[1] 刘国海[1] 

机构地区：[1]江苏大学电气信息工程学院,江苏镇江212013

出　　处：《计算机与应用化学》2016年第12期1279-1285,共7页Computers and Applied Chemistry

基　　金：江苏省自然科学金(BK20130531);江苏高校优势学科建设工程(PAPD[2011]6);江苏省政府奖学金~~

摘　　要：软测量仪表在实际应用中往往存在预测精度低、缺乏预测精度信息等问题。基于多模型方法的软测量仪表通过子模型来描述局部变化,可以有效提高软测量仪表预测精度。在本研究中,高斯过程回归(GPR)模型因其预测方差能够反映预测精度信息特性,被用于构建局部子模型。同时,基于不确定性推理方法,本文提出了基于高斯过程回归预测方差的多模型融合策略。最后,将所提方法应用于工业红霉素发酵过程数据。结果表明,与其他高斯过程回归方法相比较,所提出方法预测精度更高,95%置信区间范围更小。Many soft sensors were restricted to be applied in practical cases for bad prediction accuracy and lacking precision informa- tion of estimates. Multi-model approaches can be used to improve prediction accuracy of soft sensors because of using local models to describe local behaviors in local operating ranges. In this paper, Gaussian process regression （GPR） was used to build local models be- cause it has predicted variance which reflects the precision information of estimates. According to mechanisms of uncertain inference, a new fusion scheme based on predicted variances of local GPR models was proposed to construct a multi-model soft sensor, which can be interpreted physically. The proposed soft sensor was validated with experimental data from a practical Erythromycin fermentation process. For comparisons, several other GPR soft sensors were also investigated. Results show that the proposed method has higher prediction accuracy and smaller 95% confidence interval.

关 键 词：软测量仪表 多模型 高斯过程回归 机理建模 发酵 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]