基于互信息稀疏自编码的青霉素发酵过程软测量建模  

Soft sensor development for penicillin fermentation processes based on mutual information sparse auto-encoder

作　　者：刘聪 谢莉[1,2] 杨慧中 Liu Cong;Xie Li;Yang Huizhong(School of Internet of Things Engineering,Jiangnan University,Wuxi 214122,China;Key Laboratory of Advanced Process Control forLight Industry(Ministry of Education),Jiangnan University,Wuxi 214122,China)

机构地区：[1]江南大学物联网工程学院,江苏无锡214122 [2]江南大学教育部轻工过程先进控制重点实验室,江苏无锡214122

出　　处：《南京理工大学学报》2020年第5期590-597,共8页Journal of Nanjing University of Science and Technology

基　　金：国家自然科学基金(61403166,61773181);江苏省自然科学基金资助项目(BK20140164);中央高校基本科研业务费专项资金资助项目(JUSRP51733B)。

摘　　要：作为一类典型的间歇过程,青霉素发酵过程具有较强的非线性、时变性和不确定性。同时,菌体浓度、基质浓度和产物浓度等关键生物参数难以在线实时测量,而离线化验则需要耗费高昂的人工成本。针对这一问题,该文基于互信息加权的特征提取方法,提出一种软测量建模方法来估计青霉素发酵过程中的产物浓度。首先,基于互信息计算各个输入变量和输出变量之间的相关性,并用于加权处理稀疏自动编码器损失函数中的重构误差项,从而提取与输出更为相关的特征;然后,利用提取到的所有输入样本特征,结合最小二乘支持向量机对产物浓度进行估计。Pensim仿真平台的验证结果表明,该文所提方法能够有效提高青霉素发酵过程软测量模型的估计精度。As a class of typical batch processes,penicillin fermentation processes are characterized with strong nonlinearity,time variation and uncertainty.At the same time,key biological parameters such as biomass concentration,substrate concentration,and product concentration are difficult to be measured online in real time,while off-line analyses require high labor costs.Aiming at this problem,this paper proposes a soft sensor modeling method based on the feature extraction method with mutual information weighting to estimate the product concentration in penicillin fermentation processes.First,the correlation between each input variable and the output variable is calculated based on the mutual information,which is adopted to weight the reconstruction error term in the cost function of the sparse auto-encoder,and more relevant features with the output variable can be extracted.Furthermore,using the extracted features of all input variables,the product concentration is estimated by combining the least squares support vector machine.Validation results of the Pensim simulation platform illustrate that the proposed method here can effectively improve the estimation precision of soft sensors for penicillin fermentation processes.

关 键 词：青霉素发酵过程 互信息加权 稀疏自编码器 最小二乘支持向量机 软测量 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]