生物发酵过程变量的NN-MIV软测量模型  

Soft Sensor Model of Fermentation Process Based on NN-MIV Variable Selection

作　　者：于霜[1] 程锦翔[2] 

机构地区：[1]苏州工业职业技术学院机电工程系,江苏苏州215104 [2]江苏大学电气信息工程学院,江苏镇江212013

出　　处：《控制工程》2015年第2期312-316,共5页Control Engineering of China

基　　金：国家中小型企业创新基金项目(5512C26213202207);苏州市科技基础设施建设计划项目(SZP201303)

摘　　要：针对生物发酵过程中关键生化参量的在线检测问题,提出一种基于平均影响值的神经网络(NN-MIV)变量选择方法。发酵过程初始软测量模型含有多个辅助变量,MIV方法计算辅助变量对主变量的外部贡献率,NN方法计算辅助变量对主变量的内部贡献率,文中将两种方法综合提出了NN-MIV方法,其计算出的辅助变量对主变量的贡献率稳定性好。利用筛选出最优辅助变量建立软测量模型,对青霉素发酵过程做了数值仿真实验,与传统的变量筛选方法相比,该方法筛选出的辅助变量少,建立的软测量模型估计精度高。To measure the key biochemical variables in fermentation process online, a variable selection method based on the mean impact value of neural network is proposed. The original soft sensor of the fermentation process has too many secondary variables. The external contribution rate of secondary variables to key variables is computed by the MIV method and the internal contribution rate. The proposed method that integrated the advantages of MIV and NN method computed the contributions of secondary to key variables. The contribution has better consistency and stability. According to the NN-MIV method, the new soft sensor model is constructed and numerical simulations are done for the penicillin fermentation process. The secondary variables are fewer and the precision of soft sensor is higher using NN-MIV than the traditional method.

关 键 词：生物发酵 平均影响值 变量筛选 软测量 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]