生物发酵过程中SSVIP-NNI软测量方法  

Neural network inverse system soft-sensor method based on SSVIP in bio-fermentation process

作　　者：于霜[1] 薛迎春[1] 程锦翔[2] 

机构地区：[1]苏州工业职业技术学院机电学院,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：《计算机与应用化学》2014年第12期1475-1478,共4页Computers and Applied Chemistry

基　　金：国家中小型创新基金项目(12C26213202207);苏州市科技基础设施建设计划项目(SZP201303)

摘　　要：生物发酵过程中关键生化变量难以直接检测,提出了一种利用基于敏感性与特异性的变量投影重要性(SSVIP)方法优化神经网络逆系统(NNI)的软测量方法。根据逆系统理论建立软测量模型,采用VIP变量优选方法,对软测量模型中的辅助变量进行优化。为了进一步提高VIP方法优选变量的性能,利用模型敏感性与特异性的概念,重新定义了VIP筛选阈值,确定最优的发酵过程软测量模型的变量。构造神经网络近似最优逆系统软测量模型,实现对发酵过程中关键过程变量的估计。通过Pensim仿真平台进行实验研究,结果表明经过辅助变量优选的神经网络逆系统软测量模型具有更高的估计精度和泛化能力。It is difficult to measure biochemical variables of the complex fermentation process directly,so a neural network inverse soft sensor method based on improved variable importance in the project（SSVIP） is proposed.According to inverse system theory,inverse system soft sensor model is constructed.Using VIP variables selection method,the secondary variables of the soft sensor model are optimized.In order to improve performance of VIP variables selection method,the threshold for VIP formula can be adjusted according to the concept of sensitivity and specificity.Then the optimal soft sensor model variables are obtained.Constructed neural network to approximate the model and the key process variables are estimated.Numerical simulations show that the optimal neural network inverse soft sensor model has higher estimation accuracy and stronger generation ability.

关 键 词：变量投影重要性 内含传感器 神经网络逆系统 软测量 发酵系统 

分 类 号：TP13[自动化与计算机技术—控制理论与控制工程]