基于非线性核脊回归算法的海洋微生物发酵参数软测量  

Soft-sensing of Microorganism Ferment Parameters Based on Non-linear Kernel Ridge Regression Algorithm

作　　者：朱湘临[1] 杨建宁[1] 孙谧[2] 王跃军[2] 孙玉坤[1] 

机构地区：[1]江苏大学电气信息工程学院 [2]中国水产科学研究院黄海水产研究所

出　　处：《化工自动化及仪表》2011年第2期171-175,共5页Control and Instruments in Chemical Industry

基　　金：国家"863"基金资助项目(2007AA091602)

摘　　要：针对海洋微生物溶菌酶发酵工艺中参数实时测量,采用根据发酵过程中主辅变量关系的相关灵敏度矩阵绝对值最大元素排序选择辅助变量,运用非线性核脊回归(NKRR)算法预测主导参数的软测量方法。相关灵敏度矩阵反映主辅变量联系,根据发酵实际实验得到。改进直接优化的非线性核脊回归算法辨识发酵非线性系统,将空间Rn中主导变量和辅助变量非线性函数映射关系转化为空间H中的线性划分,运用M ercer条件的核函数求解回归方程,避免直接计算变量映射函数及其内积。通过仿真研究和比对SVM、径向基神经网络(RBFNN)两种算法的实际运行结果表明,该软测量方法具有较好的泛化性能,实时性和准确性可以满足要求。Regarding the real-time measurement of parameters in lysozyme zymotechnics of marine microorganism,a soft-sensing method was adopted.This method selected the auxiliary variable upon the correlation sensitivity matrix which reflecting the relation between the main and auxiliary variable in fermentation,and employed the non-linear kernel ridge regression（NKRR） algorithm to predict the main parameters;and thus the nonlinear function mapping relation between the main and auxiliary variable in space Rn was converted to linear division in space H,and by using Kernel Method of Mercer to solve the regression equation,the direct calculating of the variable mapping function and its inner product were averted.By simulating and comparing SVM and RBFNN algorithms,the results prove the excellent generalization performance of this improved soft-sensing method,and both real-time and accuracy can meet requirements.

关 键 词：海洋微生物发酵 软测量 非线性核脊回归算法 变量映射函数 相关灵敏度矩阵 

分 类 号：TP13[自动化与计算机技术—控制理论与控制工程]