生物发酵过程的LSSVM-Adaboost逆软测量方法  

LSSVM-adaboost inversion soft sensing method for a biologic fermentation process

作　　者：于霜[1] 丁煜函[2] 刘国海[2] 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：《计算机与应用化学》2013年第11期1265-1268,共4页Computers and Applied Chemistry

基　　金：江苏省高校优势学科建设工程资助项目([2011]6);中国博士后科学基金(20110491359);国家中小型创新基金项目(12C26213202207)

摘　　要：针对生物发酵过程关键生化量难以在线检测的问题,提出一种基于Adaboost的最小二乘支持向量机(LSSVM-Adaboost)逆的软测量建模方法。通过建立"内含传感器"模型并对其进行可逆性分析,给出"内含传感器"逆,考虑到LSSVM算法的小样本、计算速度快和Adaboost将弱学习机加权构成强学习机的优点,采用LSSVM-Adaboost方法近似难以建立精确模型的"内含传感器"逆。将训练好的逆系统串联于"内含传感器"之后构成复合的单位系统,实现不可测变量的软测量。将其应用于青霉素发酵过程菌体浓度和基质浓度的估计,仿真实验表明,所提方法有效提高了单一的LSSVM逆软测量方法对不可测变量的估计精度且具有良好的泛化能力。To solve the difficulty of on-line measuring the key biological parameters, a LSSVM-Adaboost inverse soft-sensing modeling method is proposed. ＂Inherent sensor＂ inversion model is given by constructing ＂inherent sensor＂ and its reversibility is analyzed. LSSVM method has the advantages of dealing with small sample data with quick calculatingtime, whileAdaboost can construct a strong learning machine by weighting some weak learning machines. LSSVM-Adaboost method is proposed to approximate ＂inherent sensor＂ inversion whose mathematical model is difficult to be obtained. The well trained inverse system is attached afterthe ＂inherent sensor＂ to form a composite unit system, which can fulfill the soft-sensing of immeasurable variables. The proposed soft-sensing method is applied to estimate mycelia concentration and substrate concentration in a fermentation process. Simulation results show that the proposed method has higher estimation accuracy and generalization ability than the general LSSVM soft-sensing method.

关 键 词：内含传感器 逆软测量 最小二乘支持向量机 ADABOOST 生物发酵过程 

分 类 号：TP13[自动化与计算机技术—控制理论与控制工程]