摘要
针对所有样本点均出现在最小二乘支持向量机模型中的缺陷,提出一种改进的最小二乘支持向量机回归方法.根据最小二乘支持向量机模型学习误差的大小,去除原变量空间中大部分误差较小的样本点,从而获得回归模型的“稀疏”特性,大大简化了模型复杂程度.同时,将此方法应用于生物发酵过程,建立青霉素发酵过程中产物浓度的软测量模型,实现青霉素浓度的在线预估.实验结果表明,该方法为生物发酵过程中难于在线测量质量参数的实时监测提供了一个有效的手段.
An improved least square - support vector machine(LS-SVM) regression method is proposed to overcome the drawback that all the original learning samples are in the LS-SVM model. By the learning errors of the LS-SVM model, most sample points of small errors are deleted from the original sample space, and thus the sparseness of the LS-SVM is obtained. Based on the proposed LS-SVM method, a soft sensor model is built to estimate the product concentration of the penicillin fermentation process. The experiment shows that the proposed LS-SVM is of the sparseness characteristic, and a novel procedure is provided for the realtime monitoring of quality variables, which are hard to measure on-line in fermentation processes.
出处
《控制与决策》
EI
CSCD
北大核心
2005年第11期1307-1310,共4页
Control and Decision
基金
国家自然科学基金项目(60374003)
国家973子课题项目(2002CB312200)
关键词
软测量
最小二乘支持向量机
生物发酵
青霉素浓度
Soft sensing
Least square- support vector machine
Fermentation process
Penicillin concentration