基于快速独立分量分析的苯酚浓度软测量  

Phenol Concentration Measurement Based on Fast Independent Component Analysis

作　　者：章军 杨慧中 陶洪峰 

机构地区：[1]江南大学教育部轻工过程先进控制重点实验室,江苏无锡214122 [2]无锡威泰迅电力科技有限公司,江苏无锡214125

出　　处：《化工自动化及仪表》2012年第11期1460-1464,共5页Control and Instruments in Chemical Industry

基　　金：国家自然科学基金资助(60674092);江苏省高技术研究项目资助(工业部分)(BG2006010);江苏高校优势学科建设工程资助项目;高等学校学科创新引智计划资助(B12018);江南大学博士研究生科学研究基金资助(JUDCF12030)

摘　　要：针对软测量技术中辅助变量间可能存在的复杂的内在联系给辅助变量选择及提高模型精度等方面带来的诸多不利因素,提出了一种基于负熵最大的快速独立分量分析(Fast-ICA)方法,可以有效地消除辅助变量间的信息冗余现象,得到相互独立的分量。再通过互信息分析各个分量和主导变量间的内在联系,选择和主导变量联系较大的分量用于模型训练,并将其应用到苯酚(BF)浓度的软测量中。仿真结果表明:从经过ICA分解的独立分量中提取出样本特征并用于支持向量机模型(SVM)的训练,可以有效地减少冗余信息的干扰,提高软测量模型的估计精度和泛化能力。In soft sensing technology,the complex internal relation and redundant information among secondary variables can bring negative factors to secondary variables selection and improving model accuracy.The method of fast independent component analysis(Fast-ICA) based on negative entropy biggest was proposed to effectively eliminate the information redundancy among secondary variables and to get independent components, and then according to the internal relation between the predicted variable and each component by the mutual information,to select the components which associated with the predicted variable for the model training. Both applying it to phenol(BF) concentration measurement and having it simulated show that extracting sample characteristics from independent components decomposed with ICA and having them applied to SVM can effectively reduce redundancy and can improve estimation precision of the soft sensing model.

关 键 词：软测量 独立分量分析 互信息 支持向量机 

分 类 号：TH873.7[机械工程—仪器科学与技术]