基于互信息和约减LSSVM的水泥生料细度软测量  

Soft sensor modeling of cement raw material fineness based on mutual information and Reduced LSSVM

作　　者：赵彦涛[1] 单泽宇 杨黎明 郝晓辰[1] Zhao Yantao;Shan Zeyu;Yang Liming;Hao Xiaochen(Institute of Electrical Engineering,Yanshan University,Qinhuangdao 066004,China)

机构地区：[1]燕山大学电气工程学院,秦皇岛066004

出　　处：《电子测量与仪器学报》2019年第12期173-182,共10页Journal of Electronic Measurement and Instrumentation

基　　金：河北省高等学校科学技术研究项目(QN2018083);河北省重点研发计划项目(19211602D)资助。

摘　　要：针对水泥生料粉磨工艺复杂难以解释,过程数据存在偏斜和冗余性、与生料细度映射关系难以描述等问题,提出了基于互信息(mutual information,MI)和约减最小二乘支持向量机(least square support vector machine,LSSVM)的软测量算法,在此基础上建立了水泥生料细度软测量模型。该方法首先从机理上分析生料细度的影响因素,并在此基础上采用互信息方法分析变量间的相关性,得到输入变量;然后基于KS算法对样本数据进行约减,采用最小二乘支持向量机的方法建立水泥生料细度软测量模型。最后应用某水泥厂立磨的实际运行数据进行仿真,验证了该方法的有效性。The grinding process of cement is complex and difficult to interpret. Moreover, the useful information contained in these process data is skew and redundant, and the mapping relationships between these process data and cement raw material fineness are difficult to describe. In order to deal with these problems, a soft sensor algorithm based on mutual information(MI) and reduced least squares support vector machine(LSSVM) is proposed to establish the soft sensor modeling of cement raw material fineness. At first, the influence factors of cement raw material fineness are analyzed based on the production mechanism, the mutual information method is adopted to analyze the correlation between variables, and the input variables are obtained. Then, sample data are reduced based on KS algorithm, and soft sensor model of cement raw material fineness is established based on least squares support vector machine. In the end, the simulation experiment is carried out with the actual operation data of a vertical mill in a cement plant. The simulation results verified the effectiveness of the proposed approach.

关 键 词：互信息 特征选择 样本约减 最小二乘支持向量机 水泥生料细度 软测量建模 

分 类 号：TQ172[化学工程—水泥工业] TN911[化学工程—硅酸盐工业]