引用本文:武伟宁,刘小燕,徐学奎,金姣,张美.水泥熟料质量软测量模型中的时序分析方法[J].控制理论与应用,2018,35(7):1029~1036.[点击复制]
WU Weining,LIU Xiao-yan,XU Xue-kui,JIN Jiao,ZHANG Mei.Time series analysis method for the soft measurement of cement clinker quality[J].Control Theory and Technology,2018,35(7):1029~1036.[点击复制]
水泥熟料质量软测量模型中的时序分析方法
Time series analysis method for the soft measurement of cement clinker quality
摘要点击 3259  全文点击 1227  投稿时间:2017-07-23  修订日期:2017-12-06
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2017.70501
  2018,35(7):1029-1036
中文关键词  回转窑  水泥熟料  游离氧化钙含量  软测量  时序分析  支持向量机
英文关键词  rotary kilns  cement clinker  free calcium oxide  soft measurement  time series analysis  Support Vector Machines
基金项目  国家自然科学基金项目(61374149)资助.
作者单位E-mail
武伟宁 湖南大学  
刘小燕* 湖南大学 xiaoyan.liu@hnu.edu.cn 
徐学奎 湖南大学  
金姣 湖南大学  
张美 湖南大学  
中文摘要
      针对常规水泥熟料质量软测量模型未充分考虑过程变量间时序信息的问题,本文根据水泥生产流程工艺的时滞性、连续性特点,提出了一种时序分析方法,以提高熟料中游离氧化钙含量的软测量准确度。首先根据物料的传输机理,估算出物料在各工艺设备中的停留时间,在此基础上将软测量模型的输入输出变量进行时序匹配,之后采用类高斯函数对各输入变量匹配时刻前后的时序数据进行加权,获得具有时序信息的输入-输出样本对,然后以支持向量机模型结构为例,对引入时序信息后的游离氧化钙含量软测量模型进行训练和测试,并讨论了时序参数对结果的影响。采用某水泥生产线的实际过程数据对本文模型与方法进行验证,结果表明,该模型预测值与实际的游离氧化钙含量吻合良好,能正确预测其变化趋势;将本文模型测量结果与未使用时序信息的常规支持向量机模型对比,结果表明,过程变量间的时序信息有助于提高水泥熟料质量软测量模型的精度。本文提出的时序分析方法为水泥生产等流程工业过程建立软测量模型提供了新思路。
英文摘要
      A time-series analysis method is proposed to improve the accuracy of soft measurement models for predicting the content of free calcium oxide (an indicator for cement clinker quality), by considering the process features of the cement production, such as time-delay and continuousness that are important but less investigated in literature. First, the transport mechanism of the material in each process equipment is analyzed and the corresponding residence time is calculated, based on which a time-matching strategy for inputs and output is then developed. In order to obtain input-output samples with abundant time information, the time-series of the input variable are weighted near the matched point by use of the Gaussian-like function. The proposed soft measurement model with time-series information is then trained and tested by process data sampled from a cement production line, using Support Vector Machine as an example model structure. Results show that the proposed model can predict well both the qualitative trend and the quantity of the content of free calcium oxide. Compared with conventional Support Vector Machine, the time information buried in the process variables is demonstrated to be helpful for improving the accuracy of the soft measurement model for cement clinker quality. The proposed time-series analysis method provides new thoughts to soft measurement modeling for cement production process and other continuous industrial processes.