摘要
实际生产中往往存在对产品质量影响重大、但又难以在线测量的一些参数(只能离线测量与分析)。本文提出一种基于聚类的多模型建模方法对这些难以在线测量的参数实现软测量,将相关性分析、主元分析(PCA)、聚类和多模型建模应用于软测量建模中,构建一种实现重要参数软测量的基本框架:首先,基于相关度分析进行辅助变量的选择,然后用主元分析进行数据的进一步降维,再用k-means聚类与多模型建模思想相结合。最后将提出的思想和方法应用于某精馏塔组分的软测量中,仿真结果表明,测量精度有了较大的提高。
Some parameters which had much effect on product quality but were difficult to measure on line,often existed in practical production.In this case,the modeling method of multi-model soft-sensing based on clustering was represented,which aimed at the measurement of this kind of parameters.The correlation analysis,principle component analysis（PCA）,clustering algorithm and multiple modeling were applied to the model and basic framework for important parameter soft-sensing was built.First,the secondary variables were selected based on correlation analysis.Secondly,PCA was introduced to make the dimension of data reduced further.Then a multi-model modeling method was represented,combined with k-means clustering algorithm.Finally,the proposed method was successfully used to the soft-sensing modeling of composition in butadiene distillation.The simulation results show that the output precision is increased approvingly.
出处
《化工自动化及仪表》
CAS
北大核心
2010年第1期49-52,55,共5页
Control and Instruments in Chemical Industry
