一种基于核独立元分析的非线性过程监控方法

独立元分析(ICA)在线性过程监控中得到了成功的应用,但是实际工业过程大部分是非线性的.在利用核ICA(KICA)建立过程非线性模型的基础上,根据核技巧,给出了一种高维空间分离矩阵的排序和独立元个数的选择方法,并将监控指标扩展到高雏空间,从而提出-种基于KICA的非线性过程监控方法,解决了ICA对非线性过程监控效果不理想的缺点.以田纳西一伊斯曼过程(TE过程)为例,对比了KICA与ICA的监控效果,结果证明了该方法的优越性.

Nonlinear Process Monitoring Method Based on Kernel Independent Component Analysis

Independent component analysis (ICA) has been successfully used in the linear processes monitoring,however,most of real industrial processes are nonlinear. Based on kernel ICA (KICA) model,the main contributions are as follows: firstly,a method to sort the rows of demixing matrix was introduced and the number of independent components was determined; secondly,the monitoring indices were extended into high-dimensional space by "kernel trick",and so a nonlinear process monitoring method was proposed. The proposed method avoids the disadvantages of the ICA-based method used in nonlinear process. At last,taking the monitoring of Tennessee-Eastman process (TE process) for example,KICA was compared with ICA,and the results prove the superiority of the proposed method.