摘要
针对间歇过程的建模问题,提出一种结合核独立成分分析(KICA)和稀疏核学习的软测量建模方法。首先将KICA应用于建模样本集,在高维空间提取输入变量的信息,以降低过程变量的相关性,再用稀疏核学习建立软测量模型。以估计链激酶流加发酵过程的活性菌体质量分数和链激酶质量分数为例,将基于KICA信息提取的稀疏核学习方法用于间歇过程的软测量建模。仿真结果表明,KICA信息提取能力优于传统ICA或核主元分析等其他方法,所提出的建模方法预报精度更高。
A comprehensive soft sensor modeling method using kernel independent component analysis （KICA） and sparse kernel learning （SKL） is proposed for batch process. The KICA is firstry adopted for acquiring information of input variables in the high dimensional feature space to reduce the correlation or process variables, and the model is established by using SKL. The KICA-SKL modeling method is applied to estimate the active biomass and streptokinase concentrations in a fed-batch streptokinase fermentation process. Compared to traditional ICA or kernel principal component analysis based on information extraction methods, the simulation result shows that the KICA has better information acquisition performance and the proposed modeling method can predict much more accurately.
出处
《石油化工自动化》
CAS
2012年第2期36-40,共5页
Automation in Petro-chemical Industry
基金
国家自然科学基金资助项目(61004136)
浙江省自然科学基金资助项目(Y4100457)
