摘要
针对间歇过程独特的数据特点,提出了一种基于核独立元分析(kernelICA)的局部在线建模监控方法。核独立元分析通过规范相关性将比较函数扩展到一个再生的核希尔伯特空间,并用核的方法在此空间对比较函数进行计算和寻优。对含有多种分布的过程源数据,核独立元分析是一种比独立元分析(ICA)更有效的特征提取方法。对于按批次方向展开的间歇过程历史建模数据,在每一个时间间隔点应用核独立元分析算法提取独立元用于建模,并计算I2和SPE统计量及相应的控制限。此方法不需要对未来测量值进行估计,更重要的是解决了核独立元分析不能直接处理间歇过程高维历史建模数据的难题。仿真结果验证了所提出方法的可行性和有效性,并显示出比传统MICA更好的监控效果。
A novel batch process monitoring approach based on kernel independent component analysis（kernel ICA）and local modeling was proposed.Kernel ICA is an improved independent component analysis（ICA）.It uses contrast functions based on canonical correlations in a reproducing kernel Hilbert space,and uses kernel trick to compute and optimize the contrast functions in this space.Consequently,kernel ICA is more efficient than ICA for varying source distributions.For the batch-wise unfolding normal batch process data,kernel ICA was employed to extract independent components at every time interval,and calculate the I2 and SPE statistics and their control limits.This modeling method did not need to predict the future observations,and the more important was that it could handle the problem of kernel ICA with the high dimension historical normal data.The simulation results showed the feasibility and efficiency of the new method,and the advantages of kernel ICA over conventional MICA method.
出处
《化工学报》
EI
CAS
CSCD
北大核心
2010年第5期1183-1189,共7页
CIESC Journal
基金
上海市重点学科建设项目(B504)
化工过程先进控制和优化技术教育部重点实验室~~
