间歇过程在线检测的核独立元分析及局部建模方法

Kernel ICA and Local Modeling for Online Batch Process Detecting

针对间歇过程所具有的非线性特点,提出了一种基于核独立元分析(Kernel ICA)及局部建模的在线故障检测策略.将展开的高维历史数据按时间间隔划分,应用Kernel ICA算法对每一个时间间隔点的分数据块提取独立主成分,构造检测统计量,并用非参数估计方法确定其控制限.这种建模方法机理简单,而且不需要预测过程未知的测量数据.通过对DuPont间歇聚合过程的仿真,验证了所提出方法的有效性和准确性.

To deal with the nonlinear characteristics of batch processes,a new batch process monitoring strategy based on kernel independent component analysis(Kernel ICA)and local modeling was proposed.Firstly,the high dimension unfolding history process data were divided according to time,then Kernel ICA was employed to extract independent components in every time interval,and the monitoring statistics were constructed.In addition,confidence limits of them were calculated by using kernel density estimation.The modeling method was simple and could avoid predicting the future observations.The proposed monitoring method was applied to the DuPont batch polymerization process.The feasibility and accuracy of the proposed new method were validated by the simulation results.