摘要
多尺度主元分析方法(MSPCA)将多分辨率分析(MRA)的多尺度分解数据能力和主元分析(PCA)的降低数据维数能力结合起来,为监视多个时间尺度的过程提供了强有力的工具.过程监视时,MSPCA能自动对数据滤波并调节检测控制限,使控制限最容易检测出测量变量中重要的微小变化.由于MSPCA仅能滤除随机误差,不能消除过失误差,因此,为提高统计模型的准确性,在过程监视前,首先应用PCA检测并剔除存在过失误差的数据.通过实例说明MSPCA方法监视过程的优越性.
Multiscale principal component analysis combines the ability of multiresolution analysis(MRA) to decompose measurement data into separate time scales with that of PCA to reduce the dimensionality of data, which provides a powerful tool for monitoring processes with several time scales. In process monitoring, MSPCA automatically filters the data and adjusts the detection limits for easiest detection of important and tiny changes in the measurements. Since MSPCA can only eliminate random errors and can not eliminate gross errors, consequently, in order to enhance the veracity of statistical model, this paper first apply PCA to detect and eliminate the data with gross errors prior to process monitoring. The superior performance of MSPCA for process monitoring is illustrated by examples.
出处
《沈阳化工学院学报》
2003年第3期216-220,共5页
Journal of Shenyang Institute of Chemical Technolgy
