摘要
经典PCA是一种对软测量建模数据进行误差侦破的方法,但当数据中存在单变量大误差时,该方法不能准确确定主元(PC),从而影响了误差侦破效果。针对这一情况,结合单变量误差侦破技术提出了3MAD-PCA方法。该方法首先用3MAD对数据分别进行单变量误差侦破,再利用经典PCA进行多变量误差侦破,提高了经典PCA方法的稳定性,有效实现了数据的过失误差侦破。用该方法对丙烯浓度的软测量数据进行过失误差侦破,取得了良好的效果。
Classical PCA is a method of detecting gross error for soft sensor modeling data. But this method performs badly when there is great error in the member univariate variance because the PCs (principal components) are not obtained precisely. A new method called 3MAD-PCA is presented combined with univariate detecting method. The new method firstly detects univariate error with 3MAD, then multivariate error is detected with classical PCA, as a result, the stability of classical PCA and detect gross error are improved effectively. The method is used to detect gross errors in modeling data for a propylene concentration soft sensor and good results are obtained.
出处
《计算机工程与设计》
CSCD
北大核心
2010年第1期184-186,194,共4页
Computer Engineering and Design
基金
国家863高技术研究发展计划重点基金项目(2006AA040308-02)