基于改进尺度的CDC/MVT离群点去除算法  

Outlier Detection Algorithm Based on Modified Scaling CDC/MVT

作　　者：肖应旺[1] 姚美银[1] 

机构地区：[1]华南师范大学南海校区计算机工程系

出　　处：《系统仿真学报》2008年第23期6432-6437,共6页Journal of System Simulation

基　　金：科技部"863"项目子课题--饲用酶基因工程技术研究与新产品开发中的控制与优化部分(2003AA241160).

摘　　要：针对基于多向主元分析(MPCA)(包括PCA)的统计监控模型易受建模数据中离群点影响的不足,提出了一种基于改进尺度的CDC/MVT建模数据离群点去除算法。该算法将改进尺度的CDC与MVT相结合,首先利用改进尺度得到离线建模正常数据的均值和标准差,并对数据进行中心化和标准化处理;然后利用CDC算法找出建模历史数据中最一致的一半正常点;最后用这最一致的一半正常点初始化MVT的马氏距离的均值和协方差,并通过迭代计算得到其它的正常点。将该算法应用于β-甘露聚糖酶发酵批过程离群点的去除,与其它鲁棒离群点检测算法相比,应用结果表明该算法能有效地去除建模数据中的离群点。Because statistical monitoring model based on multi-way principal component analysis was strongly affected by outlying observations, a method of extracting outliers from the modeling historical database, based on modified scaling CDC/MVT （Closest Distance to Center/ellipsoidal Multivariate Trimming） was proposed The CDC was incorporated with MVT in the proposed algorithm. Firstly, the algorithm utilized the modified scale to obtain the mean and standard deviation of the off-line modeling normal data, and carried out centering and standardization for the modeling data using the mean and standard deviation. Secondly, the most consistent half observations were extracted from the modeling historical database by the algorithm of CDC. Finally, these observations were then used to initialize the mean and covariance of Mahalanobis distance, and the other observations were gotten by the iterative calculation of Mahalanobis distance. This proposed algorithm was applied to extract outliers from β -mannanase fermentation batch process and compared with the other robust outlier detection algorithms. The application results show that the proposed algorithm could effectively extract the outliers from the modeling historical database.

关 键 词：统计监控模型 离群点去除 改进尺度的CDC/MVT算法 β-甘露聚糖酶发酵批过程 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]