摘要
针对基于多向主元分析(Multi-way Principal Component Analysis,MPCA)(包括主元分析(Principal Component Analysis,PCA)的统计监控模型易受建模数据中离群点影响的不足,通过对各种不同尺度的中心化和标准化方法及鲁棒离群点检测算法的对比研究,提出了一种基于改进尺度的中心最短距离/椭球多变量整理(Closest Distance to Center/ellipsoidal Multivariate Trimming,CDC/MVT)的建模数据离群点去除算法。该算法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他的正常点。将该算法应用于β-甘露聚糖酶发酵间歇过程离群点的去除,与其他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 （ including principal component analysis） is strongly affected by outlying observations, by the comparative research for centralization plus standardization approaches with various scale and different robust outlier detection algorithms, a method of extracting outliers from the modeling historical database, based on modified scaling CDC （Closest Distance to Center, CDC） /MVT （ellipsoidal Multivariate Trimming, MVT） is proposed. 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 observa- tions 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 Mabalanobis 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 showed that the proposed algorithm could effectively extract the outliers from the modeling historical database.
出处
《控制工程》
CSCD
北大核心
2013年第4期756-761,共6页
Control Engineering of China
基金
2011年度国家自然科学基金项目-基于RMKMFDA的间歇过程多元统计监控研究(61174123)
2009年度广东省自然科学基金项目课题-基于多元统计方法的过程监控研究(9151063101000043)
2009年度科技部"863"项目课题-高温瓦楞状陶瓷基换热器的开发研究(2009AA05Z203)
关键词
鲁棒离群点检测算法
多元统计监控建模
数据预处理
β-甘露聚糖酶发酵间歇过程
robust outlier detection algorithm
multivariate statistical monitoring modeling
data preprocessing
β-mannanase fermentation batch process
