计量学报2013，Vol.34Issue(1)：84-89,6.DOI:10.3969/j.issn.1000-1158.2013.01.19
过程控制异常值的在线检测方法研究
Method for Outlier Detection in Process Control Field
摘要
Abstract
Aiming at the characteristics of data in process industry which are large volume of data and on-line detection, an outlier detection algorithm which combines the improved RBF network and ARHMM is proposed. In the new algorithm, improved RBF network is used to model base on major data in kernel space, and then according to the residual errors,the detection results are made by kernel ARHMM. Forgetting factor and penalty factor are introduced by improved RBF network, which can make the algorithm more robust and accuracy. In order to avoid preselecting the detection threshold,KARHMM is used to detect outlier in process industry. The practicality is proved by experimentation and application,and through the comparison with AR model, it shows that the nonlinear KARHMM algorithm is more suitable for process data.关键词
计量学/过程数据/被控对象/异常数据检测/径向基函数网络/核自回归隐马尔可夫模型Key words
Metrology/Process data/Controlled objects/Outlier detection/RBF network/Kernel ARHMM分类
通用工业技术引用本文复制引用
刘芳,毛志忠..过程控制异常值的在线检测方法研究[J].计量学报,2013,34(1):84-89,6.基金项目
国家"863"计划项目(2007AA04Z194) （2007AA04Z194）
