An online outlier detection method for process control time series
Fang Liu, Mao Zhi-zhong · 2011
The ability to detect outlier online in process control filed is essential in many real-world system analysis applications. Previous algorithms require some ”clean” data to construct the statistical model at beginning, which was used to detect outlier. But actually, these clean data can not obtain at all. In this paper, we investigate a machine learning, descriptor-based approach that dose not require clean data to model, based on least square support vector outlier detection. A online window-based learn algorithm is introduced. Theoretical consideration as well as simulations on real process data demonstrate its practical efficiency.