摘要： 为了在线检测并恢复数据流中的奇异数据，该文提出了一种新颖的能够适应数据流动态变化的奇异数据识别修正方法，基于卡尔曼滤波检测下一时刻的奇异数据，引入带有尺度导引的插值小波，根据流值变化的快慢程度确定插值小波的尺度，在不降低奇异数据恢复精度的情况下，恢复奇异数据。

关键词: 数据流, 卡尔曼滤波, 奇异数据检测, 奇异数据恢复

Abstract: To adapt to online detect and restore outliers from data streams efficiently, an online detecting and restoring method for outliers over data streams, called ADR (adaptive detecting and restoring), is proposed. It applies improved Kalman filtering with the amnesia factor to identify outliers at the future timestamp first. And it introduces interpolating wavelet directed by the resolution guider, which determines interpolating resolution based on change speed of data-values, to restore outliers. It adapts to the different requested precision for outliers repairing over evolving data streams well.

Key words: data streams, Kalman filtering, outliers detection, outliers restoring

中图分类号: 

• TP311