基于知识粒度的时间序列异常检测研究

Research on Time Series Anomaly Detection Based on Knowledge Granularity

时间序列的异常检测多以相似性分析方法来处理,时间代价高昂。为减少异常检测的时间,文中围绕知识粒度方法进行研究与探讨。知识粒度在数据异常检测中应用广泛,但在时间序列的异常检测上应用较少。文中针对时间序列上下文相关异常(点)检测,提出利用知识粒度异常检测方法对于输入属性越多检测粒度越细的特性,来查找时间序列中的异常数据。实验证明,基于知识粒度的方法无需先验信息,在整个处理过程中无需事先分析历史数据,而是通过属性间的组合粒度来划分异常数据与正常数据,提高了异常检测的效率。知识粒度方法在不确定信息处理研究中的表现十分突出,文中将知识粒度在时间序列异常检测中进行应用尝试,为时间序列异常检测提供了一种新的思路。

Most of the time series＇ anomaly detections are processed with the similarity analysis,and their time complexity is rather high.In order to reduce the time of anomaly detection,it studies and discusses the method of knowledge granularity in this paper. Knowledge granularity is widely applied in the anomaly detection of data,but rarely used in anomaly detection on time series. In viewof context dependent anomaly（ point） detection in time series,the knowledge-granularity-based anomaly detection is proposed to search the anomalous data in time series,in which the more the attributes are,the finer the detection granularity is. Experiments showthat the method based on knowledge granularity does not require a priori information,partition of the abnormal data and normal data through the combination of the attributes without analysis of historical data previously,and the efficiency of anomaly detection has been improved. The knowledge granularity method is very prominent in the research of uncertain information processing. It tries to apply the knowledge granularity in the anomaly detection of time series in this paper,thus to provide a newapproach for anomaly detection of time series.