基于滑动窗口的时间序列异常检测方法

Anomaly Detecting Method for Time Series Based on Sliding Windows

针对传统滑动窗口异常检测(anomaly detection for sliding windows)中的子序列特征不能准确反映数据结构特征的问题,采用子序列斜率置信区间的方式进行解决,并提出了基于滑动窗口的时间序列异常检测方法。通过滑动窗口法将时间序列进行初始分割,提取子序列斜率的置信区间距离半径用于异常子序列的识别,并最终通过Gath-Geva聚类算法完成异常值与正常值的划分。仿真数据集检测结果表明,与以方差信息和传统斜率信息的特征提取方式相比,提出方法的查全率分别提升6.9%和46.3%。工程数据的检测实验结果表明,提出的算法能够准确识别异常数据信息,查全率和查重率都达到84%以上,验证了提出方法的工程可用性。

Aiming at the problem that the sub-sequence features in traditional anomaly detection for sliding windows cannot accurately reflect the characteristics of the data structure,the sub-sequence slope confidence interval method was used to solve the problem,and an anomaly detecting method for time series based on sliding windows was proposed.The time series were initially divided by sliding windows method,the confidence interval distance radius of the slope of the subsequence was extracted for the identification of abnormal subsequences,and finally the outlier and normal value were divided by the Gath-Geva clustering algorithm.Experimental results of simulation data set show that the recall rate of this method is improved by 6.9%and 46.3%respectively compared with the feature extraction method based on variance information and traditional slope information.Experiments results of engineering data show that the proposed algorithm can accurately identify abnormal data information,and the recall rate and precision rate are above 84%,revealing the engineering usability of this proposed method.