两阶段的多元时间序列异常检测算法

Two-stage outlier detection in multivariate time series

提出了一个两阶段的多元时间序列异常检测算法。该算法通过有界坐标系统(BCS)技术计算多元时间序列样本之间的相似性,采用基于距离的方法实现异常检测。算法第一阶段采用K-means算法对数据进行聚类,并按照一个启发式规则对其进行排序;第二阶段在聚类结果上采用循环嵌套算法进行异常检测,并通过两个剪枝规则进行高效剪枝,提高了算法的效率。在两个实际数据集上进行实验,实验结果验证了算法的有效性。

This paper proposed an efficient two-stage algorithm for detecting outliers in multivariate time series （MTS） data- sets. Used the bounded coordinate system （BCS） metric to measure the similarity between two MTS samples, and measured the outlierness of a sample by average distance to its k-nearest neighbors. It partitioned the data into clusters, and used nested loop algorithm to find top-n outliers. Utilized a heuristic and two pruning rules to quickly remove MTS samples that were not possible outlier candidates, reducing significantly the distance computation among objects. Experiments on real-world datasets show the effectiveness of the proposed algorithm.