论文部分内容阅读
多变量时间序列(MTS)在金融、医学、科学、工程等领域是非常普遍的.本文提出一种在 MTS 中识别异常模式的方法.采用自底向上的分割算法将 MTS 分割成互不重叠的子序列,使用扩展的 Frobenius 范数来计算2个MTS 子序列之间的相似性,通过 K-均值聚类将 MTS 子序列分为若干个类.根据异常模式的定义,从这若干个类中识别出异常模式.在2个实际数据集上进行实验,实验结果验证算法的有效性.
Multivariate time series (MTS) is very common in the fields of finance, medicine, science, engineering, etc. In this paper, a method of identifying anomalous patterns in MTS is proposed. The bottom-up segmentation algorithm is used to segment the MTS into non-overlapping Subsequence, using the extended Frobenius norm to calculate the similarity between two MTS subsequences, dividing the MTS subsequence into several classes by K-means clustering.According to the definition of the abnormal pattern, from the several classes Anomaly patterns were identified.Experiments were performed on two actual datasets to verify the effectiveness of the algorithm.