基于改进离群算法的多元时间序列异常检测

Multivariate time series anomaly detection based on improved outlier algorithm

为了进一步提高多元时间序列异常检测的性能与质量,提出一种基于改进离群算法的多元时间序列异常检测方法。这种方法首先通过主成分分析对每个多元时间序列数据进行特征表达;其次,通过夹角公式计算每个多元时间序列的特征矩阵与其正交坐标系之间的距离,并利用匈牙利算法得到对多元时间序列的最小距离;最后,构建基于最小距离的离群算法来实现多元时间序列异常检测。通过选取多元时间序列标准数据集对所提方法进行实验验证,结果表明,所提方法可以对多元时间序列异常状态进行准确识别,并且提高了实验的检测率,同时降低了实验的误报率和漏报率。

In order to further improve the performance and quality of multivariate time series anomaly detection,a multivariate time series anomaly detection method based on improved outlier algorithm is proposed.Firstly,characteristic representation of each multivariate time series data is carried out by principal component analysis.Secondly,the distance between the characteristic matrix of each multivariate time series and its orthogonal coordinate system is calculated by the included angle formula,and the minimum distance to the multivariate time series is obtained by Hungarian algorithm.Finally,the outlier algorithm based on the minimum distance is constructed to realize the multivariate time series anomaly detection.The proposed method is experimentally verified by selecting the standard data set of multivariate time series.The results show that the method can accurately identify the abnormal state of multivariate time series,improve the detection rate of the experiment,and reduce the false positive rate and false negative rate.