一种新的多变量时间序列数据异常检测方法

摘要

介绍了一种通过建立多变量时间序列数据相似度矩阵，对相似度矩阵进行转换以最大化数据之间的相关性，并采用随机游走模型计算数据点之间的连接系数来检测数据点上异常的方法。该方法充分利用了数据之间的相关性，有效减少了数据中不同程度噪声对异常检测的影响，检测过程中的漏报率和误报率明显减少，通过仿真实验验证了该方法的有效性。%A new method of detecting anomalies in MTS （multivariate time series） is introduced, in which a similarity matrix for MTS is set up and the similarity matrix is transformed to maximize the correlation between the data points and then the anomalous data points are detected by comparing the predefined threshold with the connectivity coefficient calculated through the random walk model. This detection method makes full use of the correlation between the data points and effectively reduces the influence of the noise. The omission rate and false alarms decrease obviously, and the simulation has tested and verified the validity of this method.