基于改进ADPP的多变量时间序列异常检测
Outlier detection based on improved ADPP for multivariate time series

针对多变量时间序列异常检测问题进行研究，提出基于改进ADPP的多变量时间序列异常检测算法IADPP. IADPP算法引入适用于多变量时间序列的张量相似性度量 S SOTPCA，并以此相似性度量构造序列集的 k －近邻图，在构造的 k －近邻图上计算多变量时间序列的异常系数. 研究结果表明，IADPP算法克服了原有ADPP算法不支持多变量时间序列和要求密度均匀的缺陷，取得了较好的检测结果.
We study the outlier detection for multivariate time series，and an approach of outlier detection for multivariate time series based on improved ADPP-IADPP is proposed. IADPP algorithm introduces tensor similarity measure S SOTPCA supporting for multivariate time series，and constructs the k － neighbor graph about the sequence set. Then，we calculate the outlier coefficient of multivariate time series on k －neighbor graph .The research results show that the proposed method overcomes the disadvantages that original ADPP does not support multivariate time series and requests uniform density，IADPP algorithm achieves a better detection results