SEQUENTIAL ALGORITHM FOR DISORDER DETECTION IN MULTIVARIATE TIME SERIES

The paper considers a set of issues related to the construction and sequential algorithms use for detecting spontaneous changes in multidimensional time series probabilistic characteristics (disorder). The study is motivated by the mathematical support problems for decision-making processes based on data from large systems multi-channel monitoring and is devoted to the analysis of the measurements multidimensional time series spatio-temporal dynamics. As an alternative to traditional approaches, new technologies for analyzing inter-channel communications are proposed. Dimension reduction technologies are used based on the data matrices presentation in the first singular basis and multiple regression in the projection space. The considered approach can be applied for interventions early detection in computer networks. The developed approach application in the analyzing the characteristics problem of a turbulent flow based on the pressure deviations measurement data at various points in the volume is demonstrated.