Cognitive technologies for analyzing multidimensional time series with changing properties

The paper considers multivariate time series with structural breaks (disorders). A structural break is an unobservable sharp change in the coefficients of the autoregressive model due to the heterogeneity of the patterns of behavior of the time series. To obtain an indicator of changes, a transition to the cognitive space of states is carried out, represented by a structured set of knowledge displayed by fuzzy states of a dynamic system specified on linguistic scales. A technique for the transition from a numerical time series to a random sequence of fuzzy states is presented. The dynamics of fuzzy states is modeled using the Kolmogorov–Chapman equation. The calculated indicator (eigenvector of the transition matrix) allows segmenting the time series into homogeneous sections and obtaining autoregressive models with constant coefficients on these sections. It is proposed to aggregate the results of autoregressive models with constant coefficients by a model with variable coefficients based on the Sugeno model. The approach under consideration allows solving the problem of modeling a time series with implicit changes in properties and helps improve the quality of decision-making based on the obtained model.