Modeling of disorders predictors for non-stationary time-series

In this work a methodology for modeling an ensemble of time series trajectories with random switches between a given set of states is formulated. An intermediate basis is introduced to predict the direction of the transition. A model of the transition predictor is proposed as the minimum distance from the current sample distribution to the intermediate baseline standards. The structure of a software package is proposed that allows for a comprehensive analysis of non-stationary time series, including finding a system of reference patterns depending on the length of the scanning sample.