Identification of non-Stationary Objects Based on Multi-Parameter Optimization Tools

Scientific and methodological foundations for optimal identification, recognition, classification, and forecasting of non-stationary objects have been developed based on the construction and application of new mechanisms for detecting and correcting information distortions through the use of statistical and dynamic characteristics. Optimization tools have been developed based on the use of an autoregressive model and the extraction of statistical parameters, correlation characteristics of random time series, as well as detection and correction of distorted information. Mechanisms for tracking optimization points are proposed based on a sliding window search and the use of correlation characteristics of the random time series. A technique for carrying out accelerated calculations of statistical parameters and correlation characteristics has been obtained. A method for searching for equivalent characteristics of the random time series identification model has been proposed, the mechanism of pseudo-gradient optimization has been studied, multiparameter analysis equations have been solved based on the backward sweep method, and their effectiveness has been tested on the basis of three-factor regression dependencies for a quasitropic process. A computational scheme has been developed for determining the equivalent relationship, solving a system of equations, inverting a three-diagonal matrix and applying the backward sweep method. A software package has been implemented in which functional modules are created in C++ in the parallel computing environment “CUDA”.