Gasanov A. Information technology modeling and forecasting of non-stationary processes based on multilevel integration. – Qualifying scientific work on the manuscript.

Information technology modeling and forecasting of non-stationary processes based on multilevel integration. – Qualifying scientific work on the manuscript. Dissertation for the degree of doctor of technical sciences, specialty 05.13.06 – Information technology. – Taras Shevchenko National University of Kyiv, Ministry of Education and Science of Ukraine, Kyiv, 2018. The dissertation work is focused on the solution of urgent scientific and applied problems of identification, modeling and forecasting of linear and nonlinear systems constructed on the basis of integration of methods and means of mathematical modeling and advanced forecasting models. The new information technology of modeling and forecasting of non-stationary processes on the basis of multilevel integration of models, methods, algorithms and software is offered which provides high adequacy of models and quality of the forecast. The approach to the automated selection of the best model of investigated process from the point of view of adequacy of parameters of model and quality achievement of estimations of forecasts which are calculated on the basis of the comparative analysis of candidate models by means of integrated criterion of quality is offered. The technique of construction of models for heteroskedastic processes, different by simplicity of realization and possibility of automation of models construction process on the basis of time series with time changing variance is offered. For revealing the heteroscedasticity the auxiliary regression model for squares of the residuals on which basis of which existence of heteroskedasticity is carried out is constructed. It allows estimating and predicting a variance changing under the complicated law. The problems of forecasting the conditional variance for one and more steps are considered. The modified methodology of modeling the time series was developed on the basis of which the new advanced technique of construction the model for error correction for two or more non-stationary time series. Thanks to this methodology there is a possibility to reach long-term balance between non-stationary macroeconomic processes. On the basis of developed models and algorithms of forecasting the information decision support system was constructed and offered. On the basis of the constructed models an analytical forecasting function has been received allowing calculating the values of the forecast for several steps ahead. The comparative analysis of the forecasting methods using data for macroeconomic indicators of Ukraine was performed and as a result the most effective method of forecasting was defined and practically applied. The problems of modeling the process of rolling pipes are considered. Advantages and disadvantages of estimation of pipes diameter using the methods of the Group Method of Data Handling (GМDH) and the generalized variable are considered. The error analysis of forecasting by these methods has shown that the errors are approximately identical. However the method of the generalized variable is simpler in realization and which advantage is also in the best adaptive properties and it does not demand difficult calculations in comparison with fuzzy polynomial GМDH. The mathematical model of inflation process in the form of autoregression model with moving average, constructed on the basis of the statistical data and high degree of adequacy is developed. Specific features of real processes with seasonal effects are studied and their modeling and forecasting are considered. The method of a combination of the forecasts received by different methods taking into account variance of estimates of forecasts and allowing to raise quality of forecasting models is offered. Examples of use of the developed methods led to modeling and forecasting of processes in hen house, and also three projects of sea crafts on the purpose of a choice of the best project. Algorithms and computing schemes of identification of objects with Kalman filter use linear and nonlinear models for adaptive definition and updating of mathematical models of the objects described by various functional dependences are considered. The method of forecasting the dynamics of a time series on the basis of sharing of a linear Kalman filter and a likelihood relation is offered. The method is used for forecasting of dynamic systems and faults of their functioning. The given method differs by high reliability of detection of refusals that is confirmed by computer modeling with use of real data.