Karamysheva A. Information Technology of Predicting for Nonstationary Time Series in Control Problems.

The dissertation is devoted to the development of new information technology forecasting non-stationary time series in control problems. An analysis of prediction methods of nonstationary time series, which revealed the weaknesses associated with the inefficiency of existing methods for solving this problem, was held. The information technology that allows to perform nonstationary time series decomposition into similar clusters and in each of them to build a local model prediction is developed. The developed module to generate nonstationary time series with given statistical characteristics allows the choice of effective methods of clustering and models of local approximation by simulation. The models of nonstationary time series analysis by extending the set of dynamic characteristics are improved. Developed further the partitioning method of nonstationary time series into clusters, which unlike existing considers the dynamic indicators of time series, allowing the identity for each obtained clusters prediction models. Developed on the basis of scientific results information technology was used for predicting short-term consumption of gas in the system dispatching gas mains in Ukraine.