FORECASTING OF NON-STATIONARY TIME SERIES BASED ON COGNITIVE ARTIFICIAL INTELLIGENCE TECHNOLOGIES

Mazurov Mikhail Efimovich, Slipchenko Aleksey Vyacheslavovich

The main properties of stationary and non-stationary time series are given. The properties of the basic ARIMA program for forecasting stationary time series are given. The main methods of forecasting non-stationary time series, which are most used at the present time, are described. These are the following forecasting methods: 1. Transformational methods of forecasting large time series; 2. Cognitive forecasting methods based on the use of cognitive maps; 3. Forecasting methods based on the use of a sliding window. An example of forecasting non-stationary time series on the stock exchange is given.

Keywords:forecasting, stationary, non-stationary time series, transformerical forecasting methods, cognitive methods, sliding window method.