Methods and algorithms of selection the informative attributes in systems of adaptive data processing for analysis and forecasting

The principles, methods and algorithms of informative attributes selection were developed for optimization of description and representation for the objects in systems of adaptive data processing, where data are non-stationary by nature. The proposed algorithms of informative attributes selection for one-dimensional time series are based on the simplified ratings of correlation, mathematical expectation, dispersion of attributes. The algorithms have been developed using dynamic properties of information, i.e. control of conditional moments and parameters of regression model on confidential intervals. The methods and algorithms of adaptive data processing were tested on an example of extensive data represented for forecasting in systems of electro-supply.