A. I. Koldaev, A. V. Lavrinenko, S. S. Kirievskiy

The model of the neural predicting wavelet-filter is offered. Methods of threshold processing of high-frequency wavelet-coefficients are considered. Differences soft and hard threshold in processing signal with noise are shown. The scheme of training of the neural predicting wavelet-filter is resulted. Influence of value of thresholds is estimated at recalculation of high-frequency wavelet-coefficients. The approximating opportunities of a traditional multilayered network and the wavelet-neural network, used in the predicting filter are investigated. The results of experimental researches showing advantage in speed of a wavelet-neural network in comparison with multilayered neural network are resulted. The parameters of the wavelet -filter with the best parameter of forecasting are certain.