Methods and algorithms of adaptive designing for neuronetworking system of processing the data with non-stationary nature

The hybrid model of data neuronetworking processing system, in which the opportunities of a wide spectrum of methods and algorithms of neural network and statistical models are combined, is researched. The original ways and algorithms of adaptation during designing of neural network are developed by escalating, selection and adjustment of activation functions. The strategy and new principles are developed for neural network output quality control. The theoretical results were approved on the basis of Koxonen model; and as result conditions of well over reduction of output data approximation error are proved.