The hybrid genetic neuronet algorithm in the identification problem parameters of digital models

Abstract

In article it is offered hybrid genetic нейросетевой algorithm in which the radially-basic neural network carries out function approximation criterion function. The scheme of use of NANOSECOND as a part of HECTARE, the scheme of preliminary data processing before training of NANOSECOND and algorithm of training is described. Results of testing of the developed algorithm on difficult synthetic problems of optimisation, such as minimisation of functions of Rastrigin and Rozenbrok with various quantity of parametres are resulted.