Tkachova T. Synthesis of neural network methods for identificatin of complicated dynamic objects

This thesis deals with the solution of a neural network identification of nonlinear dynamic objects based on an artificial neural networks (ANN) under conditions of prior and current ambiguity about the properties of objects studied and disturbances acting on them. This work presents methods of description of nonlinear objects, the main types of ANN were analysed for identification problem. A neural network realization of a Hammerstein model were synthesizes with the help of radial-basic network or general-regressive network. Results achieved are generalized into two classes of models (an additive model and a multiplicative model), that allows building mathematical models of many-dimensional nonlinear objects. Neural network adaptive observers of nonlinear dynamic objects described by the models in the state space are suggested and were proposed a quite simple procedures of teaching. Approaches to building neural network models of nonlinear dynamic objects, suggested in an imitation modeling of developed methods and procedures. The results of this thesis are confirmed by experimental research and results of its introduction.