Subbotin S. Methods of multy-dimensional nonlinear objects of control models synthesis based on the neural network approach.

The thesis is devoted to a development and research of model synthesis methods of multy-dimensional nonlinear objects of control based on the neural network approach. The new methods and software for a construction in a high-speed mode of logically transparent neural network models of objects of control are developed. It allows to automatize control and decision making processes in control systems. It also allows to extract knowledge from the data and simplify the decision-making analysis. The complex of criterions and the strategy of comparison of neural network models of objects of control and methods of their synthesis were offered. The experimental research of characteristics of developed methods and programs was conducted. The recommendations for their using in the practical problems of control are proposed.