Semi-empirical neural network models of controlled dynamical systems

Современные информационные технологии и IT-образование

Mihail V. Egorchev¹  Yury V. Tiumentsev¹ 

[1] Moscow Aviation Institute (National Research University), Russia;

关键词: Nonlinear controlled dynamical system;    semi-empirical model;    neural network based modeling;    training dataset;    aircraft;    model of motion;    identification of aircraft aerodynamic characteristics;   

DOI  :  10.25559/SITITO.2017.4.410

来源: DOAJ

【 摘 要 】

A simulation approach is discussed for maneuverable aircraft motion as nonlinear controlled dynamical system under multiple and diverse uncertainties including knowledge imperfection concerning simulated plant and its environment exposure. The suggested approach is based on a merging of theoretical knowledge for the plant with training tools of artificial neural network field. The efficiency of this approach is demonstrated using the example of motion modeling and the identification of the aerodynamic characteristics of a maneuverable aircraft. A semi-empirical recurrent neural network based model learning algorithm is proposed for multi-step ahead prediction problem. This algorithm sequentially states and solves numerical optimization subproblems of increasing complexity, using each solution as initial guess for subsequent subproblem. We also consider a procedure for representative training set acquisition that utilizes multisine control signals.

【 授权许可】

Unknown