Online optimal control for a class of nonlinear system using RBF neural networks

مرکز و کتابخانه مطالعات اسلامی به زبان های اروپایی

منو

" Online optimal control for a class of nonlinear system using RBF neural networks "
Peda Vasanta Reddy Medagam F. Pourboghrat

Document Type	:	Latin Dissertation
Language of Document	:	English
Record Number	:	52833
Doc. No	:	TL22787
Call number	:	‭3337340‬
Main Entry	:	Peda Vasanta Reddy Medagam
Title & Author	:	Online optimal control for a class of nonlinear system using RBF neural networks\ Peda Vasanta Reddy Medagam
College	:	Southern Illinois University at Carbondale
Date	:	2008
Degree	:	Ph.D.
student score	:	2008
Page No	:	74
Abstract	:	This dissertation presents an online optimal control and a modeling strategy for a class of nonlinear systems using radial basis function neural networks (RBFNN). Online optimal control technique is developed for known nonlinear systems with state measurements. Also, online modeling strategy with inverse optimal control is developed for unknown nonlinear systems with output measurements. For known nonlinear systems, optimal control has been derived using an approximate solution of the HJB equation. The HJB solution (value function) is approximated as the output of a radial basis function neural network (RBFNN) with unknown parameters (weights, centers and widths) whose inputs are the system states. The problem of solving the HJB equation is then converted to adjusting the parameters of an RBFNN. The RBFNN parameters adjustment is formulated as an estimation problem. An adaptive extended Kalman filter algorithm is then developed for estimation of parameters of the RBFNN. The proposed optimal control is then implemented in simulation studies on SISO nonlinear systems, such as DC-DC converters and permanent magnet synchronous motor (PMSM) drives. Next, an RBF neural network is developed for simultaneous modeling and state estimation of an unknown nonlinear system. The proposed nonlinear state estimation method is an augmentation of nonlinear Luenberger observer with RBF neural network. The learning rule for the neural network and the estimation of system's states are then viewed as design of a time varying Kalman filter. Generalized time-varying Riccati equation is derived for robust Kalman filter design. The stability of the corresponding neural network observer is shown using Lyapunov's method. Finally, inverse optimal control is proposed for the estimated model of the unknown system. Numerical examples show the effectiveness of the proposed method.
Subject	:	Applied sciences; Neural networks; Online control; Radial basis function neural networks; Electrical engineering; 0544:Electrical engineering
Added Entry	:	F. Pourboghrat
Added Entry	:	Southern Illinois University at Carbondale