Title :
Robust training algorithm of multi-layered neural networks for identification of nonlinear dynamic systems
Author :
Song, Q. ; Grimble, M.J.
Author_Institution :
Sch. of Electr. & Electron. Eng., Nanyang Technol. Univ., Singapore
Abstract :
The plant used is a nonlinear difference equation with a bounded deterministic disturbance. Motivated by adaptive control systems a dead zone technique is used for the nonlinear gradient descent algorithm to train multilayered feed-forward neural networks for identification of the nonlinear-dynamic system. The dead zone scheme guarantees the convergence of the neural network in the presence of disturbance. Simulation results are presented to demonstrate robustness of the algorithm
Keywords :
conjugate gradient methods; convergence; difference equations; feedforward neural nets; identification; learning (artificial intelligence); multilayer perceptrons; nonlinear differential equations; nonlinear dynamical systems; adaptive control systems; bounded deterministic disturbance; convergence; dead zone technique; disturbance; multilayered feed-forward neural network training; multilayered neural networks; nonlinear difference equation; nonlinear dynamic system identification; nonlinear gradient descent algorithm; robust training algorithm; Adaptive control; Convergence; Difference equations; Feedforward neural networks; Feedforward systems; Multi-layer neural network; Neural networks; Nonlinear dynamical systems; Robustness; Vectors;
Conference_Titel :
American Control Conference, 1997. Proceedings of the 1997
Conference_Location :
Albuquerque, NM
Print_ISBN :
0-7803-3832-4
DOI :
10.1109/ACC.1997.609162
