Neural Network-based Adaptive Robust Control of a Class of.pdf

Proceedings of the American Control Conference Chicago, Illinois June 2000 Neural Network-basedAdaptive Robust Control of a Class of Nonlinear Systems in Normal Form * J. Q. Gong Bin Yao School of Mechanical Engineering Purdue University West Lafayette, IN 47907, USA (gong2, byao}@ecn .purdue .edu Abstract may limit the approximation capability of neural networks in practice since weights can only be tuned in a relatively In this paper, Neural networks (NNs) and adaptive robust small region. As such, researchers are still keeping on look- control (ARC) design philosophy are integrated to design ing for NN structures with less restrictive conditions for the performance oriented control laws for a class of n-th or- convergence of NN weights. Fortunately, when neural net- der nonlinear systems in a normal form in the presence of works are used for control design purposes, the main focus both repeatable and non-repeatable uncertain nonlinearities. is on the performance of the closed-loop system in terms of Unknown nonlinearities can exist in the input channel also. output tracking as long as all signals are bounded. Whether All unknown but repeatable nonlinearities are approximated or not the NN weights converge to their ideal values may not by outputs of multi-layer NNs. Discontinuous projection be the key issue. As such, the NN weights can be tuned in method with fictitious bounds is used to tune NN weights on-