Stable robust control of nonlinear systems with neural networks

In the paper, neural networks trained off-line are used to model a class of nonlinear systems. Based on the assumption that system modeling errors are bounded by a state-dependent function instead of a constant, we develop a stable neural robust control scheme for such nonlinear systems. Compared with using some existing methods, a controller designed using the proposed scheme is less expensive as it is not necessary to tune online. Robustness of the controller is established in the sense that the closed loop system is globally stable in the presence of modeling errors and external disturbances. The effectiveness of the scheme is demonstrated by simulations.