A neural-model based robust controller for nonlinear systems

Tools are provided that enable the analysis of robust stability for a particular nonlinear neural model-based control scheme, and the result enables robust synthesis as well. It is shown how an uncertainty description of an off-line trained neural network can be obtained and how this can be used to analyse robustness of the adopted control strategy. It turns out that, due to the uncertainty in the network, the closed-loop system becomes uncertain within a polytopic region. Stability of the closed-loop system can be proved by finding an appropriate Lyapunov function. Finding such a Lyapunov function can be rewritten as a LMI, which is tractable from a computational point of view. It is shown how the obtained uncertainty description of the closed-loop system allows robust synthesis of the controller, one of the main goals in robust control research.