A regulator design of dynamical systems with nonlinear uncertainties using multilayered neural networks

Two control systems without using state measurements are proposed for compensating nonlinear modeling uncertainties. The designed controllers contain neural networks which provides estimates of the unknown states. Using the learning and nonlinear mapping capabilities of neural networks, the proposed control systems can be shown to accommodate a wider class of modeling uncertainties than the conventional LQR. Numerical examples are given to compare the design methods.