A stable adaptive neural-network-based scheme for dynamical system control

A stable adaptive neural-network-based control scheme for dynamical systems is presented and a continuous recurrent neural network model of dynamical systems is constructed in this paper. A novel algorithm for updating weights in the neural network, which is not derived from the conventional back propagation algorithm, is also constructed. The proposed control law is obtained adaptively by a continuous recurrent neural network identifier, but not by a conventional neural network controller. In such a way, the stability in the sense of the Lyapunov stability can be guaranteed theoretically. The control error converges to a range near the zero point and remains within the domain throughout the course of the execution. Numerical experiments for a longitudinal vibration ultrasonic motor show that the proposed control scheme has good control performance.