Design and analysis of one-step-ahead adaptive neural predictive controller

This paper aims at developing an adaptive one-step-ahead neural predictive control scheme for controlling nonlinear unknown systems. We here focus on two issues: controller design and performance analysis. A regularization cost function embodying the plant sensitivity information is proposed, which enhances the estimation accuracy of the plant Jacobian. The local linearization concept is applied to controller design. The control law is expressed explicitly with a variable parameter which may compensate for the error effect caused by the linear approximation in the neural model. The convergence theorem for neural predictor is obtained by the Lyapunov approach. The BIBO stability analysis of the closed-loop control system is also established.