Direct adaptive output tracking control using multilayered neural networks

Multilayered neural networks are used to construct nonlinear learning control systems for a class of unknown nonlinear systems in a canonical form. An adaptive output tracking architecture is proposed using the outputs of the two three-layered neural networks which are trained to approximate the unknown nonlinear plant to any desired degree of accuracy by using the modified back-propagation technique. A weight-learning algorithm is presented using the gradient descent method with a dead-zone function, and the descent and convergence of the error index during weight learning are shown. The closed-loop system is proved to be stable, with the output tracking error converging to the neighbourhood of the origin. The effectiveness of the proposed control scheme is illustrated through simulations.