An adaptive predictive control for flexible-link manipulators

An adaptive predictive control strategy for a class of nonlinear non-minimum phase system using neural network is proposed in this paper. The control strategy is based on weighted d-step ahead controller. No a priori knowledge about the nonlinearity of the system is assumed. Two separate neural networks are employed. One neural network (N-p) is used to predict the output of the system while another network (N-c) generates a proper input to minimize the tracking error. The networks are all trained and employed as online controllers and no off-line training is required. The stability of the closed loop system is shown assuming the system nonlinearity is approximated by the neural network. Simulation results for a single-link flexible manipulator is also presented.