Recurrent neural networks for nonlinear output regulation

Based on a power-series approximation method, recurrent neural networks (RNN) are proposed for real-time synthesis and auto-tuning of feedback controllers for nonlinear output regulation systems. The proposed neurocontrol approach represents a novel application of recurrent neural networks to the nonlinear output regulation problem. The proposed approach completely inherits the stability and asymptotic tracking properties guaranteed by original nonlinear output regulation systems, due to its globally exponential convergence. Excellent operating characteristics of the proposed RNN-based controller and the closed-loop nonlinear control systems are demonstrated by using simulation results of the ball-and-beam system and the inverted pendulum on a cart system. © 2001 Elsevier Science Ltd.