Neural adaptive inverse design for nonlinear systems with input unmodeled dynamics

A neural adaptive inverse compensator design method was proposed for a class of nonlinear systems with input unmodeled dynamics based on RBF neural networks. The compensator was designed using two neural networks, one to estimate the input unmodeled dynamics and another to provide adaptive inverse compensation to the input unmodeled dynamics. The method relaxes some rigorous demands to unmodeled dynamics such as relative degree zero, satisfying the small gain assumption and so on. The controller was designed using backstepping control techniques. Lyapunov theory was used to derive the tuning laws for the weight vectors of the neural networks and proved that the close-loop system is gradually stable. Simulation studies were included to demonstrate the effectiveness of the proposed method.