System identification based on DNNs with disturbance observer and application to unmanned aerial vehicles

In this paper, dynamic neural networks (DNNs) are used as the on-line identifier for a class of nonlinear systems with unknown external disturbance and unknown multiple dead zone actuators. By integrating the novel nonlinear disturbance observer with adaptive control algorithms, the parameter coupling problem between unknown dead zone and DNNs can be successfully solved and the multiple disturbances can also be rejected simultaneously. Both the observation error and the identification error can be proved to convergent to zero. Furthermore, by combining with the numerical result of an unmanned aerial vehicle (UAV) model, the effectiveness of theoretical algorithms can be fully verified.