Output feedback control for discrete-time nonlinear systems and its applications

A compound neural network (CNN) which includes a linear feed-forward neural network (LFNN) and a recurrent neural network (RNN) is constructed to identify nonaffine dynamic nonlinear systems. Because the current control input is not included in the input vector of the recurrent neural network, output feedback control laws of nonlinear systems can be easily obtained from one-step predictive models approximated by the CNN. To minimize the predictive error, the current approximation error is used in the predictive process. The computation work is small because no on-line training is required for the output feedback controller. This algorithm can be used to SISO and MIMO nonlinear system control in real time. Simulation studies have shown that this scheme is simple and has good control accuracy and robustness.