Output feedback control for nonlinear nonaffine discrete-time systems and application

A nonlinear output feedback controller is proposed in this paper based on neural networks. A compound neural network is constructed to identify the dynamic nonlinear system including single-input-single-output (SISO) system and multi-input-multi-output (MIMO) system. One part of the CNN is a linear feedforward neural network (LFNN), which as used to approximate the nonlinear system. The other part is a recurrent neural network, which can shorten the difference between the LFNN and the real nonlinear process. Because the current control input is not included in the input vector of RNN, the inverse control laws of SISO and MIMO nonlinear systems can be easily obtained by CNN approximated models. The computation work is small since no further training is required for the inverse controller. Simulation studies have shown that this scheme is simple and affects good control accuracy and robustness.