Identification and control of nonlinear systems using neural networks and multiple models

In this paper, a multiple generalized NARMA-L2 model is proposed for the identification and control of discrete nonlinear systems. It provides a global input-output representation for nonlinear systems by making use of the good local approximation property of NARMA-L2 model without encountering the curse of dimensionality problem. With the identified model, the control problem is then transformed into a constrained optimization problem based on the weighted one-step-ahead predictive control law. Simulation studies demonstrate the effectiveness of the proposed model structure.