A New Neural Network Control Architecture for a Class of Nonlinear Dynamic Systems

Neural network control strategies are considered for a class of nonlinear time varying systems. Neural networks have been shown to be viable alternatives to present day controllers for nonlinear systems. This paper reports a novel neural network architecture for the control of a class of nonlinear systems. This architecture incorporates both feedforward and feedback components using multiple networks. Implementation in simulation shows that the architecture is capable of modeling and `learning´ the nonlinear system characteristics more efficiently with very good control characteristics when compared with two other designs even for varying operating points.