An input tracking control system using a forward model on neural network

A design method for an input tracking control system using a neural network is proposed. A multi-layered neural network is first trained to model the input-output mapping of a controlled object of an unknown, discrete-time nonlinear system. This neuro emulator is used for generating Jacobian information with respect to the controlled object. The Jacobian coefficients were generated by the multi-layered neural network which had been, in advance, trained to map the input-output relation of the controlled object. The potential method was then applied for generating the input sequences under constructing an asymptotically stable gradient system. The tracking property was assured in the sense of linearization approximation of the controlled object. A simple control law is determined by Lyapunov´s method so that the output of the controlled object can follow a reference input. Simulation results show the effectiveness of the proposed method