A new method of training direct neuro-controllers

A new method of training neuro-controllers for nonlinear plants is proposed. Using this controller does not require identification of the plant or its inverse model. Direct inverse controllers discussed in the literature require the Jacobian of the plant or the sign of the Jacobian which may not be available for an unknown plant. The authors approximate the output of the plant with the output of its reference model in a model reference model in an adaptive control scheme and use the Jacobian of the reference model instead of the plant. Simulation results show a satisfactory performance.<>