Neural controllers for nonlinear systems using output feedback

In this paper, a neural controller is trained using the sign of the Jacobian for a certain class of nonlinear systems. A mathematical justification is developed to ensure the existence of a neural controller, which is fed by the output vector and its time derivatives according to a defined relative degree. Thus measuring the full states may be avoided. A training algorithm based on Marquardt-Levenberg algorithm is developed. The proposed technique has been applied to nonlinear systems to elaborate its efficiency.