Neuro-control of Nonlinear Systems with Unknown Input Constraints

This paper establishes an adaptive dynamic programming algorithm based neuro-control scheme for nonlinear systems with unknown input constraints. The control strategy consists of an online nominal optimal control and a neural network (NN) based saturation compensator. For nominal systems without input constraints, we develop a critic NN to solve the Hamilton-Jacobi-Bellman equation. Hereafter, the online approximate nominal optimal control policy can be derived directly. Then, considering the unknown input constraints as saturation nonlinearity, NN based feed-forward compensator is employed. The ultimate uniform bounded stability of the closed loop system is analyzed via Lyapunov’s direct method. Finally, simulation on a torsional pendulum system is provided to verify the effectiveness of the proposed control scheme.