Neural-Network-Based Self-Learning Disturbance Rejection Design for Continuous-Time Nonlinear Constrained Systems

Abstract

In this paper, we adopt a neural-network-based adaptive dynamic programming (ADP) method to solve the disturbance rejection problem for continuous-time nonlinear systems with control constraints. First, we define a suitable function and then derive the optimal control law with control constraints and the worst disturbance law. Besides, the constrained Hamilton-Jacobi-Isaacs equation for continuous-time nonlinear constrained systems is derived. Then, only one critic neural network is used to approximate the optimal cost function. Consequently, the approximate optimal control law and the approximate worst disturbance law are obtained. Additionally, a new updating rule is developed in the process of neural critic learning. Finally, the simulation results show that the self-learning optimal control is realized and the effectiveness of the disturbance rejection is verified by using the neural-network-based ADP method.