Multilayer Neuroadaptive Reinforcement Learning of Disturbed Nonlinear Systems via Actor-Critic Mechanism
Guichao Yang
Abstract
In this study, a novel multilayer neuroadaptive reinforcement learning algorithm via the actor-critic mechanism will be proposed for tracking control of high-dimensional uncertain nonlinear systems. In order to compensate for the modeling uncertainties, the actor multilayer neural networks and the extended state observer based disturbance estimators will be incorporated via the command filtered backstepping to approximate endogenous uncertainties and external disturbances, respectively. Specifically, the critic multilayer neural network based reinforcement signals will be introduced to further improve the approximation performance of the actor multilayer neural networks and the tracking performance. The stability of the whole closed-loop system will be guaranteed by strict theoretical proof. Finally, the expected theoretical results will be verified.