Neural-network-based robust critic learning control with advanced value iteration for continuous-time dynamical systems.

In this paper, a novel neural-network-based robust critic learning method is developed with advanced value iteration (VI) to design the controller for continuous-time nonlinear systems. The initial admissible control law is eliminated by employing the developed scheme, which significantly relaxes the starting conditions for iteration. Additionally, the approximation error is analyzed to provide guidelines for parameter selection, thereby facilitating implementation of the constructed algorithm. By introducing a relaxation factor, the convergence process is accelerated leading to the better performance than the traditional VI algorithm. The convergence of the algorithm and the stability of the system are proved in detail. Finally, three examples are presented to verify the effectiveness of the established approach.