Data-driven optimal control for a class of unknown continuous-time nonlinear system using a novel ADP method

Abstract

This paper is concerned with the optimal control problem for a class of unknown continuous-time nonlinear system. A system identification method by date-driven model is established to reconstruct the unknown system dynamic by the input-output data. Then considering the optimal control problem, a novel critic neural networks design is proposed based on the policy iteration (PI), where the updating laws of parameters are designed by the normalized gradient descent algorithm and convex optimization method. And the computational burden of cost error get reduced during the iteration procedure using the new method. Based on this adaptive dynamic programming algorithm, the weight convergence is obtained and stability is guaranteed by Lyapunov theory. Finally, two simulation examples are shown to verify the effectiveness of this novel method.