Data-based self-learning optimal control for continuous-time unknown nonlinear systems with disturbance

Abstract

In this paper, a new data-based self-learning control scheme is developed to solve infinite horizon optimal control problems for continuous-time nonlinear systems. The developed optimal control scheme can be implement without knowing the mathematical model of the system. According to the input-output data of the nonlinear systems, a recurrent neural network (RNN) is employed to reconstruct the dynamics of the nonlinear system. According to the RNN model of the system, a new two-person zero-sum adaptive dynamic programming (ADP) algorithm is developed to obtain the optimal control, where the reconstruction error and the system disturbance are considered the control input of the system. Single-layer neural networks are used to construct the critic and action networks, which are presented to approximate the performance index function and the control law, respectively. Finally, simulation results will show the effectiveness of the developed data-based ADP methods.