Adaptive neural optimal tracking control of stochastic nonstrict-feedback nonlinear systems with output constraints

Abstract

In this article, a novel adaptive optimal control method is proposed for constrained stochastic nonlinear systems in the form of nonstrict-feedback. The output constraints are handled by employing nonlinear mapping, and the controller is composed of feedforward and optimal parts. The feedforward controller is designed based on dynamic surface control technology with Gaussian properties, where a dynamic signal and linearly parametrized neural networks are employed to approximate the unknown dynamics and unavailable nonlinear functions, respectively. Using the mean value theorem and Young’s inequality, only one parameter must be online tweaked at each step. In the optimal part, the cost function is approximately obtained using adaptive dynamic programming for the optimal control law. All signals in the closed-loop systems are analytically proved to be semi-globally uniformly ultimately bounded in the probabilistic space. Numerical simulations are provided to verify the effectiveness of the proposed approach.