Adaptive nearly optimal control for a class of continuous-time nonaffine nonlinear systems with inequality constraints. (January 2017)

• Record Type: Journal Article
• Authors: Fan, Quan-Yong; Yang, Guang-Hong
• Is Part Of: ISA transactions. Volume 66(2017:Jan.)
• Page Start: 122
• Page End: 133
• Publication Date: 2017-01
• Subjects: Nonaffine nonlinear systems -- Inequality constraints -- Adaptive optimal control -- Neural network; Engineering instruments -- Periodicals; Engineering instruments; Periodicals; Electronic journals; 629.805
• DOI: 10.1016/j.isatra.2016.10.019
• Languages: English
• ISSNs: 0019-0578

Abstract

Abstract: The state inequality constraints have been hardly considered in the literature on solving the nonlinear optimal control problem based the adaptive dynamic programming (ADP) method. In this paper, an actor-critic (AC) algorithm is developed to solve the optimal control problem with a discounted cost function for a class of state-constrained nonaffine nonlinear systems. To overcome the difficulties resulting from the inequality constraints and the nonaffine nonlinearities of the controlled systems, a novel transformation technique with redesigned slack functions and a pre-compensator method are introduced to convert the constrained optimal control problem into an unconstrained one for affine nonlinear systems. Then, based on the policy iteration (PI) algorithm, an online AC scheme is proposed to learn the nearly optimal control policy for the obtained affine nonlinear dynamics. Using the information of the nonlinear model, novel adaptive update laws are designed to guarantee the convergence of the neural network (NN) weights and the stability of the affine nonlinear dynamics without the requirement for the probing signal. Finally, the effectiveness of the proposed method is validated by simulation studies. Abstract : Highlights: The optimal control problem for a class of state-constrained nonlinear systems is considered. A modified transformation technique is given to convert the constrained control problem. A suitable AC algorithm is designed without the requirement of a probing signal.