Reinforcement Learning-Based Predefined-Time Tracking Control for Nonlinear Systems Under Identifier-Critic–Actor Structure
针对存在外部干扰的非线性系统，提出一种基于强化学习的预定时间跟踪控制方案，通过标识器-评判者-执行者框架学习未知动态，并整合预设性能控制，确保跟踪误差快速收敛且收敛时间可调。
A novel reinforcement learning-based predefined-time tracking control scheme with prescribed performance is presented in this article for nonlinear systems in the presence of external disturbances. First, by employing the backstepping strategy, an adaptive optimized controller is developed under the identifier-critic-actor framework. Therein, the unknown nonlinear dynamics and the system control behavior can be learned effectively through neural networks. Moreover, aiming at obtaining the preset tracking performance, the prescribed performance control is integrated with the predefined-time control. In contrast to previous studies, the proposed scheme can not only constrain the tracking error rapidly to a prearranged vicinity of origin, but also ensure that the upper bound of convergence time can be adjusted in advance via a separate control parameter. In terms of the predefined-time stability theory, the boundedness of all system states can be proven within a predefined time. Finally, the availability and improved performances of the proposed control scheme are demonstrated by a numerical example and a single-link manipulator example.