文献类型：期刊文献

英文题名：Prescribed-time optimal tracking control for a class of stochastic systems using reinforcement learning

作者：Lin, Jie[1];Wang, Mengling[1];Yan, Huaicheng[1];Yang, Wen[1]

机构：[1]East China Univ Sci & Technol, Sch Informat Sci & Engn, 130 Meilong Rd, Shanghai 200237, Peoples R China

年份：2025

卷号：362

期号：13

外文期刊名：JOURNAL OF THE FRANKLIN INSTITUTE

收录：;EI(收录号：20252918808626);WOS:【SCI-EXPANDED(收录号:WOS:001555418900002)】；

基金：This work is supported by National Natural Science Foundation of China (No. 62333005) .

语种：英文

外文关键词：Prescribed-time control; Optimal control; Reinforcement learning; Neural networks

摘要：The study considers the issue of prescribed-time optimal tracking control for stochastic nonlinear systems with unknown dynamics. We introduce a constraint function that incorporates a new optimization function designed to adjust the convergence rate of the tracking error. By integrating this constraint function into the performance index with performance weights, we achieve optimal prescribed-time control. Using this performance index and applying Ito's lemma, we derive a new Hamilton-Jacobi-Bellman (HJB) equation that includes both diffusion terms and the constraint function. Given the complexity of solving this HJB equation analytically, we propose a reinforcement learning (RL) approach using neural networks to estimate the optimal control law and unknown dynamics. Moreover, considering that prescribed-time control for stochastic systems requires sufficiently fast state estimation, a novel adaptive identifier and its update law are developed, which ensure that the state estimation error converges to a predefined bound within the prescribed time. Finally, a Lyapunov function, incorporating the constraint function, is constructed to demonstrate that the proposed control strategy ensures convergence of the tracking error to a predefined range within a prescribed time. We validate the effectiveness of the proposed approach through a numerical simulation.