基于分布式自适应内模的多智能体系统协同最优输出调节  

Cooperative Optimal Output Regulation for Multi-agent Systems Based on Distributed Adaptive Internal Model

作　　者：董昱辰 高伟男[1] 姜钟平[2] DONG Yu-Chen;GAO Wei-Nan;JIANG Zhong-Ping(State Key Laboratory of Synthetical Automation for Process Industries,Northeastern University,Shenyang 110819,China;Department of Electrical and Computer Engineering,Tandon School of Engineering,New York University,New York NY 11201,USA)

机构地区：[1]东北大学流程工业综合自动化全国重点实验室,沈阳110819 [2]纽约大学坦登工程学院电子与计算机工程系,美国纽约NY 11201

出　　处：《自动化学报》2025年第3期678-691,共14页Acta Automatica Sinica

基　　金：国家自然科学基金(62373090);国家重点研发计划(2024YFA1012702)资助。

摘　　要：针对离散时间多智能体系统的协同最优输出调节问题,在不依赖多智能体系统矩阵精确信息的条件下提出分布式数据驱动自适应控制策略.基于自适应动态规划和分布式自适应内模,通过引入值迭代和策略迭代两种强化学习算法,利用在线数据学习最优控制器,实现多智能体系统的协同输出调节.考虑到跟随者只能访问领导者的估计值进行在线学习,对闭环系统的稳定性和学习算法的收敛性进行严格的理论分析,证明所学习的控制增益可以收敛到最优控制增益.仿真结果验证了所提控制方法的有效性.In this paper,a distributed data-driven adaptive control strategy is proposed for the problem of cooperative optimal output regulation of discrete-time multi-agent systems,in the absence of precise information of multiagent system matrices.Based on adaptive dynamic programming and distributed adaptive internal model,two reinforcement learning algorithms,value iteration and policy iteration,are introduced to learn the optimal controller by using online data,so as to achieve the cooperative output regulation of multi-agent systems.Considering that the followers can only access the estimated value of the leader for online learning,in order to prove that the learned control gain converges to the optimal control gain,this paper provides a rigorous analysis of the stability of the closed-loop system and the convergence of the learning algorithm.The simulation results verify the effectiveness of the proposed control method.

关 键 词：自适应动态规划 分布式自适应内模 强化学习 协同输出调节 多智能体系统 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]