计算机科学
同步(交流)
多智能体系统
对偶(语法数字)
强化学习
方案(数学)
控制(管理)
控制理论(社会学)
控制系统
钥匙(锁)
作者
Lina Xia,Qing Li,Ruizhuo Song,Choon Ki Ahn
标识
DOI:10.1109/tsmc.2026.3740703
摘要
This article investigates reinforcement learning (RL)-based synchronization of heterogeneous multiagent systems (MASs) over directed graphs using a dual dynamic event-triggered approach. The proposed framework has two principal advantages: 1) it learns the solutions of high-dimensional algebraic Riccati equations (AREs) without requiring global graph information; and 2) it guarantees a positive minimum interevent time (MIET) for each of the two dynamic triggering mechanisms. First, linear matrix equations are used to construct auxiliary systems and agentwise state-error systems, resolving the dimension mismatch between the leader and followers. A critic-only neural network (NN) is then integrated with two dynamic triggering laws to obtain optimal event-triggered synchronization control. Lyapunov analysis establishes closed-loop stability and yields positive MIETs for both triggering channels. Finally, heterogeneous linear systems and physically parameterized DC-motor examples demonstrate the effectiveness and resource efficiency of the proposed method.
科研通智能强力驱动
Strongly Powered by AbleSci AI