摘要
针对离散多智能体系统输出调节,提出了一种基于Q学习的最优控制策略。对于传统多智能体系统的输出调节,获取系统的精确动力学模型并依此求得其HJB方程的解为主要障碍。该策略通过智能体之间的局部通信,在不依赖系统动态模型的前提下实现了对每个智能体输出的全局最优控制。为实现对系统响应速率的优化,提出了一种新的有限时间局部误差公式,不仅保证了算法原有的全局最优性能,而且将输出同步时间缩短了近50%,并对所提算法的稳定性进行了分析。仿真结果表明,该策略在避免建立复杂系统模型和求解离散HJB方程的前提下实现了对系统的最优控制,采用更新后的有限时间局部误差公式有效缩短了收敛时间。
This paper proposed an optimal control strategy based on Q-learning for output regulation of discrete multi-agent systems.For the output regulation of the traditional multi-agent system,obtaining the exact dynamics model of the system and solving the HJB equation based on this is the main obstacle.The strategy achieved global optimal control of each agent′s output without relying on the dynamic model of the system through local communication between agents.In order to optimize the system response rate,this paper proposed a new finite time local error formula,which not only ensures the original global optimal performance of the algorithm,but also reduced the output synchronization time by nearly 50%.It analyzed the stability of the proposed algorithm.The simulation results show that this strategy achieves the optimal control of the system without building the complex system model and solving the discrete HJB equation,and the updated finite time local error formula can effectively shorten the convergence time.
作者
唐静远
魏文军
Tang Jingyuan;Wei Wenjun(School of Automation&Electrical Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China)
出处
《计算机应用研究》
CSCD
北大核心
2023年第1期204-208,共5页
Application Research of Computers
基金
光电技术与智能控制教育部重点实验室开放课题(KFKT2020-11)。
关键词
离散多智能体系统
Q学习
协同输出调节
快速收敛
discrete multi-agent system
Q-learning
cooperative output regulation
fast convergence
