基于模糊神经网络在线自学习的多智能体一致性控制  

Multi-agent Consensus Control Based on Online Self-learning Fuzzy Neural Network

作　　者：张宪霞[1,2] 唐胜杰 俞寅生 ZHANG Xian-Xia;TANG Sheng-Jie;YU Yin-Sheng(School of Mechanical and Electrical Engineering and Automation,Shanghai University,Shanghai 200444;Shanghai Power Station Automation Key Laboratory,Shanghai 200444)

机构地区：[1]上海大学机电工程与自动化学院,上海200444 [2]上海市电站自动化重点实验室,上海200444

出　　处：《自动化学报》2025年第3期590-603,共14页Acta Automatica Sinica

基　　金：国家自然科学基金(62073210)资助。

摘　　要：针对多智能体系统分布式一致性控制问题,提出一种新的融合动态模糊神经网络(Dynamic fuzzy neural network,DFNN)和自适应动态规划(Adaptive dynamic programming,ADP)算法的无模型自适应控制方法.类似于强化学习中执行者-评论家结构,DFNN和神经网络(Neural network,NN)分别逼近控制策略和性能指标.每个智能体的DFNN执行者从零规则开始,通过在线学习,与其局部邻域的智能体交互而生成和合并规则.最终,每个智能体都有一个独特的DFNN控制器,具有不同的结构和参数,实现了最优的分布式同步控制律.仿真结果表明,本文提出的在线算法在非线性多智能体系统分布式一致性控制中优于传统基于NN的ADP算法.A novel model-free adaptive control approach integrating dynamic fuzzy neural network(DFNN)with adaptive dynamic programming(ADP)algorithm is introduced to address the distributed consensus control issue in multi-agent systems.Similar to the actor-critic structure in reinforcement learning,DFNN and neural network(NN)respectively approximate control strategies and performance metrics.The DFNN actor of each agent starts from zero rules and generates and merges rules through online learning,while interacting with the agents in its local neighborhood.Ultimately,each agent has a unique DFNN controller with different structures and parameters,achieving the optimal distributed synchronization control law.Simulation results show that the proposed online algorithm outperforms traditional ADP algorithm based on NN in distributed consensus control of nonlinear multiagent systems.

关 键 词：多智能体系统 自适应动态规划 动态模糊神经网络 分布式一致性控制 在线学习 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置] TP183[自动化与计算机技术—控制科学与工程]