数据丢包下事件驱动的非线性多智能体迭代学习控制
Event-triggered iterative learning control for nonlinear multi-agent systems with data random packet dropouts
摘要点击 432  全文点击 102  投稿时间:2021-09-07  修订日期:2022-06-26
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2022.10849
  2022,39(9):1688-1698
中文关键词  非线性多智能体系统  事件驱动通信  迭代学习控制  随机链路故障  一致性
英文关键词  nonlinear multi-agent systems  event-triggering communication  iterative learning control  random link failures  consensus
基金项目  国家自然科学基金项目(61863034)资助.
作者单位E-mail
王宏伟 大连理工大学 控制科学与工程学院 wanghw@dlut.edu.cn 
李昊哲 新疆大学 电气工程学院  
中文摘要
      针对具有随机链路丢包、通信带宽受限以及模型未知的非线性多智能体一致性问题, 提出一种事件驱动的分布式无模型迭代学习控制策略. 首先建立系统的事件驱动决策机制, 给出基于输出信息的通信触发条件, 当该条件满足时触发事件, 各智能体间进行通信, 不满足条件时则不通信, 从而能够有效减少智能体间的大量通信和能量耗散. 其次, 使用伪偏导数将非线性系统沿迭代轴动态线性化, 借助邻居在前一步事件触发时的输出信息设计随机链路丢包补偿机制, 再结合事件驱动通信机制设计分布式控制协议. 在此基础上, 使用压缩映射原理分析算法收敛性能, 仿真结果表明随着迭代次数的增加, 事件触发间隔变大, 所有的智能体将完成对期望轨迹的跟踪.
英文摘要
      An even-driven distributed model-free iterative learning control strategy is proposed to solve the consensus problem of nonlinear multi-agent with random link packet loss, limited communication bandwidth and unknown dynamics. Firstly, the event-driven decision-making mechanism of the system is established, and the communication trigger condition based on output information is given. When the condition is met, the event is triggered, and the agents communicate, and when the condition is not met, the agents do not communicate, which can effectively reduce a large amount of communication and energy dissipation between agents. Secondly, the pseudo partial derivative is used to dynamically linearize the nonlinear system along the iterative axis, the random link packet loss compensation mechanism is designed with the help of the neighbor’s output information when the previous event is triggered, and then the distributed control protocol is designed combined with the event-driven communication mechanism. The convergence performance of the algorithm is analyzed by using the principle of compressed mapping. The simulation results show that with the increase of iteration times, the event trigger interval becomes larger, and all agents will complete the tracking of the desired trajectory.