数据驱动的多智能体网络鲁棒包容控制
Data-driven robust containment control of multi-agent networks
摘要点击 71  全文点击 51  投稿时间:2019-06-09  修订日期:2020-08-02
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2020.90433
  2020,37(9):1963-1970
中文关键词  数据驱动  多智能体网络  积分增强学习  包容控制
英文关键词  data-driven  multi-agent networks  integral reinforcement learning  containment control
基金项目  北京信息科技大学学科群建设项目(5121911003), 国家自然科学基金项目(61903043)资助.
作者单位E-mail
于镝 北京信息科技大学 yudizlg@aliyun.com 
中文摘要
      针对输入受限的受扰多智能体网络, 提出具有领航层、估计层、控制层和跟随者层的新型鲁棒包容控制方 案. 首先, 设计有限时间估值器获得跟随者的期望状态, 然后基于包容误差引入非均方折扣代价函数, 从而将鲁棒包 容控制问题转换成受限最优控制问题. 并应用Laypunov拓展原理证明得到的最优控制策略使得网络实现一致最终 有界稳定. 在系统动态完全未知的情况下, 采用提出的积分增强学习算法和执行器–评价器结构, 在线得到近似最 优控制策略. 仿真结果验证了理论方案的有效性和可行性.
英文摘要
      A novel robust containment control scheme is proposed for multi-agent networks with constrained input, including leaders layer, estimation layer, control layer and followers layer. At first, finite time estimators are designed to obtain the desired states of followers. Then a non-quadratic discounted cost function is introduced based on the containment errors, so the robust containment control problem is transformed into a constrained optimal control problem. Moreover, the uniform ultimate bounded stability is verified of whole networks with obtained optimal control policy according to Lyapunov extension theorem. When the dynamics of followers are completely unknown, the approximate optimal control policy is obtained online applying the developed integral reinforcement learning algorithm and actor-critic architecture. Simulation results are provided to demonstrate the effectiveness of the proposed scheme.