引用本文:王长城,戚国庆,李银伢,盛安冬.非均匀选择概率下异步随机Gossip共识算法及优化[J].控制理论与应用,2013,30(3):299~306.[点击复制]
WANG Chang-cheng,QI Guo-qing,LI Yin-ya,SHENG An-dong.Asynchronous randomized Gossip consensus algorithm with nonuniform-selected probability and optimizing[J].Control Theory and Technology,2013,30(3):299~306.[点击复制]
非均匀选择概率下异步随机Gossip共识算法及优化
Asynchronous randomized Gossip consensus algorithm with nonuniform-selected probability and optimizing
摘要点击 3935  全文点击 2537  投稿时间:2012-06-11  修订日期:2012-10-08
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2013.20663
  2013,30(3):299-306
中文关键词  多智能体系统  非均匀选择概率  随机Gossip算法  一致性  优化
英文关键词  multi-agent system  nonuniform-select probability  randomized Gossip algorithm  consensus  optimizing
基金项目  国家自然科学基金资助项目(61104186, 61273076); 江苏省自然科学基金资助项目(BK2012801).
作者单位E-mail
王长城* 南京理工大学 自动化学院 w308101484@126.com 
戚国庆 南京理工大学 自动化学院  
李银伢 南京理工大学 自动化学院  
盛安冬 南京理工大学 自动化学院  
中文摘要
      异步随机Gossip算法大都采用以均匀选择概率为基础的时间模型, 并未充分考虑网络拓扑结构对智能体获取信息的影响, 为此本文提出了一种更为合理的基于非均匀选择概率的异步随机Gossip算法. 首先给出了非均匀选择概率下的异步时间模型, 在概率意义下分析了算法的收敛性. 算法的收敛速度取决于概率化权重矩阵的第2大特征值, 并利用投影次梯度算法给出了选择概率优化方法. 仿真分析表明, 在非均匀选择概率下可通过对各智能体选择概率的优化, 改善算法的收敛速度, 并且弥补了传统的通信概率矩阵优化方法受制于网络拓扑结构的不足.
英文摘要
      The traditional asynchronous randomized Gossip consensus algorithm is founded on the basis of the uniformselected probability time model which does not consider the impact of topology on local information transfer. We introduce a more reasonable asynchronous randomized Gossip consensus algorithm with nonuniform-select probability, and analyze the convergence of the algorithm in probability sense. The convergence rate depends on the second largest eigenvalue of the probabilistic weighted matrix. An optimization algorithm for selecting probabilities is proposed by projection subgradient method. The numerical example indicates that the algorithm proposed can improve the convergence rate by optimizing the selection of probabilities for agents, and compensates for the traditional algorithm in optimizing communication matrix the disadvantages of dependence on the network topology.