随机分组策略下的分布式多智能体一致性

Distributed multi-agent consensus via a novel randomized group partition approach

中文关键词  一致性  多智能体系统  分组信息  随机网络

英文关键词  consensus  multi-agent system  group information  randomized network

中文摘要

      多智能体分布式一致性算法一般需要获得相对状态差值xi ? xj , 本文针对无法得到智能体间相对状态差 值的情况, 提出一种基于智能体分组, 通过组间信息交换来达到智能体状态一致的算法. 本文仅讨论离散情况下智 能体被随机划分为两组和多组的情况. 当存在两个随机分组时, 每个智能体都进行状态更新, 且更新量为组间的状 态差值. 此时, 系统达到期望一致的充要条件为所给出的状态更新参数应大于1. 当存在多个随机分组时, 仅通过 Gossip算法选中的两组智能体以这两组间的状态差值进行状态更新. 在这种情况下, 系统达到期望一致的充分条件 为各分组概率相等, 且状态更新参数大于1. 最后通过计算机仿真验证了结论的正确性.

英文摘要

      Distributed multi-agent consensus algorithms generally require a relative state xi ?xj . This paper presents a novel distributed scheme to solve the consensus problem if no such inter-agent relative state is available. The new scheme allows the agents divided into two or more groups, and then uses the relative group representative state as control variable. When two groups are considered, each agent updates its own state based on the relative group representative state, and the system achieves consensus in mean if and only if a weighting parameter is greater than 1. When three or more groups are introduced, only two groups using Gossip algorithm are chosen to update the agents states inside based on the relative group representative state. Then consensus in mean can be realized if the partition probability for each group is equivalent, and also the weighting parameter is greater than 1. Numerical simulations are provided to demonstrate the validity of our results.