摘要
针对分布式无人集群在动态复杂环境中因通信受限导致的局部信息传递效率低下、个体响应迟缓及全局协调困难等问题,提出一种融合复杂网络理论与多智能体强化学习的双层自适应编队框架。该框架构建通信层与结构层相结合的分布式网络模型,在通信层采用基于局部邻域信息的多智能体强化学习与去中心化策略优化,实现局部邻域信息下的高效信息共享与策略更新;结构层引入分布式拓扑重构机制,支持编队在受损恢复与任务拆分场景下的灵活调整与重构;同时在通信层与结构层间嵌入动态扰动处理机制,实现对节点失效、任务变化等事件的快速适应与重构。仿真实验表明,该方法在多种网络拓扑下的抗攻击恢复与编队拆分任务中均显著提升了任务成功率与收敛速度,具备较强的鲁棒性与适应性,为局部信息约束条件下无人集群的高效协同与稳定编队提供了有效途径与理论支持。
This paper proposed a dual-layer adaptive formation framework to address low local information transmission efficiency,slow individual response,and poor global coordination in distributed unmanned swarms under dynamic and complex environments with communication constraints.The framework built a distributed network model combining a communication layer and a structural layer.In the communication layer,it applied multi-agent reinforcement learning with decentralized policy optimization based on local neighborhood information to achieve efficient information sharing and policy updating.In the structural layer,it introduced a distributed topology reconfiguration mechanism to enable flexible adjustment and reconstruction of formations in damage recovery and task-splitting scenarios.It embedded a dynamic disturbance handling mechanism between the two layers to achieve rapid adaptation and reconfiguration for node failures and task changes.Simulation results show that the proposed method significantly improves task success rates and convergence speed in anti-attack recovery and formation-splitting tasks across multiple network topologies.The framework demonstrates strong robustness and adaptability,providing an effective approach and theoretical support for efficient collaboration and stable formation of unmanned swarms under local information constraints.
作者
崔梓涵
刘玮
谢宛真
胡棣威
Cui Zihan;Liu Wei;Xie Wanzhen;Hu Diwei(School of Computer Science&Engineering,Wuhan Institute of Technology,Wuhan 430205,China)
出处
《计算机应用研究》
北大核心
2026年第1期216-226,共11页
Application Research of Computers
基金
国家自然科学基金面上项目(52371373)
湖北省高等学校优秀中青年科技创新团队计划项目(T2023009)
武汉工程大学第十六届研究生教育创新基金资助项目(CX2024153)。
关键词
分布式无人集群
局部邻域信息
复杂网络
多智能体强化学习
去中心化策略优化
自组织编队
distributed unmanned swarm
local neighborhood information
complex networks
multi-agent reinforcement learning
decentralized strategy optimization
self-organizing formation
