电子学报2026，Vol.54Issue(3)：1048-1061,14.DOI:10.12263/DZXB.20251147
基于多智能体强化学习的无人机自组织网络动态分簇算法
Dynamic Clustering Algorithm for UAV Ad Hoc Networks Based on Multi-Agent Reinforcement Learning
摘要
Abstract
With the development of swarm intelligence collaboration and network communication technologies,UAV ad hoc networks(UANETs)have demonstrated increasingly prominent advantages in mobility,flexibility,and envi-ronmental adaptability,and have been widely applied in both military and civilian scenarios.However,due to the large net-work scale,limited node energy,and highly dynamic topology of UANETs,traditional clustering algorithms often suffer from network instability and high energy consumption.To address these challenges,this paper proposes a dynamic cluster-ing algorithm for UANETs based on multi-agent reinforcement learning(MARL).First,a clustering strategy based on ener-gy and mobility constraints is designed.By jointly considering four factors,namely residual energy,node degree,inter-node distance,and link maintenance time,a dynamic weight calculation model tailored to the characteristics of UAV networks is constructed.During cluster formation,an evaluation mechanism for the clustering states of neighboring nodes is introduced.Each node performs cluster head election according to its own weight and the clustering states of its neighbors,and follows the cluster head based on link weights,thereby effectively reducing the number of isolated nodes.Furthermore,a distributed topology optimization mechanism based on independent Q-learning(IQL)is developed,where each node acts as an indepen-dent agent and interacts with the dynamic network environment to evaluate the reward of following different cluster heads.When the link quality to the original cluster head degrades,each node autonomously optimizes its cluster affiliation decision according to the accumulated reward,thus enhancing the robustness of the cluster topology and achieving balanced energy consumption optimization across the network.Experimental results demonstrate that,compared with existing baseline algo-rithms,the proposed algorithm achieves superior performance in terms of clustering efficiency,node energy consumption,and cluster stability,providing effective technical support for the reliable operation of UANETs in highly dynamic environ-ments.关键词
无人机自组网/动态分簇/多智能体强化学习/簇首选举/拓扑优化Key words
UAV ad hoc networks/dynamic clustering/multi-agent reinforcement learning/cluster-head election/to-pology optimization分类
信息技术与安全科学引用本文复制引用
武楠,魏时雨,张婷婷,方嘉睿..基于多智能体强化学习的无人机自组织网络动态分簇算法[J].电子学报,2026,54(3):1048-1061,14.基金项目
国家自然科学基金(No.62401047) National Natural Science Foundation of China(No.62401047) （No.62401047）
