首页|期刊导航|空天预警研究学报|基于多智能体近端策略优化的动态频率指配方法

基于多智能体近端策略优化的动态频率指配方法OA

Dynamic frequency assignment method based on multi-agent proximal policy optimization

中文摘要英文摘要

针对复杂电磁环境下多设备动态频率指配场景中集中式决策实时性差、可扩展性有限的问题,提出基于多智能体近端策略优化(MAPPO)的动态频率指配方法(DFAM).该方法将每台装备建模为独立智能体,采用集中式训练、分布式执行框架,通过设计融合信干比、频率抖动与干扰协调的奖励函数,引导智能体协同避开同频和邻频干扰.仿真结果表明,DFAM的频率指配成功率可达98.6%,收敛速度较集中式DQN提升近一倍,每回合计算耗时降低约40%,且在干扰环境下保持强鲁棒性.

Aiming at the problems of poor real-time performance and limited scalability of centralized deci-sion-making in the dynamic frequency assignment of multiple equipment under complex electromagnetic environ-ments,this paper proposes a dynamic frequency assignment method(DFAM)based on multi-agent proximal poli-cy optimization(MAPPO).In this method,each piece of equipment is modeled as an independent agent,and cen-tralized training and decentralized execution framework are adopted.And then,by designing a reward function in-tegrating signal-to-interference ratio,frequency jitter,and interference coordination,the agent is guided to collab-oratively avoid co-frequency and adjacent frequency interference.Simulation results demonstrate that the DFAM achieves an assignment success rate of 98.6%,nearly doubles the convergence speed compared to centralized DQN,reduces per-round computation time by about 40%,and maintains strong robustness under interference.

彭姣丽

湖南环境生物职业技术学院 生态宜居学院,湖南 衡阳 421001

信息技术与安全科学

动态频率指配多智能体强化学习近端策略优化频谱管理

dynamic frequency assignmentmulti-agent reinforcement learningproximal policy optimiza-tionspectrum management

《空天预警研究学报》 2026 (4)

293-298,6

2025年度衡阳市指导性计划项目(57)

10.3969/j.issn.2097-180X.2026.04.011

评论