海军航空大学学报2025，Vol.40Issue(4)：528-538,11.DOI:10.7682/j.issn.2097-1427.2025.04.003
基于改进多智能体强化学习的大规模无人机集群博弈对抗
Large Scale UAV Cluster Adversarial Game Based on Improved Multi-Agent Reinforcement Learning
摘要
Abstract
In response to the problems of poor performance and difficulty in obtaining effective strategies in multi-agent reinforcement learning when the number of intelligent agents is too high and the environment is complex,combines cur-riculum learning training strategies and adopts multi-agent deep deterministic policy gradient(MADDPG)algorithm,the MADDPG based on Curriculum Learning(CL-MADDPG)algorithm framework is constructed.The innovation is the con-struction of a simulation environment that can accommodate large-scale unmanned aerial vehicle(UAV)for adversarial game,and the application of curriculum learning strategies to improve the MADDPG algorithm,thereby enhancing the ef-fectiveness of the adversarial strategy.The simulation results show that compared with traditional algorithms,the CL-MADDPG algorithm has significantly improved training convergence rate,average reward value,and adversarial victory rate.Moreover,the more UAV clusters between the two sides engage in adversarial attacks,the more significant the strate-gic superiority of the CL-MADDPG algorithm training.关键词
无人机集群/博弈对抗/多智能体强化学习/课程学习Key words
unmanned aerial vehicle swarm/adversarial game/multi-agent reinforcement learning/curriculum learning分类
信息技术与安全科学引用本文复制引用
孙世彬,王庆领..基于改进多智能体强化学习的大规模无人机集群博弈对抗[J].海军航空大学学报,2025,40(4):528-538,11.基金项目
国家自然科学基金(62373102) （62373102）
江苏省自然科学基金(BK20221455) （BK20221455）
安徽省重点研究与开发计划(2022i01020013) （2022i01020013）
