基于EPF-MADDPG算法的多导弹机动策略研究  

Research on Multi Missile Maneuvering Strategy Based on MADDPG Algorithm

作　　者：聂文川 樊志强 NIE Wenchuan;FAN Zhiqiang(Intelligent Technology Research Institute,China Electronics Technology Group Corporation,Beijing 100083,China)

机构地区：[1]中国电科智能科技研究院,北京100083

出　　处：《计算机测量与控制》2024年第2期156-161,212,共7页Computer Measurement &Control

摘　　要：随着人工智能研究的进一步加深以及在俄乌战场上相关技术的大放异彩,其在军事领域扮演的角色越来越重要;针对日益复杂的战场环境,当前的导弹突防领域存在着信息维度高、指挥反应缓慢、突防机动战术不够灵活等问题;提出了一种基于多智能体深度确定性策略梯度(MADDPG)的训练方法,用以快速制定导弹攻击机动方案,协助军事指挥官进行战场决策;同时改进算法的经验回放策略,添加经验池筛选机制缩短训练的时长,达到现实场景中的快速反应需求;通过设置多目标快速拦截策略,仿真验证了所设计的方法能够突防的机动策略优势,通过协作智能地对目标进行突防打击,并通过比较,验证了该方法相较其他算法可以提升8%的收敛速度以及10%的成功率。In recent years,with the further deepening of artificial intelligence research and the shine of related technologies on the battlefield of Russia and Ukraine,it has become more and more important in the military field.In view of increasingly complex battlefield environment,current missile penetration field has problems such as high information dimension,slow command response,and inflexible penetration maneuver tactics.A training method based on multi-agent deep deterministic strategy gradient(MADDPG)is proposed to quickly generate missile attack maneuver schemes to assist commanders in making battlefield decisions.At the same time,the experience playback strategy of the algorithm is improved,and the experience pool filtering mechanism is added to shorten the training time and meet the rapid response requirements in real scenarios.By setting the multi-target rapid interception strategy,the simulation verifies that the maneuvering strategy advantages of the designed method can penetrate defense,intelligently and collaboratively strike the target.Compared with other algorithms,the method can improve the convergence speed of 8%and success rate of 10%.

关 键 词：多智能体 MADDPG 强化学习 协同机动突防 导弹机动 

分 类 号：V211[航空宇航科学与技术—航空宇航推进理论与工程]