智能空战深度强化决策方法现状与展望  

Status and Prospect on Deep Reinforcement Learning Decision-Making Methods for Intelligent Air Combat

作　　者：张烨[1] 涂远刚 张良 崔颢[2] 王靖宇[1] Zhang Ye;Tu Yuangang;Zhang Liang;Cui Hao;Wang Jingyu(School of Astronautics,Northwestern Polytechnical University,Xi’an 710072,China;China Airborne Missile Academy,Luoyang 471009,China)

机构地区：[1]西北工业大学航天学院,西安710072 [2]中国空空导弹研究院,河南洛阳471009

出　　处：《航空兵器》2024年第3期21-31,共11页Aero Weaponry

基　　金：国家自然科学基金青年项目(52202502);中央高校基本科研业务费(D5000210857)。

摘　　要：本文聚焦于现代智能空战决策技术的发展需求,分析了智能空战场景的要素与特点,介绍了现有智能空战决策理论的研究现状,包括基于博弈理论的决策方法、先验数据驱动的决策方法、基于自主学习的决策方法,着重梳理了基于价值和基于策略的深度强化学习智能决策方法。最后,面向未来智能空战面临的各种挑战以及传统深度强化学习的局限性,展望了深度强化学习技术在空战领域的发展方向:面向集群作战的多体智能决策技术、面向广域时空的高效智能决策技术、面向复杂场景的泛化智能决策技术。This paper focuses on the development of modern intelligent air combat decision-making technology,and analyzes the elements and characteristics of intelligent air combat scenarios.It introduces the research status and practical application of existing intelligent air combat decision-making methods,including decision-making methods based on game theory,prior data-driven decision-making method,and decision-making methods based on autonomous learning,and especially focuses on deep reinforcement learning intelligent decision-making methods based on value and strategy.Finally,facing to various challenges of future intelligent air combat and the limitations of traditional deep reinforcement learning,the paper gives the future development direction of deep reinforcement learning technology in the field of air combat,which are multi-agent intelligent decision-making technology for cluster warfare,efficient intelligent decision-making technology for wide area space-time,and generalized intelligent decision-making technology for complex scenarios.

关 键 词：空战决策 人工智能 强化学习 智能博弈 集群作战 深度学习 

分 类 号：TJ760[兵器科学与技术—武器系统与运用工程]