摘要
自从协同作战的概念提出后,各军事强国在协同空战领域均取得了重大进展,协同成为提升作战能力的倍增器。近数十年来,作为解决序列问题的现代智能方法,强化学习在各领域高速发展。然而,面对高维变量问题时,传统的单智能体强化学习往往表现不佳,多智能体强化学习算法为解决复杂多维问题提出新的可能。通过对多智能体强化学习算法原理、训练范式与协同空战的适应性进行分析,提出了协同空战与多智能体强化学习的未来发展方向,为更好地把多智能体强化学习应用于协同空战提供思路。
Since the concept of cooperative operation was put forward, all military powers have made great progress in the field of cooperative air combat, and coordination has become a multiplier to enhance combat capability.In recent decades, as a modern intelligent method to solve sequence problems, reinforcement learning has developed rapidly in various fields.However, in the face of high-dimensional variable problems, the traditional single-agent reinforcement learning often performs poorly.Multi-agent reinforcement learning algorithms provide new possibilities for solving complex multi-dimensional problems.By analyzing the adaptability of multi-agent reinforcement learning algorithm principle, training paradigm and cooperative air combat, the future development direction of cooperative air combat and multi-agent reinforcement learning is proposed, which provides ideas for better application of multi-agent reinforcement learning in cooperative air combat.
作者
谢育星
陆屹
管聪
纪德东
XIE Yuxing;LU Yi;GUAN Cong;JI Dedong(Shenyang Aircraft Design&Research Institute,Shenyang 110035,China)
出处
《飞机设计》
2023年第1期6-10,共5页
Aircraft Design
关键词
协同空战
多智能体强化学习
训练范式
集中式训练分布式执行(CTDE)
coordinated air combat
multi-agent reinforcement learning
training schemes
centralized training decentralized execution(CTDE)