基于深度强化学习的多飞行器自适应协同航路规划  

Multi-aircraft adaptive cooperative route planning based on deep reinforcement learning

作　　者：杨志鹏 林松 曾长 陈子浩 毛金娣 徐哲 Yang Zhipeng;Lin Song;Zeng Chang;Chen Zihao;Mao Jindi;Xu Zhe(System Design Institute of Hubei Aerospace Technology Academy,Wuhan 430040,China;Xi'an Institute of Applied Optics,Xi'an 710065,China)

机构地区：[1]湖北航天技术研究院总体设计所,武汉430040 [2]西安应用光学研究所,西安710065

出　　处：《战术导弹技术》2024年第2期106-116,共11页Tactical Missile Technology

基　　金：国家自然科学基金(62003267)。

摘　　要：针对单一飞行器突防及作战能力受限的问题,提出一种多飞行器自适应协同航路规划方法。通过引入多智能体深度强化学习算法,构建多飞行器航路规划决策框架,对各飞行器在线航路规划指令进行求解。在此基础上,提出一种改良DL-MADDPG算法,引导智能体进行干扰学习,提升飞行器在复杂环境下的适应能力。同时,在奖励函数中分别设定协同奖励和个体奖励,有效保证多飞行器系统策略协同性和各飞行器个体策略有效性。仿真试验证明,提出的基于深度强化学习的多飞行器协同航路规划方法,具有很好的自适应性和鲁棒性,能够帮助多飞行器实现复杂多任务场景下的协同航路规划在线决策。Aiming at the problem of limited penetration and combat capability of a single aircraft,a multi-aircraft adaptive cooperative route planning method is proposed.By introducing the multi-agent deep reinforcement learning algorithm,the multi-aircraft route planning decision-making framework is constructed,and the online route planning instructions of each aircraft are solved.On this basis,an optimized DL-MADDPG algorithm is proposed to guide the aircraft for disturbance learning and enhance the adaptability of aircraft in complex environments.At the same time,the cooperative reward and individual reward are respectively set in the reward function,which can effectively ensure the cooperation of multi-aircraft system strategy and the effectiveness of individual strategy of each aircraft.Simulation results show that the proposed multi-aircraft cooperative route planning method based on deep reinforcement learning has good adaptability and robustness,and can help multi-aircraft to realize online decision-making of cooperative route planning in complex multi-mission scenarios.

关 键 词：多飞行器 协同航路规划 深度强化学习 干扰学习 复杂环境 自适应性 在线决策 

分 类 号：V448[航空宇航科学与技术—飞行器设计]