博弈学习系统在战斗机引导中的应用  

Application of game learning system for fighter guidance

作　　者：陶伟[1] TAO Wei(China Ship Development and Design Center,Wuhan 430064,China)

机构地区：[1]中国舰船研究设计中心,湖北武汉430064

出　　处：《中国舰船研究》2020年第S01期166-172,共7页Chinese Journal of Ship Research

摘　　要：[目的]为通过深度强化学习探索预警机对战斗机的引导策略,设计预警机(EWA)引导战斗机任务的博弈学习系统。[方法]该系统包括深度学习智能体、与智能体交互的战场仿真系统、博弈管理系统和分布式训练系统。针对强化学习博弈需要与环境进行大量交互的问题,在自博弈训练平台上加入分布式训练系统以提高训练效率。在分布式系统中,提出分布式训练系统中Actor和Learner解耦、各训练Learner之间定期分享更新梯度,以及择优保存并剔除无效智能体等新机制。[结果]通过该博弈学习系统,实现了红、蓝深度强化学习智能体在预警机引导战斗机任务中的博弈,获得了具有更优异表现的对抗策略,提升了预警机引导作战的能力。[结论]所做研究可为提升预警机引导作战能力提供参考。[Objectives]In order to explore the guidance strategy of early warning aircraft(EWA)for fighters through deep reinforcement learning,a game learning system for fighter guidance based on EWA is presented.[Methods]The game learning system includes a deep learning reinforcement agent,battleground simulation system which can interact with the agent,game management system and distributed training system.For reinforcement learning requires significant interaction with the environment,a distributed training system is introduced to the self-training game platform to improve training efficiency.In the distributed system,the new mechanisms include decoupling the Learner and Actor,the periodic sharing of update gradients among training learners,and selecting the best agents while eliminating invalid agents.[Results]Through the game learning system,a better EWA guidance strategy can be obtained after games between the blue agent and red agent,thereby enhancing the guidance operational capability of EWA.[Conclusions]This paper provides a reference for improving the guidance combat capability of early warning aircraft.

关 键 词：预警机 战斗机引导 深度强化学习 博弈 

分 类 号：U662.9[交通运输工程—船舶及航道工程]