一种基于深度强化学习的机动博弈制导律设计方法  

A Design Method of Maneuvering Game Guidance Law Based on Deep Reinforcement Learning

作　　者：朱雅萌 张海瑞[1] 周国峰[1] 梁卓[1] 吕瑞 Zhu Yameng;Zhang Hairui;Zhou Guofeng;Liang Zhuo;Lv Rui(China Academy of Launch Vehicle Technology,Beijing 100076,China)

机构地区：[1]中国运载火箭技术研究院,北京100076

出　　处：《航天控制》2022年第3期28-36,共9页Aerospace Control

摘　　要：针对高速机动飞行器常用的程序化机动突防方式适应性不强、突防效果不稳定的问题,提出了一种基于深度强化学习算法的机动博弈制导方法。该方法以增大交会摆脱量为任务目标,采用深度神经网络拟合飞行器的制导律,应用强化学习方法训练网络参数,得到一种以突防拦截双方的位置和速度为输入、以飞行器的需用过载为输出的智能机动博弈制导律。数学仿真验证结果表明,在连续的状态空间和动作空间中,飞行器能根据当前态势自主选择合适的制导指令。相比传统突防方式,该制导律显著提升了交会摆脱量,且突防效果更稳定。Aiming at the problems of weak adaptability and unstable penetration effect of programmed maneuvering penetration methods commonly used by high speed maneuvering aircraft,a maneuvering game guidance method based on deep reinforcement learning is proposed.By using this method,the encounter escape quantity is taken as the mission objective,deep neural network is used to fit the guidance law,reinforcement learning is used to train the network parameters and a intelligent maneuvering game guidance law is obtained with the position and velocity of both sides as the input and the demand overload of the aircraft as the output.The simulations show that the appropriate guidance instruction can be chosen by the aircraft according to the current situation in the continuous state space and action space.Compared with the traditional penetration methods,the encounter escape quantity can be significantly improved by using this guidance law,and its penetration effect is more stable.

关 键 词：飞行器 机动突防 交会摆脱量 深度强化学习 智能制导 

分 类 号：V279[航空宇航科学与技术—飞行器设计]