基于DDPG算法的导弹末制导律设计  

The Design of Missile Terminal Guidance Law Based on DDPG Algorithm

作　　者：侯贤乐 李辉[1,2] 王壮 吴昭欣 文瀚 Hou Xianle;Li Hui;Wang Zhuang;Wu Zhaoxin;Wen Han(College of Computer Science,Sichuan University,Chengdu 610065,China;Nation Key Laboratory of Fundamental Science on Synthetic Vision,Sichuan University,Chengdu 610065,China)

机构地区：[1]四川大学计算机学院,成都610065 [2]四川大学视觉合成图形图像技术国防重点学科实验室,成都610065

出　　处：《战术导弹技术》2021年第4期110-116,共7页Tactical Missile Technology

摘　　要：传统比例导引律缺少时间和视场角的限制,在面对现代空战目标机动快的情况下,拦截效果也不理想。针对该问题,提出了基于深度强化学习的智能导弹控制律设计。该算法以导弹当前态势为状态输入,考虑到导弹追击过程中时间和视场角的限制,以此设计奖励函数,在连续的状态空间和动作空间中,导弹能根据当前态势自主选择正确的制导指令,及时追上机动目标。实验表明,该算法相比传统比例导引律,具有更强的鲁棒性,同时,导弹也具有了自主决策的能力。The traditional proportional navigation law lacks the limitation of time and field angle, and the interception effect is not ideal in the case of fast maneuvering of modern air combat targets. To solve this problem, an intelligent missile control law designed based on deep reinforcement learning is proposed. This algorithm takes the current situation of the missile as the state input, and takes into account the limitations of time and field angle in the pursuit process of the missile, so as to design the reward function. In the continuous state space and action space, the missile can independently choose the correct guidance instructions according to the current situation and catch up with the maneuvering target in time. Experiments show that the algorithm is more robust than the traditional proportional guidance law, and the missile has the capability of autonomous decision.

关 键 词：导弹导引律 深度强化学习 智能化 视场角 自主决策 

分 类 号：V448.133[航空宇航科学与技术—飞行器设计]