[author_cn_name].[cn_title][J].空军工程大学学报:自然科学版,[year_id],[volume]([issue]):[start_page]-[end_page] 基于Q-network强化学习的超视距空战机动决策-BVR Air Combat Maneuvering Decision by Using Q-network Reinforcement Learning
BVR Air Combat Maneuvering Decision by Using Q-network Reinforcement Learning
中文关键词: 超视距空战  机动决策  强化学习  纳什均衡
英文关键词: beyond visual range air combat  maneuvering decision  reinforcement learning  Nash equilibrium
张强,杨任农,俞利新,张涛,左家亮 空军工程大学空管领航学院,西安,710051 
摘要点击次数: 26
全文下载次数: 33
      In consideration of the great Impact of missiles on air combat, the continuous and multidimensional state space and the weakness of traditional approaches in ignoring opponent’s strategy in the air combat, reinforcement learning is applied to 1vs1 beyond visual range (BVR) air combat maneuvering decisions. Firstly, a new reinforcement learning framework is built to decide both sides’ maneuvers. In this framework,ε-Nash equilibrium strategy is proposed to choose action, and reward function is revised by missile attack zone scoring function. Then, by using a memory base and a target network, Q-network can be trained, forming a “value network” for BVR air combat maneuvering decisions. Finally,Q-network reinforcement learning model is designed, and the whole maneuvering decision is divided into learning part and strategy forming part. In the simulation, considering that the enemy in the air combat confrontation adopts a fixed maneuver and the two sides are both agents, the former agent wins, and the latter has the advantage of the situation to win, verifying that the agent can perceive the situation of air combat and make a reasonable BVR air combat maneuver.
查看全文   查看/发表评论  下载PDF阅读器