One-on-one air combat control method based on situation assessment and DDPG algorithm

Abstract

The existing aerial combat control methods do not comprehensively consider the situation assessment based on expert knowledge and the control of aerial combat through continuous speed change. Based on the deep deterministic policy gradient(DDPG) reinforcement learning algorithm, a comprehensive reinforcement learning environment is designed that considers flight altitude limits, flight overload and flight speed limits, which is building upon the situation evaluation function as the reward function for reinforcement learning. The interaction between the DDPG algorithm and learning environment is achieved through the fully connected carrier speed control network and the environment reward network. The end condition for air combat is designed based on abnormal height and speed, missile lock time and combat time. By simulating one-on-one air combat, the effectiveness of this combat control method is validated in terms of learning under environmental constraints, situation evaluation scores and combat mode learning. The results show that the air combat control method is effective, and can provide guidance for the further development of autonomous air combat.

Keywords

• reinforcement learning
• situation assessment
• deep deterministic policy gradient
• air combat