Research on Maneuvering Control Algorithm of Short-Range UAV Air Combat Based on Deep Reinforcement Learning

With the development of autonomous control system technology of unmanned aerial vehicle (UAV), the application of UAV in air combat is increasing. In the future, the intelligent warfare represented by UAV will completely change the form of traditional warfare, and UAV will become dominant in air combat. The problem of UAV dogfight control is the most challenging research direction in short-range air combat. This paper proposes a short-range UAV combat control model based on deep reinforcement learning. Firstly, the dynamics and kinematics of UAV are modeled, and a reasonable reward function is designed for the problem of UAV dogfight. Secondly, the model uses a simulation environment to generate data sets and uses the DDPG algorithm to train the model. Aiming at the unstable training process of DDPG, a Locked parameter network restart mechanism is proposed to renew the strategic network and value network parameters. Finally, the algorithm performance is verified. Compared with traditional algorithms, the algorithm has better expressive force in the combat maneuver control of UAV.