关键词: 飞行器突防博弈对抗, 近端策略优化, 拦截器探测能力限制, 深度强化学习, 突防制导一体化

Abstract:

In the face of the development of air defense and missile defense interception technologies and equipment， attack aircraft are confronted with issues such as low battlefield survivability and poor effectiveness due to the interception by defensive weapons. To address the aircraft penetration game and countermeasures problem under the interception scenario， this research proposes an intelligent maneuvering penetration guidance strategy for aircraft that considers the limitations of the interceptor detection capabilities. Initially， the interceptor’s line-of-sight angle and detection range are defined， and the relative motion relationship between the interceptor and the attack aircraft is used to describe the evolution of the confrontation situation between the attack and the defense sides. Subsequently， a proximal policy optimization guidance method is designed based on the principles of deep reinforcement learning， constructing a Markov decision chain that guides the aircraft to actively evade the interceptor’s detection， and further optimizing the aircraft’s reward function design method to achieve precise targeting. On this basis， the convergence speed of the intelligent algorithm is addressed by introducing action exploration and generalized advantage functions. Simulation results show that the intelligent maneuvering penetration guidance strategy endows the aircraft with autonomous learning and optimization attributes， allowing it to increase the difficulty of detection for the interceptor through active evasive maneuvers， ultimately breaking through the detection capability limit of the interceptor to achieve penetration and escape. Compared with the traditional PN-sin penetration guidance method， the penetration guidance strategy proposed in this paper can maintain a higher penetration success rate in scenarios where the attack and defense sides have asymmetric maneuvering capabilities.

Key words: aircraft penetration game and countermeasures, proximal policy optimization, interceptor detection capability limitations, deep reinforcement learning, integrated penetration and guidance