Ballistic Missile Maneuver Penetration Based on Reinforcement Learning

Abstract

Ballistic missiles, as the main weapon for long-range precision fire strikes, reflect the military development level and strategic capabilities of a country. This paper focuses on the midcourse penetration process of ballistic missile maneuvers. Assuming that the interceptor missile uses a proportional guidance strategy, the reinforcement learning methods is used to train network models. The method avoids the need for traditional control theory methods to establish precise mathematical models based on controlled objects, and this reduces the difficulty of the performance model to solve the optimal analytical solution. The use of State space discretization reduce the action space, and improves the network learning efficiency. Finally, the simulation proves that reinforcement learning can greatly increase the miss distance of missile maneuver penetration.