关键词: 比例导引, 蒙特卡洛强化学习, Q-learning强化学习, 导航比

Abstract: In order to improve the guidance performance of missiles, on the basis of proportional guidance, Monte Carlo reinforcement learning and Q-learning reinforcement learning are used respectively to design the navigation ratio. The first navigation ratio design method using Monte Carlo reinforcement learning only roughly segments a missile's flight process, whose algorithm is simple with strong engineering usability. The second navigation ratio design method using Q-learning reinforcement learning further subdivides the guidance environment by using flight time, line-of-sight rate, expected encounter time and target characteristics, and adaptively adjusts the navigation ratio of proportional guidance according to the changes of environment and state, so as to obtain the best flight guidance strategy. Based on a certain type of air defense missile, the navigation ratio design is carried out by using the above methods，and the batch trajectories are randomly selected from the whole airspace trajectory library for simulation calculation, which is then compared with the traditional empirical design. The simulation results show that the navigation ratio designed with reinforcement learning can significantly reduce the miss distance of boundary trajectories, indicating that the proposed design method can effectively improve the guidance and interception capabilities of the missile.

Key words: proportionalnavigation, MonteCarloreinforcementlearning, Q-learningreinforcementlearning, navigationratio