关键词: 群目标协同拦截, 近端策略优化, 多智能体强化学习, 集中式评价-分布式执行, 深度学习

Abstract:

The attack defense confrontation and interception between unmanned clusters is an important operational scenario in the future intelligent war. Aiming at the problem of cooperative interception of game confrontation against aircraft cluster attacks， a multi-agent deep reinforcement learning cooperative interception strategy based on the near end strategy optimization method is proposed. Combining the single agent near end strategy optimization algorithm with the centralized evaluation distributed execution algorithm architecture， a multi-agent reinforcement learning intelligent maneuver strategy is designed. On this basis， to solve the problem of slow algorithm convergence， the generalized dominance function is introduced to improve the convergence performance of the algorithm. Simulation results show that the multi aircraft intelligent cooperative interception strategy endows the UAV with the attribute of autonomous learning， which can intelligently and autonomously assign interception tasks according to the real-time battlefield situation， and improves the algorithm convergence rate by constraining the update range of the strategy. Through continuous iterative self-learning， this strategy can realize the autonomous optimization of game interception strategy. Improve collaborative interception efficiency by self-learning in different scenarios.

Key words: multi-target cooperative interception, proximal policy optimization, multi-agent reinforcement learning, centralized evaluation-distributed execution, deep learning