Abstract: Aiming at the decision making problems of multi coupling tasks such as cooperative target assignment and penetration trajectory planning in UAV cluster countermeasure in complex scenes, an intelligent decision making method for multi coupling tasks in UAV cluster countermeasure is proposed. Firstly, aiming at the problems of multi coupling tasks and large decision making space in UAV cluster countermeasure, combined with the advantages of centralized and hierarchical architectures, a hybrid deep reinforcement learning architecture for multi coupling tasks is designed, which can improve the cooperation between the multi coupling tasks and the effectiveness of cluster countermeasure. Secondly, for the sparse reward problem of sequential decision making in trajectory planning, a trajectory construction method is designed. Thirdly, aiming at the scene uncertainty problem under the strong countermeasure conditions, based on the UAV cluster red blue countermeasure simulation platform, a red blue game training method based on multiple random scenes is designed, which can enhance the generalization of the strategy network. Finally, by comparing with the traditional method, the centralized architecture method and the hierarchical architecture method, the simulation results show that the effectiveness and the advanced nature of the proposed method are verified.

Key words: Deep reinforcement learning, Intelligent decision making; UAV cluster countermeasure, Cooperative target assignment, Penetration trajectory planning