Abstract: Aiming at the problem that the traditional command and control network may be difficult to deal with the cooperative operation and decentralized and resilent command and control of huge unmanned clusters in mosaic warfare, a command and control network framework of Mosaic warfare based on multi-agent deep reinforcement learning, and the multi-agent modeling are proposed, which abstracts the cooperative operation at the level of unmanned platform and within the cluster into global and local command and control network models, dominated by combat tasks and battlefield situation. Based on multi-agent deep reinforcement learning algorithm, the command and control networks under mosaic warfare are modeled, and the autonomous learning and cooperation strategies of unmanned platform and the cluster can be simulated and analyzed. The mosaic warfare is rehearsed under the background of UAV operations in the conflicts between A and Y countries. The results show that the proposed command and control networks can realize decentralized operations, accelerate the closure of OODA ring and improve the command and control and enhance the survivability of killing network.