面向即时响应的卫星在轨分布式协商智能任务规划OA北大核心CSTPCD

The mission planning of LEO remote sensing constellation is a complex multi-objective op-timization problem.At present,there are some problems in satellite mission planning research based on deep reinforcement learning,such as small scale of test data constellation,single optimization objective,repeated task arrangement and poor model adaptability.To solve the above problems,the CON_DQN(Contract network and Deep Q Network,CON_DQN)algorithm is proposed in this paper,which adopts the master-slave on-orbit distributed negotiation mechanism,the slave satellite makes decisions based on the planning,and the master satellite makes multi-objective optimization decisions from the aspects of priority,resource cost and load balancing based on the deep reinforcement learning algorithm,and pro-cesses on-orbit distributed negotiation intelligent mission planning for instant response.Aiming at the scene where the user demand reaches the key observation area dynamically at high frequency,the simu-lation experiment of different scale task sets of 100-star constellation is carried out.The results show that the proposed algorithm has a fast response speed and can achieve higher task benefits.