Autonomous collaborative observation method for time-sensitive moving target tracking by satellite swarms

Time-sensitive moving targets exhibit characteristics such as positional uncertainty and dynamics, rendering traditional satellite swarm planning algorithms inadequate. Therefore, proposing a reasonable, rapid, and effective observation task planning method has become a key issue. Using reinforcement learning techniques, this paper proposes an autonomous collaborative observation method for time-sensitive moving targets with multiple satellites and various payloads.Considering the dynamic updates of the target's position, the observation process is treated as numerous discretized subtasks. By inputting the initial target position, the target positions within the subtasks are updated and shared among the satellite swarm through broadcasting. For each subtask, the satellite calculates its own observation conditions and centrally processes them within the leader-satellite. This paper addresses the characteristics of task discretization by utilizing an Improved DDQN(IDDQN) method to plan sub-tasks, ultimately generating a planning result for the entire observation task of the moving target. In addition, the design of the planning objectives comprehensively considers factors such as tracking duration, observation quality, energy consumption, and memory usage. Simulation experiments show that the global optimization effect of this method, especially the optimization effect of task allocation, is better than that of CNP method. It solves the problem that ICLPSO method cannot be used for dynamic targets, significantly improves the computational efficiency, and is also capable of solving untrained scenarios which can meet the timeliness requirements for observing time-sensitive moving targets, thus indicating promising application prospects.