Satellite-ground Station Scheduling Using Pointer Networks and Reinforcement Learning

This study addresses the satellite-ground station scheduling problem, a typical NP-hard combinatorial optimization challenge, by proposing an end-to-end deep reinforcement learning framework that integrates a pointer network with the REINFORCE algorithm, a policy gradient method. The task-ground station allocation is modeled as a sequential decision-making problem under strict visibility constraints. The network architecture comprises a bidirectional recurrent encoder to extract the temporal features of tasks and a decoder that leverages pointer attention mechanisms to generate allocation decisions while incorporating hard constraints for visibility and soft penalties for load balancing. For training, the REINFORCE algorithm is utilized, with optimization based on a composite reward signal that balances task completion, conflict mitigation, and load distribution. Experiments conducted using a simulated multi-satellite, multi-ground-station dataset demonstrate that the proposed strategy achieves zero-conflict allocation with a near-optimal load balance, outperforming baselines. Overall, this study provides a scalable, rule-free intelligent decision-making solution for autonomous satellite scheduling.