論文ID: 2024EAL2106
Agile Earth observation satellite (AEOS) mission planning (AEOSMP) problem aims to optimize observation efficiency by selecting and scheduling tasks from the Earth's surface, subject to complex resource constraints. Increased flexibility of AEOS presents challenges for autonomous mission planning and scheduling. Deep reinforcement learning (DRL) and clustering tasks are two approaches to enhance the autonomy and observation efficiency of AEOSMP. This letter introduces two innovative algorithms to tackle the AEOSMP problem: the Sequential Clique Clustering and PPO Planning algorithm (SCC-PPO) and the Simultaneous Clustering and Planning PPO Algorithm (SCP-PPO). SCC-PPO initially partitions the mission tasks into cliques, followed by planning. In contrast, SCP-PPO combines clustering and planning into a single, concurrent process. Numerical simulations reveal that SCP-PPO enhances the observation reward by 1.01% to 11.43% compared to SCC-PPO.
