.png){kind=icon}
返回
Multisatellite Scheduling via Reinforcement Learning-Based Mixed Integer Linear Programming
.png){kind=icon}
PRE
AI
DOI：10.1109/TAES.2025.3632780.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Agile satellites possess advanced Earth observation capabilities and highly flexible attitude maneuvering, rendering their scheduling problems increasingly crucial. As the number of such satellites continues to grow, mission scheduling has become considerably more complex, with more frequent observation conflicts, tighter resource coupling, and higher computational demands. To address these challenges, this article addresses the agile satellite scheduling problem by formulating a mixed integer linear programming model and solving it using the branch-and-bound (B&B) algorithm. Specifically, we improve the B&B algorithm's efficiency by leveraging reinforcement learning (RL) to optimize variable selection during the branching process. To enhance training robustness, we propose a novel Q-network framework that integrates a superior Q-network as a competitive regularizer. This framework constrains the learned policy to align with demonstration-based behaviors, forming a conservative online RL strategy. Numerical experiments on different scenarios demonstrate that the proposed method efficiently obtains high-quality solutions, outperforming traditional approaches in scheduling performance.
Keyword:
Branch-and-bound (B&B)
mixed integer linear programming (MILP)
multisatellite scheduling
reinforcement learning (RL)