EUVIMEDEUVIMED entdecken
Crossref

Deep Reinforcement Learning and Edge Computing for Multisatellite On-Orbit Task Scheduling

Qiangqiang Jiang, Peng Han, Xu Xin, Kang Chen

IEEE Transactions on Aerospace and Electronic Systems · 2025-10 · 61 (5) · 14143-14159

DOI: 10.1109/taes.2025.3583276 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Aerospace and Electronic Systems, 2025-10. Verfasst von Qiangqiang Jiang und 3 weiteren Autor:innen. Die Arbeit wurde bisher 4-mal zitiert (Crossref) und nennt 39 Literaturangaben. Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Today’s satellite applications often require the immediate processing of exploding data, which has driven a shift to on-orbit computing needs. Due to the limited onboard resources and energy supply, satellite edge computing is developed for intersatellite and satellite–ground collaborations. However, it is a challenge to achieve efficient multisatellite on-orbit computation considering the harsh satellite-to-ground transmission. Therefore, this article proposes an edge-computing-enabled multisatellite on-orbit computation scheduling (MSOCS) through multiagent deep reinforcement learning (MADRL). First, we formulate the MSOCS problem, where on-orbit computation tasks are described as a directed acyclic graph, and the intermittent satellite-to-ground link is established as connectable and disconnectable constraints. Second, a multiagent proximal policy optimization-based algorithm is developed by modeling the MSOCS problem via a partially observable Markov decision process. To cope with the dynamic change of observational input, we design a problem-specific embedding constructor, which is also utilized as the structural basis of the policy and value networks. Finally, practical remote sensing data processing missions are employed as the case study to conduct evaluation experiments. Simulation results validate the scheduling effectiveness of the proposed MSOCS algorithm in the test instances with different data scales, which surpasses other MADRL approaches. Moreover, our MSOCS improves the efficiency of on-orbit computation by 70.48% in time and 68.33% in energy, compared with the baseline without scheduling optimization.

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Jiang, Q., Han, P., Xin, X., & Chen, K. (2025). Deep Reinforcement Learning and Edge Computing for Multisatellite On-Orbit Task Scheduling. IEEE Transactions on Aerospace and Electronic Systems, 61(5), 14143-14159. https://doi.org/10.1109/taes.2025.3583276

Vancouver

Jiang Q, Han P, Xin X, Chen K. Deep Reinforcement Learning and Edge Computing for Multisatellite On-Orbit Task Scheduling. IEEE Transactions on Aerospace and Electronic Systems. 2025;61(5):14143-14159. doi:10.1109/taes.2025.3583276

BibTeX
@article{Jiang2025,
  author = {Jiang, Qiangqiang and Han, Peng and Xin, Xu and Chen, Kang},
  title = {Deep Reinforcement Learning and Edge Computing for Multisatellite On-Orbit Task Scheduling},
  journal = {IEEE Transactions on Aerospace and Electronic Systems},
  year = {2025},
  volume = {61},
  number = {5},
  pages = {14143--14159},
  doi = {10.1109/taes.2025.3583276}
}
Kostenlos starten →