Topology Planning for LEO Satellites Based on Deep Reinforcement Learning

Chang He, Wei Liu*, Zhengbi Yong, Zhiyuan Xu, Qihao Zhou, Shilei Li*, Dawei Shi

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

The Low Earth Orbit (LEO) satellite network has made significant advancements in global communication services, offering low latency and extensive coverage. Effective topology planning has the dual benefit of enhancing network Quality of Service and reducing signal delay, while simultaneously optimizing resource utilization. The advent of deep reinforcement learning (DRL) methodologies offers novel approaches to inter-satellite topology planning in LEO constellations. In this work, a DRL topology planning algorithm based minimum spanning tree (DQN-MST) is proposed and compared with two traditional graph optimization algorithms. The results demonstrate that DQN-MST reduces the total network link length by 38.20% and 62.62% compared to the graph optimization algorithms, confirming the effectiveness of DQN-MST.

Original languageEnglish
Title of host publication2024 IEEE 3rd Industrial Electronics Society Annual On-Line Conference, ONCON 2024
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9798331540319
DOIs
Publication statusPublished - 2024
Externally publishedYes
Event3rd IEEE Industrial Electronics Society Annual On-Line Conference, ONCON 2024 - Beijing, China
Duration: 8 Dec 202410 Dec 2024

Publication series

Name2024 IEEE 3rd Industrial Electronics Society Annual On-Line Conference, ONCON 2024

Conference

Conference3rd IEEE Industrial Electronics Society Annual On-Line Conference, ONCON 2024
Country/TerritoryChina
CityBeijing
Period8/12/2410/12/24

Keywords

  • deep reinforcement learning (DRL)
  • inter-satellite
  • Low Earth Orbit (LEO)
  • topology planning

Fingerprint

Dive into the research topics of 'Topology Planning for LEO Satellites Based on Deep Reinforcement Learning'. Together they form a unique fingerprint.

Cite this