Scholarlink

<p>With the approach of the 6G era, the core position of computing power in the digital industry has become more and more prominent, and the rapid development of computing-intensive services centred on satellite networks. Satellite communications achieve high bandwidth, extensive coverage, seamless connectivity, and ubiquitous computing through the fusion of large-scale low-Earth orbit constellations and high-Earth orbit satellites, laying a solid foundation for the Internet of Everything. However, due to its diverse and unevenly distributed demands, and high time-varying nature, the satellite network faces a major challenge of optimal allocation of computing power and network resources. Traditional resource allocation strategies are difficult to adapt to the dynamically changing demands, resulting in low resource utilization. To solve this problem, this paper innovatively proposes a layered satellite computing network architecture based on software-defined networking, models the resource allocation problem as a virtual network embedding problem, and designs an algorithm based on deep reinforcement learning. Experimental results demonstrate that compared to the baseline algorithm, this algorithm improves the revenue-cost ratio by approximately 22.6% and exhibits superior performance in terms of the acceptance rate of virtual network requests and the average long-term revenue. This fully validates its outstanding advantages in improving resource utilization efficiency and meeting complex business demands.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep reinforcement learning-based resource orchestration algorithm for SDN-enabled satellite computing networks

  • 期刊论文

  • 发表日期：

    2026年2月23日

查看全文
  • Jian Fan,
  • Yi Zhang,
  • Jian Wang,
  • Ning Chen,
  • Lizhuang Tan,
  • Peiying Zhang

摘要

With the approach of the 6G era, the core position of computing power in the digital industry has become more and more prominent, and the rapid development of computing-intensive services centred on satellite networks. Satellite communications achieve high bandwidth, extensive coverage, seamless connectivity, and ubiquitous computing through the fusion of large-scale low-Earth orbit constellations and high-Earth orbit satellites, laying a solid foundation for the Internet of Everything. However, due to its diverse and unevenly distributed demands, and high time-varying nature, the satellite network faces a major challenge of optimal allocation of computing power and network resources. Traditional resource allocation strategies are difficult to adapt to the dynamically changing demands, resulting in low resource utilization. To solve this problem, this paper innovatively proposes a layered satellite computing network architecture based on software-defined networking, models the resource allocation problem as a virtual network embedding problem, and designs an algorithm based on deep reinforcement learning. Experimental results demonstrate that compared to the baseline algorithm, this algorithm improves the revenue-cost ratio by approximately 22.6% and exhibits superior performance in terms of the acceptance rate of virtual network requests and the average long-term revenue. This fully validates its outstanding advantages in improving resource utilization efficiency and meeting complex business demands.