Energy-Optimized Inter-Satellite Collaborative Computing Offloading Strategy in Low Earth Orbit Satellite Networks
Year: 2025
Type: article
Abstract: 随着低地球轨道卫星网络的扩展，用户计算需求不断提高，推动计算任务从地面云计算中心向卫星边缘迁移。在任务卸载过程中，如何合理分配计算资源、优化功耗管理并实现卫星星座的协同工作，已成为提升计算性能和能效的关键挑战。现有研究普遍假设每颗卫星能够获取整个星座的状态信息，但忽视大规模卫星网络中获取全局信息时产生的延迟和资源浪费。为解决这些问题，本文提出一种星间协同计算卸载流程，并结合双深度Q网络算法，设计星间协同计算卸载策略。通过在每颗卫星上自主感知周围卫星状态并进行任务调度，卫星可以根据实际情况决定是否卸载任务或转交给其他卫星，从而实现协同计算。实验结果表明，所提策略在任务能耗和完成率方面均优于传统卸载算法，展现了显著的性能提升。With the rapid expansion of Low Earth Orbit (LEO) satellite networks, the growing computational demands of users are driving the migration of computation tasks from terrestrial cloud centers to the LEO satellite edge. However, during the task offloading process, how to allocate computing resources efficiently, optimize power consumption, and achieve collaborative operation within the satellite constellation has become a key challenge in improving computing performance and energy efficiency. Existing studies generally assume that each satellite can access the status information of the entire constellation, but they overlook the delays and resource wastage that arise from obtaining global information in large-scale satellite networks. To address these issues, this paper proposes an inter-satellite collaborative computing offloading process and, combined with the Deep Double Q-Learning (DDQN) algorithm, designs a collaborative computing offloading strategy. By autonomously sensing the status of surrounding satellites and scheduling tasks on each satellite, the satellite can decide whether to offload tasks or transfer them to other satellites for further scheduling, thus achieving collaborative computing. Experimental results show that the proposed strategy significantly outperforms traditional offloading algorithms in terms of task energy consumption and completion rate, demonstrating a significant performance improvement.
Source: Modeling and Simulation
Author: 业超 黄
Language: English
Subfield: Aerospace Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4409513659