@article{PANG2026, 
author = {Guimei PANG and Zhenyu NA and Wen LIU and Bin LIN and Qing GUO},
title = {A task-priority-based intelligent computing-aware routing method for LEO satellite networks},
year = {2026},
journal = {Acta Aeronautica et Astronautica Sinica},
volume = {47},
number = {18},
keywords = {low earth orbit satellite, task priority, dueling-DDQN, computing-aware routing, delay optimization},
url = {https://www.sciopen.com/article/10.7527/S1000-6893.2026.33624},
doi = {10.7527/S1000-6893.2026.33624},
abstract = {To address the challenges of limited satellite-ground bandwidth and heterogeneous Quality of Service (QoS) requirements for tasks in Low Earth Orbit (LEO) satellite networks, an intelligent computing-aware routing method based on task priority is proposed. First, considering satellite computing and storage resources, a resource-constrained LEO satellite network model is constructed. The sun outage phenomenon is incorporated to dynamically evaluate link availability, thereby capturing the impact of space environmental disturbances on routing decisions. Second, a QoS classification and priority scheduling mechanism is established to characterize the heterogeneous requirements of tasks in terms of latency and packet loss rate. Subsequently, a joint optimization model for computation offloading and data transmission is formulated with the objective of minimizing the end-to-end average delay. Furthermore, the optimization problem is transformed into a Markov Decision Process to enable reinforcement learning-based solutions. A task-priority queue is introduced in task processing and data forwarding, and QoS constraints are incorporated to dynamically filter candidate action space. Meanwhile, task-priority weights are embedded into the reward function. Based on these designs, a Dueling Double Deep Q-Network (Dueling-DDQN)-based computing-aware routing decision algorithm is designed to achieve adaptive optimization of computation offloading and routing decisions. Simulation results demonstrate that, under a constellation of 288 satellites, the proposed method outperforms the Proximal Policy Optimization algorithm under different traffic loads. In particular, the average end-to-end delay is reduced by 4.28%–14.53%, and the task computing ratio is increased by 1.44%–2.89%. Moreover, the proposed method maintains good applicability across different constellation scales.}
}