ST-DSQN: A Task Offloading Method for Low Earth Orbit Satellite Networks

The task offloading problem in low earth orbit (LEO) satellites has become an active research area. To address the limitation that current traditional and improved reinforcement learning methods fail to jointly incorporate temporal and spatial stream information in task offloading decision-making processes, this paper proposes a Spatio-Temporal Dual-Stream Q Network (ST-DSQN) method. The proposed dual-stream network architecture simultaneously handles both temporal and spatial information within a unified deep Q-network framework. Experimental results demonstrate that, compared with other algorithms, the ST-DSQN approach achieves significant improvements in critical metrics such as average task offloading delay, energy consumption, and task success rate.