星地融合网络中一种基于强化学习的任务卸载方法

eploying edge servers in satellite-terrestrial integrated networks can provide users with high-quality computing services. This paper investigates the task offloading problem in satellite-terrestrial integrated networks, first, considering the resource competition among users caused by constrained communication-computing resources in satellite-terrestrial Mobile Edge Computing(MEC) networks, along with the conflict between user quality of service and overall edge network performance, we establish a two-layer game model encompassing both inter-user competition and user-edge network interaction. A multi-objective total utility function is designed to optimize the overall utility across both transmission and computation phases. Second, to address the uneven sample distribution in the experience replay pool, we propose an improved prioritized experience replay-based Soft Actor-Critic(SAC) algorithm, which optimizes the sampling mechanism by introducing sample value evaluation to accelerate model training. Finally, simulation results demonstrate that compared with the Proximal Policy Optimization(PPO) algorithm and other baseline methods, the proposed algorithm can effectively enhance the overall system utility.