Enabling a satellite network with edge computing capabilities can complement the advantages further of a single terrestrial network and provide users with a full range of computing service. Satellite edge computing is a potentially indispensable technology for future satellite-terrestrial integrated networks. In this article, a three-tier edge computing architecture consisting of the terminal–satellite–cloud is proposed, where tasks can be processed at three planes and intersatellites can cooperate to achieve on-board load balancing. Facing varying and random task queues with different service requirements, we formulate the objective problem of minimizing the system energy consumption under the delay and resource constraints, and jointly optimize the offloading decision, communication, and computing resource allocation variables. Moreover, the distribution of resources is based on the reservation mechanism to ensure the stability of the satellite-terrestrial link and the reliability of computation process. To adapt to the dynamic environment, we propose an intelligent computation offloading scheme based on the deep deterministic policy gradient (DDPG) algorithm, which consists of several different deep neural networks (DNNs) to output both discrete and continuous variables. Additionally, by setting the selection process of legal actions, the simultaneous decisions on offloading locations and allocating resources under multitask concurrency is realized. The simulation results show that the proposed scheme can effectively reduce the total energy consumption of the system by ensuring that the task is completed on demand, and outperform the benchmark algorithms.

中文翻译：

---

具有协作计算卸载的卫星边缘计算：智能深度确定性策略梯度方法


让卫星网络具备边缘计算能力，可以进一步补充单一地面网络的优势，为用户提供全方位的计算服务。卫星边缘计算是未来星地一体化网络潜在不可或缺的技术。本文提出了一种由终端-卫星-云组成的三层边缘计算架构，任务可以在三个平面上处理，星间可以协作实现星上负载均衡。面对不同服务需求的变化和随机任务队列，我们​​制定了在延迟和资源约束下最小化系统能耗的目标问题，并联合优化卸载决策、通信和计算资源分配变量。而且，资源分配基于预留机制，保证星地链路的稳定性和计算过程的可靠性。为了适应动态环境，我们提出了一种基于深度确定性策略梯度（DDPG）算法的智能计算卸载方案，该算法由多个不同的深度神经网络（DNN）组成，以输出离散和连续变量。另外，通过设置法律行为的选择过程，实现了多任务并发下的卸载位置和资源分配的同时决策。仿真结果表明，所提方案能够在保证任务按需完成的情况下有效降低系统总能耗，且性能优于基准算法。