| 摘要: |
| 针对空天地网络中计算资源受限的边缘服务器在处理大量任务时，面临过载导致任务完成时间和用户能耗增加的问题，提出了一种基于深度强化学习的三层协同任务卸载和资源分配方案，以任务完成时间和用户能耗建立任务开销函数，在计算资源的约束下联合优化用户卸载决策、用户传输功率、子载波分配和计算资源分配。首先采用拉格朗日乘子法优化计算资源分配，然后使用深度强化学习求解卸载决策、用户发射功率和子载波分配，最后通过交替迭代的方法得到优化解。仿真结果表明，与DQN(Deep Q-learning Network)、DDQN(Double DQN)、DDPG(Deep Deterministic Policy Gradient)等方案相比，所提方案任务开销分别下降约19%、10%和13%。 |
| 关键词: 空天地一体化网络 移动边缘计算 计算卸载 资源分配 深度强化学习 |
| DOI：10.20079/j.issn.1001-893x.240110001 |
|
| 基金项目:重庆市自然科学基金创新发展联合基金（市教委）项目(CSTB2023NSCQ-LZX0076) |
|
| Task Offloading Strategies for Space-Air-Ground Edge Computing Networks |
| YU Xiang,QU Yuanyu,YANG Lu |
| (School of Communications and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China) |
| Abstract: |
| In the space-air-ground integrated edge computing network,a large number of computational tasks can lead to overloading of edge servers,which increases the completion time and energy consumption of user tasks.To solve the problem,a three-tier collaborative task offloading and resource allocation scheme based on deep reinforcement learning is proposed,which creates a task overhead function with task completion time and user energy consumption,and jointly optimizes the user offloading decision,user transmission power,subcarrier allocation and computational resource allocation under the constraints of computational resources.First,the Lagrange multiplier method is used to optimize the computational resource allocation.Secondly,deep reinforcement learning is used to solve the offloading decision,user transmission power and subcarrier allocation,and finally the optimized solution is obtained by an alternating iteration method.The simulation results show that,compared with that of Deep Q-learning Network(DQN),Double DQN(DDQN) and Deep Deterministic Policy Gradient(DDPG),the mission overhead of the proposed scheme declines by approximately 19%,10% and 13%. |
| Key words: space-air-ground intergated network mobile edge computing computational offloading resource allocation deep reinforcement learning |
