MEC系统中基于多智能体深度增强学习的协作式任务卸载和资源分配

Cooperative Offloading and Resource Allocation with Multi-Agent Deep Reinforcement Learning in MEC

摘要:

移动边缘计算(MEC)通过将计算任务卸载到MEC服务器上,在缓解智能移动设备计算负载的同时,可以降低服务时延。然而目前在MEC系统中,关于任务卸载和资源分配仍然存在以下问题:1)边缘节点间缺乏协作；2)计算任务到达与实际环境中动态变化的特征不匹配；3)协作式任务卸载和资源分配动态联合优化问题。为解决上述问题,文章在协作式MEC架构的基础上,提出了一种基于多智能体的深度确定性策略梯度算法(MADDPG)的任务卸载和资源分配算法,最小化系统中所有用户的长期平均成本。仿真结果表明,该算法可以有效降低系统的时延及能耗。

Abstract:

Mobile edge computing (MEC) has emerged as a key technology to alleviate the computation workloads and decrease service latency for computation-intensive applications by offloading the tasks to MEC servers. However, the existing computation offloading and resource allocation studies present some several problems:poor collaboration between edge servers; mismatch between the computational task arrival and the dynamic characteristics in the real environment; and the dynamic joint optimization problem of the collaborative task unload and resource allocation. To solve such issues, based on the collaborative MEC framework, a multi-agent based deep deterministic policy gradient (MADDPG) is proposed for task unloading and resource allocation to minimize the overall long-term average cost. Simulation results reveal that the proposed scheme can reduce the delay and energy consumption.