摘要
在用户设备、边缘计算服务器和云服务器构成的三层移动网络体系中,如何高效地进行任务卸载是一个重要的问题。针对移动边缘计算(MEC)中多用户多服务器环境下的长时延和高能耗问题,提出一种基于深度强化学习(DRL)算法的三层移动网络架构,将卸载决策问题模型化为约束条件下的最优化问题。结合深度强化学习理论,利用改进的A3C(IA3C)算法求解。仿真结果表明:与深度Q网络(DQN)、全本地卸载算法、全边缘卸载算法相比,在设备数量、MEC计算能力和用户数据量三个方面,提出的卸载策略均能更有效地降低总成本。
In a three-tier mobile network system composed of user equipment,edge computing servers,and cloud servers,how to efficiently perform task offloading is an important issue.For the problems of long latency and high energy consumption in the multi-user and multi-server environment in mobile edge computing(MEC),a three-layer mobile network architecture based on the Deep Reinforcement Learning algorithm is proposed to model the offloading decision problem as optimization problem under constraints.Combined with deep reinforcement learning theory,the improved Asynchronous Advantage Actor-Critic algorithm(IA3C)is used to solve it.Simulation results show that compared with the deep Q network(DQN),the Full Local offloading algorithm,and the edge offloading in short,Full Offloading algorithm,in terms of the number of devices,MEC computing power and user data volume,the proposed Unloading strategies can reduce the total cost more effectively.
作者
葛海波
赵其实
车虹葵
李照宇
GE Haibo;ZHAO Qishi;CHE Hongkui;LI Zhaoyu(School of Electronic Engineering,Xi’an University of Posts and Telecommunications,Xi’an 710121,China)
出处
《传感器与微系统》
CSCD
北大核心
2022年第8期60-63,67,共5页
Transducer and Microsystem Technologies
基金
陕西省重点产业创新链(群)—工业领域项目(2019ZDLGY13—03—02)
陕西省重点产业创新链(群)项目(S2019—YF—ZDCXL—ZDLGY—0098)。
关键词
移动边缘计算
三层移动网络体系
任务卸载
异步优势动作评价
深度强化学习
mobile edge computing(MEC)
three-tier mobile network system
task offloading
asynchronous advantage actor-critic(A3C)
deep reinforcement learning(DRL)
