5G边缘计算环境下工作流任务的卸载策略  

Workflow task offloading strategy in 5G edge computing environment

作　　者：葛海波 冯安琪 王妍 GE Haibo;FENG Anqi;WANG Yan(School of Electronic Engineering,Xi'an University of Posts and Telecommunication,Xi'an 710121,China)

机构地区：[1]西安邮电大学电子工程学院,陕西西安710121

出　　处：《传感器与微系统》2020年第8期130-133,137,共5页Transducer and Microsystem Technologies

基　　金：国家自然科学基金青年科学基金资助项目(61805198);陕西省自然科学基金资助项目(2011JM8038);陕西省重点产业创新链(群)项目(S2019-YF-ZDCXL-ZDLGY-0098);西安邮电大学研究生创新基金资助项目(CXJJ2017013,CXJJLZ2019029)。

摘　　要：针对5G移动边缘计算(MEC)场景下的卸载策略高时间复杂度问题,提出结合深度学习的计算卸载算法HCOGA。该算法将移动服务建模为工作流模型,通过改进非支配遗传算法联合优化工作流模型的开销。为适应5G实时环境,HCOGA算法通过深度神经网络拟合改进非支配遗传算法输入、输出间的非线性关系,构成完备的卸载预测模型,可以快速、准确地做出卸载决策。仿真结果表明:HCOGA算法运行时间较短,能有效降低系统开销。Aiming at the high time complexity problem of the offloading strategy in 5 G mobile edge computing(MEC)scenarios,a computing offloading algorithm HCOGA combined with deep learning is proposed.This algorithm models mobile services as a workflow model,and jointly optimizes the overhead of the workflow model through improving non-dominated sorting genetic algorithms(NSGA-Ⅱ).In order to adapt to the 5 G real-time environment,HCOGA algorithm fits non-linear relationship between input and output of NSGA-Ⅱthrough deep neural network to form a complete offloading prediction model,which can make offloading decisions quickly and accurately.Simulation results show that the HCOGA algorithm has a shorter running time and can effectively reduce system overhead.

关 键 词：移动边缘计算 计算卸载 工作流 5G 深度学习 改进非支配遗传算法 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]