改进深度强化学习算法的计算卸载策略  

Improved computational offloading strategy of deep reinforcement learning algorithm

作　　者：葛海波 弓海文 宋兴 李顺 孙奥 GE Haibo;GONG Haiwen;SONG Xing;LI Shun;SUN Ao(School of Electronic Engineering,Xi'an University of Posts and Telecommunications,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：《西安邮电大学学报》2021年第6期9-16,共8页Journal of Xi’an University of Posts and Telecommunications

基　　金：陕西省自然科学基金项目(2011JM8038);陕西省重点产业创新链(群)项目(S2019-YF-ZDCXL-ZDLGY-0098)。

摘　　要：为了降低移动边缘计算(Mobile Edge Computing,MEC)系统的成本、提高计算效率,提出了一种改进深度强化学习算法的计算卸载策略。在任务卸载执行的时延中引入排队时延的计算,利用优先经验重放(Prioritized Experience Replay,PER)方法对历史经验赋予优先级,优先采样高优先级的经验,以提高学习效率,快速、准确地做出合理卸载决策。仿真结果表明,与相关经典策略对比,改进算法的计算效率较高,系统总成本较低。In order to reduce the cost of the mobile edge computing(MEC)system and improve the computational efficiency,a computational offloading strategy to improve the deep reinforcement learning algorithm is proposed.Introduce the calculation of queuing delay in the execution delay of task offloading,use the prioritized experience replay(PER)method to give priority to historical experience,and prioritize sampling of high-priority experience to improve learning efficiency,so as to make reasonable unloading decisions fast and accurately.The simulation results show that compared with the related classic strategies,the improved algorithm has higher computational efficiency and lower total system cost.

关 键 词：移动边缘计算 卸载决策 排队时延 深度强化学习 优先经验重放 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]