北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2020, Vol. 43 ›› Issue (2): 29-39.doi: 10.13190/j.jbupt.2019-104

• 论文 • 上一篇    下一篇

基于平均场博弈的超密集网络边缘缓存和删除分配研究

王孟哲1, 滕颖蕾1, 宋梅1, 韩丹涛2, 张勇1   

  1. 1. 北京邮电大学 电子工程学院, 北京 100876;
    2. 机械工业仪器仪表综合技术经济研究所, 北京 100055
  • 收稿日期:2019-05-31 发布日期:2020-04-28
  • 通讯作者: 滕颖蕾(1983-),女,副教授,E-mail:lilytengtt@gmail.com. E-mail:lilytengtt@gmail.com
  • 作者简介:王孟哲(1994-),男,硕士生.
  • 基金资助:
    国家自然科学基金项目(61771072);科技部重点研发基金项目(2018YFB120022)

Mean-Field Game Based Edge Caching and Deleting Allocation in Ultra-Dense Networks

WANG Meng-zhe1, TENG Ying-lei1, SONG Mei1, HAN Dan-tao2, ZHANG Yong1   

  1. 1. School of Electronic Engineering, Beijing University of Posts and Telecommunications, Beijing 100876, China;
    2. Instrument Technology and Economy Institute, Beijing 100055, China
  • Received:2019-05-31 Published:2020-04-28

摘要: 超密集网络设备数目庞大导致缓存分配算法复杂度极高,频繁地缓存和删除同样的内容导致的系统不稳定,为此,提出了基于平均场博弈(MFG)的分布式缓存分配算法和基于李雅普诺夫漂移加惩罚(DPP)方法的分布式删除分配算法.MFG方法使缓存分配算法的复杂度与基站数目无关.DPP方法将具有时间相关性的删除分配问题解耦成为每个时刻的问题,并求解得到了兼顾系统稳定性和网络开销优化的删除分配策略.仿真结果表明,MFG方法能够使网络最优控制策略快速收敛,并且在超密集场景下得到明显低于基本缓存分配方法的网络开销;李雅普诺夫DPP方法能够实现兼顾网络开销优化的网络缓存和删除稳定性.

关键词: 超密集网络, 平均场, 平均场博弈, 李雅普诺夫理论, 漂移加惩罚

Abstract: A distributed caching allocation algorithm based on mean-field game (MFG) and a distributed deleting allocation algorithm based on Lyapunov drift-plus-penalty (DPP) method were proposed to solve the problem of caching allocation algorithms' extremely high complexity caused by lots of devices in ultra-dense network and the problem of system instability caused by caching and deleting same content frequently. The caching allocation algorithm's complexity independent of the number of base stations is made by MFG. The time-correlated deleting allocation problem into problems each time slot is decoupled by DPP. Thereafter the deleting allocation policy for the tradeoff between system stability and minimizing network cost gets solved. Simulation shows that MFG can both make the network optimal control strategy converge quickly and save network cost obviously compared to baseline caching allocation method under ultra-dense scenario. Lyapunov DPP method can ensure network caching and deleting stability while minimizing network cost.

Key words: ultra-dense network, mean-field, mean-field game, Lyapunov theory, drift-plus-penalty

中图分类号: