中国科学技术大学学报 ›› 2018, Vol. 48 ›› Issue (12): 996-1011.DOI: 10.3969/j.issn.0253-2778.2018.12.004

• 原创论文 • 上一篇    下一篇

基于强化学习的无线网络自组织性研究

王 超   

  1. 中国科学技术大学电子工程与信息科学系,安徽合肥 230026
  • 收稿日期:2018-05-07 修回日期:2018-06-10 出版日期:2018-12-31 发布日期:2018-12-31
  • 通讯作者: 沈聪
  • 作者简介:王超,男,1992年生,博士生. 研究方向:强化学习算法在无线通信网络中的应用. E-mail: wch15@mail.ustc.edu.cn
  • 基金资助:
    国家自然科学基金重点项目(61631017),国家自然科学基金(61572455)资助.

Research onself-organization of wireless network based on reinforcement learning

WANG Chao   

  1. Department of Electronic Engineering and Information Science, University of Science and Technology of China, Hefei 230026,China
  • Received:2018-05-07 Revised:2018-06-10 Online:2018-12-31 Published:2018-12-31

摘要: 传统无线通信技术逐渐无法满足5G通信系统中日益复杂的需求,而无线自组织网络(self-organizing network, SON)相关技术的引入为5G网络智能化管理提供了一套扩展性良好的解决方案.强化学习算法在SON中的应用,为无线网络架构提供了更为广泛的感知能力和更完备的优化能力.本文以强化学习算法在 SON 的技术方面的进展为重点,对现有的相关文献进行综述.首先,文章将 SON 的相关应用分为自配置、自优化、自愈合三大模块,对每个模块的具体实例进行分析;然后,从强化学习算法的可扩展性、复杂度、鲁棒性、收敛性等多重参数标准角度,对不同网络场景中涉及的算法进行对比,并总结一般性准则;最后,阐述了强化学习算法应用于未来网络可能遇到的挑战,总结并展望了自组织网络未来的发展方向.

关键词: 无线自组织网络(SON), 强化学习(RL), 5G

Abstract: Traditional wireless communication technologies are gradually unable to meet the increasingly complex requirements of the 5G system. The technologies related to self-organizing network (SON) provide scalable solutions for network intelligent management. The implementation of reinforcement learning (RL) algorithms in SON illustrates its capability on network recognition and optimization. In this paper, three modules in SON and their applications were introduced, which were self-configuration, self-optimization and self-healing. Then, related RL algorithms from different criteria were evaluated, such as scalability, complexity, robustness and convergence. Finally, this research was summarized by analyzing the challenges associated with the application of RL in future wireless networks and identifying the directions for future research.

Key words: self-organizing network (SON), reinforcement learning (RL), 5G