基于深度强化学习的网络切片资源管理算法  

Resource management algorithm for network slicing based on deep reinforcement learning

作　　者：王菲菲 王兰 郑斯辉 陈翔[2,4] WANG Feifei;WANG Lan;ZHENG Sihui;CHEN Xiang(College of Electronic and Information Engineering,Shenzhen University,Shenzhen Guangdong 518060,China;Research Institute of Tsinghua University in Shenzhen,Shenzhen Guangdong 518057,China;Shenzhen International Graduate School,Tsinghua University,Shenzhen Guangdong 528055,China;School of Electronics and Information Technology,Sun Yat-sen University,Guangzhou Guangdong 510006,China)

机构地区：[1]深圳大学电子与信息工程学院,广东深圳518060 [2]深圳清华大学研究院,广东深圳518057 [3]清华大学深圳国际研究生院,广东深圳528055 [4]中山大学电子与信息工程学院,广东广州510006

出　　处：《太赫兹科学与电子信息学报》2024年第7期792-799,共8页Journal of Terahertz Science and Electronic Information Technology

基　　金：深圳市基础研究重点资助项目(JCYJ20200109143016563)。

摘　　要：随着第五代通信技术(5G)的发展,各种应用场景不断涌现,而网络切片可以在通用的物理网络上构建多个逻辑独立的虚拟网络来满足移动通信网络多样化的业务需求。为了提高移动通信网络根据各切片业务量实现资源按需分配的能力,本文提出了一种基于深度强化学习的网络切片资源管理算法,该算法使用两个长短期记忆网络对无法实时到达的统计数据进行预测,并提取用户移动性导致的业务数据量动态特征,进而结合优势动作评论算法做出与切片业务需求相匹配的带宽分配决策。实验结果表明,相较于现有方法,该算法可以在保证用户时延和速率要求的同时,将频谱效率提高约7.7%。With the development of the 5th Generation Mobile Communication Technology(5G),various application scenarios continue to emerge.Network slicing can construct multiple logically independent virtual networks on a common physical network to meet the diverse service requirements of mobile communication networks.In order to enhance the ability of mobile communication networks to allocate resources on demand according to the traffic of each slice,this paper proposes a network slicing resource management algorithm based on deep reinforcement learning.The algorithm uses two Long Short-Term Memory(LSTM)networks to predict statistical data that cannot be reached in real time,and extracts dynamic characteristics of business data volume caused by user mobility,and then makes bandwidth allocation decisions that match the needs of slice services in combination with the Advantage Actor-Critic(A2C)algorithm.Experimental results show that compared with existing methods,this algorithm can improve the spectral efficiency by about 7.7%while ensuring the user's delay and rate requirements.

关 键 词：5G 网络切片 深度强化学习 资源分配 

分 类 号：TN929.5[电子电信—通信与信息系统]