电信科学 ›› 2023, Vol. 39 ›› Issue (9): 63-75.doi: 10.11959/j.issn.1000-0801.2023181

• 专题:网络智能化与生成式人工智能 • 上一篇    

自智网络全栈部署技术研究与实践

薛飞1, 陈彬1, 刘静2, 梁晓扬2, 朱琳2, 王凤2, 李天1, 张靓1, 陈贞贞1, 李潇1   

  1. 1 中国移动通信集团广东有限公司,广东 广州 510632
    2 中国移动通信有限公司研究院,北京 100053
  • 修回日期:2023-09-06 出版日期:2023-08-01 发布日期:2023-08-01
  • 作者简介:薛飞(1983- ),男,中国移动通信集团广东有限公司高级工程师,主要研究方向为自智网络高阶演进、网络 AI 应用开发和网络AI训练平台等
    陈彬(1980- ),男,现就职于中国移动通信集团广东有限公司,主要研究方向为自智网络高阶演进、网络智能化规划等
    刘静(1991- ),女,现就职于中国移动通信有限公司研究院,主要研究方向为九天网络智能化平台规划和需求管理、MLOps训推一体应用等
    梁晓扬(1984- ),男,现就职于中国移动通信有限公司研究院,主要研究方向为自智网络平台架构、MLOps以及大模型在自智网络领域的应用
    朱琳(1983- ),女,中国移动通信有限公司研究院网络智能化能力及应用产品线CEO、高级工程师,长期从事通信网络智能化理论、算法、关键技术、产品研发及应用落地相关工作
    王凤(1985- ),女,现就职于中国移动通信有限公司研究院,主要研究方向为网络智能化一线综合应用样板间设计、通信网络大模型等
    李天(1981- ),男,现就职于中国移动通信集团广东有限公司,主要研究方向为网络异常检测算法和应用、九天网络智能化平台边缘节点建设、AI能力编织技术等
    张靓(1980- ),女,现就职于中国移动通信集团广东有限公司,主要研究方向为自智网络全栈部署体系、MLOps训推一体
    陈贞贞(1990- ),女,现就职于中国移动通信集团广东有限公司,主要研究方向为自智网络体系架构、自智网络成效评估、MLOps训推一体等
    李潇(1982- ),女,现就职于中国移动通信集团广东有限公司,主要研究方向为自智网络应用开发等

Research and practice on technologies for full stack deployment of autonomous networks

Fei XUE1, Bin CHEN1, Jing LIU2, Xiaoyang LIANG2, Lin ZHU2, Feng WANG2, Tian LI1, Liang ZHANG1, Zhenzhen CHEN1, Xiao LI1   

  1. 1 China Mobile Group Guangdong Co., Ltd, Guangzhou 510632, China
    2 China Mobile Research Institute, Beijing 100053, China
  • Revised:2023-09-06 Online:2023-08-01 Published:2023-08-01

摘要:

自智网络通过构建智能化的网络基础设施,实现网络自主管理、自优化和自修复。自智网络分成能力建设和能力部署两个关键阶段,目前业界较少关注能力部署。首先系统研究了自智网络的能力部署阶段,随后介绍自智网络架构,然后提出了 3 项自智网络全栈部署核心技术,最后通过异常检测、智慧机房和设备巡检等案例验证核心技术的有效性。对自智网络能力部署进行了系统探讨,对运营商实施网络智能化转型具有重要的参考价值。

关键词: 自智网络, 全栈部署, 训推一体, 云边协同部署, AI能力编织

Abstract:

Autonomous networks achieve network self management, self optimization, and self repair by building intelligent network infrastructure.Autonomous network was divided into two key stages: AI model building and AI model deployment.However, the industry paid less attention to AI model deployment.The deployment phase of autonomous networks was systematically studied.Firstly, it elaborated on the independent deployment mode and full stack deployment mode of autonomous networks, and pointed out that full stack deployment was the main direction.Secondly, a detailed introduction was given to the full stack architecture with “five layers, dual domains, and four closed-loops”, which achieved full life cycle intelligence through a layered closed-loop design of resources and processes.Then, three core technologies for independent innovation were proposed: AI model training and inference integration to achieve rapid iterative updates of models, AI fabric technology to achieve customized application by rapid construction, and AI model cloud-edge collaborative deployment technology to achieve efficient application.Finally, the effectiveness of these three core technologies was verified through cases such as anomaly detection, smart telecommunication rooms, and equipment inspections.The deployment of autonomous networks was systematically explored, especially in terms of architecture design and core technology innovation, which had important reference value for telecommunication operators’ network digital transformation.

Key words: autonomous network, full stack deployment, training and inference integration, cloud edge collaborative deployment, AI fabric

中图分类号: 

No Suggested Reading articles found!