摘要
5G网络发生故障可能影响全网稳定性和服务质量,其故障发现与修复是5G网络运维管理的关键之一。基于历史故障发生前的网元关键绩效指标(Key Performance Indicator,KPI)变化以及设备告警信息、指标统计特征、自动异常检测、指标关联特征、告警编码特征等,提出了5G核心网网元多维特征融合故障预警方法,构建了多维特征空间训练故障预警模型。最后,针对现网运行数据,选取了6类网元并进行了实验验证,实验结果表明,该模型对各种类型网元的预警效果有不同程度提升,平均F1值提升了18%。
Faults in 5G network operation may affect the stability and service quality of the whole network.Fault detection and fixing are keys to network operation and maintenance. Based on KPI(Key Performance Indicator) changes of network elements before the occurrence of historical faults, as while as equipment alarm information, indicator statistical features, automatic anomaly detection, indicator correlation features,alarm coding features, etc., this paper proposes a machine learning method based on feature fusion for fault warning, and constructs a multi-dimensional feature space training fault warning model. Finally, according to current network operation data, 6 types of network elements are selected and verified by experiments. The experimental results indicate that the warning effect of the model on various types of network elements is improved to varying degrees, and a state-of-the-art improvement with 18% of F1 score is proved.
作者
韦强申
宋勇
李红霞
王希栋
叶晓舟
欧阳晔
WEI Qiangshen;SONG Yong;LI Hongxia;WANG Xidong;YE Xiaozhou;OUYANG Ye(Telco Artificial Intelligence Labs,AsiaInfo Technologies(China)Co.,Ltd.,Beijing 100193,China)
出处
《通信技术》
2022年第3期394-403,共10页
Communications Technology
关键词
5G核心网
故障预警
多特征融合
机器学习
5G core network
fault warning
multi-feature fusion
machine learning