基于无监督学习的无线网络性能异常检测方法  

Performance Anomaly Detection Method Based on Unsupervised Learning

作　　者：张乐[1] 吴艳芹[1] 杨昊 张平[3] 胡华伟[2] ZHANG Le;WU Yanqin;YANG Hao;ZHANG Ping;HU Huawei(Research Institute of China Telecom Corporation Limited,Beijing 102209,China;Fujian Branch of China Telecom Corporation Limited,Fuzhou 350001,China)

机构地区：[1]中国电信股份有限公司研究院,北京102209 [2]中国电信股份有限公司福建分公司,福建福州350001 [3]中国电信股份有限公司研究院云网运营技术研究所,北京102209

出　　处：《无线电通信技术》2022年第4期758-762,共5页Radio Communications Technology

基　　金：国家部委基金资助项目。

摘　　要：在无线网络的实际运营中,性能异常检测主要依靠人工规则和阈值,对网络容量和覆盖等进行判断,检测手段单一,难以适应复杂多变的无线网络变化趋势。针对该问题,就无线移动网络性能异常的诊断识别问题,给出了三类通用的检测方法,分别为基于统计特征的异常检测、基于密度的异常检测以及基于聚类的异常检测,并选取现网性能指标数据,对三种算法进行评估分析,结果表明,基于聚类的异常检测算法在对无线网络诊断识别上效果最好。In the actual operation of wireless network,abnormal performance detection mainly relies on manual rules and thresholds to judge network capacity and coverage.And the detection method is monotonous,which is difficult to adapt to the complex and changeable trend of wireless network evolution.To facilitate the diagnosis of wireless mobile network performance anomaly recognition problem,this paper gives three common detection methods,respectively,based on statistical characteristics,density,and clustering.And existing network performance data is selected to evaluate three algorithms,results show that the Clustering based anomaly detection algorithm has the best performance in wireless network diagnosis and identification.

关 键 词：无线网络 无监督学习 性能异常检测 性能劣化 

分 类 号：TN919.23[电子电信—通信与信息系统]