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

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.