网络切片中物理节点的分布式在线异常检测方法

Distributed online anomaly detection method for physical nodes in network slicing

网络切片中的异常检测问题是实现网络切片自动化管理的重要研究内容,针对网络切片中物理节点的异常检测问题,提出了基于支持向量数据描述的分布式在线物理节点异常检测方法。基于支持向量数据描述建立了一种分布式的物理节点异常检测模型;通过引入随机近似函数,解决了数据分布式存储场景下的核函数计算问题,从而实现观测数据的切片内处理;基于随机梯度下降法,提出了一种在线的物理节点异常检测算法,保证了模型动态更新并减轻了异常数据导致的模型性能下降。在不同条件下进行了仿真分析,仿真结果表明,该方法可在避免切片间观测数据传输的同时,有效利用网络切片中虚拟网络功能的无标签观测信息检测物理节点异常。

The problem of anomaly detection in network slicing is an important part of research to achieve automated management of network slicing.Aiming at the anomaly detection of physical nodes in network slicing,this paper proposes a distributed online physical node anomaly detection method based on support vector data description.Firstly,a distributed physical node anomaly detection model is established based on support vector data description.Secondly,by introducing random approximation function,the kernel function calculation problem is solved,so as to realize the intra-slice processing of observation data;Finally,based on stochastic gradient descent method,an online physical node anomaly detection algorithm is proposed to ensure dynamic updating of the model and mitigate the model performance degradation caused by abnormal data.Simulation analysis was performed under different conditions,and the simulation results show that the proposed method can effectively exploit unlabeled observation data of virtual network functions to detect physical node anomalies while avoiding observation data transmission between slices.