基于深度学习的网络异常检测和智能流量预测方法

边缘计算场景下,边缘设备时刻产生海量蜂窝流量数据,在异常检测任务中针对直接对原始数据检测异常存在的计算冗余问题,提出基于特征降维的蜂窝流量数据异常检测方法.该方法在全局范围内利用LSTM自编码器提取流量数据特征和标识异常网格,然后在存在可疑异常的网格使用K?means聚类进行局部异常确认,结果表明可以更好地检测出不同活...

Research on Network Anomaly Detection and Intelligent Traffic Prediction Method Based on Deep Learning

In order to solve the problems of data redundancy and calculation redundancy in anomaly detection of large-scale cellular traffic data,a traffic anomaly detection scheme based on feature dimension reduction is proposed.The method uses LSTM auto-encoder to extract low-dimensional features of traffic data,and then identifies the grid where an exception exists.Then,the k-means clustering approach is used to confirm local anomalies in the identified grids.This method can efficiently detect missed anomalies in different areas.At the same time,in the traffic data prediction task,only the spatiotemporal correlation is considered,while the correlation between different services is ignored.We propose a multi-data set joint prediction method,and introduce an attention mechanism to learn the relevance between different businesses.Theoretical analysis and numerical simulations verify the effectiveness of the scheme.