基于扩张卷积神经网络的异常检测模型  

Anomaly Detection Model Based on Extended Convolutional Neural Network

作　　者：高治军[1] 曹浩东 韩忠华[1] GAO Zhijun;CAO Haodong;HAN Zhonghua(School of Electrical and Control Engineering,Shenyang Jianzhu University,Shenyang,China,110168)

机构地区：[1]沈阳建筑大学电气与控制工程学院,辽宁沈阳110168

出　　处：《沈阳建筑大学学报（自然科学版）》2024年第4期738-744,共7页Journal of Shenyang Jianzhu University：Natural Science

基　　金：国家重点研发计划项目(2018YFF0300304-04);辽宁省重点研发计划项目(2020JH2/10100039);辽宁省教育厅高等学校基本科研项目重点项目(LJKZ0583);沈阳市中青年科技创新人才支持计划项目(RC200026)。

摘　　要：目的提出一种基于DCNN-MiLSTM的异常检测模型,解决传统的网络异常检测模型难以处理具有时序特征网络流量数据的问题。方法对原始流量数据的时间标签进行重定义,利用扩张卷积神经网络对整体特征进行提取,同时引入Mogrifier LSTM网络,对时序信息进行深层次挖掘。结果与其他异常检测模型相比,DCNN-MiLSTM模型的准确率达到99.12%,召回率为98.94%,F_(1)值为99.03%,各项指标均优于其他常见模型,提升了检测异常网络流量数据的能力。结论DCNN-MiLSTM模型可以更好地处理具有时序特征的流量,捕捉流量数据中的时间依赖关系和趋势,更有效地检测和识别出异常数据。A DCNN-MiLSTM-based anomaly detection model is proposed to solve the problem that traditional network anomaly detection models are difficult to handle network traffic data with temporal characteristics.The timestamps of the original traffic data are redefined,and the overall features are extracted by using an expansive convolutional neural network,while the Mogrifier LSTM network is introduced for deeper mining of temporal information.Compared with other anomaly detection models,the DCNN-MiLSTM model achieves an accuracy of 99.12%,a recall of 98.94%,and a F_(1)of 99.03%,which are better than other common models in all metrics,and improves the ability of detecting anomalous network traffic data.The DCNN-MiLSTM model can better deal with traffic flows with temporal characteristics,capture the time in traffic data dependencies and trends in traffic data,and more effectively detect and identify anomalous data.

关 键 词：网络异常检测 扩张卷积神经网络 标签重定义 时序特性 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]