摘要
当前,为满足多种应用场景的各项指标需求,5G网络引入HTTP/2协议以提高网络功能数据传输速率和并发能力,然而针对HTTP/2协议的低速DoS攻击具有流量峰值低、攻击过程隐蔽等特点,严重威胁网络安全。通过分析现有低速DoS攻击原理和识别算法的不足,提出一种基于卷积神经网络的HTTP/2协议低速DoS识别方法。首先,提取HTTP/2控制帧字节级别数据构建流量特征灰度图;其次,设计具有卷积计算、池化降维和全连接dropout的卷积神经网络,并将特征灰度图输入到神经网络中训练调优;最后,将训练好的模型用于低速DoS流量识别。仿真结果表明,所提方法在分类准确性、泛化性等方面优于现有识别分类算法,为5G网络安全提供更好的防护。
To meet the requirements of various indicators in a variety of application scenarios,HTTP/2 protocol is introduced into 5G networks to improve the data transmission rate and concurrency of network functions.However,slow rate DoS attacks against the HTTP/2 protocol have low traffic peaks and concealed attack processes characteristics,which seriously threaten network security.By analyzing the existing slow rate DoS attack principles and identification algorithms deficiencies,a slow rate DoS recognition of HTTP/2 protocol based on convolutional neural network is proposed.Firstly,the byte-level data of HTTP/2 control frame is extracted to construct a traffic grayscale map.Then,we design a convolutional neural network with convolution calculation,pooling dimensionality reduction and fully connected dropout.Besides,the feature gray map is input into the neural network for training and tuning.Finally,the trained model is used for slow rate DoS traffic recognition.Simulation results show that the proposed method is superior to existing recognition and classification algorithms in classification accuracy and generalization,and provides better protection for 5G network security.
作者
张奕鸣
刘彩霞
刘树新
潘菲
石灏苒
ZHANG Yiming;LIU Caixia;LIU Shuxin;PAN Fei;SHI Haoran(Information Engineering University,Zhengzhou 450001,China)
出处
《信息工程大学学报》
2022年第4期392-401,共10页
Journal of Information Engineering University
基金
国家科技重大专项资助项目(2018ZX03002002)。