基于深度学习与开集识别技术的对抗式DDoS攻击检测技术  

Adversarial DDoS Attack Detection Based on Deep Learning and Open Set Recognition Techniques

作　　者：吴志祥 刘莉丹 高博 Wu Zhixiang;Liu Lidan;Gao Bo(China Unicom Heilongjiang Branch,Haerbin 150001,China)

机构地区：[1]中国联通黑龙江分公司,黑龙江哈尔滨150001

出　　处：《邮电设计技术》2024年第8期18-23,共6页Designing Techniques of Posts and Telecommunications

摘　　要：网络已成为现代生活不可或缺的一部分,但也面临着诸多的安全风险,特别是分布式拒绝服务(DDoS)攻击。利用人工智能(AI)技术可应对DDoS攻击带来的挑战。基于CNN-Geo和CycleGAN技术,提出一种包含一个增量学习模块的防御模型,该增量学习模块能够训练未知流量并不断提高模型的防御能力。该模型可以识别偏离学习分布的未知攻击,评估结果表明其准确度超过98.16%,增强了对现实场景中不断演变的DDoS攻击策略的检测和防御能力。The Internet has become an integral part of modern life,but it also faces many security risks,especially Distributed Denial of Service(DDoS)attacks.The use of artificial intelligence(AI)technology can address the challenges posed by DDoS attacks.It proposes a defense model based on CNN-Geo and Cycle GAN techniques,which includes an incremental learning module that is able to train unknown traffic and continuously improve the model's defense capability.This model can identify unknown attacks that deviate from the learning distribution,and the evaluated results show that its accuracy is more than 98.16%,which enhances the ability to detect and defend against the evolving DDoS attack strategies in real scenarios.

关 键 词：DDOS AI 开放集识别 CNN-Geo CycleGAN 增量学习 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]