基于集成学习的入侵检测系统对抗攻击检测  

Adversarial attacks detection in intrusion detection systems with ensemble learning

作　　者：李青青 张凯 李晋国 赵健 LI Qing-qing;ZHANG Kai;LI Jin-guo;ZHAO Jian(College of Computer Science and Technology,Shanghai University of Electric Power,Shanghai 201306,China)

机构地区：[1]上海电力大学计算机科学与技术学院,上海201306

出　　处：《计算机工程与设计》2025年第3期850-856,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(U1936213)。

摘　　要：为解决已有对抗攻击检测方案无法保证正常数据检测精度和使用数据集过时的问题,提出一种基于集成学习的防御方案AADEL,是一个双模块的训练框架,结合集成学习和对抗训练技术,提高模型的鲁棒性,保持对正常样本的检测精度。基于CICIDS2018基准数据集实现AADEL模型,实验结果表明,相比于传统解决方案,AADEL在正常样本和对抗样本上都有更好检测性能。To solve the problem that existing adversarial attack detection schemes cannot guarantee the detection accuracy of normal data and use outdated datasets,an ensemble learning-based defense scheme,AADEL,was proposed as a two-module trai-ning framework that combined ensemble learning and adversarial training techniques to improve the robustness of the model while maintaining the detection accuracy on normal samples.The AADEL model was implemented based on the CICIDS2018 benchmark dataset.Experimental results show that AADEL has better detection performances on both normal and adversarial samples compared to traditional solutions.

关 键 词：机器学习 入侵检测系统 对抗攻击 集成学习 对抗训练 对抗检测 网络安全 

分 类 号：TP309.2[自动化与计算机技术—计算机系统结构]