摘要
针对DDoS分布式拒绝服务攻击早期检测的时效性与准确性,提出一种基于时空特征融合的LSTM-GAN混合预测模型。通过构建双通道特征提取模块同步捕获网络流量数据的时间和空间关联特征,实现攻击特征的跨维度融合。在对抗训练框架下,通过引入GAN生成对抗网络机制,借助生成器模拟攻击流量演变模式,驱动判别器提升对攻击初期流量变异系数小于5%、持续时间不足10 s的微小波动特征的敏感性。该方法可在攻击流量未形成显著峰值时实现早期预警,为主动式网络安全防护提供新的技术路径。
It addresses the timeliness and accuracy requirements for early detection of DDoS attacks by proposing an LSTM-GAN hybrid prediction model based on spatiotemporal feature fusion.A dual-channel feature extraction module is constructed to synchronously capture temporal and spatial correlation characteristics in network traffic data,achieving cross-dimensional fusion of attack features.Under an adversarial training framework,the introduction of a GAN mechanism enables the generator to simulate the evolution patterns of attack traffic,thereby driving the the discriminator to enhance its sensitivity to the tiny fluctuation features of initial attack traffic with a coefficient of variation<5%and a duration<10 seconds.This method facilitates early warning when attack traffic has not yet formed significant peaks,providing a novel technical approach for proactive cybersecurity defense.The proposed methodology offers a new pathway for implementing preventive network security protection strategies.
作者
杨飞
周晗
由志远
王新
Yang Fei;Zhou Han;You Zhiyuan;Wang Xin(China Information Technology Designing&Consulting Institute Co.,Ltd.,Beijing 100048,China;Anhui University of Science and Technology,Hefei 230041,China)
出处
《邮电设计技术》
2025年第9期14-19,共6页
Designing Techniques of Posts and Telecommunications
