Research on Generative AI-Driven Network Attack Sample Generation and Intelligent Detection Technology

Abstract

Traditional intrusion detection methods have limited ability to identify unknown attacks and attack variants, especially when high-quality attack samples are insufficient. To address this bottleneck, this paper proposes a proactive defense framework based on generative AI for network attack sample generation and intelligent detection. The framework first constructs a CGAN-VAE hybrid attack sample generator to synthesize realistic and diverse malicious traffic. A coevolutionary adversarial training mechanism is then designed so that the generator and detector are optimized iteratively through dynamic game learning. Finally, an active learning strategy filters high-value samples and forms a closed loop of “generation–detection–enhancement”. Experimental results show that the generated samples reach a feature coverage area of 31.5, representing a 95.9% improvement over SMOTE, and the trained detector achieves 91.7 % recall and 92.3% F1-score for unknown attack detection. The framework is applicable to industrial control networks, wireless monitoring systems, and electromagnetic-compatible manufacturing infrastructures, where secure data links and reliable antenna-based communication are essential for preventing disruptions in automated production systems.