通信学报2026，Vol.47Issue(1)：156-173,18.DOI:10.11959/j.issn.1000−436x.2026011
基于RL-WGAN的5G网络异常数据生成方法
RL-WGAN based method for 5G network anomalous data generation
摘要
Abstract
To address the challenges of data scarcity,protocol complexity,and stealthiness of dynamic attacks in 5G net-work anomaly detection,an anomalous data generation method based on a reinforcement learning Wasserstein generative adversarial network(RL-WGAN)was proposed.By integrating the dynamic reward mechanism of reinforcement learning with protocol constraints,a multi-stage joint optimization framework was constructed.A hierarchical parsing strategy was designed to resolve the general packet radio service turneling protocol user plane(GTP-U)protocol encapsulation bottle-neck,enabling precise extraction of traffic features.An innovative protocol compliance reward function was combined with Wasserstein adversarial loss to ensure that generated data approximates real traffic in both protocol semantics and statistical distribution.Bidirectional temporal modeling was adopted to enhance the generator's capability to capture dynamic traffic evolution patterns.Experimental results demonstrate that both the distributional fidelity and protocol compliance of the gen-erated samples are significantly enhanced.This effectively mitigates the problem of training data scarcity for anomaly detec-tion models,providing a robust data augmentation solution for dynamic security in 5G networks.关键词
强化学习/生成对抗网络/5G网络安全/异常数据生成/数据增强Key words
reinforcement learning/generative adversarial network/5G network security/anomalous data generation/data augmentation分类
信息技术与安全科学引用本文复制引用
宁兆龙,邹道远,周力,欧阳瑞崎,熊炫睿..基于RL-WGAN的5G网络异常数据生成方法[J].通信学报,2026,47(1):156-173,18.基金项目
国家自然科学基金资助项目(No.62171449,No.62272075) （No.62171449,No.62272075）
重庆市自然科学基金资助项目(No.CSTB2024NSCQ-JQX0013,No.CSTB2024NSCQ-QCXMX0058,No.CSTB2025NSCQ-LZX0050)The National Natural Science Foundation of China(No.62171449,No.62272075),The Natural Science Founda-tion of Chongqing(No.CSTB2024NSCQ-JQX0013,No.CSTB2024NSCQ-QCXMX0058,No.CSTB2025NSCQ-LZX0050) （No.CSTB2024NSCQ-JQX0013,No.CSTB2024NSCQ-QCXMX0058,No.CSTB2025NSCQ-LZX0050）
