沈阳工业大学学报2026，Vol.48Issue(3)：103-110,8.DOI:10.7688/j.issn.1000-1646.2026.03.14
融合ResNet和SRU的网络数据流实时异常流量检测技术
Real-time abnormal traffic detection technology for network data streams based on ResNet and SRU
摘要
Abstract
[Objective]With the rapid development of the industrial internet,detecting abnormal traffic in network data streams has become a critical task for ensuring network security.Traditional machine learning models struggle with feature extraction and generalization,making it difficult to handle high-dimensional,diverse,and massive network traffic data.To address these challenges,this study proposed a novel abnormal traffic detection method combining a simple recurrent unit(SRU)with an improved residual network(ResNet).This method aims to enhance detection accuracy and efficiency through spatiotemporal feature extraction while mitigating issues such as overfitting and the gradient vanishing problem,thus offering a more efficient and reliable solution for network abnormal traffic detection.[Methods]A deep learning model integrating SRU and the improved ResNet was constructed.The SRU network handled data screening and temporal feature extraction,enabling efficient parallel computation via reset and forget gates,which significantly boosted training speed.The improved ResNet adopted an atrous residual structure,expanding the receptive field with atrous convolution to enhance feature extraction and alleviate gradient vanishing.By combining these networks,both spatial and temporal features of network traffic data were captured comprehensively.Experiments were conducted on the KDD Cup 99 dataset for binary classification to evaluate the model's performance.[Results]The experimental results show that the ResNet-SRU model achieves a classification accuracy of 98.89％and a precision of 98.66％on the KDD Cup 99 dataset.Compared to methods such as CNN-LSTM,ResNet-GRU,and CNN-GRU,it achieves approximately a 1％improvement.During training,the model demonstrates faster convergence and superior stability.It outperforms comparative models in accuracy,precision,recall,and AUC,highlighting its effectiveness and robustness in abnormal traffic detection.Although training and testing times are slightly longer,the significant improvement in detection performance justifies this trade-off.[Conclusions]The abnormal traffic detection method based on ResNet and SRU shows remarkable advantages in processing high-dimensional network traffic data.By integrating the advantages of atrous residual structures for spatial modeling with SRU's temporal feature extraction,it effectively overcomes the limitations of traditional models in feature extraction and generalization,enhancing detection accuracy and efficiency.However,the model's parameter scale and computational cost still require optimization.Future research will focus on lightweight model design,improving detection performance for imbalanced samples,and further reducing computational overhead to enhance its practical application value.关键词
异常流量检测/长短记忆网络/空洞残差/时空融合/序列模型/特征优化Key words
abnormal traffic detection/long short-term memory network/atrous residual/spatiotemporal fusion/sequence model/feature optimization分类
信息技术与安全科学引用本文复制引用
周保红,张玉松,刘道君,沈柯言,石磊..融合ResNet和SRU的网络数据流实时异常流量检测技术[J].沈阳工业大学学报,2026,48(3):103-110,8.基金项目
水利部重大科技项目(SKS-2022120) （SKS-2022120）
湖北省自然科学基金创新发展联合基金重点项目(2022CFD027) （2022CFD027）
中国长江电力股份有限公司科研项目(2422020006). （2422020006）
