计算机科学

计算机科学 ›› 2026, Vol. 53 ›› Issue (7): 230-241.doi: 10.11896/jsjkx.250600078

• 数据库 & 大数据 & 数据科学 • 上一篇    下一篇

CBT-AD:CNN-BiLSTM-Transformer混合架构的时序异常检测模型

许建1,2, 陈仕杰1, 冯健聪1, 杨庚1,2   

  1. 1 南京邮电大学计算机学院 南京 210023
    2 大数据安全与智能处理省高校重点实验室(南京邮电大学) 南京 210023
  • 收稿日期:2025-06-12 修回日期:2025-10-08 出版日期:2026-07-15 发布日期:2026-07-10
  • 通讯作者: 许建(xuj@njupt.edu.cn)
  • 基金资助:
    国家自然科学基金(62372244)

CBT-AD:CNN-BiLSTM-Transformer Hybrid Model for Time Series Anomaly Detection

XU Jian1,2, CHEN Shijie1, FENG Jiancong1, YANG Geng1,2   

  1. 1 School of Computer Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,China
    2 Jiangsu Key Laboratory of Big Data Security and Intelligent Processing(Nanjing University of Posts and Telecommunications),Nanjing 210023,China
  • Received:2025-06-12 Revised:2025-10-08 Published:2026-07-15 Online:2026-07-10
  • About author:XU Jian,born 1980,Ph.D,associate professor,is a member of CCF(No.P5359M).His main research interests include artificial intelligence and cyberspace security.
  • Supported by:
    National Natural Science Foundation of China(62372244).

PDF (PC)

27

摘要： 现有基于深度学习的时序数据异常检测方法在局部与全局特征协同建模能力以及长序列计算效率方面存在不足。为此,提出一种基于混合架构的异常检测模型。首先,在多尺度特征提取中,采用深度可分离卷积捕获局部细节特征,结合动态池化增强突发异常响应能力。其次,设计双向时序建模机制,通过BiLSTM融合双向上下文特征,并采用门控Dropout抑制长序列训练过程的过拟合风险。进一步,设计分层稀疏全局注意力模块,利用局部窗口注意力捕捉时频特征,借助多头机制与残差连接优化梯度传播稳定性。最后,提出了一种动态特征融合方法,并结合分块处理框架实现检测精度与计算效率的协同优化。在4个公开时序数据集上的实验结果表明,所提模型的各项性能指标较现有方法提升显著,且展现出较好的鲁棒性与泛化能力。

关键词: 时序异常检测, 混合神经网络, 多尺度特征融合, 分层稀疏注意力

Abstract: Existing deep learning-based time series anomaly detection methods exhibit limitations in both the collaborative mode-ling of local and global features,as well as in computational efficiency for long sequences.To address these issues,this paper proposes a hybrid-architecture anomaly detection model.Firstly,for multi-scale feature extraction,depthwise separable convolution is employed to capture local detail features,combined with adaptive pooling to enhance the response capability to burst anomalies.Secondly,a bidirectional temporal modeling mechanism is designed,which integrates bidirectional context features through Bi-LSTM and utilizes gated Dropout to mitigate the risk of overfitting during long-sequence training.Furthermore,a hierarchical sparse global attention module is designed,leveraging local window attention to capture time-frequency features,while employing multi-head mechanisms and residual connections to optimize gradient propagation stability.Finally,a dynamic feature fusion method is proposed and integrated with a chunk-based processing framework to achieve collaborative optimization of detection accuracy and computational efficiency.Experimental results on four public time-series datasets demonstrate that the proposed model achieves significant improvements across various performance metrics compared to existing methods,while also exhibiting strong robustness and generalization capability.

Key words: Time series anomaly detection, Hybrid neural network, Multi-scale feature fusion, Layered sparse attention

中图分类号: 

  • TP183
[1]REN L,JIA Z,LAILI Y,et al.Deep Learning for Time-Series Prediction in IIoT:Progress,Challenges,and Prospects[J].IEEE Transactions on Neural Networks and Learning Systems,2023,35(11):15072-15091.
[2]ZAMANZADEH DARBAN Z,WEBB G I,PAN S,et al.Deep learning for time series anomaly detection:A survey[J].ACM Computing Surveys,2024,57(1):1-42.
[3]YAN P,ABDULKADIR A,LULEY P P,et al.A comprehensive survey of deep transfer learning for anomaly detection in industrial time series:Methods,applications,and directions[J].IEEE Access,2024,12(1):3768-3789.
[4]MOHAMMADI FOUMANI N,MILLER L,TAN C W,et al.Deep learning for time series classification and extrinsic regression:A current survey[J].ACM Computing Surveys,2024,56(9):1-45.
[5]YANG Y P,WANG S T.Research on malicious traffic classification algorithm based on CNN combined with BiGRU[J].Computer Science,2024,51(11A):231100106-9.
[6]YE T,QIN W,LI Y,et al.Dense and small object detection in UAV-vision based on a global-local feature enhanced network[J].IEEE Transactions on Instrumentation and Measurement,2022,71:1-13.
[7]JIN M,LIU Y,ZHENG Y,et al.Anemone:Graph anomaly detection with multi-scale contrastive learning[C]//Proceedings of the 30th ACMInternational Conference on Information & Knowledge Management.2021:3122-3126.
[8]HOCHREITER S,SCHMIDHUBER J.Long short-term memory[J].Neural Computation,1997,9(8):1735-1780.
[9]ZHANG Y,WANG J,CHEN Y,et al.Adaptive memory networks with self-supervised learning for unsupervised anomaly detection[J].IEEE Transactions on Knowledge and Data Engineering,2022,35(12):12068-12080.
[10]LIU P,QIAN W,ZHANG H,et al.Automatic sleep stage classification using deep learning:signals,data representation,and neural networks[J].Artificial Intelligence Review,2024,57(11):301-369.
[11]ZHUANG X B,NIU B,LIN Z J,et al.Multi-parameter deception jamming detection method of parallel CNN-Transformer neural network based on gating mechanism[J/OL].http://kns.cnki.net/kcms/detail/11.4494.TN.20250402.1524.004.html
[12]WANG H,DI X,WANG Y,et al.An Intelligent Digital Twin Method Based on Spatio-Temporal Feature Fusion for IoT Attack Behavior Identification[J].IEEE Journal on Selected Areas in Communications,2023,41(11):3561-3572.
[13]MU X T,CHENG K,SONG A X,et al.Privacy-preserving federated learning against Byzantine attacks[J].Chinese Journal of Computers,2024,47(4):842-861.
[14]PIAO X,CHEN Z,MURAYAMA T,et al.Fredformer:Fre-quency debiased transformer for time series forecasting[C]//Proceedings of the 30th ACM SIGKDD Conference on Know-ledge Discovery and Data Mining.2024:2400-2410.
[15]PAN J,WANG X H.Time series prediction model integrating multi-scale features and attention mechanism[J].Computer Science,2026,53(2):180-186.
[16]CHEN W,LI L,LI J.Multimodal selective state space model-based time series classification for electricity theft detection[J].Expert Systems with Applications,2025,278:127364.
[17]YIN T,ZHANG Z,HOU S,et al.MSAnomaly:Time Series Anomaly Detection with Multi-scale Augmentation and Fusion[C]//International Conference on Advanced Data Mining and Applications.Singapore:Springer,2024:356-371.
[18]CARMONA C U,AUBET F X,FLUNKERT V,et al.Neural contextual anomaly detection for time series[J].arXiv:2107.07702,2021.
[19]MOORE A,MORELLI D.conDENSE:Conditional Density Estimation for Time Series Anomaly Detection[J].Journal of Artificial Intelligence Research,2024,79:801-824.
[20]GOSWAMI M,CHALLU C,CALLOT L,et al.Unsupervised model selection for time-series anomaly detection[J].arXiv:2210.01078,2022.
[21]SCHMIDL S,NAUMANN F,PAPENBROCK T.AutoTSAD:Unsupervised Holistic Anomaly Detection for Time Series Data[J].Proceedings of the VLDB Endowment,2024,17(11):2987-3002.
[22]GHOJOGH B,TOUTOUNCHIAN M A.Probabilistic Classification by Density Estimation Using Gaussian Mixture Model and Masked Autoregressive Flow[J].arXiv:2310.10843,2023.
[23]ZHOU H,ZHANG S,PENG J,et al.Informer:Beyond efficient transformer for long sequence time-series forecasting[C]//Proceedings of the AAAI Conference on Artificial Intelligence.2021:11106-11115.
[24]XU J,WU H,WANG J,et al.Anomaly transformer:Time series anomaly detection with association discrepancy[J].arXiv:2110.02642,2021.
[25]TULI S,CASALE G,JENNINGS N R.Tranad:Deep trans-former networks for anomaly detection in multivariate time series data[J].arXiv:2201.07284,2022.
[26]YUE Z,WANG Y,DUAN J,et al.Ts2vec:Towards universal representation of time series[C]//Proceedings of the AAAI Conference on Artificial Intelligence.2022:8980-8987.
[27]YANG Y,ZHANG C,ZHOU T,et al.Dcdetector:Dual attention contrastive representation learning for time series anomaly detection[C]//Proceedings of the 29th ACM SIGKDDConfe-rence on Knowledge Discovery and Data Mining.2023:3033-3045.
[28]DARBAN Z Z,WEBB G I,PAN S,et al.CARLA:Self-supervised contrastive representation learning for time series anomaly detection[J].Pattern Recognition,2025,157:110874.
[29]LEE S,PARK T,LEE K.Soft contrastive learning for time series[J].arXiv:2312.16424,2023.
[30]ZHAO B C,MA J J,CUI L,et al.Anomaly detection of photovoltaic power generation based on improved VMD-XGBoost-BiLSTM combined model[J].Computer Engineering,2024,50(3):306-316.
[31]NING Z,MIAO H,JIANG Z,et al.Using Multi-Scale Convolution Fusion and Memory-Augmented Adversarial Autoencoder to Detect Diverse Anomalies in Multivariate Time Series[J].Tsinghua Science and Technology,2024,30(1):234-246.
[32]LIU J,LI Q,AN S,et al.EdgeConvFormer:DynamicGraph CNN and Transformer based Anomaly Detection in Multivariate Time Series[J].arXiv:2312.01729,2023.
[33]ZHANG S,DU M,ZHANG G J.Research on TemperatureForecasting Model Based on BiLSTM-Attention Algorithm[C]//International Conference on Image,Vision and Intelligent Systems.Singapore:Springer,2025:461-472.
[34]ZHOU T,MA Z,WEN Q,et al.Fedformer:Frequency enhanced decomposed transformer for long-term series forecasting[C]//International Conference on Machine Learning.PMLR,2022:27268-27286.
[35]BAI N,WANG X,HAN R,et al.PAFormer:anomaly detection of time series with parallel-attention transformer[J].IEEE Transactions on Neural Networks and Learning Systems,2023,36(2):3315-3328.
[36]LIU J,LI Q,AN S,et al.EdgeConvFormer:Dynamic GraphCNN and Transformer based Anomaly Detection in Multivariate Time Series[J].arXiv:2312.01729,2023.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
渝ICP备16013121号-4
地址：重庆市两江新区洪湖西路18号    邮编：401121  
电话：023-63500828（编辑部） 023-67039612（会议/专题） 023-67039623（财务/发票）
E-mail：jsjkx12@163.com
本系统由北京玛格泰克科技发展有限公司设计开发