Mamba-ADCL： time series anomaly detection method based on contrastive learning and Mamba

doi:10.11772/j.issn.1001-9081.2025080958

Abstract

Abstract:

To address the issues of the existing time series anomaly detection methods， such as the heavy reliance on labeled data and the insufficient modeling capability in complex scenarios， a Mamba-ADCL （Anomaly Detection in time series via Contrastive Learning with Mamba） was proposed. In the method， a dual-branch Mamba structure was designed， in which a shared-weight framework and a contrastive learning strategy were employed to enhance the feature discrepancies between normal and abnormal samples effectively， thereby improving model robustness and generalization. Meanwhile， a representation method with collaborative time-domain and frequency-domain modeling was introduced， in which time series data was transformed into frequency representations and combined with time-domain data to construct a dual-view contrastive learning mechanism jointly， thereby enhancing the recognition ability for abnormal patterns. Experimental results on four public datasets， MSL， SMAP， SWaT， and PSM show that the F1-score of Mamba-ADCL is improved by 2.22， 0.22， 3.86， and 0.35 percentage points compared with that of AnomalyTrans （Anomaly Transformer）， and by 2.43， 0.52， 0.70， and 0.28 percentage points compared with that of TFMAE （Temporal-Frequency Masked AutoEncoder）， which verifying the effectiveness of Mamba-ADCL in time series anomaly detection.

Key words: Mamba, contrastive learning, time series anomaly detection, self-supervised learning, frequency domain

摘要：

针对现有的时间序列异常检测方法普遍存在对标注数据依赖性强、在复杂场景下的建模能力不足等问题，提出一种基于对比学习与Mamba的时间序列异常检测方法Mamba-ADCL（Anomaly Detection in time series via Contrastive Learning with Mamba）。该方法设计双分支Mamba结构，利用共享权重结构和对比学习策略，有效扩大正常与异常样本间的特征差异，提升模型鲁棒性与泛化能力；同时，提出时域与频域协同建模的表征方法，将时间序列数据转换为频域表示，并与时域数据共同构建双视角的对比学习机制，从而增强对异常模式的识别能力。在4个公开数据集MSL、SMAP、SWaT和PSM上的实验结果表明，Mamba-ADCL的F1-score相较于AnomalyTrans（Anomaly Transformer）分别提升了2.22、0.22、3.86和0.35个百分点，相较于时频掩码自编码器（TFMAE）分别提升了2.43、0.52、0.70和0.28个百分点，验证了该方法在时间序列异常检测中的有效性。

关键词: Mamba, 对比学习, 时间序列异常检测, 自监督学习, 频域

CLC Number:

TP391.4

Zhihong ZHAO, Xushuo LIU. Mamba-ADCL： time series anomaly detection method based on contrastive learning and Mamba[J]. Journal of Computer Applications, 2026, 46(10): 3199-3205.

赵志宏, 刘栩硕. 基于对比学习与Mamba的时间序列异常检测方法Mamba-ADCL[J]. 《计算机应用》唯一官方网站, 2026, 46(10): 3199-3205.

Figures/Tables 9

References 30

[1]	Pang G， Shen C， Cao L， et al. Deep learning for anomaly detection： a review［J］. ACM Computing Surveys， 2022， 54（2）： No.38.
[2]	Yang Y， Zhang H， Li Y. Pipeline safety early warning by multifeature-fusion CNN and LightGBM analysis of signals from distributed optical fiber sensors［J］. IEEE Transactions on Instrumentation and Measurement， 2021， 70： No.2514213.
[3]	Yamanishi K， Takeuchi J I. A unifying framework for detecting outliers and change points from non-stationary time series data［C］// KDD 2002. New York： ACM， 2002： 676-681.
[4]	Moshtaghi M， Bezdek J C， Leckie C， et al. Evolving fuzzy rules for anomaly detection in data streams［J］. IEEE Transactions on Fuzzy Systems， 2015， 23（3）： 688-700.
[5]	Canizo M， Triguero I， Conde A， et al. Multi-head CNN-RNN for multi-time series anomaly detection： an industrial case study［J］. Neurocomputing， 2019， 363： 246-260.
[6]	Park D， Hoshi Y， Kemp C C. A multimodal anomaly detector for robot-assisted feeding using an LSTM-based variational autoencoder［J］. IEEE Robotics and Automation Letters， 2018， 3（3）： 1544-1551.
[7]	Vaswani A， Shazeer N， Parmar N， et al. Attention is all you need［C］// NeurIPS 2017. Red Hook： Curran Associates Inc.， 2017： 6000-6010.
[8]	Xu J， Wu H， Wang J， et al. Anomaly Transformer： time series anomaly detection with association discrepancy［PP/OL］. V5. arXiv （2022-06-29）［2025-05-03］..
[9]	Zhang C， Zhang K， Zhang C， et al. How does SimSiam avoid collapse without negative samples？ a unified under-standing with self-supervised contrastive learning［PP/OL］. arXiv （2022-03-30）［2025-05-03］..
[10]	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： No.110874.
[11]	Gu A， Goel K， Ré C. Efficiently modeling long sequences with structured state spaces［PP/OL］. V3. arXiv （2022-08-05）［2025-05-03］..
[12]	Gu A， Dao T. Mamba： linear-time sequence modeling with selective state spaces［PP/OL］. V2. arXiv （2024-05-31）［2025-05-03］..
[13]	Cai X， Zhu Y， Wang X， et al. MambaTS： improved selective state space models for long-term time series forecasting［PP/OL］. arXiv （2024-05-26）［2025-05-03］..
[14]	Liang A， Jiang X， Sun Y， et al. Bi-Mamba+： bidirectional Mamba for time series forecasting［PP/OL］. V3. arXiv （2024-06-27）［2025-05-03］..
[15]	Zhang W， Huang J， Wang R， et al. Integration of Mamba and Transformer-MAT for long-short range time series forecasting with application to weather dynamics［C］// ICECCE 2024. Piscataway： IEEE， 2024： 1-6.
[16]	Wu Y， Meng X， Hu H， et al. Affirm： interactive mamba with adaptive Fourier filters for long-term time series forecasting［J］. Proceedings of the AAAI Conference on Artificial Intelligence， 2025， 39（20）： 21599-21607.
[17]	Geiger A， Liu D， Alnegheimish S， et al. TadGAN： time series anomaly detection using generative adversarial networks［C］// Big Data 2020. Piscataway： IEEE， 2020： 33-43.
[18]	Li Y， Peng X， Zhang J， et al. DCT-GAN： dilated convolutional Transformer-based GAN for time series anomaly detection［J］. IEEE Transactions on Knowledge and Data Engineering， 2023， 35（4）： 3632-3644.
[19]	Fang Y， Xie J， Zhao Y， et al. Temporal-frequency masked autoencoders for time series anomaly detection［C］// ICDE 2024. Piscataway： IEEE， 2024： 1228-1241.
[20]	郑育靖，何强，张长伦，等. 基于GRU-Attention的无监督多变量时间序列异常检测［J］. 山西大学学报（自然科学版）， 2020， 43（4）： 756-764.
	Zheng Yujing， He Qiang， Zhang Changlun， et al. GRU-Attention based unsupervised multivariate time series anomaly detection［J］. Journal of Shanxi University （Natural Science Edition）， 2020， 43（4）： 756-764.
[21]	黄训华，张凤斌，樊好义，等. 基于多模态对抗学习的无监督时间序列异常检测［J］. 计算机研究与发展， 2021， 58（8）： 1655-1667.
	Huang Xunhua， Zhang Fengbin， Fan Haoyi， et al. Multimodal adversarial learning based unsupervised time series anomaly detection［J］. Journal of Computer Research and Development， 2021， 58（8）： 1655-1667.
[22]	段雪源，付钰，王坤. 基于VAE-WGAN的多维时间序列异常检测方法［J］. 通信学报， 2022， 43（3）： 1-13.
	Duan Xueyuan， Fu Yu， Wang Kun. Multi-dimensional time series anomaly detection method based on VAE-WGAN［J］. Journal on Communications， 2022， 43（3）： 1-13.
[23]	Hundman K， Constantinou V， Laporte C， et al. Detecting spacecraft anomalies using LSTMs and nonparametric dynamic thresholding［C］// KDD 2018. New York： ACM， 2018： 387-395.
[24]	Abdulaal A， Liu Z， Lancewicki T. Practical approach to asynchronous multivariate time series anomaly detection and localization［C］// KDD 2021. New York： ACM， 2021： 2485-2494.
[25]	Mathur A P， Tippenhauer N O. SWaT： a water treatment testbed for research and training on ICS security［C］// CySWater 2016. Piscataway： IEEE， 2016： 31-36.
[26]	Zhou B， Liu S， Hooi B， et al. BeatGAN： anomalous rhythm detection using adversarially generated time series［C］// IJCAI 2019. California： IJCAI， 2019： 4433-4439.
[27]	Su Y， Zhao Y， Niu C， et al. Robust anomaly detection for multivariate time series through stochastic recurrent neural network［C］// KDD 2019. New York： ACM， 2019： 2828-2837.
[28]	Li Z， Zhao Y， Han J， et al. Multivariate time series anomaly detection and interpretation using hierarchical inter-metric and temporal embedding［C］// KDD 2021. New York： ACM， 2021： 3220-3230.
[29]	Shen L， Li Z， Kwok J T. Timeseries anomaly detection using temporal hierarchical one-class network［C］// NeurIPS 2020. Red Hook： Curran Associates Inc.， 2020： 13016-13026.
[30]	Wu H， Hu T， Liu Y， et al. TimesNet： temporal 2D-variation modeling for general time series analysis［PP/OL］. V3. arXiv （2023-04-12）［2025-05-03］..

数据集	维度	样本数		异常比例/%
数据集	维度	训练集	测试集	异常比例/%
MSL	55	58 317	73 729	10.5
SMAP	25	135 183	427 617	12.8
SWaT	51	495 000	449 919	12.1
PSM	25	132 481	87 841	27.8

数据集	维度	样本数		异常比例/%
数据集	维度	训练集	测试集	异常比例/%
MSL	55	58 317	73 729	10.5
SMAP	25	135 183	427 617	12.8
SWaT	51	495 000	449 919	12.1
PSM	25	132 481	87 841	27.8

模型	MSL			SMAP			SWaT			PSM
模型	P	R	F₁	P	R	F₁	P	R	F₁	P	R	F₁
Trans-ADCL	93.48	75.08	83.28	92.12	89.42	90.75	96.91	77.46	86.10	98.06	97.58	97.82
w/o SW	94.71	96.14	95.42	95.82	94.77	95.29	95.93	99.76	97.81	98.15	97.13	97.64
w/o SG	94.50	91.98	93.22	96.09	97.69	96.88	95.81	99.18	97.47	98.18	97.82	98.00
w/o Freq	94.00	82.07	87.63	94.84	84.38	89.30	93.65	98.27	95.90	98.09	94.28	96.15
w/o Time	94.60	88.53	91.47	94.50	91.38	92.91	93.53	95.28	94.40	98.28	97.79	98.03
Mamba-ADCL	95.16	96.46	95.81	96.18	97.65	96.91	95.95	100.00	97.93	98.14	98.34	98.24

模型	MSL			SMAP			SWaT			PSM
模型	P	R	F₁	P	R	F₁	P	R	F₁	P	R	F₁
Trans-ADCL	93.48	75.08	83.28	92.12	89.42	90.75	96.91	77.46	86.10	98.06	97.58	97.82
w/o SW	94.71	96.14	95.42	95.82	94.77	95.29	95.93	99.76	97.81	98.15	97.13	97.64
w/o SG	94.50	91.98	93.22	96.09	97.69	96.88	95.81	99.18	97.47	98.18	97.82	98.00
w/o Freq	94.00	82.07	87.63	94.84	84.38	89.30	93.65	98.27	95.90	98.09	94.28	96.15
w/o Time	94.60	88.53	91.47	94.50	91.38	92.91	93.53	95.28	94.40	98.28	97.79	98.03
Mamba-ADCL	95.16	96.46	95.81	96.18	97.65	96.91	95.95	100.00	97.93	98.14	98.34	98.24

模型	MSL			SMAP			SWaT			PSM
模型	P	R	F₁	P	R	F₁	P	R	F₁	P	R	F₁
LSTM-VAE	85.49	79.94	82.62	92.20	67.75	78.10	76.00	89.50	82.20	73.62	89.92	80.96
BeatGAN	89.75	85.42	87.53	92.38	55.85	69.61	64.01	87.46	73.92	90.30	93.84	92.04
LSTM	85.45	82.50	83.95	89.41	78.13	83.39	86.15	83.27	84.69	76.93	89.64	82.80
OmniAnomaly	89.02	86.37	87.67	92.49	81.99	86.92	81.42	84.30	82.83	88.39	74.46	80.83
InterFusion	81.28	92.70	86.62	89.77	88.52	89.14	80.59	85.58	83.01	83.61	83.45	83.52
THOC	88.45	90.97	89.69	92.06	89.34	90.68	83.94	86.36	85.13	88.14	90.99	89.54
AnomalyTrans	92.09	95.15	93.59	94.13	99.40	96.69	91.55	96.73	94.07	96.91	98.90	97.89
TimesNet	89.41	75.26	81.73	89.98	55.28	68.48	91.58	95.32	93.42	98.46	95.89	97.16
TFMAE	92.92	93.84	93.38	94.25	98.62	96.39	97.42	97.05	97.23	97.66	98.25	97.96
Mamba-ADCL	95.16	96.46	95.81	96.18	97.65	96.91	95.95	100.00	97.93	98.14	98.34	98.24