📑PatchAD: A Lightweight Patch-based MLP-Mixer for Time Series Anomaly Detection

This is the implementation of PatchAD: A Lightweight Patch-based MLP-Mixer for Time Series Anomaly Detection.

The paper has been accepted by TBD 2025 and appear in early access. For now, the latest preprint is available on arXiv. We added further experiments and analyses. Thank you for your interest!

If you have any questions, please feel free to contact me or raise an Issue.

📰Abstract

Anomaly detection in time series analysis is a pivotal task, yet it poses the challenge of discerning normal and abnormal patterns in label-deficient scenarios. While prior studies have largely employed reconstruction-based approaches, which limits the models' representational capacities. Moreover, existing deep learning-based methods are often not sufficiently lightweight. Addressing these issues, we present PatchAD, our novel, highly efficient multi-scale patch-based MLP-Mixer architecture that utilizes contrastive learning for representation extraction and anomaly detection. With its four distinct MLP Mixers and innovative dual project constraint module, PatchAD mitigates potential model degradation and offers a lightweight solution, requiring only 3.2MB. Its efficacy is demonstrated by state-of-the-art results across nine datasets, outperforming over 30 comparative algorithms. PatchAD significantly improves the classical F1 score by 50.5%, the Aff-F1 score by 7.8%, and the AUC by 10.0%. The code is publicly available.

🛠Architecture

📊Overall Performance

💾Datasets

You can download all datasets here. (Thanks for DCdetector repo and its authors.)

🔧Installation

You can refer to requirements.txt to install all the packages.

pip install -r requirements.txt

🔧Quick start

1. You should download the datasets into ABSOLUTE/PATH/OF/DATASET.
2. The dataset structure should be like 'dataset_struct.txt'
3. Run the scripts below.
4. Note that the most important is you should change the parameter of --data_path.
5. We use --model_save_path and --res_pth for model and result saving.
6. TODO

Train

python main_ad.py --anormly_ratio 0.9 -ep 3 --data_path ABSOLUTE/PATH/OF/DATASET --batch_size 128 --mode train --data_name PSM --win_size 105 --stride 1 --patch_size [3,5,7] --patch_mx 0.1 --d_model 60 --e_layer 3 -lr 0.0001

Test

python main_ad.py --anormly_ratio 0.9 -ep 3 --data_path ABSOLUTE/PATH/OF/DATASET --batch_size 128 --mode test --data_name PSM --win_size 105 --stride 1 --patch_size [3,5,7] --patch_mx 0.1 --d_model 60 --e_layer 3 -lr 0.0001

Citation

If you find this repo useful, please cite our paper🥰.

@ARTICLE{zhong2024patchad,
  author={Zhong, Zhijie and Yu, Zhiwen and Yang, Yiyuan and Wang, Weizheng and Yang, Kaixiang and Chen, C. L. Philip},
  journal={IEEE Transactions on Big Data}, 
  title={PatchAD: A Lightweight Patch-Based MLP-Mixer for Time Series Anomaly Detection}, 
  year={2025},
  volume={},
  number={},
  pages={1-15},
  keywords={Mixers;Time series analysis;Anomaly detection;Contrastive learning;Big Data;Data models;Feature extraction;Transformers;Training;Measurement;Time series;anomaly detection;data mining;lightweight;MLP},
  doi={10.1109/TBDATA.2025.3596745}}

Updation

1. 2025-5-24. Update ablations' models.