Skip to content
forked from huankoh/CST-GL

Multivariate Time-Series Anomaly Detection with GNN.

Notifications You must be signed in to change notification settings

canghaiwuya/CST-GL

 
 

Repository files navigation

CST-GL

This is a PyTorch implementation of the paper: Correlation-aware Spatial-Temporal Graph Learning for Multivariate Time-series Anomaly Detection (TNNLS).

Requirements

The model is implemented using Python3.7 with dependencies specified in requirements.txt

Data Preparation

Water Treatment datasets

Follow the instruction in https://itrust.sutd.edu.sg/ to download SWaT and WADI datasets. Following Deng & Hooi (2021), we pre-process and downsample the original dataset. See the jupyter notebook in generate_data folder.

Server Machine datasets

Download the SMD dataset from https://github.com/zhhlee/InterFusion. Pre-process and place it to the data folder. Refer to generate_data foler.

Anomaly Detection

Water Treatment datasets

  • SWaT
python run.py --data data/swat --expid swat --delays [0,6,30,60,120,180,360] --num_nodes 51 --subgraph_size 15 --normalization_window 28000 --pca_compo 5
  • WADI
python run.py --data data/wadi --expid wadi --delays [0,6,30,60,120,180,360] --num_nodes 127 --subgraph_size 30  --normalization_window 200 --pca_compo 100

Server Machine datasets

  • SMD

Few examples from server machine of different groups. For more information about dataset grouping please refer to Su et al. (2019).

Machine-1-1

python run.py --data data/machine-1-1  --expid machine-1-1 --delays [0,1,5,10,20,30,60] --num_nodes 38 --subgraph_size 10 --pca_compo 8

Machine-2-7

python run.py --data data/machine-2-7 --expid machine-2-7 --delays [0,1,5,10,20,30,60] --num_nodes 38 --subgraph_size 10 --pca_compo 8

Machine-3-4

python run.py --data data/machine-3-4 --expid machine-3-4 --delays [0,1,5,10,20,30,60] --num_nodes 38 --subgraph_size 10 --pca_compo 26

Citation

If you find this research useful, please cite our paper:

@article{zheng2023correlationaware,
      title={Correlation-aware Spatial-Temporal Graph Learning for Multivariate Time-series Anomaly Detection}, 
      author={Yu Zheng and Huan Yee Koh and Ming Jin and Lianhua Chi and Khoa T. Phan and Shirui Pan and Yi-Ping Phoebe Chen and Wei Xiang},
      year={2023},
      journal={IEEE Transactions on Neural Networks and Learning Systems}
}

About

Multivariate Time-Series Anomaly Detection with GNN.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 75.6%
  • Jupyter Notebook 24.4%