Privacy-Preserving Lightweight Time-Series Anomaly Detection for Resource-Limited Industrial IoT Edge Devices (TII 2025)
This repository provides a PyTorch implementation of PPLAD (paper).
The recommended requirements for PPLAD are specified as follows:
- torch==1.13.0
- numpy==1.26.4
- pandas==2.2.2
- scikit-learn==1.5.1
- matplotlib==3.9.2
- statsmodels==0.14.2
- tsfresh==0.20.3
- hurst==0.0.5
- arch==7.0.0
The dependencies can be installed by:
pip install -r requirements.txtThe datasets can be obtained and put into datasets/ folder in the following way:
- Our model supports anomaly detection for multivariate time series datasets.
- We provide the SKAB dataset. If you want to use your own dataset, please place your datasetfiles in the
/dataset/<dataset>/folder, following the format<dataset>_train.npy,<dataset>_test.npy,<dataset>_test_label.npy.
There are six files/folders in the source
-
data_factory: The preprocessing folder/file. All datasets preprocessing codes are here.
-
main.py: The main python file. You can adjustment all parameters in there.
-
metrics: There is the evaluation metrics code folder.
-
model: PPLAD model folder
-
solver.py: Another python file. The training, validation, and testing processing are all in there
-
requirements.txt: Python packages needed to run this repo
- Install Python 3.9, PyTorch >= 1.4.0
- Download the datasets
- To train and evaluate PPLAD on a dataset, run the following command:
python main.py @ARTICLE{10908726,
author={Chen, Lei and Xu, Yepeng and Li, Ming and Hu, Bowen and Guo, Haomiao and Liu, Zhaohua},
journal={IEEE Transactions on Industrial Informatics},
title={Privacy-Preserving Lightweight Time-Series Anomaly Detection for Resource-Limited Industrial IoT Edge Devices},
year={2025},
volume={21},
number={6},
pages={4435-4446},
keywords={Anomaly detection;Image edge detection;Data models;Computational modeling;Gaussian distribution;Data privacy;Adversarial machine learning;Industrial Internet of Things;Cloud computing;Informatics;Data privacy-preserving;edge devices;industrial Internet of Things;lightweight anomaly detection;resource-limited;similarity discrepancy},
doi={10.1109/TII.2025.3538127}}