Explainable Multivariate Time Series Anomaly Detection by Feature Graph Structure Learning

The programming language used is Python 3.10 with Pytorch 2.0.1, CUDA 11.7, and PyTorch Geometric Library 1.5.0.

How to run the code?

After downloading the code, you can run

python3 main_run.py

directly for categorical clustering. We suggest adjusting the hyperparameters multiple times to achieve better results.

What are the scripts used for?

(1)make_dataset: Data processing. Help us prepare the training set.

(2)my_models: Define the network structure of FGEAD.

(3) utils: Contains functions for data processing and model evaluation.

Several toolkits may be needed to run the code

(1) pytorch (https://anaconda.org/pytorch/pytorch)

(2) sklearn (https://anaconda.org/anaconda/scikit-learn)

(3) transformers (https://anaconda.org/conda-forge/transformers)

datasets

24070103_train.csv: Contains only normal multivariate time series data during tobacco drying process, used to train the model. list. txt: Contains all variable names (in Chinese) during the tobacco drying process. Merged_csv: Contains multivariate time series data of anomalies during tobacco drying process, used to test model performance.