AD-TFM-AT-Model

Detection of incipient faults in power distribution network (LSTM/Adaptive Wavelet trasnform/Attention).

Requirement

• Python 3.6
• Tensorflow-gpu 1.14.0
• Keras 2.25

Train and test the model

Run command below to train and test the model:

python test_tf.py  

Experiment

Datasets are obtained from a small Incipient Fault dataset in Power Distribution (IFPD) system from [1] (https://dx.doi.org/10.21227/bwjy-7e05), and a relatively large dataset logged by State Grid Corporation of China in AnHui Province (SGAH) from [2] (https://github.com/smartlab-hfut/SGAH-datasets.git).

   device: Tesla V100   
   dataset: IFPD and SGAH   
   optimizer: Adam(lr=0.001, eps=1e-08)  
   batch：800 

These are the result for the incipient fault detection in two datasets.

Metrics	Accuracy	Precision	Recall	F1score
IFPD	0.97	0.97	0.96	0.96
SGAH	0.99	0.97	0.98	0.98

Fig.1 ROC of AD-TFM-AT model on IFPD.

Fig.2 ROC of AD-TFM-AT model on SGAH.

Reference

If you use the codes or the datasets, please cite the following papers:

@unknown{unknown,  
author = {Li, Qiyue and Deng, Yuxing and Liu, Xin and Sun, Wei and Li, Weitao and Li, Jie and Liu, Zhi},  
year = {2022},  
month = {05},  
pages = {},  
title = {Autonomous Smart Grid Fault Detection},  
doi = {10.48550/arXiv.2206.14150}  
}  

@article{li2022resource,  
title={Resource Orchestration of Cloud-edge based Smart Grid Fault Detection},  
author={Li, Jie and Deng, Yuxing and Sun, Wei and Li, Weitao and Li, Ruidong and Li, Qiyue and Liu, Zhi},  
journal={ACM Transactions on Sensor Networks (TOSN)},  
year={2022},  
publisher={ACM New York, NY}  
}

Copyright

See LICENSE for details.