Partial Discharge Recognition of Transformers Based on Data Augmentation and CNN-BiLSTM-Attention Mechanism

Partial discharge (PD) is a commonly encountered discharge-related fault in transformers. Due to the unique characteristics of the environment where PD occurs, challenges such as difficulty in data acquisition and scarcity of samples arise. Convolutional neural networks (CNNs) are widely used in pat...

Full description

Saved in:
Bibliographic Details
Published in:Electronics (Basel) 2025-01, Vol.14 (1), p.193
Main Authors: Fu, Zhongjun, Wang, Yuhui, Zhou, Lei, Li, Keyang, Rao, Hang
Format: Article
Language:English
Subjects:
Accuracy
Adaptive sampling
Algorithms
Artificial neural networks
Classification
Data acquisition
Data augmentation
Deep learning
Discharge
Fault diagnosis
Feature extraction
Fourier transforms
Generative adversarial networks
Neural networks
Pattern recognition
Signal processing
Spatiotemporal data
Time series
Wavelet transforms
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Partial discharge (PD) is a commonly encountered discharge-related fault in transformers. Due to the unique characteristics of the environment where PD occurs, challenges such as difficulty in data acquisition and scarcity of samples arise. Convolutional neural networks (CNNs) are widely used in pattern recognition because of their strong feature extraction capabilities. To improve the recognition accuracy of PD models, this paper integrates CNN, bidirectional long short-term memory (BiLSTM), and an attention mechanism. In the proposed model, CNN is employed to extract local spatial and temporal features, BiLSTM is utilized to extract global bidirectional spatial and temporal features, and the attention mechanism assigns adaptive weights to the features. Additionally, to address the issues of sample scarcity and data imbalance, an improved GAN is introduced to augment the data. The experimental results demonstrate that the CNN-BiLSTM-attention method proposed in this paper significantly improves the prediction accuracy. With the help of GAN, the proposed method achieves a recognition accuracy of 97.36%, which is 1.8% higher than that of the CNN+CGAN(Conditional Generative Adversarial Network) method and 5.8% higher than that of thetraditional recognition model, SVM, making it the best-performing method among several comparable methods.
ISSN:2079-9292
2079-9292
DOI:10.3390/electronics14010193