.png){kind=icon}
返回
A partial discharge pattern recognition method based on multi-scale adaptive denoising network and Stacking Ensemble Learning
.png){kind=icon}
PRE
AI
DOI：10.1016/j.epsr.2024.111392.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Partial discharge (PD) serves as a critical indicator of insulation deterioration in electrical equipment, and accurate recognition of PD faults is essential. Existing PD pattern recognition methods lack automatic denoising capabilities and the ability to learn multi-scale fault features. Additionally, these methods primarily rely on single classification models, which limits their ability to learn from a diverse feature space. To address these issues, a novel Multi-scale Attention Adaptive Denoising Network (MAADNet) and a Stacking Ensemble Learning Framework with diverse model integration (DMIStacking) are proposed in this paper. MAADNet integrates a Multi-scale Feature Learning Module (MFLM), and an Adaptive Denoising Module (ADM). The MFLM employs multiple dilated convolutions with different dilation rates to extract multi-scale features. The ADM integrates the CBAM module and the soft thresholding function, working together to achieve adaptive denoising based on the characteristics of the input PD signals, thereby avoiding errors caused by manually setting denoising parameters. Furthermore, to learn diverse feature representations and further improve the recognition accuracy of PD fault diagnosis, the DMIStacking model is developed, which employs deep learning models MAADNet and Transformer as base-learners, together with the machine learning model XGBoost, and utilizes SVM as meta-learner. Experimental results on both our on-site PD dataset and a public PD dataset validate the proposed MAADNet and DMIStacking outperforms other state-of-the-art methods, exhibiting outstanding pattern recognition performance.
Keyword:
Deep learning
Fault diagnosis
Machine learning
Multi-scale
Partial discharges
Pattern recognition
Stacking Ensemble Learning
期刊
IF:
4.2
论文数:
1.2W
被引数:
2.2W
机构
引用论文
.png){kind=icon}
分享收藏Random Forest Based Optimal Feature Selection for Partial Discharge Pattern Recognition in HV Cables
OAAI
OAAI分享收藏
查看更多内容