Fig 1.
The framework for partial discharge recognition in power transformers by synergistically combining variational mode decomposition (VMD) optimization with deep spatiotemporal feature learning.
Fig 2.
Data acquisition and preprocessing process.
Fig 3.
VMD-based signal feature extraction optimization.
Fig 4.
The proposed hybrid deep learning architecture.
Table 1.
STFT feature performance metrics.
Fig 5.
STFT feature performance metrics.
Table 2.
Wavelet scattering transform results.
Fig 6.
Wavelet scattering transform results.
Table 3.
VMD-CNN performance across all test conditions.
Fig 7.
VMD-CNN performance.
Table 4.
Model robustness under adverse conditions.
Table 5.
Standard LSTM performance metrics.
Fig 8.
Standard LSTM performance.
Table 6.
Transformer model performance across all conditions.
Fig 9.
Transformer model performance.
Table 7.
Attention-LSTM detailed performance.
Fig 10.
Attention-LSTM performance.
Fig 11.
TDOA triangulation performance.
Table 8.
TDOA triangulation localization performance.
Fig 12.
U-net CNN performance.
Table 9.
U-Net CNN localization performance across all conditions.
Fig 13.
Hybrid model performance.
Table 10.
Hybrid model detailed localization performance.
Table 11.
Extended comparison with state-of-the-art methods.