Table 4 Performance comparison of data-driven and statistical methods on various metrics. The best result for each metric within each category is highlighted in bold.
From: Dual graph attention network for robust fault diagnosis in photovoltaic inverters
Data driven methods | Statistical-based methods | ||||||||
|---|---|---|---|---|---|---|---|---|---|
Method | F1-score | Test acc | Precision | Recall | Method | F1-score | Test acc | Precision | Recall |
ANN46 | 0.889 | 91.21% | 0.900 | 0.879 | SVM47 | 0.803 | 85.37% | 0.820 | 0.789 |
CNN48 | 0.903 | 92.43% | 0.914 | 0.885 | KNN49 | 0.784 | 84.12% | 0.792 | 0.780 |
RNN50 | 0.912 | 94.12% | 0.922 | 0.905 | RF51 | 0.830 | 87.11% | 0.846 | 0.816 |
GAT52 | 0.918 | 95.18% | 0.930 | 0.898 | DT53 | 0.825 | 86.53% | 0.831 | 0.809 |
GRU + Attention54 | 0.914 | 93.74% | 0.921 | 0.906 | BC | 0.834 | 87.82% | 0.843 | 0.828 |
TCN55 | 0.918 | 94.28% | 0.926 | 0.910 | |||||
Transformer56 | 0.938 | 95.63% | 0.944 | 0.931 | |||||
ResNet-1D57 | 0.912 | 93.91% | 0.919 | 0.905 | |||||
InceptionTime58 | 0.946 | 96.07% | 0.951 | 0.941 | |||||
LightGBM + SHAP59 | 0.856 | 88.92% | 0.861 | 0.852 | |||||
DualGAT (proposed) | 0.971 | 97.35% | 0.979 | 0.982 | |||||
