Author Affiliations
School of Earth Science and Engineering, Hohai University, Nanjing 211100, Jiangsu, Chinashow less
Fig. 1. Model of ResPNet
Fig. 2. Model of Residual block
Fig. 3. Model of Respblock
Fig. 4. Model of Gblock
Fig. 5. Ensemble neural network framework for prediction of photovoltaic modules
Fig. 6. Training loss of each model
Fig. 7. Validation accuracy of each model
Fig. 8. Feature layer visualization information of underlying feature information enhancement module. (a) Feature maps information of ResNet-50; (b) feature maps information of ResPNet[D]
Fig. 9. Feature layer visualization information of Respblock module. (a) Feature maps information of ResPNet[D]; (b) feature maps information of ResPNet[P]
Fig. 10. Feature layer visualization information of Gblock module. (a) Feature maps information of ResPNet[P]; (b) feature maps information of ResPNet[G]
Fig. 11. Test results of different photovoltaic module fault detection models. (a) ResNet; (b) ResPNet
Fig. 12. Test results of cascaded photovoltaic module fault detection
Fault name | Fault description | Number of images | Number of images in train and validation dataset | Number of images in test dataset |
---|
Cell | There is a single square hot spot in the photovoltaic module | 1877 | 1689 | 188 | Cell-multi | There are multiple square hot spots in the photovoltaic module | 1288 | 1160 | 128 | Cracking | Cracks exist on the surface of the photovoltaic module | 940 | 845 | 95 | Hot-spot | A single hot spot exists on the surface of the photovoltaic module | 249 | 224 | 25 | Hot-spot-multi | There are multiple hot spots on the surface of the photovoltaic module | 246 | 222 | 24 | Shadowing | Shadows created by plants,artifacts between adjacent rows | 1056 | 950 | 106 | Diode | Bypass diode short circuit causes that 1/3 photovoltaic module cannot operate normally | 1499 | 1349 | 150 | Diode-multi | The short circuit of the bypass diode causes that 2/3 of the photovoltaic module cannot operate normally | 175 | 158 | 17 | Vegetation | The plants shade the photovoltaic module | 1639 | 1476 | 163 | Soiling | Dust and garbage may block the surface of the photovoltaic module | 204 | 183 | 21 | Offline-module | The entire photovoltaic module fails to operate properly | 827 | 744 | 83 | No-anomaly | Trouble-free | 10000 | 9000 | 1000 |
|
Table 1. Description of photovoltaic module and dataset partitioning
Fault name | Image data |
---|
Cell | | Cell-multi | | Cracking | | Hot-spot | | Hot-spot-multi | | Shadowing | | Diode | | Diode-multi | | Vegetation | | Soiling | | Offline-module | | No-anomaly | |
|
Table 2. Type of photovoltaic module images
Dataset | Model | /% |
---|
Infrared Solar Modules | ResNet-50 | 83.2 | ResPNet[D] | 83.8 | ResPNet[P] | 84.4 | ResPNet[G] | 84.6 |
|
Table 3. Ablation experiment results of infrared image classification accuracy of photovoltaic modules
Model | /% | /% | /% | /% | Speed /ms |
---|
ShuffleNetV2[19] | 76.8 | 59.8 | 54.1 | 56.8 | 4.9 | MobileNetV2[20] | 78.3 | 63.0 | 57.2 | 59.9 | 4.0 | ResNet-50[14] | 83.2 | 74.2 | 67.9 | 70.9 | 5.1 | ResNet-101[14] | 83.5 | 74.3 | 67.0 | 70.4 | 12.3 | EfficientNetV7[21] | 81.4 | 68.7 | 63.6 | 66.1 | 26.9 | Alves et al.模型[11] | 78.8 | — | — | — | — | CNN-SVM[12] | 82.9 | — | — | — | — | Le et al.模型[13] | 84.1 | — | — | — | — | ResPNet[G] | 84.6 | 75.8 | 67.6 | 71.5 | 30.8 |
|
Table 4. Model accuracy of infrared image classification in photovoltaic modules
Model | No. of model | Ac /% |
---|
Proposed by Le et al. [13] | 15 | 85.9 | ResPNet[G] | 3 | 85.9 |
|
Table 5. Integrated models for infrared image classification accuracy of photovoltaic module