Solar energy, in the form of photovoltaic (PV) panels, is important for achieving clean energy solutions. The photovoltaic health index must be monitored and improved because of the high demand for green energy. Unfortunately, defective solar cells are a significant source of performance degradation in photovoltaic (PV) systems. Experts often manually analyze electroluminescence (EL) images by visually inspecting them, which is personal, time-consuming, and requires extensive expertise. This work presents a comparative analysis of YOLOv8 and an Improved YOLOv5 for an automatic PV defect detection system in EL images in which Global Attention Module (GAM) is incorporated into the traditional YOLOv5s model for better object representation. Adaptive Feature space fusion (ASFF) was added to YOLOv5’s original structure for feature fusion. The Distance Intersection over Union (Non-Maximum) Suppression (DIoU-NMS) is aggregated to produce a more accurate bounding box. The ELDDS1400C5 dataset was used to train and evaluate the proposed system. Experiments on the ELDDS1400C5 test set revealed that the Improved YOLOv5 algorithm achieved a mean Average Precision of 76.3% (mAP@0.5), which is a 2.5% improvement over the standard YOLOv5 algorithm for detecting faults in PV modules in EL images. Furthermore, the experimental results demonstrated that Test Time Augmentation (TTA) significantly increased the mAP@0.5 to 77.7%, surpassing the YOLOv8 model, which achieved 77.5% under the same conditions.

Solar power is a clean and sustainable energy source that does not emit greenhouse gases or other atmospheric pollutants. The inherent variability in solar energy due to random fluctuations introduces novel attributes to the power generation and load dynamics of the grid. Consequently, there has been growing attention to developing an accurate forecast model using various machine and deep learning techniques. Temporal attention mechanisms enable the model to concentrate on the critical components of the input sequence at each time step, thereby enhancing the accuracy of the prediction. The suggested GRU–temporal fusion transformer (GRU-TFT) model was trained and validated employing the “Daily Power Production of Solar Panels” Kaggle dataset. Furthermore, an innovative loss function termed DILATE is introduced to train the proposed model specifically for multistep and nonstationary time series forecasting. The outcomes have been subjected to a comparative analysis with alternative algorithms, such as neural basis expansion analysis for interpretable time series (N-BEATS), neural hierarchical interpolation for time series (N-HiTS), and extreme gradient boosting (XGBoost), using several evaluation metrics, including the absolute percentage error (MAE), mean square error (MSE), and root mean square error (RMSE). The model presented in this study exhibited significant performance improvements compared with traditional statistical and machine learning techniques. This is evident from the achieved values of MAE, MSE, and RMSE, which were 1.19, 2.08, and 1.44, respectively. In contrast, the machine learning approach utilizing the Holt–Winters method for time series forecasting in additive mode yielded MAE, MSE, and RMSE scores of 4.126, 29.105, and 5.3949, respectively.

This article focuses on the optimization of tilt angle for solar panels in the Gulf Cooperation Council (GCC) countries. The tilt angle is a crucial factor that affects the amount of solar radiation received by the solar panel. The study uses a mathematical model to calculate the optimal tilt angle based on the latitude and longitude of the location and compares the results with the PVWatts calculator. The importance of selecting the optimal tilt angle for maximizing solar energy production is emphasized. The study finds that the mathematical model and PVWatts calculator are in good agreement, except for the negative tilt angles calculated by the model. The article also provides monthly, seasonally, and yearly irradiance values for the GCC countries calculated using the PVWatts calculator. It suggests that changing the tilt angle 12 times during the year can enhance the power output by 10.9%. The results show that the monthly automated angle has a maximum in December and a minimum in June, with Oman having the highest maximum angle of 50.5 and the lowest minimum angle of -10.58. Kuwait has the highest maximum angle of 58.33 and the lowest minimum angle of -2.75. The results are verified by the PVWatts calculator, showing a good similarity with a percentage of error around 3%. The study highlights the importance of selecting the optimal tilt angle to achieve maximum solar energy production in the GCC countries. The mathematical model and PVWatts calculator can serve as tools to calculate the optimum tilt angle for solar panels.

Proper surveillance and maintenance of photovoltaic (PV) systems are crucial to ensure continuous power generation and prevent operational downtimes. However, manual analysis of electroluminescence (EL) images is subjective, time-intensive, and requires significant expertise. To address this issue, a comprehensive deep learning architecture has been developed for the semantic segmentation of 29 different features and defects within EL images of PV panels. The SegNet architecture encoder has been replaced with the VGG16 encoder, which incorporates pre-trained weights to leverage transfer learning during the feature extraction stage. A Convolutional Block Attention Module (CBAM) block has also been introduced to enhance the decoder’s ability to generate fine-grained segmentations. Additionally, the suggested architecture has been evaluated through the application of three different loss functions: weighted categorical cross-entropy loss, categorical cross-entropy, and focal loss. The Attention-Based SegNet architecture proposed with a weighted categorical cross-entropy loss exhibits superior performance in terms of accuracy, F1 score, intersection over union (IoU), precision, recall, mean IoU (mIoU), specificity, Jaccard index, and Dice coefficient. It achieves a Dice coefficient of 0.9408 and an mIoU of 0.9101, outperforming the state-of-the-art SEiPV-Net trained on the same dataset by 8.77% and 4.97%, respectively.