Search

This study presents a comprehensive approach to developing models for decomposing global solar radiation in to its diffuse and direct components, a critical step in solar energy applications. Utilizing data from multiple geographical regions and Köppen-Geiger climate zones, the research ensures a broad understanding of the environmental factors affecting solar radiation. The study meticulously selects predictors through multiple feature selection techniques, focusing on capturing essential information while minimizing redundancy. Single-predictor models are developed to provide insights into the relationships between selected predictors and the diffuse fraction, while two multi-predictor models demonstrates the potential for more accurate estimations by leveraging the collective predictive power of multiple variables. One of the developed models, called Nested, performs better than other state-of-the-art universal models, while performing slightly worse than the best climate-specific model in terms of three performance metrics. The research lays a solid foundation for future studies and practical applications in the solar energy sector, emphasizing the need for further testing of techniques and procedures for decomposition model development.

This work aims to give insight into the challenges of performing aerial daylight drone Electroluminescence (EL) inspections in PV farms. In this context, we have compared an EL inspection process using a tripod to record the images to an aerial drone EL inspection system. We focus on the challenges of the “Stop and Go” procedure for recording EL images with the drone, which consists of stopping the drone module by module during the inspection flight. The drone inspection tests resulted in an average module inspection time of 24 seconds per module and around 8 minutes per PV string. Regarding image processing challenges, drone inspection requires an added module tracking algorithm due to the drone motion which introduces some artifacts in the image but does not lower the image quality and diagnostic information of the images compared to stationary daylight EL. Images acquired with both the drone and tripod fail in identifying microcracks in the images but do well in detecting larger B and C cracks. For obtaining the best possible image quality, it is essential to have a calibrated camera, avoid shadows from the drone when recording, and avoid taking images with fast-changing irradiance.

To perform outdoor daylight electroluminescence (EL) and photoluminescence (PL) of PV modules for defect and fault detection, it is necessary to filter away any other light source apart from the solar cell's luminescence emission. Image processing is required in broad daylight to increase the signal-to-noise ratio digitally. However, it is reported in the literature that an optical (often bandpass) filter in front of the camera lens or sensor is also needed to overcome the 4-5 orders of magnitude more intense sunlight. The specs of such a "daylight filter" are often mentioned but are rarely the same between research groups and inspection companies. To our knowledge, a study has not addressed the main factors necessary to define an optimal filter for daylight luminescence imaging of PV modules. In this work, we evaluated the spectroscopic characteristics of four filter configurations. We performed daylight EL imaging to estimate their performance in signal-to-noise ratio and EL image quality.

Electroluminescence imaging is a powerful technique for PV fault diagnosis and for this reason there is a growing interest of the O&M industry to make it widely available for in site inspections instead of randomly choosing samples throughout the PV plant. Few groups in the world disclose methodologies for daylight EL (dEL) image acquisition of installed PV modules, using different equipment and image processing procedures. In this work, we aim to compare methodologies from two research groups that have been working on practical ways to perform robust dEL imaging in parallel. We present the results from acquisitions made with similar but not identical equipment, weather conditions and PV module technologies. Image processing on the same datasets from both laboratories is performed using their own methods, indicating if the image processing methods and approach provide consistent results or not. While each laboratory has developed its own signal-to-noise ratio markers, both present good consistency with signal at an exposure time analysis, but not always a direct correspondence to the image quality. This intercomparison will lead to improvements on each laboratory procedure, expand the discussion to other groups and initiate a discussion on the establishment of dEL image quality and SNR metric baselines.

With the growth of photovoltaic plant installations in quantity and size, the operations & maintenance industry calls for more efficient and accurate ways to inspect the PV modules health. To expand inspection viability, Daylight Electroluminescence (dEL) plays a big role, as EL allows accurate diagnosis of PV panels deep into the cell level. Obtaining EL images in the field during direct sunlight without covers is challenging but effective and robust using a modulated electrical signal and image processing. The thresholds required by the modulation techniques and a detailed description of how the sunlight influences the capture of the EL images are not clearly described in the research or technical literature. In this case study, our goal is to report how sun irradiance in the plane of the array influences dEL image acquisition of PV panels under direct sunlight and to better specify the irradiance limit for when modulation of the forward bias is necessary instead of constant bias. All this considering reaching the minimum acceptable EL image quality for PV diagnosis using previously available and new signal-to-noise ratio (SNR) markers.

Product integrated photovoltaic (PIPV) systems, can be severely impacted by shading in the surroundings as their collector area is typically small. To ensure the proper functioning of these devices, methodologies to estimate solar irradiance and near shading accurately and efficiently in built up locations are needed. Digital Surface Models (DSM) have proven to be a good alternative to estimate solar irradiance on building surfaces and urban areas. However, high-accuracy DSM are scarce, with limited availability and tend to show the maximum elevation of the obstructions elements disregarding their real shape. We propose a methodology with onsite LiDAR measurements to improve the accuracy to which the solar irradiance can be estimated in urban environments. This methodology produces horizon matrices which can reduce the estimated irradiance’s RMSE by 25-40% compared to fisheye sky imaging methodologies, and by 45-75% to the ones obtained with low-resolution DSMs. However, at locations with low shading profiles the differences are negligible.

In this work, we present and validate a machine learning approach for estimating power loss in monocrystalline silicon PV cells due to cell cracks. The primary focus is selection and evaluation of robust and accurate prediction-features derived from electroluminescence (EL) images of solar cells. To achieve this, we have experimentally created training and validation EL and I-V measurements datasets of cells with two to up to five busbars, with different extents of cell crack degradation. We have evaluated simple features derived from the cells' luminescence histogram, which are easy to compute; these are the cell inactive area, the mean, standard deviation (STD), skewness, and kurtosis. Learning models are then trained against cell I-V measurements and validated. This cell-level approach has the advantage of being scalable so that it could be used to estimate the power loss of a full PV module based on a module EL image that is segmented in cell images. Similarly, this bottom-up approach could facilitate the development of more generalizable machine algorithms, where just one model can be used to simulate power no matter the size of the module, number, layout, or type of the cracked cells. As a baseline, a linear regression approach was attempted, which yielded a R2 score of 0.85 and a maximum error of 0.26. These results were subsequently improved by applying a 2nd degree polynomial linear regression and a custom non-linear regression to 0.869 of R2 and 0.17 of maximum error. The model selection has been chosen from a statistical exploratory analysis of the dataset, where it has been demonstrated that mono-crystalline cells of with different number of busbars behave differently in terms of histogram features correlations with the output power, and that adding a classifier variable as the busbar number as a predictor it is possible to account for the non-linearity of each cell type and model it accordingly.