article · Scientific Reports
Efficient cooling is vital for maintaining photovoltaic panel performance, yet standard temperature probes struggle to capture spatial temperature differences across panel surfaces. An alternative approach applies deep learning models to thermal imaging videos to evaluate panel cooling dynamics non-invasively. A U-Net network separates solar panels from surrounding background elements in thermal video footage. Two regression models, a three-layer feedforward neural network and a convolutional neural network, were developed to estimate cooling percentages from individual images. The convolutional neural network outperformed the feedforward network, reaching a mean absolute error of 1.2 per cent and an R-squared value of 0.95. The work also assesses practical implementation needs, including hardware requirements, integration into current infrastructure, and economic viability, highlighting opportunities for cost savings and improved energy output in large photovoltaic installations.
Overheating reduces the electrical output of solar panels, but measuring cooling performance across an entire panel using physical probes is difficult. Using computer vision and thermal video offers a non-invasive way to measure cooling efficiency accurately. This allows solar plant operators to detect thermal problems, improve maintenance schedules, and maximise electricity generation from renewable energy assets.
Targeted at operators and maintenance providers of large-scale photovoltaic installations, this approach provides non-invasive thermal monitoring using existing infrastructure and standard thermal cameras. Because the work evaluates hardware requirements, economic viability, and integration alongside model testing, it sits at an applied and tested stage, though wider deployment depends on expanding to larger datasets and formal industry trials.
AI-generated from the published abstract. Always read the original work before citing.
Efficient cooling systems are critical for maximizing the electrical efficiency of Photovoltaic (PV) solar panels. However, conventional temperature probes often fail to capture the spatial variability in thermal patterns across panels, impeding accurate assessment of cooling system performance. Existing methods for quantifying cooling efficiency lack precision, hindering the optimization of PV system maintenance and renewable energy output. This research introduces a novel approach utilizing deep learning techniques to address these limitations. A U-Net architecture is employed to segment solar panels from background elements in thermal imaging videos, facilitating a comprehensive analysis of cooling system efficiency. Two predictive models-a 3-layer Feedforward Neural Network (FNN) and a proposed Convolutional Neural Network (CNN)-are developed and compared for estimating cooling percentages from individual images. The study aims to enhance the precision and reliability of heat mapping capabilities for non-invasive, vision-based monitoring of photovoltaic cooling dynamics. By leveraging deep regression techniques, the proposed CNN model demonstrates superior predictive capability compared to traditional methods, enabling accurate estimation of cooling efficiencies across diverse scenarios. Experimental evaluation illustrates the supremacy of the CNN model in predictive capability, yielding a mean square error (MSE) of just 0.001171821, as opposed to the FNN's MSE of 0.016. Furthermore, the CNN demonstrates remarkable improvements in mean absolute error (MAE) and R-square, registering values of 1.2% and 0.95, respectively, whereas the FNN posts comparatively inferior numbers of 3.5% and 0.85. This research introduces labeled thermal imaging datasets and tailored deep learning architectures, accelerating advancements in renewable energy technology solutions. Moreover, the study provides insights into the practical implementation and cost-effectiveness of the proposed cooling efficiency monitoring system, highlighting hardware requirements, integration with existing infrastructure, and sensitivity analysis. The economic viability and scalability of the system are assessed through comprehensive cost-benefit analysis and scalability assessment, demonstrating significant potential for cost savings and revenue increases in large-scale PV installations. Furthermore, strategies for addressing limitations, enhancing predictive accuracy, and scaling to larger datasets are discussed, laying the groundwork for future research and industry collaboration in the field of photovoltaic thermal management optimization.
This page summarises published work. The authoritative version sits with the publisher.
DOI: 10.1038/s41598-024-81101-x
Is something wrong with this record? Report it or request removal.
Discussion
Have you built on this work, tried to replicate it, or seen it applied in practice? Share what you know. Verified researchers and MARATTO™ domain experts can open a discussion, and any member can reply. Contributions are reviewed before they appear.
No discussion yet. Open the first thread.
New to MARATTO™? Create a free account.
