Machine learning prediction of photovoltaic-thermal system with V-shaped perforated fins

Ahmad Alqatamin, Jinzhan Su,
Machine learning prediction of photovoltaic-thermal system with V-shaped perforated fins,
Applied Thermal Engineering,
Volume 280, Part 3,
2025,
128344,
ISSN 1359-4311,
https://doi.org/10.1016/j.applthermaleng.2025.128344.
(https://www.sciencedirect.com/science/article/pii/S1359431125029369)
Abstract: Accurate prediction of hybrid photovoltaic and thermal systems remains challenging due to the nonlinear behavior of photovoltaic cells, complex thermal and fluid interactions, and limited datasets in previous machine learning studies. This work develops a hybrid computational fluid dynamics and machine learning framework to improve the design and long-term planning of such systems. A water-cooled photovoltaic and thermal system with perforated V-shaped fins was simulated using high-fidelity computational fluid dynamics to generate more than 1000 operating cases. These data trained and compared four machine-learning models—support vector regression, random forest, multilayer perceptron, and self-organizing feature map. The best model was coupled with local weather records to forecast hourly performance in Xi’an, China, for 2025 to 2027. Support vector regression achieved a coefficient of determination of 0.998 for electrical power prediction. The finned, water-cooled design lowered solar cell temperature by up to 31.6 K relative to a module without cooling and delivered annual electrical-efficiency gains of 7.09 %, 7.82 %, and 7.51 % for 2025, 2026, and 2027, respectively. Peak thermal efficiency reached 33.0 % in summer. These results indicate that combining detailed simulation with data-driven prediction provides accurate, location-specific forecasts and a practical way to optimize photovoltaic and thermal modules with enhanced cooling fins. This study extends prior work by uniting a perforated V-shaped fin design with long-term forecasting and by quantifying benefits with validated, high-accuracy predictions.
Keywords: PV; PVT-PVSFs (Photovoltaic-thermal with Perforated V-shape fins); Machine learning (ML); Thermal management; Computational fluid dynamics (CFD)