This study provides a comparative analysis of datadriven models for predicting the performance of solar thermal collectors (STCs), using a flat-plate STC installed at the University of Applied Sciences and Arts of Western Switzerland in Sion as a case study. Considering the challenges posed by the intermittent nature of solar energy, the study evaluates static and dynamic data-driven modeling approaches, including polynomial regression, autoregressive with exogenous inputs, autoregressive moving average with extra input, nonlinear autoregressive with exogenous inputs (NARX), and long short-term memory (LSTM) neural networks. Experimental data collected under varied environmental conditions were utilized to develop and test the models. Results show that dynamic modeling methods, especially NARX and LSTM, outperform the static modeling approach, achieving higher accuracy in capturing temporal dependencies and nonlinearities. The findings demonstrate that utilizing two meteorological inputs—solar irradiance and ambient tempera-ture—can provide consistent thermal power predictions, simplifying model complexity while ensuring reliability. The research highlights the effectiveness of data-driven modeling methods to predict the performance of STCs, which is essential for optimizing their operation and ensuring proper integration into renewable energy systems.
