Machine Learning Approaches for Predicting and Optimising Thermal Performance of Solar Collectors

Abstract

The application of machine learning techniques in the renewable energy sector has revolutionised the predictive modelling and optimisation of solar thermal systems. Solar air heaters with artificial roughness and phase change materials are widely used to enhance thermal performance, necessitating accurate thermo-hydraulic performance predictions for efficient design and operation. This study employs the XG-Boost machine learning model to predict thermo-hydraulic performance values based on experimental data obtained for polygonal and trapezoidal rib configurations at different relative roughness pitch ratios (p/e) and Reynolds numbers (Re = 2000 to 20000) in the absorber plate of a solar air heater. The experimental results reveal that a polygonal rib configuration with p/e = 7.5 exhibits the highest thermo-hydraulic performance value of 2.95 at Re = 4000, while the trapezoidal rib with p/e = 7.5 achieves a thermo-hydraulic performance of 2.93 at the same Reynolds number. The results were trained in the proposed machine learning model to validate the accuracy of predicted results. The coefficient of determination R2, mean absolute percentage error, root mean squared error and mean absolute percentage error matrix were considered for the training and testing dataset. The model effectively captures the nonlinear thermal behaviour, achieving R2 = 0.976, mean absolute error at 0.045, root mean squared error at 0.056 and mean absolute percentage error of 2.42%, demonstrating superior predictive capability. Further, this study provides the significance of plethora machine learning algorithms, data pre-processing strategies, feature selection and hybrid machine learning models for performance optimisation in solar air heaters. Additionally, few case studies highlight the practical implementation of machine learning in solar energy systems, demonstrating its potential for accurate prediction of thermal efficiency in real time assessment. Besides, it outlines future research directions, challenges, and emerging opportunities to advance machine learning applications in sustainable energy solutions. This „review-experimental” comparative article serves as a valuable resource for researchers and industry professionals, driving innovation in solar thermal energy systems through data-driven intelligence.