Machine learning based prediction of thermal performance for an evacuated tube solar air heater with pierced twisted tape

Evacuated tube collector solar air heater (ETCSAH) is suited to low temperature solar heating, but its thermal and thermo-hydraulic performance depends on nonlinear interactions among operating, environmental and geometrical parameters. Although machine learning (ML) has been applied to predict outlet air temperature and thermal efficiency in solar thermal systems, little attention has focused on simultaneously predicting outlet air temperature, thermal efficiency and thermo-hydraulic efficiency of an ETCSAH fitted with a pierced twisted tape (PTT) insert from experimentally measured operating and geometrical parameters. Hence, this study experimentally explored the effect of PTT on ETCSAH thermal performance and developed comparative ML based models to predict it. The input parameters were: Solar irradiance; Mass flow rate; Ambient temperature; Wind speed; Pitch length; Hole diameter; Pressure drop and Temperature rise. Algorithms developed and tested include Support Vector Machine Regression (SVM), Stochastic Gradient Descent Regression (SGDR), and Extra Trees Regression (ET), used to predict outlet air temperature, thermal efficiency and thermo-hydraulic efficiency. Experimental results showed that outlet air temperature, temperature rise and thermal efficiency decreased as mass flow rate increased, whereas heat recovery rate increased. SVM gave the best prediction ability. For the testing dataset, SVM achieved R 2 = 0.98256, for outlet air temperature; R 2 = 0.98791 for thermal efficiency; and R 2 = 0.98502 for thermo-hydraulic efficiency. The results illustrate that ML models can be used to quickly estimate the ETCSAH performance in the experimentally tested operating region and can be used for performance evaluation and optimization, thus lowering the number of repeated experimental evaluations.