This study develops reliable and robust machine learning (ML) models to predict the outlet air temperature and humidity and thermal efficiency of a solar air heater (SAH). Also, the application of predictive models for optimal control of the SAH operation is proposed. For this, the work contains three main parts: (a) a vertically-mounted symmetrical SAH was installed outside of a building room and operated throughout the winter of 2022. (b) By conducting experiments for five air mass flow rates, a large dataset with more than 62,500 sample points was collected. (c) Six input features containing time, environmental-related attributes, and SAH variables were applied to develop several state-of-the-art ML algorithms. To figure out the most accurate models for predicting output variables, the dataset was partitioned into three parts. Also, various modeling performance evaluation criteria were calculated and compared on the validation and test sets. Among these models, the gradient boosting machine algorithm based on LightGBM implementation achieved the best degree of generalization in modeling the target variables. The results demonstrated that the developed models obtained the lowest R-squared and the highest mean absolute percentage error of 0.9827 and 2.95%, respectively, on the test set. Moreover, the offline analysis of SAH operation based on the proposed control scheme demonstrated that 350 kWh of thermal energy can be generated during the application in the one-year winter season, 24% more than SAH operation without a model-based control strategy.
