Performance augmentation and machine learning-based modeling of wavy corrugated solar air collector embedded with thermal energy storage: Support vector machine combined with Monte Carlo simulation

At present, artificial intelligence methods have been effectively utilized for predicting the complex performance of storage-based solar thermal technologies for cooling/heating applications. It is crucial to have accurate energy storage-based sustainable system estimation, which would contribute to increased operational time, thus maximizing the overall efficiency of solar energy-based storage systems. Hence, this work introduces a comparative experimental investigation and support vector machine (SVM) modeling on a wavy corrugated solar air collector (WCSAC) with and without using a paraffin wax storage container. Experiments on the WCSAC were performed under three air flow rates of 0.540, 1.68, and 3.72 kg/min and two paraffin layer thicknesses of 2 cm and 4 cm, respectively. Moreover, improved SVM models implemented in MATLAB software are developed for predicting the thermal performance parameters; including air temperature ratio, average air temperature, convective heat transfer coefficient, and energy efficiency for the WCSAC. The optimal solution of the SVM modeling is developed by incorporating the Karush-Kuhn-Tucker conditions and several kernel functions. In addition, a sensitivity analysis is also conducted to explore the significance of model input parameters (air inlet temperature, time, solar irradiance, air flowrate, PCM layer thickness) on the output parameters prediction using the Monte Carlo simulation technique. The experimental results presented that the daily energy efficiencies of the WCSAC equipped with 4 cm paraffin layer thickness are 24.0 %, 20.39 %, and 16.37 % higher than that of the WCSAC without PCM at airflow rates of 3.72, 1.68, and 0.54 kg/min, respectively. Moreover, the connective heat transfer coefficient of the WCSAC with PCM is more than 1.20 times that yielded with the WCSAC without PCM. Additionally, the SVM simulations showed that the optimal solution of the SVM model is developed by incorporating the Karush-Kuhn-Tucker conditions and Lagrangian function kernel function, which revealed a superior accuracy with the highest coefficient of determination of 0.990 and 0.950 for training and test processes, respectively.