بررسي و ارزيابي عملكرد كلكتور فتوولتائيك-حرارتي با استفاده از روش هوش مصنوعي

Investing an‎d eva‎luating the performance of photovoltaic-thermal collector using artificial intelligence

مهندسي مكانيك

Abstract In this research, a photovoltaic-thermal collector equipped with a spiral tube, using water as the working fluid, has been numerically simulated in COMSOL software. The geometry of the panel plates an‎d the spiral tube has been modeled in accordance with the validation reference paper, an‎d the physics of heat transfer an‎d turbulent fluid flow have been applied. The input parameters are: solar radiation intensity, ambient temperature, wind speed, fluid inlet temperature, an‎d fluid mass flow rate. The output parameters are: thermal efficiency, average panel surface temperature, an‎d fluid outlet temperature. By applying initial conditions, boundary conditions, an‎d input values, the corresponding outputs are extracted by the software. To eva‎luate the uniformity of the panel surface temperature distribution, the relevant index has been calculated an‎d the effect of each input parameter on it has been analyzed. In order to train a prediction model based on machine learning, the input values have been varied within specified ranges, an‎d using a parametric study in the software, a dataset containing the corresponding inputs an‎d outputs has been generated. The effect of each input on thermal efficiency an‎d average panel surface temperature has also been investigated. After normalization an‎d cross-validation, the obtained data have been prepared as vectors consisting of five input features an‎d one output feature (thermal efficiency). Specifically, the share of training an‎d testing data from these 123 data samples is 80% (98 data samples) an‎d 20% (25 data samples), respectively. Then, five common machine learning models have been trained, an‎d their performance has been eva‎luated using indicators such as the coefficient of determination, mean absolute error, an‎d overfitting criterion. The results indicate that the Ridge regression model has the best performance among the models an‎d is capable of predicting the system behavior with high accuracy an‎d minimal error. Specifically, the prediction error an‎d coefficient of determination for the test data in this model are 0.0122 an‎d 0.8728, respectively. The feature importance plot also indicates that solar radiation intensity, ambient temperature, an‎d wind speed have the greatest impact on thermal efficiency, with contributions of 45%, 28%, an‎d 24%, respectively. Additionally, the fluid inlet temperature an‎d mass flow rate parameters together account for 5% of the impact on thermal efficiency. Overall, the developed model can predict the thermal efficiency of the collector with appropriate accuracy by receiving the input parameters, thus reducing the time an‎d cost of numerical computations in subsequent analyses. Keywords: Photovoltaic–thermal collector, Serpentine tube, Thermal efficiency, Average panel surface temperature, Temperature uniformity of panel surface, Machine learning model

5

155050