Performances of Solar Vapor Compression Refrigeration Systems: Comparison of Simulations between an Auto-Regressive with eXogeneous Variables (ARX) and an Artificial Neural Network (ANN)

Abstract

In this study, a solar vapor compression refrigeration system was modeled by using the two machine learning approaches, Auto-Regressive with eXogeneous Variables (ARX) and an Artificial Neural Network (ANN). The performances of these two approaches were compared. The system being modeled was composed of a vapor compression refrigerator unit, two 300 W solar modules, two 12 V batteries with a capacity of 200 Ah (each) and a charge controller. Ten experiments were carried out using this system. Bottles of water were used as cooling loads. Data collected from the experiments that used loads of 10, 30, 50, 70 and 100 liters of water were used to build the models, and the experiments with the loads of 20, 40, 60, 80 and 90 liters of water were used to test the performance of the models. The differences between the experimental load temperatures and the predicted load temperatures for these models, were used as indicators of the performance of the models, which were stated as the percentage of the root mean square difference relative to the mean measured values (RMSD) and the percentage of mean bias difference relative to a mean measured values (MBD). The RMSD of the ARX model was 4.3% and the MBD was 0.6%. The RMSD of the ANN model was 66.1% and the MBD was 20.1%. These results show that the ARX learning approach performed better than the ANN learning approach for this system.