Thermodynamic analysis of variable speed refrigeration system using artificial neural networks

► Thermodynamic performance modeling of variable speed refrigeration system was carried out using artificial neural networks. ► Backpropagation learning algorithm with two different variants used in the network. Logistic sigmoid transfer function was used. ► Eighty percent of data patterns for training and twenty percent for testing procedure were used. ► The best fitting training data set was found to be with eight neurons in the hidden layer. ► Statistical error values were found to be within acceptable uncertainties. Predicted values were very close to actual values. This study presents thermodynamic performance modeling of an experimental refrigeration system driven by variable speed compressor using artificial neural networks (ANNs) with small data sets. Controlling the rotational speed of compressor with a frequency inverter is one of the best methods to vary the capacity of refrigeration system. For this aim, an experimental refrigeration system was designed with a frequency inverter mounted on compressor electric motor. The experiments were made for different compressor electric motor frequencies. Instead of several experiments, the use of ANNs had been proposed to determine the system performance parameters based on various compressor frequencies and cooling loads using results of experimental analysis. The backpropagation learning algorithm with two different variants was used in the network. In order to train the neural network, limited experimental measurements were used as training and test data. The best fitting training data set was obtained with eight neurons in the hidden layer. The results showed that the statistical error values of training were obviously within acceptable uncertainties. Also the predicted values were very close to actual values.