A Comparative Study of Multiple Regression and Artificial Neural Network models for a domestic refrigeration system with a hydrocarbon refrigerant mixtures

Abstract

Abstract Present paper discus with the development models such as Multiple Regression Analysis (MRA) and Artificial Neural Networks (ANN) for a domestic refrigeration system to predict the performance of a system using Hydrocarbon Refrigerant Mixtures (HCRM). The experiments conducted with the changing evaporator temperature (Te), different hydrocarbon refrigerant mass (mr), and variation in capillary tube length (Lc). The ANN model working with the feed backward back-propagation and Levenberg-Marquardt training algorithm is used to develop the model. This model is used to predict the performance considerations such as power consumption, refrigeration effect and COP. Similarly to develop an MRA and it has been checked for adequacy and significance. Finally, the ANN model is an excellent agreement to a multiple regression model with the coefficient of determinations (R2) of 0.989. The minimum correlation coefficient obtained with the ANN method is higher than the maximum correlation coefficient obtained from the MRA. Also, the value of the error of the ANN method is much less than the MRA method indicating the higher confidence coefficient of the ANN. These values indicate that the developed ANN model is more reliable and accurate than the multiple regression analysis. Based on the above study reveals that the ANN approach is a better technique than MRA for performance and prediction of domestic refrigeration system with high accuracy