An optimized ANN for the performance prediction of an automotive air conditioning system

Abstract

This article presents the prediction of the thermal performance of an automotive air conditioning system (AACS) by using an artificial neural network (ANN). The ANN has predicted the cooling capacity, compression work, and coefficient of performance (COP) of the AACS for a range of input parameters like refrigerant charge, compressor speed, and blower speed under a steady state. The ANN, optimized for a 3–10–3 configuration with the Levenberg-Marquardt algorithm, has shown a good agreement with the experimental values with a correlation coefficient higher than 0.999, mean relative error (MRE) between 5.0% and 6.49%, and low range of root mean square error (RMSE) and error index (EI). The impact of normalized and unnormalized data along with the type of input parameters on the model performance is also observed with a large number of experimental data. This investigation shows that a suitably designed ANN can provide better accuracy and higher reliability. It can be used as a predictive tool for an AACS that generally has a wide variation of operating conditions.