A General Hybrid GMDH–PNN Model to Predict Thermal Conductivity for Different Groups of Nanofluids

Abstract

In this study, a general model for estimating the nanofluids (NFs) thermal conductivity by using a hybrid group method of data handling polynomial neural network (GMDH–PNN) has been investigated. NFs thermal conductivity was modeled as a function of nanoparticle size and volume fraction, nanoparticle and base fluid thermal conductivity, and base fluid temperature. For this purpose, a network that contains 6 hidden layers with 2 inputs in each layer and with training algorithm of least squares regression has been applied. The results showed a good accuracy for estimating the thermal conductivity of NFs with a root mean squared error (RMSE) of 0.03027 for 118 systems containing 1929 training data sets. Furthermore, the RMSE for 27 systems containing 244 data as the validation sets was 0.02843 and also mean absolute percentage errors (MAPE) for training and validation data sets were 4.47 and 4.59%, respectively. Moreover, the proposed hybrid GMDH–PNN model was compared with different models from literature for different groups of NFs. The results indicated an improvement in prediction of thermal conductivity with lower errors compared to the previous models.