A comparative study in the prediction of thermal conductivity enhancement of nanofluids using ANN-MLP, ANN-RBF, ANFIS, and GMDH methods

In this work, four types of data mining methods, namely adaptive neuro-fuzzy inference system, artificial neural network—multilayer perceptron algorithm (ANN-MLP), artificial neural network—radial basis function algorithm (ANN-RBF), and group method of data handling (GMDH) have been used to predict the enhancement of the relative thermal conductivity of a wide range of nanofluids with different base fluids and nanoparticles. The total number of experimental data used in this work is 483 from 18 different nanofluids. The input parameters are thermal conductivity of base fluid and nanoparticles, volume fraction percent, the average size of nanoparticles, and temperature. Although the results showed that all four models are in relatively good agreement with experimental data, the ANFIS method is the best. The average absolute relative deviations (AARD%) between the experimental data and those of obtained using ANFIS, ANN-MLP, ANN-RBF, and GMDH methods were calculated as 2.7, 2.8, 4.2, and 4.3, respectively, for the test sets and as 1.1, 2.4, 3.9, and 4.5, respectively, for the training sets. Comparison between the predictions of the proposed ANN-MLP, ANN-RBF, ANFIS, and GMDH models and those predicted by traditional models, namely Maxwell and Bruggeman models showed that much better agreements can be obtained using the four models especially ANFIS model. Accordingly, the ANFIS method can able us to predict the relative thermal conductivity of new nanofluids in different conditions with good accuracy.