Using a Multilayer Perceptron Network for Thermal Conductivity Prediction of Aqueous Electrolyte Solutions

In this study, a multilayer perceptron (MLP) network is proposed to predict the thermal conductivity (λ) of an electrolyte solution at atmospheric pressure, over a wide range of temperatures (T) and concentrations (x) based on the molecular weight (M) and number of electrons (n) of the solute. The accuracy of the proposed artificial neural network (ANN) was evaluated through performing a regression analysis on the predicted and experimental values of various aqueous solutions, some of which were not used in the network training. The comparison of the developed MLP network to other correlations recommended in the literature indicates that the proposed neural network outperforms other alternative methods, with respect to accuracy as well as extrapolation capabilities. Besides, others’ conductivity correlations are usually suggested for a specific electrolyte solution and a limited range of temperatures and concentrations, while such limitations do not exist for the proposed MLP network.