Artificial neural network modelling of liquid thermal conductivity for alcohols

This study investigates the applicability of artificial neural networks (ANNs) as an efficient tool for the description of thermal conductivity of liquid alcohols for a broad range of temperatures. The proposed multilayer perceptron has 1 hidden layer with 43 neurons, determined according to the constructive approach. The model developed was trained and validated on the set of data gathered, showing that the accuracy of the ANN model is higher than that of other approaches proposed in the literature. The experimental or experimental and predicted thermal conductivity data of alcohols were taken from the database due to the ‘DIPPR Database’. The ability of the ANN method to reproduce the original data was tested for 26 alcohols in the liquid phase at reduced temperatures ranging from 0.30 to 0.90. The maximum absolute deviations between experimental and calculated thermal conductivity data points are generally less than 0.0110%, while the average absolute deviations are usually less than 0.0016%. This study shows that the model used is a good alternative to estimating thermal conductivity of alcohols with excellent precision.