Estimation of thermal conductivity of pure Liquids by using artificial neural networks

A Feed-forward back propagation neural network is proposed to predict thermal conductivity of pure liquids at a wide range of temperatures based on their molecular weight, critical temperature, critical pressure. The weights and biases are optimized to minimize error between experimental calculated thermal conductivity data. Based on results, the best structure for neural network is logsig transfer function for hidden layer with 16 neurons in this layer. Results show that the model was able to estimate the liquids thermal conductivity satisfactorily. Average relative deviation percent for compounds is in range of 0.02% to 2%. Results show that optimum neural network architecture is able to predict thermal conductivity of pure liquids level of accuracy ARD % of 0.514, R2 of 0.9972 and RMS of 9.7×10-4.

Keywords:

Artificial Neural Network , Thermal conductivity , Levenberg-marquardt , Transfer function , Liquid conductivity