Neural network-based correlations for the thermal conductivity of propane

An alternative approach, exploiting neural networks, is proposed to develop thermal conductivity correlation of propane for the first time. In order to test the accuracy of the proposed technique and demonstrate its utility in fitting the thermal conductivity surface of propane, we have established a thermal conductivity correlation in terms of temperature and density, and then compared its predictions with those obtained by the conventional method. The results obtained are so impressive that the neural network correlation has lower overall average absolute deviations (AADs) in each data set. The requirement of using a high accuracy equation of state (EoS) for the correlations which include density as a variable has been avoided by developing thermal conductivity equations as a function of temperature and pressure. For this purpose, three neural network models have been constructed for the liquid, vapour, and supercritical phases. It is found that neural network approach produces a much better correlation for the liquid region while the predictions of the other two models are in substantial agreement with the traditional results. Consequently, neural networks offer a powerful tool for the development of thermal conductivity correlations of fluids, no matter whether an EoS is used or not. © 2007 Elsevier B.V. All rights reserved.