Prediction of thermal conductivity for ionic liquids using neural networks

A feed-forward three-layer neural network is used to predict thermal conductivity (k) of ionic liquids (ILs) at atmospheric pressure over a wide range of temperatures based on their critical temperature (Tc), critical pressure (Pc), acentric factor(ω) and molecular weight (MW). Since the experimental critical properties for ionic liquids are scarce in literature, in the present work we used the outputs of group contribution methods which published in pervious researches. Thermal conductivity data at several temperatures taken from the literature for 24 ILs with 294 data points which 200 data for 20 ILs were used for training the network. The accuracy and capabilities of the designed network were tested by predicting thermal conductivities for situations not considered during the training process. To show performance of selected optimal network statistical analyses is done by computing MRE, MSE and R2 (regression coefficient). The results demonstrate that the optimal configuration and the variables considered allow estimating the thermal conductivity of ILs with acceptable accuracy for engineering calculations.

کلیدواژه ها:

Thermal conductivity ، Ionic liquids ، neural networks