ANN-derived equation and ITS application in the prediction of dielectric properties of pure and impure CO2

High-performing equation has been step-wisely extracted from artificial neural network (ANN) simulation and subsequently applied for the prediction of the dielectric properties of pure and impure CO2. Data of relative permittivity (epsilon(r)) for pure and impure CO2 were used in the ANN to train different ANN structures so that the network can recognise and predict CO2 property under different conditions. Analyses of the results from the training showed that single-layer ANN model [3-6-1] outperformed others. From this best-performing ANN structure, a single mathematical equation was extracted that can be employed in predicting epsilon(r) for pure CO2 and CO2-ethanol mixture, even without access to ANN software. Using this ANN-based mathematical model, predictions of the relative permittivity (epsilon(r)) for pure CO2 and CO2-ethanol mixture were performed, under different temperatures and pressures, and at different ethanol concentrations. Under similar conditions, the output of the model provides good match with the original experimental epsilon(r). With increment in ethanol concentration, the model correctly predicted the rise in epsilon(r) for the mixture. Also, it was shown that the epsilon(r) rises with an increase in pressure but decreases with a rise in temperature. The work showed the reliability and applicability of the ANN in characterizing and predicting the dielectric property of pure CO2 as well as its mixture or impurities. The model developed and the techniques demonstrated in this work offers immense benefits and guides for researchers, who may want to explore the behaviours of a pure compound and its mixtures impurities using ANN, as well as those interested in derived mathematical model from statistical computation tool like ANN. (C) 2017 Elsevier Ltd. All rights reserved.