Use of artificial neural networks for prediction of phase equilibria in the binary system containing carbon dioxide

Present work investigated the potential of artificial neural network (ANN) model to correlate the bubble and dew points pressures of binary systems containing carbon dioxide (CO2) and hydrocarbon systems as functions of reduced temperature of non-CO2 compounds, critical pressure, acentric factor of non-CO2 compounds and CO2 composition. In this regards, five binary systems at the temperature and pressure ranges of 263.15–393.15 K at 0.18–12.06 MPa were used to examine the feasibility of cascade-forward back-propagation ANN model. In this regard, the collected experimental data were divided in to two different subsets namely training and testing subsets. The training subset was selected in a way that covers all the ranges of the experimental data and operating conditions. Then, the accuracy of the proposed ANN model was evaluated through a test data set not used in the training stage. The optimal configuration of the proposed model was obtained based on the error analysis including minimum average absolute relative deviation percent (AARD %) and the appropriate (close to one) correlation coefficient (R2) of test data set. The obtained results show that the optimum neural network architecture was able to predict the phase envelope of binary system containing CO2 with an acceptable level of accuracy of AARD % of 2.66 and R2 of 0.9950 within their experimental uncertainty. In addition, comparisons were done between the Peng–Robinson (PR) equation of state (EOS) and ANN model for three different binary systems including CO2 + 1-hexene, CO2 + n-Hexane, and CO2 + n-butane. Results show that developed optimal ANN model is more accurate compared to the PR EOS.