Rigorous Modeling of Solubility of Acid in Supercritical Carbon Dioxide Using Connectionist approach: Comparison between ANN and density based modeling

Abstract

ABSTRACT Recently, supercritical fluid has a wide application in different industries; due to this fact, researchers have a great tendency to investigate these types of fluids and their properties. The solubility, as the important properties of supercritical, is a function of different parameters. Since the experimental tests and investigations are costly and requires lots of time, a computational investigation was designed to estimate the solubility acids with different characteristic in supercritical CO2 (SC-CO2). A well-trained multilayer network was used to estimate the solubility of different acids in SC-CO2 as a function of various parameters. These factors or parameters are pressure, temperature, molecular weight of acid, carbon number, hydrogen number, and values of constant for acid dissociation. For the training and testing step of the modeling, 180 data points from the literature were utilized. The comparison between experimental data and estimated values of acid solubility illustrated that the predicting connectionist model for acid solubility estimation of acids in SC-CO2 has high accuracy based on reported statistical parameters, so the aforementioned model is reliable for estimation of solubility. This model also was compared with the density-based model, and the ANN results demonstrate that the proposed model is reliable than the density-based model. R2 (coefficient of determination) and average relative absolute deviation (AARD) corresponded to ANN and density-based modeling are 0.9125 and 1.2688, 0.9989, and 0.9759, respectively. The mentioned parameters are the statistical witness for the superiority of ANN against the density-based model.