The phase equilibria of binary mixtures of CO2+alcohols are crucial for the design and optimization of supercritical fluid applications with recently growing reliance on CO2 as the primary solvent at its supercritical conditions. Essentially, alcohols, as cosolvent, contribute to the solvating characteristic of supercritical CO2, which is de facto required for (polar) compound extraction in different industries. However, the proposed, less time-consuming alternatives to experimentally measuring the vapor-liquid equilibrium (VLE) data have been successful in such non-ideal mixtures only to a certain extent. In particular, the observed high deviation in existing models from the experimental values in correlating VLE data necessitates developing an intelligent, self-learning, unique model with more accuracy as well as applicability to a notably broader range of conditions. To this end, we have employed a novel Least-Squares Support Vector Machine (LSSVM) approach as a problem-independent, general-purpose tool. In this research, the LSSVM model is optimized with a Coupled Simulated Annealing (CSA) optimization algorithm. The hybrid model is based on a comprehensive databank of 531 reliable experimental VLE data that cover binary systems of CO2+13 types of short to medium chained alcohols at temperature and pressure overall ranges of 293.15–432.45K and 5.2–213.91bar, respectively. The statistical and graphical error analyses vividly demonstrate the supremacy and robustness of the developed CSA-LSSVM model, particularly compared to the conventional SRK and PR equations of state. However, minor deviations in mixtures containing Pentanol and Hexanol are observed, despite the overall model accuracy that strongly satisfies engineering purposes. We hypothesize that uncertainty in the corresponding experimental data could be the cause; we prove this by applying a Leverage approach, while studying the validity and applicability domain of the model. Furthermore, detailed sensitivity analysis is conducted through a relevancy factor, which suggests that the CO2 mole fraction is the most influential parameter on the studied binary mixture bubble/dew point pressures.
Read full abstract