Phase equilibrium data are essential for the proper design and operation of most chemical processes. When experimental data are unavailable, thermodynamic models, such as group contribution methods, are used to predict phase equilibrium. The accuracy of these models in predicting infinite-dilution activity coefficients (γ∞) of aqueous systems is questionable. Moreover, model development is hampered by a lack of (a) γ∞ data at temperatures above 300 K, and (b) γ∞ data for water in hydrocarbon systems. The database contains over 1400 data points at temperatures ranging from 283 to 373 K were used for this study. The data include both direct and indirect measurements for a diverse organic subset.

The modeling efforts in this study focused on developing quantitative structure-property relationship (QSPR) models for the prediction of infinite-dilution activity coefficient values ( ) of hydrocarbon-water systems. Specifically, case studies were constructed to investigate the efficacy of (a) QSPR models using multiple linear regression analyses and non-linear neural networks (b) a theory-based QSPR model, where the Bader-Gasem activity coefficient model derived from a modified Peng-Robinson equation of state (EOS) is used to model the phase behavior, and (c) a single parameter Peng-Robinson EOS. QSPR neural networks are then used to generalize the EOS interaction parameters.

In general, the use of non-linear QSPR models developed in this work were satisfactory and compared favorably to the majority of predictive models found in literature; however, these literature models did not account for temperature dependence. The Bader-Gasem activity coefficient model fitted with QSPR generalized binary interactions was capable of providing accurate predictions for the infinite-dilution activity coefficients of hydrocarbons in water. Careful validation of the model predictions over the full temperature range of the data considered yielded absolute average deviations of 3.4% in ln and 15% in , which is about twice the estimated experimental uncertainty. The results from this study further demonstrate the effectiveness of theory-framed QSPR modeling of thermophysical properties.