Non-random two, liquid (NRTL) and universal quasi chemical (UNIQAUC) are activity coefficient models used widely in phase equilibria calculations. These models have two to three adjustable parameters that are determined through regression of experimental data for a specific system. Generalizing these models can reduce the time, money and effort required to carry out experiments.
This work focuses on application of a theory-framed quantitative structure-property relationship (QSPR) modeling approach. Proven theoretical frameworks are employed to develop the behavior models, and QSPR is used to generalize the substance-specific parameters of the models. This novel approach, we find, is more effective than the typical efforts to develop generalized models by using directly QSPR techniques and represents a significant contribution to current modeling techniques. Further, our theory-framed QSPR modeling represents a departure from the standard fragment-based methodology of group contributions. Upon further confirmation and development, the application of this novel approach would constitute a significant development both in scope and efficacy, as evidenced by the substantial reduction in prediction errors relative to well-established methods.
Our newly developed NRTL-QSPR model provides a priori predictions for the NRTL binary model parameters. A database of 578 binary systems was employed to develop this QSPR model. Molecular descriptor reduction and network architecture optimization were performed using evolutionary algorithms combined with neural networks. The model predicts temperature and pressure within 6% and 0.5% AAD (average absolute deviation), respectively. The overall property predictions result in significant improvement compared to the group contribution method (UNIFAC), which resulted in 18% AAD for pressure predictions.
See more of this Group/Topical: Engineering Sciences and Fundamentals