Black Box Modeling of Hydrocarbon Physiochemical Properties and its Approach to Fuel Optimization

Abstract

An energy pathway of great interest is gas-to-liquid (GTL) technologies, which converts natural gas to valuable chemicals and fuels. Three powerful regression methods were implemented, to create models for accurate predictions of physical properties– density, freezing point, flash point, and heat content. With the use of experimental training data, three distinct techniques were performed and analyzed: artificial neural networks, support vector machine (SVM) and Kriging modeling.

For further accuracy, optimal simulation settings were elucidated through repeated runs and rigorous testing, with substantial increases in performance in low performing models. Most models generated were accurate with good trends, except freezing point. A formulation package called DataMine, coded in R, was created for current work and future endeavors.