Predicting fuel properties using chemometrics: a review and an extension to temperature dependent physical properties by using infrared spectroscopy to predict density

Although the use of chemometric methods to predict fuel quality properties has received wide attention over the past three decades, as seen from the review included with this article, no studies were found about predicting temperature dependent properties of fuels. Since our research is focused on determining thermodynamic properties, rather than quality properties, taking temperature dependencies into account became even more important. To determine if accurate predictions could be obtained over a range of temperatures, the densities of over 300 fuel samples (mostly narrow boiling range oil fractions, considered here as pseudocomponents) were measured and predicted. An alternative fuel (a phenol-rich oil shale oil) was studied because the property prediction methods developed for conventional petroleum samples often give poor results for this and other alternative fuels. The temperature dependence of density for these fuel samples was modelled using a linear equation based on the density at 20 °C and the slope of the density-temperature relationship. Support vector regression was used to predict these parameters for each sample from its infrared spectrum. Then these parameters were used to predict the densities at other temperatures. Densities spanned the range from 0.713 to 1.088 g/cm3, and the root mean squared error of the predicted values was 0.004660 g/cm3, which is a relative error of less than 1%. In addition to the experimental portion, a literature review is included, which contains an assessment of the accuracy of chemometric methods for predicting many fuel properties.