Mid-infrared Attenuated Total Reflectance (MIR-ATR) Predictive Models for Asphaltene Contents in Vacuum Residua: Asphaltene Structure–Functionality Correlations Based on Partial Least-Squares Regression (PLS-R)

In this research, an original way to obtain models able to predict the asphaltene content in vacuum residua by mid-infrared attenuated total reflectance (MIR-ATR) spectroscopy by reduction of X variables was developed. Partial least-squares regression (PLS-R) was used to reach this goal. A total of 69 samples for calibration and 18 samples for external prediction were used. It was demonstrated that dimensional reduction of the dependent variables greatly improves the prediction power of models. This methodology was evaluated in three processes of modeling, and 18 predictive models are reported. The model with better predictability was constructed with 35 spectral intensities, in which errors of calibration, validation, and prediction were 1.354, 2.095, and 1.24, respectively, and in which regression coefficients of calibration, validation, and prediction were 0.9799, 0.9720, and 0.9838, respectively. The spectral intensities in the best model are part of nine clearly differentiated functionalities. On the basis of this, it was possible to infer structural characteristics of the asphaltenes that are part of the studied vacuum residua; it means that structure–functionality correlations were reached.