Prediction of Long-Residue Properties of Potential Blends from Mathematically Mixed Infrared Spectra of Pure Crude Oils by Partial Least-Squares Regression Models

Abstract

Research has been carried out to determine the feasibility of partial least-squares (PLS) regression models to predict the long-residue (LR) properties of potential blends from infrared (IR) spectra that have been created by linearly co-adding the IR spectra of crude oils. The study is the follow-up of a recently developed method to predict LR and short-residue properties from IR spectra and which is currently the subject of PCT patent application WO 2008/135411 filed by Shell International Research Maatschappij B.V. It is found that the PLS prediction models for seven different LR properties [i.e., yield long on crude (YLC), density (DLR), viscosity (VLR), sulfur content (S), pour point (PP), asphaltenes (Asph), and carbon residue (CR)] enabled us to predict the LR properties of 16 blends in two ways. The first predictions were carried out on the IR spectra recorded from the physically prepared blend samples. Next, IR spectra were submitted to the PLS models that were created mathematically by linearly co-adding the IR spectra of the corresponding crude oils in the appropriate weight ratio. Minor differences in the real and artificial blend spectra have been observed, which have been assigned to nonlinear effects. However, preprocessing of the spectra, by subsequently taking the first derivative, multiplicative scatter correction (MSC), and mean centering (MC), resulted in predicted LR property values of the two parallel sets that are largely the same. It implies that mimicking blend spectra by mathematically mixing the IR spectra of crude oils is a valuable, fast, clean, and cheap alternative for the “dirty” and elaborate preparation and testing methods of real blends currently used in the laboratory. Besides, the method can be used as a rapid screening tool for a large series of potential blends.