| dc.description.abstract | Parameters such as API gravity and sulfur and nitrogen concentrations are important to determine crude oil quality. API, S and N grade determinations were evaluated by ASTM D-7042, ASTM D-5453 and ASTM D-4629 methods, respectively. As an alternative to official methods, the determination of these parameters can be performed using Fourier transform infrared spectroscopy with attenuated total reflectance (ATR-FTIR) associated with multivariate calibration methods. The first derivative with a Savitzky-Golay filter (SG) and multiplicative scattering correction (MSC) data transformation methods were evaluated. The initial construction of calibration models was set up using randomly selected samples for the calibration and prediction sets and partial least squares algorithm (PLS). The iPLS, biPLS and siPLS (iToolbox for MATLAB) were evaluated as variable selection methods for each oil property. The iPLS algorithm produced the best calibration model for sulfur using 335 independent variables and a root mean square error of prediction (RMSEP) of 517 mg kg-1. The siPLS algorithm presented the best models for API gravity and total nitrogen content, with RMSEP values of 1.98° and 321 mg kg-1 and number of variables equal to 168 and 167, respectively. The results showed that multivariate calibration models are a viable alternative for the determination of API, S and N in crude oils. In addition, the application of the variable selection tools allowed the optimization of the method, making it more robust, simpler and faster. | |
