Determination of gasoline origin by distillation curves and multivariate analysis

Abstract

Pattern recognition (PCA – principal component analysis) and classification (LDA – linear discriminant analysis) were employed to determine the origin of various samples of gasoline commercialized in the state of Minas Gerais, Brazil. With this in view, distillation curves were performed following ASTM D86 standard method, and PCA demonstrated that a small number of variables dominate the total data variability since the first three principal components (PCs) accounted for 87% of total variability. LDA was constructed using the origin declared in the invoices and the distances between groups were used to determine the similarity of the samples. Refineries REPLAN/REVAP presented the lowest distance value and REDUC/REGAP, the highest. About 80% of the samples, whose origins were not declared, were classified as belonging to the REGAP group, with 95% probability of correct classification.