Contents lists available at ScienceDirect Fuel journal homepage: www.elsevier.com/locate/fuel Full Length Article Comprehensive and multidimensional tools for crude oil property prediction and petrochemical industry refinery inferences Daniella L. Vale a, ⁎ , Paula F. de Aguiar a , Lize Mirela S.L. de Oliveira b , Gabriela Vanini a , Vinicius B. Pereira a , Larissa O. Alexandre a , Giovani S.C. da Silva b , Luiz André Mendes b , Alexandre O. Gomes b , Débora A. Azevedo a, ⁎ a Universidade Federal do Rio de Janeiro, Instituto de Química, Ilha do Fundão, Rio de Janeiro, RJ 21941-909, Brazil b Petrobras/Cenpes, Ilha do Fundão, Rio de Janeiro, RJ 21941-909, Brazil GRAPHICAL ABSTRACT ARTICLE INFO Keywords: Comprehensive bidimensional gas chromatography Multiway prediction Crude oil Multiway partial least squares ABSTRACT The association of comprehensive bidimensional gas chromatography with time-of-flight mass spectrometry (GC × GC-TOFMS) with high-order chemometric, N-way partial least squares (N-PLS), is an analytical innova- tion for the characterization of complex samples such as crude oil. The N-PLS method was applied to calibrate third-order data for sets of crude oil samples using whole oil comprehensive bidimensional chromatograms. The calibration model for API gravity had a bias equal to -5.8 × 10 -3 and R 2 Cal of 0.9808 and WAT model had coefficient of determination for calibration model equal to 0.9436 and a bias of 8.7 × 10 -3 . The results obtained by the decomposition of 11 components for API gravity were 99.79% for the X data and 98.08% for the Y data. The root mean square error for calibration (RMSEC) was equal to 0.81 and 1.01, while the root mean square error for prediction (RMSEP) was equal to 1.96 and 1.97 for the API gravity model and WAT, respectively and the explained variance obtained by decomposition in 9 components for WAT was 99.90% for the X data and 94.36% for the Y data. In the calibration models, all the errors for each sample were below 3.0 and 2.5 for °API and WAT, respectively. For the prediction set that was used to validate the model, the errors for each sample were below 3.0 and 3.2 for °API and WAT, respectively. The data indicates improvements for the correlation of petroleomic properties, thus allowing for the simultaneous prediction of certain properties instead of traditional analyses for each property when making inferences in the refining process. This application allows automation of the responses generated using crude oil samples without the need for pretreatment or fractionation steps; in addition, only one drop of each sample is required. This analytical application leads to cost reductions compared https://doi.org/10.1016/j.fuel.2018.01.109 Received 26 September 2017; Received in revised form 16 January 2018; Accepted 26 January 2018 ⁎ Corresponding authors. E-mail address: daniellavale@iq.ufrj.br (D.L. Vale). Fuel 223 (2018) 188–197 Available online 20 March 2018 0016-2361/ © 2018 Elsevier Ltd. All rights reserved. T