Multi-output chemometrics model for gasoline compounding

Computational models may help to reduce research cost by predicting properties of alternative blends. Nowadays, most efforts focus on prediction a few for sets gasoline samples. However, there are no reports able classification samples with multiple output measured in real life refinery plants. In this work, Information Fusion (IF), Perturbation Theory (PT), and Machine Learning (ML) algorithm (IFPTML) was used model production data >230,000 outcomes gathered from petroleum plant. IF-pre-processing phase assembled the working dataset 44 physicochemical vs. 574 input variables 4 lines distributed 26 blocks including 14 different streams 23 operations carried out PT-calculation quantifies effect perturbations (deviations) all using PT Operators. Last, ML-analysis involved Linear Discriminant Analysis (LDA) Artificial Neural Networks (ANN) training. IFPTML-LDA presented AUROC = 0.936 overall Sensitivity Sn Specificity Sp ≈ 84–91% training validation sets. internal control experiment we obtained an IFPTML-FT-NIR similar 86–97%, >25,000 values 16 FT-NIR technique; demonstrating robustness changes experimental techniques used. This could be useful design new alternatives blends (biofuels, refuse-derived fuels, etc.) lower environmental impact.