Unified multi-parameter predictive modeling of crude oil and its distilled fractions through Artificial Neural Networks

This article proposes an alternative approach for assessing four quality parameters of petroleum and its main products, namely API gravity, total sulfur, nitrogen content, and basic nitrogen content. Such primary measures are determined experimentally using four ASTM protocols. The proposed method is based on the use of a chemical ionization Fourier transform Orbitrap (R) mass spectrometer (CI-HRMS) and a multivariate model that allows for the simultaneous assessment of a broad range of parameters, regardless of the volatility of the complex sample (crude oil, diesel, gas oil, naphtha, and kerosene). The composition assignment is performed stepwise based on the analysis of the monoisotopic masses and isotope patterns to generate chemically meaningful identifications. The predominant chemical composition consisted of classes of compounds containing sulfur, nitrogen, and oxygen. The analysis time was about 6 min, due to a fast automated analysis that uses fingerprinting information to predict the features of crude oil and its derivatives. The developed Artificial Neural Network (ANN) models are independent of the volatility of the sample and can be applied to both raw petroleum and its products. The ANN models afford acceptable predictions. The values of the global coefficient of determination (R2) ranged from 0.86 to 0.97, which were calculated between predicted and reference values - obtained from ASTM methods. The four quality parameters are estimated by convenient CI-HRMS method and such measurements are demonstrated to be a viable alternative to the classic ASTM methods otherwise only available by laborious and low sample throughput experimental techniques.