Prediction of the Composition and Proportion of Blended Crude Oil Using Near-Infrared Spectroscopy

Crude oil blending is a process to optimize the properties of blended crude oil for production processes. However, quickly and accurately obtaining the components and properties of blended crude oil is challenging due to its complex composition. In this study, we combined near-infrared (NIR) spectroscopy technology and convolutional neural networks (CNNs) to analyze complex blended crude oil qualitatively and quantitatively. We first simulated the spectral data of blended crude oil based on the types and proportions of single-type crude oils used in the blending process. The first-order derivative processing, wavelength selection, and principal component analysis were used to extract spectral information. Then, we established qualitative and quantitative analysis models for the blended crude oil using the high computational efficiency and strong feature extraction capabilities of CNNs. The experimental results demonstrated that the models could predict the composition and proportion of single-type crude oils composing blended crude oil, offering a convenient, fast, and accurate analytical approach for optimizing crude oil blending processes.