Enhanced Characterization of Heptanes Plus Fraction in Crude Oil Using Machine Learning Techniques

Abstract Characterization of crude oil is important for describing the phase behavior of the fluid, which aids in determining the productivity of the reservoir. Crude oil consists of numerous hydrocarbon components, with the heavier ones typically grouped under the C7+ fraction. This fraction is often characterized by a single set of critical properties for PVT calculations, which can lead to inaccuracies. To address this, it is essential to employ advanced characterization techniques that more accurately capture the properties of the heavier crude oil components. The primary objective of this study is to apply various machine learning methods to replicate traditional C7+ characterization approaches, aiming to improve accuracy and representation. Also, clustering techniques were used to group hydrocarbon components independently from the conventional lumping techniques. The results indicated that both boosting and neural network models achieved an average accuracy with R2 values exceeding 0.85 in predicting the critical pressure, critical temperature, and acentric factor for each lumped group. Consequently, these models can effectively replace traditional methods of characterizing C7+ fractions. The application of these models can substantially reduce computational time while maintaining high accuracy. On the other hand, the clustering approach also demonstrated promising results, employing techniques such as K-means and hierarchical clustering to group hydrocarbon components based on their mole fractions. Notably, both methods effectively divided the C7+ fraction into distinct groups, yielding unique compositions for each group. By evaluating various performance metrics, both clustering methods were proven to be efficient in accurately grouping hydrocarbon components.