Flow Assurance-Asphaltene Basics

Recently we started a discussion about how machine learning algorithms could assist in mitigating flow assurance problems. In this article, I briefly refer to Asphaltene which could lead to severe problems once precipitated such as 1) fluid path blockage. 2) wettability alteration. 3) relative permeability reduction.

Introduction

Asphaltenes are generally described as poly-disperse class of organic solids that are constructed of various aromatic structures combined with aliphatic chains. They have the heaviest molecules that could be solved in some aromatic solvents such as Xylene. They precipitate from crude oil due to changes in oil composition, pressure and temperature. During EOR methodologies such as gas injection (Nitrogen - Natural gas - Carbon dioxide), Asphaltene precipitation could be a probable situation, thus the estimation of Asphaltene onset pressure (AOP) is vital in order to avoid its undesired consequences.

Different approaches have been carried out to estimate the amount of Asphaltene precipitation and they could be classified into four groups:

1. Molecular thermodynamic models, which presume polymeric structure of asphaltene molecules.
2. Colloidal models, which assume that the solubility of asphaltene molecules occur due to attachment to resins.
3. Scaling equation model, which are simple correlations based on experimental data.
4. Artificial Intelligence models, which depend on data and do not account for the chemical process of precipitation itself.

ML Research

1. (Garreto et. al. 2013) mentioned that determination of apshaltene precipitation is difficult due to its high viscosity, so that dilution is favoured yet it should be noted that asphaltene phase behaviour could alternate due to the change in composition because of the added solvents. They claimed that most past models that dealt with this topic did not represent real conditions of oil production. Thereafter, they carried out a study to find out the dependency of the model developed by Gonzalez et. al. (2010) and concluded that phase behaviour of crude oil diluted with solvents could be maintained if adequate solvents or a mixture of them were used accordingly.
2. (Gharbani et. al. 2016) introduced a support vector machine regression model that is optimized with genetic algorithms (GA-SVR) using MATLAB to predict the amount of asphaltene precipitation and compared it to the most common scaling models. They claimed that their model was outperforming on both training and testing datasets and achieved R2 score of 99.5%.
3. This year, (Tanzikeh et. al. 2022) studied the prediction of AOP and bubble-point pressure to optimize gas injection for EOR purposes. They investigated different ML algorithms such as 1) support vector machine. 2) extra trees. 3) k-nearest neighbours. Next, they compared the performance of models against the polulrly used methodology of PC-SAFT (Perturbed-chain statistical associating fluid theory) and declared that ML model does not outperform the mentioned methodology although it achieves R2-score of 0.988. On the other hand, they referred to the optimistic results achiever and asserted the fact that those models are very useful especially when there is lack of data as the proposed technique requires fewer inputs, which are (reservoir pressure and temperature - API -GOR - SARA fraction - fluid molecular weight - monophasic composition - composition of the injected gas).

References

1. Garreto, M., Gonzalez, G., Ramos, A., & Lucas, E. (2010). Looking for a model solvent to disperse asphaltenes.
2. Garreto, M. S. E., Mansur, C. R. E., & Lucas, E. F. (2013). A model system to assess the phase behavior of asphaltenes in crude oil.?Fuel,?113, 318-322.
3. Ghorbani, M., Zargar, G., & Jazayeri-Rad, H. (2016). Prediction of asphaltene precipitation using support vector regression tuned with genetic algorithms.?Petroleum,?2(3), 301-306.
4. Tazikeh, S., Davoudi, A., Shafiei, A., Parsaei, H., Atabaev, T. S., & Ivakhnenko, O. P. (2022). A Comparison between the Perturbed-Chain Statistical Associating Fluid Theory Equation of State and Machine Learning Modeling Approaches in Asphaltene Onset Pressure and Bubble Point Pressure Prediction during Gas Injection.?ACS omega,?7(34), 30113-30124.