Machine learning modelling of dew point pressure in gas condensate reservoirs: application of decision tree-based models.

In gas condensate reservoirs, the dew point pressure (PDew) plays a significant role in gas and liquid assessment, reservoir characterisation, surface facility design, and reservoir simulation. Although field and laboratory measurements of PDew give accurate results, both approaches are time-consuming and resource-intensive; hence, a fast and accurate determination of PDew is very important. Equation of states (EoS) and empirical correlations are other alternative methods that are used for PDew determination. However, these methods are unable to fully capture the non-linear and complex relationships between fluid composition and PDew. Machine Learning (ML) methods, as reliable tools, have emerged in different aspects of engineering. In this study, for the first time, the application of different decision tree-based methods for the prediction of PDew is investigated. A comprehensive database, containing 681 samples (almost all the available experimental data set of pure and impure samples published from 1942 to 2018), is collected from open literature and different decision tree-based methods namely Decision Tree (DT), Random Forest (RF), Gradient Boosting (GB), and Extremely Randomised Tree (ET) are used for modelling. The statistical analysis of developed models' performance showed that the ET method yields the best predictions by Root Mean Squared Error (RMSE) and R2 values of 441 psi and 0.9227, respectively for the testing dataset. Moreover, the results show that the novel ET model has a better performance compared with existing models in the literature and EoSs for the prediction of PDew of gas condensate reservoirs. [ABSTRACT FROM AUTHOR]