Permeability prediction of petroleum reservoirs using stochastic gradient boosting regression

Permeability prediction of petroleum reservoirs using stochastic gradient boosting regression

Forecasting

Gasoline

Nuclear magnetic resonance

Oil wells

Regression analysis

Petroleum reservoir engineering

Oil field development

Petroleum reservoirs

Iterative methods

Learning systems

Stochastic systems

Well testing

Petroleum geology

Stochastic models

Errors

Mean square error

Random processes

Estimation

Nuclear magnetic logging

Reservoir permeability is a crucial parameter for reservoir characterization and the estimation of current and future production from hydrocarbon reservoirs. Permeability can be conventionally estimated from traditional approaches such as core analysis and well-test data, which are time-consuming and expensive. Many scientists tried to estimate permeability from nuclear magnetic resonance (NMR) logs utilizing complex mathematical equations that may achieve imprecise approximation of the permeability values. Gradient boosting generates additive regression models by successively fitting a straightforward base learner to present pseudo-residuals using least squares at every iteration. The execution speed and accuracy of gradient boosting can be considerably enhanced by applying randomization process. Besides, the randomization process improves robustness against over fitting of the base learner. So, the novelty of the current study is the development of a Stochastic Gradient Boosting (SGB) regression model to predict the permeability of petroleum reservoirs based on well logs. Besides benchmark machine learning techniques are used to predict reservoir permeability from well logs. The correlation coefficient (R), relative absolute error (RAE), mean-absolute error (MAE), root mean-squared error (RMSE), and root relative squared error (RRSE) are utilized to check the overall fitting between measured permeability versus predicted ones. It is found that Stochastic Gradient Boosting model achieved overall better performance than the other models. 2020, Springer-Verlag GmbH Germany, part of Springer Nature.

2020

Subasi, Abdulhamit

El-Amin, Mohamed F.

Darwich, Tarek

Dossary, Mubarak

journalArticle

18685137

10.1007/s12652-020-01986-0