Porosity Prediction for Carbon Dioxide Storage Assessment Using Boosting Ensemble Machine Learning Algorithms. Case Study: Darling Basin, Australia

Machine learning has been utilised to estimate porosity in the subsurface formations for the past couple of decades in reservoir engineering. However, with the newfound interest in carbon sequestration to drive the oil and gas industry towards carbon net zero, estimating porosity of potential carbon storing formations has become a vital part in estimating the carbon storage capacity. Boosting ensemble machine learning algorithms are capable of reducing underfitting, which is an advantage over many traditional machine learning algorithms. In this study, three regression-friendly ensemble algorithms, Adaboost regression, gradient boost regression and extreme gradient boost regression were used to predict porosity of sandstone layers in Darling basin. The dataset was cleaned and fed into the machine learning algorithms developed using Python programming language. Models' performances were evaluated using coefficient of determination
$(\mathbf{R}^{2})$
. Results showed that, extreme gradient boost regression algorithm performs the best during porosity prediction with a test-model
$\mathbf{R}^{2}$
value of 0.9321. Further, gradient boost regression algorithm yielded an
$\mathbf{R}^{2}$
value of 0.8804 while Adaboost regression yielded the least
$\mathbf{R}^{2}$
value of 0.8129. Moreover, results suggested that extreme gradient boost regression algorithm can successfully be used in predicting porosity of subsurface formations, during the assessment phase of carbon dioxide storage.