Porosity prediction using ensemble machine learning approaches: A case study from Upper Assam basin

Abstract

Porosity is an important petrophysical parameter that determines the amount of fluid, including oil, water, and gas contained within the rock. In petroleum exploration, it plays a crucial role in reservoir characterization, petrophysical studies, and geological analysis. However, in practice, porosity logs do not enumerate in the recorded logs for the entire well, either due to instrumental error or borehole environmental issues. Thus, the prediction of the porosity logs is essential to perform better petrophysical analysis and reservoir characterization. In the present study, we emphasize the implementation of machine learning (ML) approaches in predicting porosity logs at the missing data interval. Here, we predicted the porosity log from the sonic, bulk density, and gamma log. Initially, using the existing data from the well, we prepared the training sample. Further, using the extreme gradient boosting (XGBoost) and random forest (RF) ML approaches, we obtained the mapping function between the input logs (bulk density, gamma, and p-wave velocity logs) and output log (neutron porosity) of the training sample. Subsequently, using the mapping functions of the trained RF and XGBoost models, we obtained the porosity at the test wells at the missing data intervals. The outcome of the present study is satisfactory enough as we have seen the correlation coefficient values of the actual and predicted porosity vary between 0.8 and 0.9. Thus, ensemble machine learning approaches could provide other petrophysical parameters at the missing data interval.