ABSTRACT

Machine learning models form a major branch of artificial intelligence. The application of these models into petrophysical analysis has been widely adopted as it provides a more economical and efficient means of obtaining petrophysical properties. For this project, Permeability prediction was carried out using machine learning models and ten well log data from wells of the Niger Delta X-Fields. Several machine learning models were trained with the well data from well011. The models used were the Ridge Regression model, Bagging Regressor model, ExtraTrees Regressor model and the Xgboost model. The model that best predicts the permeability was selected and then used to predict missing permeability logs of ten (10) other well log data. The available logs were the Caliper, Gamma, Res_Deep (Resistivity), Density, PHIE (Porosity), SW (Water Saturation), VSH (Volume of Shale), and PERM (Permeability) logs. The extra tress regression model was selected as it produced a mean absolute percentage error of about 0.76% and a mean square error of less than 10% and also performed best during the training of all the models. This in turn enabled the prediction of permeability logs for the aforementioned amount of wells with a very low percentage error. Predictions were carried out using mainly the Gamma ray and Porosity logs as they provide a very strong correlation to permeability. The model performance across all the wells was deemed acceptable as its predictions corresponds with possible reservoir formations which were determined from low Gamma Ray Log readings which indicates a low percentage of shale thereby giving off relatively higher permeability predictions.