PETROPHYSICAL EVALUATION USING MACHINE LEARNING MODELS FOR THE PREDICTION OF POROSITY

A key branch of artificial intelligence is machine learning models. The incorporation of these models into petrophysical analysis has gained popularity since it gives a more cost-effective and efficient method of acquiring petrophysical parameters. Porosity prediction was performed for this study utilizing machine learning models and 10 well log data from Niger Delta X-Fields wells. The well data from well 02 was used to train four machine learning models. The Ridge Regression model, Bagging Regressor model, ExtraTrees Regressor model, and Xgboost model were employed. The model that predicted porosity the best was chosen and used to forecast missing permeability logs from nine (9) other well log data sets. The available log data include Caliper, Gamma, Res_Deep (Resistivity), Density, PHIE (Porosity), SW (Water Saturation), VSH (Volume of Shale), and (Permeability) logs. The bagging model was selected as it was the most effective, with a mean absolute error of 0.003, a root mean squared error of 0.010, and a mean absolute percentage error of 2.3%. This in turn enabled the prediction of porosity logs for the aforementioned amount of wells with a very low percentage error. Predictions were carried out using mainly the Permeability and Density logs as they provide a very strong correlation to Porosity. It was discovered that the difference between AIC value and mean absolute error value cannot be used as the only method of model evaluation; hence, the entire error margin, as well as the visualization using subplots must be taken into consideration when evaluating model performance. It should also be noted that, the percentage error of the various models differ slightly; however, the model with the smallest error margin should be used.