Machine learning models for generating the drilled porosity log for composite formations

Determining the porosity of the drilled formation is a significant task for formation evaluation purposes for further implementation in petroleum reservoir simulation and estimating the hydrocarbon reserves. The cost and time are common limitations for the rock porosity determination by the conventional logging way. To overcome these restrictions, this research presents a new approach for porosity estimation from the drilling data by employing the vision of the fourth industrial revolution. The study utilized three machine learning methods (support vector machine (SVM), adaptive network-based fuzzy inference system (ANFIS), and functional network (FN)) for developing rock porosity prediction models from the drilling parameters for composite types of formation lithology that include sandstone and carbonate formations in addition to shale rock. The drilling data covered the penetration rate, rotating speed, torque, weight on bit, mud pumping rate, and pressure of standpipe with 3817 measurements for developing the models. The three models were optimized for enhancing better accuracy level, and the models were evaluated by calculating the coefficient of correlation (R), variance account for (VAF), average absolute percentage error (AAPE), and root mean square error (RMSE) between the real porosity measurements and the predicted ones. The obtained prediction for the developed models presents a strong prediction for the rock porosity for the three developed models with R greater than 0.92, VAF higher than 0.85, AAPE less than 8.34%, and RMSE below 0.02. For further validation, the models were tested for an unseen dataset of 1000 points which covered the same drilled formation types, and the results confirmed the high level of accuracy. In addition, the study proposed FN-based equation for porosity prediction from the drilling data while drilling composite lithology formations.