Artificial Intelligence Models for Real-Time Bulk Density Prediction of Vertical Complex Lithology Using the Drilling Parameters

Practically, the rock bulk density is measured either through logging while drilling tools or wireline logging techniques. However, these measurements are not always available, which necessitates using different empirical correlations. But these correlations have considerable limitations, which restricted their reliability and accuracy. This work aims to develop several artificial intelligence models for real-time bulk density prediction of complex lithology while drilling. The support vector machine (SVM), functional networks (FN), and random forest (RF) techniques were applied using the drilling parameters as inputs. A vertical well of 2912 data points of complex lithology containing sand, shale, and carbonate was used for model development. The developed models were validated using different dataset from another well. The results demonstrated that the three developed models predicted the bulk density with high matching accuracy. The SVM approach gives correlation coefficient (R) values of 0.999 and 0.984 and average absolute percentage error (AAPE) values of 0.113 and 0.512% in training and testing processes. The FN technique has R values of 0.980 and 0.979 and AAPE of 1.115 and 1.163%, while the RF-based model results in R values of 0.996 and 0.992 and AAPE values of 0.453 and 0.635% for training and testing, respectively. The validation process indicated the reliability and robustness of the constructed models with R values of 0.992, 0.977, and 0.990 and AAPE of 0.366, 1.224, and 0.733% for SVM, FN, and RF approaches, respectively. Each developed model can predict inexpensively the bulk density for multiple lithology types in real-time at high matching accuracy.