Bulk density prediction in missed intervals of Nubian reservoir using multi-machine learning and empirical methods

This paper presents a comprehensive study on predicting bulk density in missed intervals of the Nubian reservoir in the Sirt Basin, Libya, leveraging both empirical and machine learning methodologies. Bulk density is one of the most significant and crucial parameters for rock physics modeling, geomechanical analysis, and reservoir characterization; however, this measurement is not present in all intervals of the Nubian reservoir in the Sirt Basin to predict an accurate, reliable prediction and save cost. Empirical equations such as Gardner, Lindseth, and Khandelwal models, alongside machine learning algorithms including random forest (RF), multi-layer perceptron (MLP), and support vector machine (SVM), are employed using conventional logs that were collected from four vertical wells. The data set undergoes a pre-processing step before being divided into 50%, 20%, and 30% for training, testing, and validation, respectively. The optimization is performed using the Grid search CV function. Based on the findings, using machine learning rather than empirical models to predict bulk density is more effective. The machine learning model achieves a higher correlation coefficient above 0.89 and lower mean absolute error than the empirical approaches. Conclusively, a predicted bulk density by supervised machine learning approaches can be used as a reference in all intervals that lack the density log.