Application of Bagging Ensemble Machine Learning Models to Predict Porosity of Sandstone Formations Using Well Log Data

Machine learning (ML) which is a subset of artificial intelligence is being used in the field of upstream oil and gas industry to enhance its sustainability by reducing the time consumed and capital spent in the course of subsurface characterization which key parameters such as porosity is evaluated. Over the years traditional ML models, such as artificial neural networks, support vector machines and decision trees have been used in porosity prediction. A new breed of ML developed by combining ML base models known as ensemble ML has been used in recent studies in many engineering fields, such as petroleum engineering, civil engineering, and chemical engineering. Bagging is a type of ensemble ML that has the capability to reduce overfitting which is a weakness seen in traditional ML algorithms. In this study, usability of two regression-friendly bagging ensemble ML models; random forest regression (RFR) and extra tree regression (ETR) were used to investigate the prediction of porosity in sandstone rich formations. The models were developed using a dataset acquired from Volve oilfield in North Sea. Several data preprocessing methods, such as data arrangement, outlier removal and removing missing values were applied on the dataset. Further, the dataset was divided in the ratio of 60:20:20 for training, validating, and testing the models. The final dataset consisted of 900 datapoints with three input features. Here, well log data was used as inputs while calculated porosity was used as targets. The RFR and ETR algorithms yielded predictions with R² of 0.8168 and 0.8056 respectively. However, RFR’s computational time was considerably higher, and it was more than twice that of ETR. Out of the three input features resistivity logs had the highest influence on the output for both the algorithms. Based on the performance and time consumption results, it was observed that bagging ensemble models have the potential to be used as a tool in the petroleum industry to predict the porosity of sandstone formations.