An Inferable Machine Learning Approach for Reservoir Lithology Characterization Using Drilling Data

Abstract

Abstract Reservoir lithology is a key factor in petroleum exploration and petrophysical calculations. It is of utmost importance as it serves as a foundation for reservoir characterization and formation evaluation. Accurate estimation of the reservoir permeability, porosity, and water saturation, is greatly dependent on accurate identification of the reservoir lithology. Ideally, the reservoir lithology is determined by obtaining physical samples of the reservoir. This process is however very expensive and time-consuming, hence the wide adoption of well log responses for identifying the reservoir lithology. Most Machine learning approaches are imminently built to render good classification, and some have been adapted to probability estimation. The purpose of this study is to demonstrate how machine learning can be used to estimate the probability of reservoir lithology with the use of drilling data. The drilling data used in this research is from the Volve oil field in Stavanger, Norway. The preprocessed data consisted of pump pressure, surface torque average, rotation per minute of drill bit, mudflow rate, total gas content, effective circulation density, pump stroke rate, lithology type, and weight on bit. The data was split into 80% for training and 20% for the test set. Feature selection was done using expert domain knowledge. The three lithology characteristics captured by the data include sandstone, claystone, and marl. Intelligent models are algorithms designed to learn from large volumes of data and draw valuable insights from them. Examples are neural networks, logistic regression, and Random Forest. In this study, we are primarily interested in probabilistic prediction rather than label classification or a deterministic prediction. The problem was treated as a probability estimation problem using logistic regression, Decision trees, and Random Forest models. Decision Trees are a type of supervised machine learning where the data is continuously split according to a certain parameter. Logistic regression is a supervised learning classification algorithm used to predict the probability of a target variable. Random Forest is an ensemble learning method for classification and regression that operates by constructing multiple decision trees at training time. The probabilistic classifier predicts a probability distribution over a set of lithology classes using drilling data. The stratified k-fold cross validation technique was used for model comparison on the training data. The performance of models was evaluated using the metrics- accuracy score, the area under the receiver operating characteristic curve (AUC), precision, recall and f1 score. The AUC score was considered to be the best evaluation metric for the task. We relied on the receiver operating characteristic curve (ROC) and the area under the curve (AUC) to evaluate the performance of the models. The higher the AUC, the better the ability to distinguish between the lithology classes. The logistic regression, Decision trees, and Random Forest models achieved ROC AUC scores of 0.7547, 0.8747, and 0.9932 respectively. The results revealed that the Random Forest model outperformed the other models. The Random Forest model achieved a ROC AUC score of 98.59% on the test dataset indicating its capability to estimate the probability of having a reservoir lithology with a high confidence level. This study resulted in the application of machine learning techniques to develop models capable of estimating the probability of a reservoir lithology in the absence of a reservoir sample. The models were developed by fitting logistic regression, Decision trees, and Random Forest machine-learning algorithms to a drilling dataset. The results revealed that the models performed satisfactorily in estimating the probability of a reservoir lithology. The Random Forest model outperformed the other models. Therefore, in the absence of a reservoir sample, the probability of a reservoir lithology can be estimated using the model. These predictions can be used for compatibility tests between formation and bit, improved bit selection programs, and drilling rate optimization. The accurate predictions from the model will be very useful for drilling planning and bit optimization thereby reducing drilling costs. Lithology characterization based on drilling data is also important for real-time geosteering in the oil and gas industry.