Ensemble Machine Learning Model for Predicting Rock Drillability Rate for Composite Lithology

Drilling operations for oil and gas wells are considered one of the highest operating costs for the petroleum industry, and hence, implementing the new technology-based systems is highly required for cost reduction and efficient functionality performance for the drilling system. This paper introduces a successful application of machine learning to develop a drillability rate prediction model while drilling complex lithologies by employing the ensemble technique. The study presents a procedural methodology for developing the machine learning model using three learners named functional network (FN), radial basis function (RBF), and support vector machines (SVM) that include two base learners (FN, and RBF) and one for meta strong learners (SVM). Two data sets were utilized from two vertical wells within the same field that penetrated the same sequence geology of drilled formations that covered anhydrite, carbonates, and abrasive sandstone that are commonly interbedded with silt, mudstone, and shale layers. One data set was used for training and testing the models (5000 data points), while the models were validated through an unseen data set. The data features include real-time sensor data along with conventional log data for predicting the penetration rate across the composite geology sections. The obtained results showed that developing an ensemble model (SVM-Meta) boosted the accuracy performance for predicting the rock drillability rate through the testing and validation stages. The correlation coefficient showed higher than 0.95 for the ensembled model while it has a maximum of 0.91 for base learners during the validation phase. The ensemble machine learning technique succeeded to overcome one of the challenges for predicting the rate of penetration for drilling complex geology formations.