Year: 2025
Type: article
Abstract: The key to a successful and economically viable drilling operation lies in optimum drilling aimed at completing drilling activity within the shortest possible period with the least damage to equipment, formation, and environment. However, the increase in complexity requirements for drilling activities has led to several problems such as poor rate of penetration (ROP), which results in an increase in non-productive time (NPT) and in additional associated costs. In this study, Random Forest (RF) from scikit-learn package and Linear Regression (LR) from XlStat software were utilized for machine learning evaluation for ROP prediction. A 150 dataset from a Niger-Delta oilfield was utilized to develop a model which was statistically evaluated and validated, featuring important parameters detected. Study results show the RF regression model recorded R2 of 99.12%, MSE of 9.38%, RMSE of 3.06% and MAE of 1.8%, while LR yielded R2, MSE, MAE and RMSE of 77.5%, 25.42%, 2.56%, and 5.2% respectively. Validation results present RF recorded values closer to the actual ROP in comparison to the ones found using the Linear Regression (LR) model. From the results, feature importance study, plastic viscosity, rotary speed, and standpipe pressure had the highest impact on ROP.
Authors: A. Kerunwa, C. F. Dike, A. H. Ezekiel, C. E. Okalla
Institution: Federal University of Technology Owerri
Language: English
Subfield: Ocean Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4409506223