Sequential Data Approach for Rate of Penetration Prediction Using Machine Learning Models: A Case Study the Offshore Volve Oil Field, North Sea, Norway

Abstract

In drilling operations, the Rate of Penetration directly impacts drilling efficiency. Accurately predicting this rate is crucial as it affects the cost-effectiveness of the process. In this study, ensemble machine learning models such as Random Forest, Gradient Boosting, Extreme Gradient Boosting, Adaptive Boosting, and K-Nearest Neighbor were used to predict ROP using a Volve dataset from the Norwegian North Sea provided by Equinor Company. To simulate real-world scenarios, especially in cases with limited amounts of data, we utilized sequential data or depth-dependent data to trained and tested various machine learning models using sequential split such as continuous learning and sliding window split techniques. The model’s performances were evaluated by calculating mean absolute error. The results of this study showed that the ensemble machine learning models, especially Extreme Gradient Boosting with sliding window split technique outperformed in predicting the Rate of Penetration. Moreover, the various sizes of the sliding window were studied to find the optimal size.