| dc.description.abstract | Accurate bottom hole pressure (BHP) prediction during drilling with and without kick play crucial roles in maintaining the safety and efficiency of drilling operations in the oil and gas industry. Traditional methods for BHP prediction face challenges in accuracy and real-time responsiveness. This bachelor thesis explores the application of machine learning (ML) techniques to improve BHP prediction during scenarios of normal drilling and kick in drilling operations.
To generate a diverse dataset for analysis, the Openlab Drilling Simulator developed by NORCE research institution is utilized. The simulator provides data on BHP, standpipe pressure (SPP), flow rate, hook load, weight on bit (WOB), and other relevant parameters. A total of 500 simulations are created, with 66.6\% representing normal drilling and 33.3\% representing gas kicks. Random values for flow rate, RPM, and WOB are used to simulate normal drilling conditions, while additional random values for mass rate and mass of the kick are introduced for kick simulations.
Minimal preprocessing is required as the data from the simulator is relatively clean. The primary preprocessing step involves removing unique values and unnecessary features. Following preprocessing, machine learning algorithms are applied for BHP prediction. Random Forest (RF), XGBoost (XGB), AdaBoost (ADA), and Long Short-Term Memory (LSTM) models are used for both normal drilling and drilling while kick. Feature selection and sensitivity analysis are performed to optimize the ML models.
Three different data splitting approaches are tested to determine the optimal method for training and evaluating the ML models. The best data splitting method is then used for further evaluation of the best ML model. The performance of each model is assessed based on the mean squared error (MSE) and overall prediction compared to the actual values. The thesis aims to identify the best ML model together with the optimal split for BHP prediction under normal drilling conditions and the best ML model for BHP prediction when a kick is occurring.
This bachelor thesis contributes to the growing body of research on the application of machine learning techniques in the oil and gas industry. By comparing and evaluating different ML models for BHP prediction,the study provides valuable insights into the effectiveness of these methods in enhancing the safety and efficiency of drilling operations. | |