Abstract
This research is proposing a new machine-learning approach for predicting static bottom hole pressure (SBHP) using localized injection and production ratio (IPR) data and related parameters. The methodology identifies key features, compares several machine learning techniques, and demonstrates the potential of the suggested approach to improve the precision and effectiveness of SBHP estimate using the available dataset. The goal is to offer an innovative methodology that can help with better SBHP forecasting using the data already available.
The machine learning approach involves a comprehensive methodology for forecasting SBHP in individual wells. Key parameters considered in this study include well depth, reservoir and wellbore properties, production and injection rates, time, and various flow and pressure data.
This research incorporated various types of features, including static, dynamic, numerical, and categorical data. To ensure optimal performance of the machine learning model, the data underwent preprocessing. The methodology employed decision trees, random forests, and gradient boosting as machine learning algorithms. The evaluation of the model involved the utilization of the coefficient of determination (R-squared), mean squared error (MSE), and root mean squared error (RMSE). Training the model relied on a dataset comprising more than 13,000 data points gathered from diverse fields and wells.
The results of this study indicate the precision of machine learning algorithms in forecasting static bottom hole pressure (SBHP) using individual well injection/production ratio and other essential parameters. By combining static, dynamic, numerical, and categorical the proposed approach achieved a remarkable 97% accuracy in SBHP prediction. Decision trees, random forests, and neural networks, which were the machine learning algorithms employed in this research, demonstrated encouraging outcomes in SBHP forecasting. Notably, gradient boosting regressor algorithm exhibited superior performance, delivering highly accurate SBHP predictions.
