Predictive Modelling of Flowing Bottom Hole Pressure in Sandstone Reservoir Formations Using Machine Learning

Abstract Flowing Bottom Hole Pressure (FBHP) serves as a critical parameter in reservoir engineering, providing insights into reservoir behaviour and aiding in production optimization. Leveraging production data from a sandstone reservoir. This study investigates the utilization of machine learning models to predict Flowing Bottom Hole Pressure (FBHP). Support Vector Regressor (SVR), Multilayer Perceptron (MLP), and Random Forest Regression (RFR) algorithms were independently trained with various hyper-parameters to ensure optimal model selection and parameter combinations. Six different combinations of the available input production data were utilized for training and testing purposes. SVR demonstrated superior performance in FBHP prediction for the Sandstone formation, exhibiting a coefficient of determination (R2) close to 0.9999 across all six scenarios for both training and testing datasets. Additionally, other regression analyses proved effective in predicting FBHP for the Sandstone formation, achieving R2 values exceeding 90% with the input production dataset scenarios. These high R2 values underscore the accuracy and reliability of the predictive models, indicating their proficiency in capturing the complex relationships between input features and FBHP. By accurately forecasting FBHP, these models offer valuable insights for reservoir management and production optimization strategies. They enable engineers and operators to make informed decisions, leading to enhanced hydrocarbon recovery and increased operational efficiency. The robustness of the machine learning models demonstrates their potential to revolutionize reservoir engineering practices, paving the way for more effective and sustainable oil and gas production operations.