Key points are not available for this paper at this time.
Abstract This paper presents the development of a machine learning model for predicting daily oil production volume in barrels (bbl) based on pressure-related data. The Volve field production data from 2008 to 2016, operated by Equinor, was used for training and testing the model. An exploratory data analysis was conducted to identify outliers and anomalies in the dataset, and the most relevant features for the prediction task were selected based on statistical methods and domain knowledge. The selected features included pressure-related parameters such as average downhole pressure, average production tubing size, average annulus pressure, average choke size in percentage, average wellhead pressure, production choke size, and onstream hours. To prevent overfitting and bias, features with high correlation were identified and dropped, and the remaining features were evaluated using Mutual Information (MI) scores. Various algorithms, including Convolutional Neural Networks (CNN), Feed-Forward Neural Network (FNN), Linear Regression (LR), XGBOOST, and Long Short-Term Memory (LSTM) Networks, were tested for training and evaluating the predictive performance of the model. The LSTM model outperformed the other models with an average metric score of 104, demonstrating superior ability to generalize trends and learn minimal signals. The model was then deployed on a web app for the prediction of oil production volume based on pressure-related data. The Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) were used as metrics to evaluate the performance of each model. The results indicate that the developed machine learning model is highly effective for predicting daily oil production volume using pressure-related data. This approach has the potential to significantly improve production optimization in the oil and gas industry, offering a robust and reliable tool for operational decision-making.
Ajibade et al. (Mon,) studied this question.