A Deep Learning Framework for Accurate Liquid Rate Prediction & Back Allocation in Oil Production

Abstract Accurately predicting production rates for individual wells in commingled production systems is a crucial yet challenging task in the oil and gas industry. Traditional allocation methods often fail to adapt to the dynamic nature of reservoir conditions and operational behaviours, resulting in significant inaccuracies. This research introduces a deep learning approach to improve the back allocation of oil production, leveraging advanced machine learning algorithms to enhance prediction precision and flexibility. The study involved comprehensive data collection and preprocessing of historical production data, with critical features such as choke size, flowing tubing head pressure (FTHP), separator pressure, and temperature engineered to capture intricate relationships between parameters. Deep learning models, alongside other machine learning techniques like Gradient Boosting, K-Nearest Neighbours (KNN), Decision Tree, and XGBoost, were trained and evaluated using metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), and R2. Among these, XGBoost exhibited exceptional predictive accuracy, achieving an MAE of 5.7549, an MSE of 211.8709, and an R2 of 0.8595. Case studies utilizing data from Niger Delta peripheral fields further validated the models, demonstrating high accuracy and reliability even in scenarios without dependable well tests. Additionally, a Streamlit application was developed to allow users to adjust parameters interactively and dynamically predict liquid rates. This application enhances practical utility, enabling real-time adjustments and predictions, and showcasing significant improvements over traditional methods in real-world case studies. By integrating deep learning into the back allocation process, this research offers a robust and adaptive methodology for dynamically predicting production rates. This innovative approach addresses the challenges of conventional allocation methods, offering a powerful solution for dynamic and accurate hydrocarbon production allocation in complex commingled systems.