Machine Learning for Predicting and Optimizing Well Rates in Gas-Lifted Wells with Web Application Integration

Abstract Accurate prediction and optimization of well production rates are critical for maximizing efficiency and enhancing hydrocarbon recovery in gas-lift operations. This study presents a robust machine learning (ML) framework for predicting oil, water, and gas production rates and optimizing oil rates using advanced predictive and optimization techniques. A dataset comprising 12,538 samples was analyzed, and six ML algorithms, including Kernel Ridge Regression (KRR), Random Forest, and Artificial Neural Network (ANN), were evaluated. Data pre-processing involved scaling for KRR and ANN, while hyperparameter tuning was performed using GridSearchCV. Artificial Neural Network emerged as the best-performing model, achieving R2 scores of 0.9959, 0.9972, and 0.9977 for oil, water, and gas rates, respectively, with corresponding RMSE values of 74.56 STB/D, 39.26 STB/D, and 0.03 MMSCF/D. Optimization of oil production rates was conducted using a Genetic Algorithm, identifying optimal operating parameters, such as a gas injection rate of 2.756 MMSCF/D and a choke inner diameter of 2.224 inches, resulting in an optimized oil rate of 7,404.02 STB/D from an initial maximum rate of 6,051.93 STB/D. Feature importance analysis highlighted parameters such as Pressure Drop across the Choke, Inner Diameter of Choke, Gas Injection Rate and Wellhead Temperature as critical drivers of oil production in gas-lift operations. To bridge the gap between research and application, an interactive web application was developed, enabling real-time prediction and optimization of oil rates. This innovative approach demonstrates the potential of machine learning and optimization to revolutionize gas-lift operations, delivering significant improvements in production efficiency and operational decision-making in the oil and gas industry.