Improved Model for Predicting Water Production in a Typical Dry Gas Well

Water production is a major challenge in many gas wells leading to a substantial reduction in the net gas deliverability, high operational costs associated with large-scale water separation facilities, increased energy consumption for lifting fluids, and significant logistical and regulatory burden of produced water disposal. This study focused on the development of a more accurate model for predicting and diagnosing water production in dry gas wells to enhance reservoir management and optimize production strategies. The model was developed by investigating the effectiveness of various machine learning models in predicting water production rates from dry gas wells using data sourced from Niger Delta. Five algorithms were trained and validated on a dataset comprising 249 daily records across eight dry gas wells in three reservoirs. The results showed CatBoost as an effective tool for petroleum reservoir forecasting demonstrating its superiority over widely used algorithms such as Random Forest, XGBoost, KNN and Support Vector Regression in predicting water production. Furthermore, the model predictions revealed that permeability had the strongest positive correlation with water production, whereas reservoir temperature and dew point pressure exhibited significant negative correlations. Overall, the findings underscore the superiority of ensemble-based models in capturing the complex, nonlinear relationships inherent in the Niger Delta field data. The developed predictive model is crucial for engineers to size equipment properly and avoid both under-capacity (leading to facility bottlenecks) and over-design (leading to wasted capital). Further research will focus on expanding the dataset to include a wider range of operational and reservoir parameters to enhance the model robustness and applicability.