Perpustakaan Digital ITB

Liquid loading in gas wells significantly hampers gas extraction operations, reducing production by up to 50% and causing well shutdowns. Traditional methods like Turner's and Li’s Equation rely on empirical observations and manual interventions, which are often time-consuming and inaccurate. Accurate prediction of liquid loading is crucial for enabling engineers to take preventive actions before severe impacts on production occur. Early identification of potential liquid loading allows engineers to implement strategies to mitigate its effects, maintaining consistent gas flow and optimizing well performance. This research aims to develop advanced predictive models for liquid loading using machine learning techniques combined with Decline Curve Analysis (DCA). By leveraging historical production data, these models will determine the status of liquid loading in gas wells, predict critical rates of liquid loading in gas wells, and forecast the timing of liquid loading events, thus enhancing gas well performance and enabling timely preventive actions. The developed models using a total of 391 data points from 309 wells, sourced from both literature and field data. Gradient boosting was selected as the best algorithm, achieving an accuracy of 92.9% for the classification model and an R-square of 0.869 with an MAE of 603.25 for the regression model. The final model was applied to well TMO-1, determining its status as "unloaded," with a critical rate of 9047.22 Mscfd. Liquid loading in TMO-1 Well is forecasted to occur in 70 days (exponential), 95 days (hyperbolic), and 108 days (harmonic).