Compositional modeling of gas-condensate viscosity using ensemble approach.

Area of Science:

• Petroleum Engineering
• Physical Chemistry
• Data Science

Background:

• Gas-condensate reservoirs experience liquid dropout near wellbores as pressure drops below the dew point.
• Accurate estimation of liquid viscosity is crucial for predicting production rates in these reservoirs.
• Existing models often rely on the solution gas oil ratio (Rs), which is challenging to measure accurately.

Purpose of the Study:

• To develop accurate predictive models for gas condensate viscosity.
• To eliminate the reliance on the solution gas oil ratio (Rs) as an input parameter.
• To utilize a comprehensive dataset and advanced machine learning techniques for improved viscosity prediction.

Main Methods:

• Utilized a comprehensive database of 1370 laboratory data points for gas condensate viscosity.
• Applied and compared several intelligent techniques: Ensemble methods, Support Vector Regression (SVR), K-Nearest Neighbors (KNN), Radial Basis Function (RBF), and Multilayer Perceptron (MLP).
• Developed models using temperature, pressure, and condensate composition as input parameters, excluding Rs.

Main Results:

• The Ensemble method demonstrated the highest accuracy with an Average Absolute Percent Relative Error (AAPRE) of 4.83%.
• Other models showed varying accuracies: SVR (4.95%), KNN (5.45%), MLP-BR (6.56%), MLP-LM (7.89%), and RBF (10.9%).
• Reservoir temperature negatively impacted viscosity, while the mole fraction of C11 positively influenced it, as determined by parameter relevancy analysis.

Conclusions:

• Accurate gas condensate viscosity models were developed using temperature, pressure, and composition.
• The Ensemble method provides a highly accurate and practical approach for viscosity prediction in gas-condensate reservoirs.
• Identified and reported suspicious laboratory data points using the leverage technique for data quality enhancement.