Developing a cost-effective tool for choke flow rate prediction in sub-critical oil wells using wellhead data

Area of Science:

• Petroleum Engineering
• Machine Learning Applications
• Data Science

Background:

• Accurate prediction of oil production rates is vital for optimizing crude oil output and operational efficiency in the petroleum industry.
• Wellhead chokes significantly influence flow performance, making their accurate modeling crucial for production management.

Purpose of the Study:

• To develop and optimize machine learning (ML) models for predicting oil production rates through wellhead chokes.
• To identify key parameters influencing choke flow performance and provide an interpretable prediction framework.

Main Methods:

• Utilized a comprehensive dataset from a Middle Eastern petroleum production facility, including parameters like GOR, choke size, BS&W, THP, and API.
• Applied robust data preprocessing with Monte Carlo Outlier Detection (MCOD) and trained Gradient Boosting Machine (GBM) models using 198 data points with 5-fold cross-validation.
• Optimized GBM models using advanced algorithms including Self-Adaptive Differential Evolution (SADE), Evolution Strategy (ES), Bayesian Probability Improvement (BPI), and Batch Bayesian Optimization (BBO).

Main Results:

• The Self-Adaptive Differential Evolution (SADE) algorithm demonstrated superior performance in optimizing Gradient Boosting Machine (GBM) models.
• Key performance metrics such as Average Absolute Relative Error (AARE%), R-squared (R²), and Mean Squared Error (MSE) were used for model evaluation.
• SHAP (SHapley Additive exPlanations) analysis revealed choke size and wellhead pressure (THP) as the most influential parameters in predicting oil production rates.

Conclusions:

• Developed a data-driven framework for accurate and interpretable prediction of oil production rates via wellhead chokes.
• The study highlights the significant impact of choke size and THP, providing valuable insights for production optimization.
• The proposed ML approach offers a robust solution for enhancing operational efficiency in the oil and gas sector.