Speaker
Description
An oil flow-rate prediction methodology is presented that leverages field production data and data-driven modeling as its basis for calculations. The approach utilizes Hassi Messaoud’s extensive well-test database to refine existing empirical correlations and machine-learning models under real operating conditions. Results are evaluated against conventional multiphase flow models, which are often limited by operat-ing-range constraints, demonstrating improved accuracy in predicting oil flow rates downstream of chokes when direct measurements are unavailable.
This study adopts a data-driven refinement approach combining empirical correlations and machine-learning techniques to improve oil flow-rate estimation in the Hassi Messaoud (HMD) field. An extensive well-test database containing over 61,000 meas-urements is first analyzed and classified by flow regime and gas–oil ratio (GOR). Two widely used Gilbert correlation variants are then adapted separately for each flow re-gime and refined across different GOR intervals. In parallel, multiple machine-learning models are developed and trained using the same dataset. Model performance is evaluated against real field data to assess predictive accuracy and robustness under field operating conditions.
The adapted empirical correlations demonstrate improved performance in estimating oil flow rates across both flow regimes in the HMD field. Refinement based on field-specific GOR intervals leads to accuracy improvements exceeding 50% in certain GOR ranges where standard correlations typically fail. Among the machine-learning models tested, the Artificial Neural Network (ANN) achieves the best performance, with correlation coefficients of approximately 0.80 for both flow regimes when vali-dated against real field data.
The results highlight that conventional multiphase flow models and generic correla-tions are often limited by their operating envelopes and do not fully capture the com-plex multiphase behavior observed in mature fields such as HMD. By leveraging a large, high-quality field database, existing methods can be effectively adapted to local reservoir and production characteristics. The study concludes that field-specific cali-bration significantly enhances oil flow-rate prediction reliability downstream of chokes, particularly in situations where direct measurements are unavailable, thereby supporting more accurate production monitoring and decision-making.
The novelty of this work lies in the large-scale, field-specific refinement of established empirical correlations and machine-learning models using more than 61,000 real well-test data points. Rather than proposing new equations, the study demonstrates how existing methods can be systematically adapted to real-field conditions, providing practicing engineers with a practical, low-cost approach to improve virtual flow-rate estimation and reduce reliance on frequent field measurements.
| Country | Norway |
|---|---|
| Acceptance of the Terms & Conditions | Click here to agree |
