Data driven methodology for model selection in flow pattern prediction.

Area of Science:

• Fluid Dynamics
• Chemical Engineering
• Petroleum Engineering

Background:

• Accurate multiphase flow parameter determination is crucial for industries like oil and gas.
• Existing methods include data-based algorithms and mechanistic models, each with limitations.
• Data-based methods lack explanatory power for two-phase flow characteristics.

Purpose of the Study:

• To propose a novel data-driven methodology for selecting closure relationships within unified mechanistic models for two-phase flow.
• To enhance the prediction accuracy of multiphase flow parameters.

Main Methods:

• Developed a decision tree-based model using a large dataset (27,670 points).
• Applied a data-driven methodology for closure relationship selection.
• Validated the model for flow pattern prediction using an independent dataset (9,224 observations).

Main Results:

• The closure relationship selection model demonstrated high accuracy in classifying flow regimes.
• Intermittent flow achieved the highest accuracy (86.32%), while annular flow had the lowest (49.11%).
• Global accuracy loss was less than 10% compared to direct data-based algorithms.

Conclusions:

• The proposed data-driven methodology effectively selects closure relationships for unified mechanistic models.
• This approach offers a balance between prediction accuracy and the explanatory power of mechanistic models.
• The method shows promise for improving multiphase flow analysis in industrial applications.