Permeability modelling in a highly heterogeneous tight carbonate reservoir using comparative evaluating...

Area of Science:

• Petroleum Geoscience
• Reservoir Engineering
• Machine Learning Applications

Background:

• Permeability modelling is crucial for reservoir simulation but challenging in heterogeneous carbonate formations.
• Traditional static rock typing and fitting-based models often fail due to porosity-permeability data scattering in carbonates.
• Petrophysical well log data offer continuous and comprehensive information for improved permeability prediction.

Purpose of the Study:

• To develop and compare machine learning (ML) and fitting models for predicting permeability in un-cored intervals of a tight carbonate reservoir.
• To assess the efficacy of ML models against traditional exponential and statistical fitting approaches using well log data.
• To enhance the accuracy of permeability distribution for improved 3D reservoir modeling and simulation.

Main Methods:

• Utilized porosity, permeability, and petrophysical well log data from two oil wells in a tight carbonate reservoir.
• Developed predictive permeability models using machine learning algorithms (Random Forest) and fitting techniques.
• Compared the performance of ML models against exponential and statistical fitting models using Root Mean Squared Error (RMSE).

Main Results:

• The integrated ML permeability model, specifically Random Forest, demonstrated superior performance over fitting-based methods.
• Achieved low RMSE values for horizontal and vertical permeability predictions: 3.7 and 4.5 (well 2), and 1.7 and 0.86 (well 4).
• Demonstrated that incorporating well log data via ML improves permeability modeling by integrating reservoir rock physics.

Conclusions:

• Machine learning, particularly Random Forest, offers a robust solution for accurate permeability prediction in complex, heterogeneous carbonate reservoirs.
• The developed ML models provide a more reliable method for permeability estimation compared to conventional techniques.
• This research enhances reservoir characterization, leading to improved 3D permeability models for more accurate dynamic reservoir simulations.