Modeling Permeability Using Advanced White-Box Machine Learning Technique: Application to a Heterogeneous Carbonate...

Area of Science:

• Petroleum Geoscience
• Reservoir Engineering
• Machine Learning Applications in Earth Sciences

Background:

• Reservoir rock permeability is critical for hydrocarbon field development, yet accurate prediction is challenging in heterogeneous formations.
• Conventional petrophysical rock typing methods often fail to provide accurate permeability correlations for complex reservoirs.
• Costly reservoir rock samples necessitate reliable predictive models for permeability estimation.

Purpose of the Study:

• To develop a universal, accurate permeability prediction model for a heterogeneous carbonate reservoir.
• To compare the performance of novel machine learning algorithms against conventional and existing data-driven methods.
• To identify the key parameters influencing permeability prediction in the studied reservoir.

Main Methods:

• Classified the reservoir into two petrophysical zones using K-nearest neighbors based on porosity, pore throat radius (r35), and connate water saturation (Swc).
• Developed a universal permeability prediction model using modified group modeling data handling (GMDH) and genetic programming (GP) with porosity, r35, and Swc as inputs.
• Conducted parameter importance analysis on the developed machine learning models.

Main Results:

• Conventional zone-specific permeability prediction showed limitations in accuracy for the heterogeneous reservoir.
• The universal models developed using GP and GMDH achieved high prediction accuracy (R² of 0.95 and 0.99, respectively).
• The radius of pore throats at mercury saturation of 35% (r35) was identified as the most impactful feature for permeability prediction.

Conclusions:

• Novel machine learning algorithms (GMDH and GP) provide superior and universal permeability prediction in heterogeneous carbonate reservoirs compared to traditional methods.
• The developed models offer a reliable and accurate alternative to costly core analysis for permeability estimation.
• The study highlights the importance of pore throat radius as a primary driver of permeability in such formations.