Extending Dual Porosity Models to Anisotropic Fractured Reservoirs Using Physics-Guided Neural Networks

The non-orthogonal fractures in conventional Warren and Root (WR) induce directional permeabilities and anisotropic flow patterns that cannot be captured using simple fracture properties. In this paper, we use the proposed equivalent fracture aperture, obtained via a correction coefficient (η), to account for this effect. Previous studies estimated this coefficient by calibrating to static field data. The main contribution is to treat this coefficient as a physics-dependent parameter rather than a calibration value to bridge fracture-scale flow with the continuum model. A data-driven approach is developed to quantify η as a function of fracture geometry and reservoir-scale properties. Moreover, a dataset of ۲,۴۷۸ simulation samples was generated using COMSOL Multiphysics, covering a range of fracture and reservoir properties. In addition, an artificial neural network (ANN) was trained and optimized on this dataset to learn the nonlinear relation with the correction coefficient, achieving a predictive accuracy of R² = ۰.۹۹۴۶. By using the ANN-predicted correction coefficient in the WR approach, a multiscale bridge between fracture-scale physics and dual-porosity models can be achieved. Furthermore, the cubic relationship between permeability and fracture aperture (kθ = η۳ k) highlights the importance of accurate η estimation, as any error in η propagates results into large permeability uncertainties. Ultimately, the numerical results showed that the dependency of the correction coefficient η on fracture orientation is much greater than its dependency on either fracture size and matrix shape or system size. Also, using this simple correction factor, the applicability of conventional dual-porosity models to anisotropic fractured reservoirs with non-orthogonal fractures is extended.

کلیدواژه ها:

Naturally Fractured Reservoirs ، COMSOL ، Anisotropy ، Artificial Neural Network (ANN) ، Fracture Aperture Correction Coefficient