Extended Multiple Interacting Continua (E-MINC) Model Improvement with a K-Means Clustering Algorithm Based on an Equi-dimensional Discrete Fracture Matrix (ED-DFM) Model

DOĞAN M. O.

MATHEMATICAL GEOSCIENCES, 2023 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Basım Tarihi: 2023
  • Doi Numarası: 10.1007/s11004-023-10110-9
  • Dergi Adı: MATHEMATICAL GEOSCIENCES
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Aerospace Database, Agricultural & Environmental Science Database, Aquatic Science & Fisheries Abstracts (ASFA), Communication Abstracts, Compendex, Geobase, Metadex, zbMATH, Civil Engineering Abstracts
  • Orta Doğu Teknik Üniversitesi Adresli: Evet

Özet

Modeling fluid flow in fractured reservoirs can be complicated, not only because of the permeability differences between fractures and matrix, but also due to the complex network of fractures. If the exact fracture geometry is known, accurate flow models can be obtained by discrete fracture matrix (DFM) models. However, implementing the DFM model for densely fractured reservoirs is practically infeasible because of the computational cost and time limitations. To overcome this problem, dual/multi-continuum models such as dual-porosity (DP), multiple interacting continua (MINC), and extended MINC (E-MINC) have been developed. In these upscaling methods, fracture-to-matrix exchange flows are calculated using shape factors, volume fractions, and transmissibility exchange parameters. The main bottleneck of dual/multi-continuum models is in estimating parameters for exchange flow in complex fracture networks. In this study, the E-MINC model parameters are improved by using a K-means clustering algorithm in Python, taking the equi-dimensional DFM (ED-DFM) model as a reference solution. The E-MINC module is developed under MATLAB LiveLink for COMSOL. Pressure levels for the determination of volume fractions and transmissibility of interacting continua are optimized using a Python scikit-learn K-means clustering algorithm. It is shown that for bar-type fractures, both the E-MINC K-means model and DP model with the Vermeulen equation provide better results than DP models having constant shape factors. For a naturally fractured domain, the E-MINC model with K-means clustering provides better results than the E-MINC model with uniformly distributed iso-pressure levels. The transient results of E-MINC K-means and ED-DFM models on both small- and large-scale domains are very similar.