Poromechanics of Fractured/Faulted Reservoirs During Fluid Injection Based on Continuum Damage Modeling and Machine Learning

Fethi Abbassi, Ali Karrech, Md Saiful Islam, Abdennour C. Seibi

Research output: Contribution to journalArticlepeer-review

Abstract

The reactivation of faults is governed by the variation of effective stresses in the fault's plane. These effective stresses are dependent on the total stresses (solely related to the regional geological stresses and lithology) and pore pressure (strongly affected by rock properties, fluid content, and saturation conditions). Injecting fluids into a reservoir formation may change the distribution of pore pressures, which influences the effective stresses and may cause the reactivation of existing faults, which has a wide range of consequences. This study investigated the reactivation of preexisting faults due to fluid injection into hydrocarbon reservoirs at different pressures and temperatures. A 3D model containing a continuous normal fault that divides the domain into two compartments was used. A user-defined constitutive model based on continuum damage mechanics implemented as a Fortran subroutine to predict the behavior of fractured and faulted reservoirs was used. A parametric analysis was performed to examine the influence of geometric parameters, such as the fault dip angle, reservoir characteristics, and fluid injection parameters. A machine learning approach based on artificial neural networks (ANNs) is incorporated to predict the enhanced oil recovery using fluid injection. The results predicted by the ANN were further confirmed by numerical modeling.

Original languageEnglish
Pages (from-to)413-430
Number of pages18
JournalNatural Resources Research
Volume32
Issue number1
Early online date30 Oct 2022
DOIs
Publication statusPublished - Feb 2023

Fingerprint

Dive into the research topics of 'Poromechanics of Fractured/Faulted Reservoirs During Fluid Injection Based on Continuum Damage Modeling and Machine Learning'. Together they form a unique fingerprint.

Cite this