Modeling Natural Fracture Networks Using Improved Geostatistical and Geomechanical Inferences
Open Access
- Author:
- Chandna, Akshat
- Graduate Program:
- Energy and Mineral Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- October 09, 2020
- Committee Members:
- Sanjay Srinivasan, Dissertation Advisor/Co-Advisor
Sanjay Srinivasan, Committee Chair/Co-Chair
Zuleima T Karpyn, Committee Member
Hamid Emami-Meybodi, Committee Member
Chris J Marone, Outside Member
Mort D Webster, Committee Member
Mort D Webster, Program Head/Chair - Keywords:
- Fracture Characterization
Multiple Point Statistics
Geostatistics
Geomechanics
Machine Learning
Stochastic Simulation
Modeling Natural Fracture Networks
Teapot Dome - Abstract:
- Natural fractures control the flow of subsurface fluids and the transport of chemical species. Estimation of a reservoir’s production potential depends largely on accurate physical and mathematical modeling of the existing fracture networks. However, significant uncertainties associated with the prediction of the spatial locations and connectivity of fracture networks is likely due to lack of sufficient data to model them. Therefore, stochastic characterization of these fractured reservoirs becomes necessary. Stochastic methods rely on extracting statistics from outcrop images or subsurface reservoirs. Many commonly used stochastic methods, use the two-point covariance to describe the statistical relationship between different data points and that fails to account for higher order relationships that may describe complex fracture features. Representing the termination and intersection of fractures accurately while accounting for variations in mechanical properties of rocks can be best achieved by constraining the fracture models to multiple point statistics (MPS). But since, MPS algorithms are purely based on statistical inferences from outcrop images, the models produced, fail to account for the physical processes that dictate the propagation and termination of fractures. These processes are better represented in geomechanical models of the fracturing process. However, full-physics numerical models are computationally inefficient for modeling fractures at a reservoir scale while accounting for material heterogeneities. More importantly, geomechanical simulations yield deterministic results, thus failing to represent the inherent uncertainties in input properties and paleo stress conditions. This motivates a hybrid modeling process incorporating both MPS based simulation as well as fracture network characteristics derived from geomechanical simulations. This research entails development of a new, improved, efficient and robust non- gridded multiple point statistics (MPS) algorithm that outputs propagation angles for each simulation fracture node constrained to pattern information inferred from analogs. The information inferred from analog fracture models is subsequently augmented with information from a classification algorithm constrained to geomechanical considerations. This helps in representation of fracture terminations and intersections for multiple fracture sets based on the statistical inferences drawn from the training image and geomechanical simulations, without having to specify arbitrary termination or branching rules. This algorithm is characterized by non-gridded training images, spatially flexible templates used within a fracture propagation framework. The templates are self- adjusting according to the fracture node being simulated and thus need not be fixed and defined a-priori. In order to facilitate geomechanical characterization, a number of small-scale high fidelity finite discrete element method (FDEM) based forward models are executed and the relationship between prevailing stress conditions and the fracture propagation direction is inferred using Machine Learning (ML) approaches. The inferred information from these sources are then combined to predict the angle of orientation of the propagating fractures. The result is a fracture network modeling approach that is computationally efficient. The method extends the capabilities of statistical fracture modeling approaches by accounting for the physical process of fracture propagation. Application of this combined MPS and machine learning based geomechanical algorithm to a synthetic case as well as on a case study using an outcrop dataset from the Teapot Dome in Wyoming, demonstrates the ability of the algorithm to effectively and efficiently capture the general orientation and interconnectivity of fractures from the training image.