Hybrid Data/Mechanics Based Stochastic Subsurface Characterization

  • Mohammad Burhan Abdulla

Student thesis: Doctoral Thesis

Abstract

Subsurface characterization underpins many decisions made at every stage of subsurface resources utilization, such as geothermal systems and hydrocarbon exploitation. Nonetheless, such a task is a challenge and affected by inevitable sources of uncertainties, making decisions related to the subsurface highly complicated. Therefore, for the purpose of making well-informed decisions, stochastic subsurface characterization is ideally sought. In this work, we develop a set of novel hybrid stochastic subsurface characterization tools based on field data and mechanics. The work consists of three major components: (1) a probabilistic geologic layer identification model: GLI, (2) a stochastic discrete fracture network (DFN) model: EllipFrac and (3) a stochastic fracture initiation and propagation model – FracProp. Two tools were developed for GLI: (a) Artificial neural networks (ANN) based and (b) optimized multivariate Gaussian distributions based. The former method consists of a novel approach to convert the ANN decisions from deterministic to probabilistic ones, focusing on quantifying interpolation uncertainties. The latter tool targets uncertainties related to the interpolation and to the locations of the interfaces between geologic layers. Both tools are demonstrated through an application to the characterization of the subsurface of Masdar city, UAE. They are used to identify geologic layers and to create hazard maps. The remaining set of tools focuses on rock fracture characterization. The first one is EllipFrac, a stochastic DFN developed to represent natural fracture networks. EllipFrac is based on site-specific data and uses a series of stochastic processes to represent orientations, sizes, and locations of fractures and represent natural rock fracture systems. EllipFrac is demonstrated through an application to the Yates oil field, Texas, USA. Finally, FracProp, a mechanics based stochastic model to predict the initiation and propagation of induced fractures using compounded directional probability density functions. FracProp is qualitatively validated on experiments conducted at the rock mechanics lab at MIT and the simulated fractures are in agreement with the test results.
Date of AwardJul 2018
Original languageAmerican English

Keywords

  • subsurface; stochastic
  • rock fractures networks; induced fractures; uncertainty quantification.

Cite this

'