Date of Award
12-2024
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Environmental Engineering and Earth Science
Committee Chair/Advisor
Lawrence C. Murdoch
Committee Member
Scott DeWolf
Committee Member
Leonid N. Germanovich
Committee Member
Ronald W. Falta
Abstract
Understanding reservoir properties and structure is essential to managing subsurface operations, such as geologic storage of CO2 or production of water, hydrocarbons or heat. Previous work demonstrated the potential of analyzing strain data during transient well tests by estimating hydraulic diffusivity using a type-curve-like approach. The following dissertation aims to pursue this potential by developing and evaluating methods for interpreting strain tensor data to improve subsurface characterization. The research advances the classic, intuitive method of using type-curves to interpret well tests, but it also brings together modern computational methods from machine learning, proxy modeling and Bayesian inversion, and it utilizes field data from transient well tests from a 530-m-deep sandstone reservoir at the North Avant Field, Osage County, Oklahoma.
The dissertation is presented in three chapters, which are currently either published or in review. Chapter II describes a type-curve-like method that uses classic Agarwal analysis to interpret strain measurements during recovery tests. Machine learning techniques were used to relate the semi-log slopes to the reservoir parameters. This approach enables the interpretation of strain data during recovery to improve reservoir characterization.
Chapter III evaluates methods for characterizing lateral boundaries in reservoirs. Lateral boundaries control the capacity of a reservoir during recovery and storage projects. A type-curve-like method was developed to interpret the shallow strain tensor at a single location to characterize the distance, orientation and location of the boundary. The boundary distance from an active well and its orientation were obtained through a plot-features analysis, while the horizontal strain tensor rotation constrains the boundary location.
Type-curve analyses are limited to idealized scenarios and unable to characterize reservoir complexities, so a computational workflow was developed in Chapter IV to handle the general situations. This method uses machine learning to reduce the computational requirements and make it practical to use Bayesian methods to invert strain data measured during well testing. The inversion results will improve how strain tensor data can be used to guide decision-making during the recovery of resources or the storage of wastes in the subsurface.
Recommended Citation
Roudini, Soheil, "Interpreting Strain Caused by Transient Well Testing" (2024). All Dissertations. 3849.
https://open.clemson.edu/all_dissertations/3849
Author ORCID Identifier
0000-0002-2256-1808
Included in
Environmental Engineering Commons, Geological Engineering Commons, Geotechnical Engineering Commons, Oil, Gas, and Energy Commons, Water Resource Management Commons
