Shale Formation Characterization Based on Acyclic Pore Model and Permeability Prediction Using Machine Learning

In shales, a significant fraction of the pore sizes is within the 100-nm range and the transport properties under nano-size confinements deviate from those measured under unconfined conditions at identical pressures and temperatures. This dissertation aims to characterize different effective transport properties of shales at the core scale. The tree-like model is used to capture the effective pore connectivity at the core scale because it mimics various petrophysical measurements. We use the mercury injection capillary pressure measurements as a means of characterizing the pore connectivity. We first propose a simple relation based on the tree-like model to estimate matrix permeability. The estimated permeability is presented in a tabular format for ease of use. We modify the flow behavior by implementing models from nanofluidics for a single conduit. We then determine various petrophysical properties at the core scale, including the effective slippage factor, the critical properties (critical temperature, critical pressure), and the compressibility (Z) factor. Unlike previous research, we do not use a single size averaged from the pore size distribution to examine the deviation due to pore proximity. Instead, we implement the effective size to account for the connectivity of the pores at the core scale. The slippage factor at the core scale is analyzed for the wetting and the nonwetting phases in terms of governing parameters such as pore pressure and wettability. We implement various models from nanofluidics and examine their outcomes at the core scale. The effective critical properties at the core scale are characterized using the pore-throat and pore-body size distributions. The characterized properties are relevant to the fluid flow (displacement) and to the storage. The deviations in the critical properties also modify the compressibility (Z) factor. The deviation of the compressibility factor at the core scale is presented in a ratio form. This dissertation also discusses the use of machine learning to capture the matrix permeability in carbonates. A machine learning model is trained to predict the permeability. The proposed model is an improvement over the conventional models and further advances the implementation of physics in modeling flow behavior in porous media.