Characterization of petrophysical properties of organic-rich shales by experiments, lab measurements and machine learning analysis

Abstract

The increasing significance of shale plays leads to the need for deeper understanding of shale behavior. Laboratory characterization of petrophysical properties is an important part of shale resource evaluation. The characterization, however, remains challenging due to the complicated nature of shale. This work aims at better characterization of shale using experiments, lab measurements, and machine leaning analysis. During hydraulic fracturing, besides tensile failure, the adjacent shale matrix is subjected to massive shear deformation. The interaction of shale pore system and shear deformation, and impacts on production remains unknown. This work investigates the response of shale nanoscale pore system to shear deformation using gas sorption and scanning electron microscope (SEM) imaging. Shale samples are deformed by confined compressive strength tests. After failure, fractures in nanoscale are observed to follow coarser grain boundaries and laminae of OM and matrix materials. Most samples display increases in pore structural parameters. Results suggest that the hydrocarbon mobility may be enhanced by the interaction of the OM laminae and the shear fracturing. Past studied show that the evolution of pore structure of shale is associated with thermal maturation. However, the evolution of shale transport propreties related to thermal maturation is unclear due to the difficulty of conducting permeability measurement for shale.This work studies evolution of permeability and pore structure measurements using heat treatment. Samples are heated from 110°C to 650°C. Gas sorption and GRI (Gas Research Institute) permeability measurements are performed. Results show that those petrophysical parameters, especially permeability, are sensitive to drying temperature. Multiscale pore network features of shale are also revealed in this study. Characterizing fluids in shale using nuclear magnetic resonance (NMR) T₁-T₂ maps is often done manually, which is difficult and subjected to human decisions. This work proposes a new approach based on Gaussian mixture model (GMM) clustering analysis. Six clustering algorithms are performed on T₁1-T₂ maps. To select the optimal cluster number and best algorithm, two cluster validity indices are proposed. Results validate the two indices, and GMM is found to be the best algorithm. A general fluid partition pattern is obtained by GMM, which is less sensitive to rock lithology. In addition, the clustering performance can be enhanced by drying the sample