Page 184 - Machine Learning for Subsurface Characterization

Dimensionality reduction and clustering techniques Chapter 6 159

focus in this study is to describe EOR efficiency in terms of above-mentioned
processes 2 and 3.
Physical properties and engineering practices governing the EOR process can
be assigned specific weights for purposes of ranking/screening reservoirs and
hydrocarbon-bearing zones in terms of the efficacy of these reservoirs/zones
for EOR using gas injection [4–7]. Rivas et al. [8] developed a ranking
characteristic parameter to overcome the “binary characteristics” of
conventional reservoir screening method and applied the method to rank
reservoirs for EOR using gas injection, which was subsequently used and
modified by Diaz et al. [9], Shaw and Bachu [10], and Zhang et al. [11].Oil
saturation index (OSI) has also been used to describe the potential
producibility of shale formations. OSI is a simple geological normalization of
oil content to TOC because kerogen has strong affinity to oil. For example,
80 mg of oil can be retained by 1 g of kerogen, which reduces the
producibility of the formation [12]. OSI requires laboratory measurements to
get the oil and TOC weight fractions. To overcome this requirement, Kausik
et al. [13] introduced carbon saturation index (CSI) and reservoir producibility
index (RPI) based entirely on the downhole logging tool measurements. CSI is
the weight ratio of carbon in light oil to TOC. Unlike OSI, CSI only considers
light oil. RPI is formulated by multiplying the light oil content and CSI.
Compared to CSI, RPI accounts for organic richness that differentiates the
reservoir qualities of organic-rich and organic-lean intervals.

1.3 Objectives
Hydrocarbon recovery potential of various flow units in the shale formation can
be quantified from well logs to facilitate efficient reservoir development and
management plans. We develop three log-based EOR-efficiency indices,
namely, the ranking (R) index, microscopic displacement (MD) index, and
K-means clustering (KC) index, to identify flow units suitable for light-
hydrocarbon injection along the length of a well in shale formation. The
R-index is a modification of Rivas et al.’s [8] reservoir ranking method and
implements Jin et al.’s [14] findings from the laboratory investigation of
miscible gas injection. On the other hand, MD-index is the ratio of positive
to negative factors affecting miscible gas injection. MD-index involves a
dimensionality reduction technique called factor analysis and a novel method
to calculate the volume of miscible free oil in the presence of pore-
confinement effect common in nanoporous shales. Finally the KC-index is
obtained by K-means clustering of the available well logs.
Index is used to track variations in a phenomenon or process that cannot be
captured in other ways. An index aggregates multiple features (physical
properties, parameters, or attributes) to generate composite statistics that
quantify the effects of changes in individual or group of features on the process
or phenomenon of interest. Indices facilitate the summarization and ranking of
the observations. Features implemented in an index can be differentially