Page 205 - Machine Learning for Subsurface Characterization

Dimensionality reduction and clustering techniques Chapter 6 175

method, we can now calculate the portion of in situ oil, residing in a certain pore
size distribution, that can achieve miscibility with a specific injected fluid under
a certain reservoir pressure and temperature.
Fig. 6.6 shows that the MMP is around 4000 psi in unconfined pores when
the injected hydrocarbon is 100% C1. MMP decreases with an increase in the
longer-chain carbon content. In the shale formation under investigation, the
average reservoir pressure is above 4000 psi, which means miscibility is not
a problem for light-hydrocarbon injection in the shale formation. All free oil
in all the pore sizes of the shale formation under investigation can achieve
miscibility when the reservoir pressure is maintained above 4000 psi during
the miscible displacement.

4.2.4 Step 4: Compute the MD-index
MD-index is calculated using Eq. (6.5). This calculation requires estimates of
miscible free oil-filled porosity corrected for pore-confinement effect
(described in Step 3) and those of free water-filled and bound fluid-filled
porosities calculated using the factor analysis (described in Step 2). The
kerogen volume fraction needed in Eq. (6.5) is acquired from Quanti-ELAN
inversion provided by Schlumberger. All these data are normalized and
transformed using the heuristic Eq. (6.2). Such a transformation eliminated
situations where the denominator in Eq. (6.5) becomes small making the MD-
index unreasonably large.
The computed MD-index across the 200-ft depth interval of the upper,
middle, and lower shales is shown in the second last track of Fig. 6.2. Based
on the MD-index, the middle section has relatively better EOR potential
when using light-hydrocarbon injection. Furthermore, the upper and lower
shales of the formation exhibit low EOR potential for light-hydrocarbon
injection. Around 50% and 20% of the middle shale exhibit good and
intermediate EOR potential, respectively. Compared with R-index, the MD-
index shows a similar trend; however, the MD-index better differentiates
layers of low EOR potential in the middle shale.

5 Methodology to generate the K-means clustering (KC) index
5.1 Description of the KC-index

This index categorizes all the depths into one of the four groups, which
indicate low, low-intermediate, high-intermediate, or high EOR potential
for light-hydrocarbon injection. This index is computed using K-means
clustering technique popular in various machine learning tasks for
unsupervised learning. Input logs (features) from a single depth are
arranged as n coordinates representing a point in an n-dimensional space.
The K-means algorithm processes all input logs from all the depths and