Page 209 - Machine Learning for Subsurface Characterization

178 Machine learning for subsurface characterization

few depths in the middle shale that exhibits KC-index of 3 and the upper and
lower shales exhibit KC-index of 1 for the entire depth.
The KC-index, R-index, and MD-index are shown in Fig. 6.2 for purposes
of comparison. The KC-index shows good agreement with the MD-index and
R-index. Data points of upper and lower shales, where MD-index indicates
low miscible-recovery potential, exhibit KC-index of 1, which is the least
suitable for the light-hydrocarbon injection. KC-index in the middle
shale shows a trend consistent with MD-index, as shown in Fig. 6.2.
Layers in the middle shale with high values of MD-index and R-index
have KC-index of 3 and 4, whereas layers with lower MD-index value
have KC-index of 2. According to KC-index, around 40% and 50% of
middle shale will demonstrate high and intermediate oil-recovery
potentials when using light-hydrocarbon injection, respectively.

6 Limitations

There are few limitations of the proposed approach, especially those related to the
assumptions of fluid properties, reservoir properties, and derivation of mean pore
throat diameter in the absence of relevant laboratory data. Laboratory
experiments and field studies are needed to further validate the predictions and
input data related to the three indices. These indices are not based on actual
measurements on fluid mobility, fluid displacement, and fluid trapping. MD-
index is designed for formations where pore sizes are predominantly in the
nanoscale range. For formations where pore sizes are large, pore-confinement
effect will not produce a significant effect on the miscibility. MD-index is
based on the analysis of microscopic displacement efficiency. However,
parameters such as formation geometry, injector-producer orientations, API
gravity, and fracture orientations that significantly influence miscible injection
mechanism cannot be adequately addressed in the proposed method. The
indices do not include the effects of (1) natural and induced fracture
properties, (2) formation damage due to fracturing fluid, (3) gravity and
capillary effects, and (4) temporal alterations in the spatial distribution of
reservoir properties during production on the recovery potential. Few
limitations of the R-index are as follows: (1) It is difficult to design
experiments or develop theoretical formulations to determine the weights to be
assigned to the ranking parameters, and (2) the interdependence of the ranking
parameters is ignored. For KC-index, one limitation is that the cluster centers
do not represent a real sample; as a result, the properties corresponding to the
cluster center are unreliable in terms of their absolute value. Nonetheless, the
relative values of cluster centers are indicative of EOR potential along the
well length of the shale formation.

7 Conclusions

Wireline log-derived EOR-efficiency indices can be implemented for the
identification of flow units in shale formations that are suitable for EOR