Page 187 - Machine Learning for Subsurface Characterization

162 Machine learning for subsurface characterization

3 Methodology to generate the ranking (R) index
3.1 Description of the R-index

Rivas et al. [8] developed a ranking characteristic parameter to rank reservoirs/
zones according to their suitability for miscible oil displacement when
performing gas injection. Shaw and Bachu [10] and Zhang et al. [11]
successfully applied Rivas et al.’s [8] method to rank reservoirs for purposes
of EOR. Ranking characteristic parameters are obtained by comparing the
actual reservoir parameters with fictive optimum and worst reservoir
parameters using an exponentially varying function. The optimum and worst
reservoir parameters are obtained through numerical simulation. We modified
Rivas et al. [8] ranking method into the R-index, which is computed by
processing well logs acquired in the shale formation of interest. In doing so,
we demonstrate a procedure for index construction to understand and quantify
the complex process of enhanced oil recovery due to light hydrocarbon injection.

3.2 Calculation of the R-index

First the formation of interest is partitioned into flow units. Following that, we
need to decide the optimum and worst values for the parameters governing the
EOR efficiency of the light-hydrocarbon injection into those flow units. Each
property (j) of the flow unit (i) is normalized as follows:

P i, j P o, j

(6.1)
X i, j ¼
P w, j P o, j

where P i, j is the magnitude of property (j) in the flow unit (i) being ranked and
P o, j and P w, j are the optimum and worst values of the property (j), respectively,
such that P o, j and P w, j represent properties of two artificial flow units. One
artificial flow unit has the best properties for miscible oil displacement using
light-hydrocarbon injection, and the other has the worst properties. We use
the maximum and minimum values of a property in the entire formation
under investigation as the optimum and worst values to obtain the P o, j and
P w, j . The transformed variable X i, j ranges between 0 and 1. To better
emphasize the properties that are not conducive for the proposed EOR, X i, j
is transformed to A i, j using a heuristic exponential equation expressed as
4:6X 2
A i, j ¼ 100e i, j (6.2)
where A i, j ranges from 1 to 100, such that A i, j ¼ 100 represents the best value of
a specific property. Following that, A i, j is weighted by w j , which represents the
relative importance of each reservoir property in influencing the EOR
efficiency of the light-hydrocarbon injection. We use the values of weight w j
as published by Jin et al. [14]. Weighted grade of property (j) in the flow
unit (i) is expressed as

W i, j ¼ A i, j w j (6.3)