American Oil and Gas Reporter - June 2019 - 70

SpecialReport: Artificial Lift Technology

Method Guides Development Decisions
By Bruno de Ribet
AUSTIN, TX.-A variety of new tools
for interpreting the subsurface at different
resolutions has become available to geoscientists in recent years. At the same
time, geoscientists have access to a multitude of local and global measurements
of the subsurface-much more data than
one person can process and consume simultaneously.
While using statistical tools to help
extract information from vast amounts
of raw data is not new, technological advancements have allowed this methodology to quickly surpass anything imaginable
only a few years ago. Big data analytical
tools enable users to handle huge amounts
of data at unprecedented speed, and predictive analytics are being used as forecasting tools in many fields to ramp asset
performance and accelerate exploration,
field development and production decision
making cycles.
Several innovative predictive methods
are applicable to the challenges of oil
and gas exploration and production. These
include deep learning and other machine
learning-based methods for enriching
subsurface models. As demonstrated in
an application in the Permian Basin Wolfcamp play, machine learning has been
integral to developing a new approach to
resolving reservoir facies heterogeneities
in seismic data. In the Permian case study,
the primary challenge encountered by
the oil and gas operator was a thin and
laterally discontinuous reservoir (oil-filled
packstone).
Despite having collected a high-resolution, state-of-the-art 3-D seismic survey
with usable frequencies up to 138 Hertz,
and even after generating seismic attribute
volumes to aid with the interpretation,
the operator was not able to manually
generate an interpretation that matched
the rock-type interpretation at the wells.
Therefore, the decision was made to supplement human interpretation with the
machine learning methodology.
To evaluate the quality of a reservoir
and gain a more realistic measure of its
behavior, geoscientists try to achieve accurate facies distribution mapping. Predicting rock-type quality distribution enables geoscience professionals to better
understand depositional processes in order
to help them optimize the drilling decision
70 THE AMERICAN OIL & GAS REPORTER

making process.
A standard approach to understanding
reservoir quality is to perform seismic inversion to predict elastic properties. However, this solution may suffer from a
"nonuniqueness" problem and it may be
difficult to separate different facies, since
reservoir quality is not linearly correlated
with seismic data, which needs the introduction of uncertainty measurements. The
true integration of well and seismic data
always has been a challenge because of
their different responses and resolutions.
Machine Learning Method
To resolve these ambiguities, new machine learning methods change the applicability of seismic data from exploration
to becoming a valuable prospect development tool. While automatic unsupervised classification methods enable a
valid geologic interpretation embedded
in the seismic data for exploration or
infill development well positioning, this
new supervised approach delivers the
most probable facies and probability associated with each. The strength of this
method is based on the system's ability
to integrate different types of data (core,
wireline and seismic).
The developed technique assumes the
existence of a relationship between a
seismic response at a given point and the
rock-type distribution around that point.
However, no model has been established
yet, and the mathematical formulation of
such an operator is complex. Its determination would imply a long empirical
process to evaluate the consequence of
rock-type distribution on seismic response.
Estimating the number of parameters is
challenging and is a function of the geological context, measurement constraints
and experimental design. Therefore, the
workflow creates an operator using learning techniques.
As with any learning method, the democratic neural network association
(DNNA) technique utilized needs a representative dataset to build a robust operator. Unfortunately, rock type is not available as a volume and must be approximated by the lithofacies distribution defined within vertical windows along the
borehole.
The initial phase in this method is to
use facies logs constructed from well
data as the main source for describing

the quality of the reservoir in terms of
lithology, hydrocarbon saturation or rock
type. In this sequence, another machine
learning method, called multiresolution
graphic clustering, is used. This method
defines clusters of different resolutions,
helping to differentiate between homogeneous and laminated geologic contexts.
The goal is to generate a probabilistic
facies model from the seismic data. An
association of naïve neural networks,
each with a different learning strategy, is
run simultaneously to predict facies and
avoid biasing any of the neural network
architectures. To train the neural networks,
facies at the well and seismic data extracted
along the wellbore in the interval of
interest are used as input data (the hard
training dataset). A major benefit of this
technology is its ability to combine the
full dimensionality of the prestack data,
which carries more information, with any
type of seismic attributes.
A second phase introduces seismic
data away from the borehole (soft data).
The neural networks "vote" on their integration to enrich the initial training
dataset in order to update the model.
Adding the soft data avoids overlearning
from a limited dataset.
The last step is to propagate the final
neural network model on the full seismic
dataset to generate probabilistic facies
models composed of three different volumes: most probable facies, maximum
probability for all facies, and probability
for each. Analysis of the facies and associated probability distribution provides
valuable insights into prospect uncertainties
and seismic data reliability for prediction.
East Soldier Mount Area
The East Soldier Mount study area is
located about 125 miles northeast of
Midland, in the Eastern Shelf of the
Permian Basin. The packstones were
formed at the time of deposition during
Lower Permian/Wolfcamp/Unayzah time
(295 million years ago) when the area
was shallow marine, and organisms were
inhabiting the mounds and subtidal zones.
The study area contains thin and laterally
discontinuous oil-filled packstones in
both the Upper and Lower Wolfcamp
intervals.
Bioturbation and oolitic shoals caused

American Oil and Gas Reporter - June 2019

Table of Contents for the Digital Edition of American Oil and Gas Reporter - June 2019