Reservoir Characterization - F&a
Reservoir Characterization - F&a
Reservoir Characterization - F&a
Scrivener Publishing
100 Cummings Center, Suite 541J
Beverly, MA 01915-6106
Scope: This series’s mission is to publish peer-reviewed, original research seeking sustainable methods
of worldwide energy production, distribution, and utilization through engineering, scientific, and
technological advances in the areas of both fossil fuels and renewable energy. Energy issues cannot
be addressed in isolation, without attention to the economy and the environment. Thus, this series
introduces the “E cubed” concept, addressing sustainability with a three-pronged emphasis on energy,
economy and environment, publishing research in all of these areas, and their intersections, as they
apply to global energy sustainability. This unique multi-disciplinary theme allows the introduction
of new and cutting-edge processes and technologies across all areas of energy production, transportation,
and transmission, including fossil fuels and renewable energy and their intersection with the economy
and environment. Papers are invited on any individual topic or those which are interdisciplinary.
Publishers at Scrivener
Martin Scrivener ([email protected])
Phillip Carmical ([email protected])
Reservoir Characterization
Fred Aminzadeh
This edition first published 2022 by John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, USA
and Scrivener Publishing LLC, 100 Cummings Center, Suite 541J, Beverly, MA 01915, USA
© 2022 Scrivener Publishing LLC
For more information about Scrivener publications please visit www.scrivenerpublishing.com.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or
transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or other-
wise, except as permitted by law. Advice on how to obtain permission to reuse material from this title
is available at http://www.wiley.com/go/permissions.
For details of our global editorial offices, customer services, and more information about Wiley prod-
ucts visit us at www.wiley.com.
ISBN 9781119556213
Set in size of 11pt and Minion Pro by Manila Typesetting Company, Makati, Philippines
10 9 8 7 6 5 4 3 2 1
Contents
Foreword xix
Preface xxiii
Part 1: Introduction 1
1 Reservoir Characterization: Fundamental
and Applications - An Overview 3
Fred Aminzadeh
1.1 Introduction to Reservoir Characterization? 3
1.2 Data Requirements for Reservoir Characterization 5
1.3 SURE Challenge 7
1.4 Reservoir Characterization in the Exploration,
Development and Production Phases 10
1.4.1 Exploration Stage/Development Stage 10
1.4.2 Primary Production Stage 11
1.4.3 Secondary/Tertiary Production Stage 11
1.5 Dynamic Reservoir Characterization (DRC) 12
1.5.1 4D Seismic for DRC 13
1.5.2 Microseismic Data for DRC 14
1.6 More on Reservoir Characterization and Reservoir
Modeling for Reservoir Simulation 15
1.6.1 Rock Physics 16
1.6.2 Reservoir Modeling 17
1.7 Conclusion 20
References 20
v
vi Contents
What is reservoir characterization? As you will see from this book, this is
a very advanced topic so let’s break it down a bit and start form the basics.
What is a reservoir? This is ‘a place where something is kept in store’. And
what is characterization? That is ‘to describe the character or quality’ all
according to the Webster dictionary. So, we are arrived at: ‘describe the
character of something that’s kept in store’. It seems relatively benign and
easy but ‘the devil is in the details’ is perhaps the best way to get the readers
intrigued and immersed in this topic. So, we are left wondering what are
these details where the devil resides? And here starts the story…..
In fact, a better wording would be ‘Subsurface Reservoir Characterization’
or SRC. There have been on the order of thousands of studies in reser-
voir characterization over the life time of this field. As such, this topic has
evolved and matured with many learnings. As illustrated in this book,
there are now well established and tested workflows SRC and I’d like to go
over some aspects of these understandings and workflows.
First, it is key to understand that SRC is a continuously changing,
multi-discipline and multi-scale topic. For continuously changing a good
example would be the recent impact of say machine learning methods. I
have learned that if our data quality is good enough and there are physical
relationships between reservoir data and properties, machine learning can
be an excellent way to quickly uncover relationships in a multi-variable
universe. However, once again, even here, the devil is in the details….
Multi-discipline is a word we easily use but have difficulty implementing.
In many projects the geologist is tasked with building a static reservoir
model and then passing it on to the reservoir engineer to build a dynamic
model and history match production. However, it has been challenging
to form a loop versus a linear workflow or for the dynamic model to be
updated with new static information or cover a range of possible models
that fit the data…… As for multi-scale, the discipline involves integration
of data from a wide range of data, say, nanometer (electron microscope),
to centimeter (cutting and core samples), to decimeter (well log), to meter
xix
xx Foreword
(seismic) scale. Spatially most of these data are acquired within a small
portion of one or several wells and geophysical data gives the capability to
extrapolate away from the wells with lower resolution. Due to uncertainties
in the data, rapid variations in the subsurface, and sparse sampling multi-
scale integration can be a challenging task. There is a good discussion of
“SURE Challenge” in the book where the author addresses the above men-
tioned challenges of integration incolving multitude of data set with differ-
ent Scale, Unvertainty, Resoultion and Environemnet. It is suggested that
different AI and Data Analytics techniques may be best equipped to handle
the SURE Challenge.
The second component can be categorized into input data quality
(informally ‘garbage in, garbage out’). Any workflow that is lets say cutting
edge cant work without high quality input data. Further, it may cause mis-
interpretation that a workflow is ‘not’ a good workflow or appropriate sim-
ply because the input data was the culprit. The input data in fact starts from
data acquisition, then to data processing and finally to data interpretation
and integration. One of the pitfalls along the way is to simply obtain the
data as an interpreter and not be aware of lets say the ‘history’. An exam-
ple would be to apply amplitude based seismic analysis to data that non-
amplitude preserving processing was applied to (Automatic Gain Control
or AGC would be a simple example). However, the same could be happen-
ing with say well-log or production data. The good news is that over time
in every SRC related discipline data quality has been improving with not
only better tools but also more frequent data acquisition during the life of a
reservoir. Further, over time we have learned to build much better process-
ing tools that provide high quality data for the integration component. The
net result of this has improved our ability to conduct integrated studies and
quantitative products. One example of this near to my heart is joint seis-
mic inversion of PP reflected waves with PS (or converted) reflected waves
from a reservoir. We have seen that with improved acquisition and process-
ing, the joint PP/PS inversion can substantially improve pre-stack seismic
inversion providing a stable S-impedance as well as a P-impedance that can
provide valuable information such as formation properties, porosity, Total
Organic Carbon (TOC,) fluid types, and time-lapse reservoir pressure and
saturation changes over the life of the reservoir. Such improvements are
going on in all the subsurface disciplines thanks to modern acquisition and
more diverse data with higher quality.
This book is an excellent resource for beginners in SRC to get an over-
view of the topic and for expert to study most recent advances in their
own and related disciplines. The book covers a wide range of topics from
conventional to unconventional reservoirs, from geology to geophysics to
Foreword xxi
xxiii
xxiv Preface
Fred Aminzadeh
Santa Barbara, California
September 22, 2021
Part 1
INTRODUCTION
1
Reservoir Characterization: Fundamental
and Applications - An Overview
Fred Aminzadeh *
Abstract
This article provides a brief overview of reservoir characterization at different
stages of a field from exploration to development to production and post pri-
mary production. It demonstrates the challenges associated with integration of
different data types. It also shows how “Dynamic Reservoir Characterization”
can assist in monitoring of the field for various well stimulation processes such
as enhanced oil recovery as well as reservoir stimulation. Different sections of
this entry attempt to highlight different aspects of reservoir characterization,
as an exploration tool, development tool, production tool and monitoring tool.
As reservoirs age, different measures are taken to extend their productive life.
This includes different types of reservoir stimulation and enhanced oil (or gas)
recovery.
*Email: [email protected]
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (3–22) © 2022
Scrivener Publishing LLC
3
4 Reservoir Characterization
Structural Stratigraphic
model model
Well and
seismic
data
Integration
Production
of 4D seismic Integration Forecasts
data y of production
tor
his data asts
no ching o rec
t f
ma CCS ctio
n
h
wit Inje
Figure 1.1 Different components of reservoir characterization, from Fornel and Estublier
[5].
crack-----fissure------fracture------fault-----------------------------------------
pore--lamina--bed/set--facies--compartment--paraseq.--deltas--basin--
ultrasonic--sonic logging------------
cross-hole--vsp------------
high-resolution seismic-----
reflection seismic----------------
earthquake seismology
gravity/magnetic
remote sensing data
Figure 1.2 Wide range of physical scale for different data types associated with different
geological and reservoir features.
Reservoir Characterization: Fundamental and Applications 7
Scale,
Cores Uncertainty
n
Resolution
tio
Sc
lu
ale
Environment
so
/C
Re
ov
y/
Well Logs
er
ag
r ta
Experience Pyramid
e
Ce
Seismic Data
Experience
Millimeters
Cores
Well
Logs SR2020’s World
Increasing Vertical Resolution
Bore Hole
Seismic
Surface
Seismic
10’s of
Meters
Figure 1.4 Areal coverage of well data is complemented by the larger areal sampling
of the geophysical methods. VSP vertical seismic profile and Crosswell seismic fill a
resolution “gap” between sonic log measurements and vertical seismic profiles. Courtesy
of SR2020 (now Optasens).
Reservoir Characterization: Fundamental and Applications 9
SEM
0.0001 The integration gap
light
microscopy
0.001
core
0.01
FMI
Vertical resolution, m
0.1
horizontal well log
well
1 log
cross- VSP
surface well
10 seismic
controlled-source
electomagnetics
100
grav-mag
1000
10 000 1000 100 10 1 0.1 0.01 0.001 0.0001
Low resolution Spatial resolution, m High resolution
Figure 1.5 Vertical and spatial resolution of various geophysical, well logs and laboratory
measurements. From www.agilegeoscience.com (left), and Optaense (right).
10 Reservoir Characterization
data and well log data on one side of the scale and 3D seismic data on the
other side.
In general. The resolution of different data types for reservoir charac-
terization and description varies considerably. Figure 1.5 illustrate such a
large variability for core to log to borehole geophysics and seismic, grav-
ity magnetics data and control source electromagnetics, among others.
This further demonstrates the importance finding a solution to the “SURE
Challenge” for reservoir characterization and other E&P problems. Also,
see Ma et al. [8] addressing integration of seismic and geologic data for
modeling petrophysical properties.
top reservoir
OWC
OWC
(a)
top reservoir
(b)
Figure 1.6 Time-lapse seismic response changes caused by different positions of oil-
water contact (OWC) in Gullfaks field Tarbert reservoir. (a) Oil-water contact level
before production and (b) after production for 10 years. Zones with changes in seismic
impedance are circled. http://accessscience.com.
14 Reservoir Characterization
i.e., acquiring time lapse seismic data and computing the differences in the
seismic attributes between the measurements the changes in the distribu-
tion of reservoir fluid properties between wells could often be detected.
The changes in time lapse seismic data are due to the acoustic imped-
ance variation (in this case, mostly the compressional velocity change),
caused by the reservoir production or other changes (such as water or CO2
injection in the EOR process). Acoustic impedance is the product of veloc-
ity (V) and density (ρ), for time lapse seismic amplitudes are influenced
by the incompressibility (Krock) of the reservoir rock and the production-
generated changes in the incompressibility of the pore fluids (∆Kfluids).
The reservoir rocks must therefore be sufficiently compressible so that
there is a prominent and measurable contribution from the pore fluids.
Soft compressible rocks like unconsolidated sands (younger in geologic
times) are ideal for time lapse seismic while rigid or incompressible reser-
voir rocks such as carbonates do not lend themselves for effective applica-
tion of this technology.
The acquisition and processing parameters for different vintage 3D seis-
mic should either be the same or necessary calibration should be applied
to make them consistent. That is the seismic response should be identical
when no changes in the geologic formation due to injection of production
has taken place. Some of these difficulties may be mitigated by using per-
manent sensors in wells and recording time lapse data.
C B'
A D'
400
C'
B A'
800
Depth (m)
1200
1600
2000
100
50 D 80
60
100 40
20
150
Figure 1.7 Use of conventional seismic, well log data and MEQ data to create a 3-D
fracture distribution volume, using a neural network by integrating “Fracture Zone
Identifier” or FZI attributes, Maity and Aminzadeh [9].
TN
Φ, vshale, facies...
3.5
TN
0.2
3
Vp, Vs (kft/sec)
2.0 2.1
Vp, Vs (kft/sec)
2.5 2 1.5
Avg. pressure
2
1.0 1.2
1
Vp, Vs, ρ
Figure 1.8 The entire process of reservoir model updating through 4D seismic modeling
and reservoir simulation, (from Meadows, [11]).
Description of rock and fluid properties between the well control points
requires understanding of the linkage of bulk and seismic properties to
each other and their changes with geologic age, burial depth, and location.
This connection allows us to understand and model the petrophysical and
geometrical properties which give rise to the seismic signal. Rock physics
requires a knowledge and understanding of geophysics, petrophysics, geo-
mechanics and the causes of distribution of fluids in the subsurface reser-
voir between wells. From seismic fluid monitoring we can obtain valuable
information about reservoir fluid movements and geologic reservoir het-
erogeneities. The results can also resolve seal integrity issues and guide the
optimum placement of wells in complex reservoirs.
Rock physics uses sonic, density and dipole sonic logs to establish a rela-
tionship between the geophysical data and the petrophysical properties. In
‘80s and ‘90s many oil companies had their own rock physics laboratories.
Because of the longer-range objectives and the need to assemble large data-
bases, today such laboratories are found primarily within five or six univer-
sities and a few service companies. The focus of rock physics analysis started
with estimating porosity and permeability of sandstones and carbonates.
Today, much of the research is focused on unconventional reservoirs and
on estimating rock strength or “fracability” and the presence of total organic
carbon. For some detailed discussion on the value of rock physics analysis in
various aspects of reservoir characterization and reservoir property estima-
tion see Dvorkin and Nur [5] and Castagna et al. [4].
Integration of 3D seismic interpretation with well measurements provides
a powerful tool for characterization a reservoir for the 3D distribution of rock
properties and the geometric framework of the reservoir. While the cores,
wireline logs and outcrops provide the vertical resolution it is only geophys-
ical data like 3D seismic data that can provide detailed spatial information
between the wells for the geological model. Since 3D seismic is a measurement
made at the surface of the earth, the subsurface interpretation using seismic
data can be done only after proper calibration with available well information.
Seismic reflection data provide the gross acoustic properties within a volume
of rock and do not have the vertical resolution of wireline logs.
Figure 1.9 Reservoir modeling process workflow. The process takes control of the
data within its modeling framework and integrates the various types of data attributes.
Courtesy: Roxar-Emerson.
−600
−620
−640
−660
−680
−700
−720
0 10 20 30 40 50
Geophysical Monitoring
Reservior Modeling
Seismic Modeling
Figure 1.10 Integrated reservoir modeling, fluid simulation update and reiteration
by incorporating geophysical monitoring data. http://www.co2care.org/Sections.
aspx?section=538.5.
20 Reservoir Characterization
1.7 Conclusion
Reservoir Characterization Is an important step in the entire life cycle of
the reservoir. Reservoir Characterization is aimed at assessing reservoir
properties and its condition, using the available data from different sources
such as core samples, log data, seismic surveys (3D and 4D) and produc-
tion data. This is done in different stages of the E&P process from high
grading reservoirs in exploration to their delineation, for their develop-
ment, as well as their description for optimum production to assessing
their evolution in their stimulation for enhance oil/gas recovery to extend
their economic life. An integrated approach for reservoir characteriza-
tion bridges the traditional disciplinary divides, leading to better han-
dling of uncertainties and improvement of the reservoir model for field
development. Among the main difficulties in reservoir characterization is
what I call “SURE” Challenge. The display here demonstrates the com-
plications involved in integrating different data types with different Scale,
Uncertainty, Resolution and Environment.
References
1. Aminzadeh, F., 2005, Meta-Attributes: A new concept detecting geologic fea-
tures and predicting reservoir properties, Second International Congress on
Geosciences Merida, Mexico September 2005
2. Aminzadeh, F. and Dasgupta, S., 2013 Geophysics for Petroleum Engineers,
Elsevier.
Reservoir Characterization: Fundamental and Applications 21
Abstract
The objective of this study is to assess the accuracy of the empiricalequations in
estimating the shear wave velocity and elastic modulus of rock sample at reservoir
pressure condition. The evaluated relations are Gassman, Greenberg and Castagna
which have been in use by researchers for a long time and have shown acceptable
results. The plug sample investigated in this study is taken from Berea sandstone
reservoir southwest of Australia, which is a known reference sandstone for this
type of study. This plug in the laboratory was flooded with supercritical carbon
dioxide fluid, saturated and pressured under axial, radial and pore pressure com-
parable to oil reservoirs pressures, after which elastic wave velocity and elasticity
modules were determined. Then, using empirical relationships such as Gassman,
Greenberg and Castagna and measured values of P-wave velocity, shear wave
velocity, and elasticity coefficients were estimated. Comparison of theoretical val-
ues versus experimental values shows they compare very well, only in some cases
a difference between estimated and experimental values for the coefficients of elas-
ticity has been observed. We believe that the difference is due to the assumptions
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (25–46) © 2022
Scrivener Publishing LLC
25
26 Reservoir Characterization
that were made in those theories. The shear wave modulus didn’t remain constant
during fluid saturation. Also, the measured bulk modulus and the calculated val-
ues based on Gassman formula did not compare very well. This difference was
observed to be larger at higher pressures.
2.1 Introduction
Study of propagation of shear and compressional waves give useful infor-
mation and constitutional characteristics of hydrocarbon reservoirs, such
as lithology and pore fluid type. This information is very important for res-
ervoir development and recovery, and especially for future decision mak-
ing. On the other hand, the behavior of reservoir rocks geomechanics, play
an important role in the design and implementation of drilling, production
planning and sustainability of oil and gas wells.
Having physical geology information such as density, porosity, com-
pressional and shear wave velocities are required to successfully perform
the above-mentioned projects. This is usually the case that the informa-
tion about shear wave velocity is not readily available compared with other
data. Therefore, theoretical or experimental approaches are necessary to
estimate this velocity.
In geomechanical evaluation of hydrocarbon reservoirs, several meth-
ods can be used to estimate shear wave velocity and elastic constants.
Conventional methods for estimating the shear wave velocity can be
Laboratory
measurements Core analysis
Methods to
Empirical relationships Greenberg - Castagne equations -
estimate shear
gassmn equations and...
wave velocity
Figure 2.1 Common methods for estimating the shear wave velocity.
Comparison Between Shear Wave Velocity and Elastic Modulus 27
2.2 Methodology
2.1.2 Estimating the Shear Wave Velocity
A major part of the seismic signal analysis in regards to rock physics mod-
els relates shear wave velocity to mineralogy and porosity. Rock physics
analysis based on logs and cores and the relation of these to the geological
model, leads to the establishment of a relationship between velocity and
porosity. Formulation of the relation between rock velocity and rock prop-
erties like porosity was initiated by Gassman [18] and revised later on by
Mavko and Mukerji [19] and Mavko et al. [20]. Other studies on this sub-
ject include Wyllie et al. [21], Raymer et al. [22], Castagna et al. [11], Han
[23], Raiga-Clemenceau and colleagues [24], Eberhart [25], and the critical
porosity model of Wang and Nur [26].
Greenberg - Castagna model is utilized in this study to estimate the
shear wave velocity of a rock sample. Greenberg and Castagna (1993) pre-
sented an empirical formula for multi-mineral rocks saturated in brine:
−1 −1
L Ni
L
Ni
1
VS =
2 ∑ ∑ Xi aijV j P +
∑ ∑ Xi
aijV − j P
, (2.1)
i =1 j =0 i =1 J =0
L
∑X =1 i
i =1
Comparison Between Shear Wave Velocity and Elastic Modulus 29
M = K + 43 µ( )
(2.2)
The usual process is initiated by replacing the primary fluid with a fluid
with similar sets of velocities and rock densities, compared with the pri-
mary fluid. These velocities are usually obtained from logs, but sometimes
they may also be the results of theoretical models. In this study, the velocity
of the wave that has passed through the primary fluid (in our case, super-
critical dioxide is injected into the water) is obtained through laboratory
measurements. But the removal of the existing fluid effects and replacing
it by the common fluid (brine) has been achieved through the following
steps (Dvorkin, 2003):
In the first stage the effective bulk modulus of pore fluid composition,
(K fluid ) is calculated using:
where, Sgas, Soil, Sbr, indicate gas, oil and brine saturation and Kgas, Koil, Kbr,
correspond to the apparent modulus of gas, oil, and brine. In the next step,
bulk modulus of rock (Klog), is calculated by equation (2.4):
The compressional wave velocity after removal of the primary fluid and
replacing it with brine is obtained by:
In this case, when the shear wave data is not available, compressional
wave modulus (Mlog) is calculated from charts (logs) using the following
relation:
Comparison Between Shear Wave Velocity and Elastic Modulus 31
The compressional wave modulus of the dry rock (Mdry) is also calcu-
lated using compressional wave modulus of the rock’s minerals:
( ( ) )
K = ρ Vp2 − 4 3 Vs2
(2.13)
32 Reservoir Characterization
G = ρ × Vs2 (2.14)
3Vp2 − 4Vs2
E = ρ × Vs2 × (2.15)
Vp2 − Vs2
A1 = (π (OD )2 − π ( ID )2 )
σ = Paxial = Pax ∗ A1 / A2 (2.16)
A2 = π r 2
Pr
OD PP Pax
Pax
Pr
Different parts of this chamber which is in direct contact with the fluid,
are neutral to chemical reaction. At the ends of this chamber, two caps are
placed were the transducers have been installed. As seen in Figure 2.3, the
receiver and transmitter are in direct contact with the sample.
The laboratory system is equipped with a back pressure regulator (BPR),
to make it possible to simulate the behavior of real reservoir pressure. For
axial and confining pressures, a hand pump has been used, and to achieve
pore pressure, a hydraulic pump has been utilized. This test has been per-
formed at room temperature. Signals are stored in a computer equipped
with particular software which has the capability to integrate into the oscil-
loscope. Shown in Figure 2.4a is the complete flooding system device with
three different fluid injection capsule, each with different kinds of fluids
and the oven chamber that maintains the desired test temperature (46 °C
in this experiment). Figure 2.4b indicates a picture of the core holder and
connected transducers.
It should be noted that the injected fluid in this experiment is CO2
supercritical, which has been injected under 1300 psi pressure and a
Transduce
Cap
Figure 2.3 Schematic, placement of sample with transducer and the top cap.
(a) (b)
Figure 2.4 (a) The core flooding system, (b) Image of the holder connected to the core
and transducers.
34 Reservoir Characterization
4000
3500
3000
Velocity (m/s)
2500
2000
1500
1000
1000 1500 2000 2500 3000 3500 4000 4500 5000
Effective pressure (psi)
Vp-experimental (m/s) Vs-experimental (m/s)
Figure 2.5 Compressional and shear wave velocity vs different effective pressure for super
critically saturated core sample.
4100
4000
3900
Velocity (m/s)
3800
3700
3600
3500
3400
3300
1500 2000 2500 3000 3500 4000 4500 5000
Effective pressure (psi)
Vp-experimental (m/s) Vp-estimate (m/s)
Figure 2.6 P-wave velocity (experimental and estimated) at different effective pressures.
Comparison Between Shear Wave Velocity and Elastic Modulus 35
2120
2080
Velocity (m/s)
2040
2000
1960
1920
1500 2000 2500 3000 3500 4000 4500 5000
Effective pressure (psi)
Vs-experimental (km/s) Vs-estimate (km/s)
Figure 2.7 S-wave velocity (experimental and estimated) at different effective pressures.
4
Y = 3.47 X – 8200
R² = 0.96
3.9
Vp-estimated (km/s)
3.8
3.7
3.6
3.4 3.42 3.44 3.46 3.48 3.5
Vp-experimental (km/s)
Figure 2.8 Cross plot of estimated P-wave velocities vs. laboratory measurements.
2.08
Y = 0.71 X + 0.57
2.06
R² = 0.96
Vs-estimate (km/s)
2.04
2.02
2
1.98
1.96
1.94
1.92 1.96 2 2.04 2.08 2.12
Vs-experimental (km/s)
Figure 2.9 Cross plot of the estimated S-wave velocities vs. laboratory measurements.
Comparison Between Shear Wave Velocity and Elastic Modulus 37
Figures 2.10 and 2.11 show the cross plots of the experimental and
estimated compressional and shear wave velocities. Mathematical rela-
tionship between two sets of wave velocities were obtained. As it shows
90% correlation observed in both modes between compressional and
shear wave velocity. The important note is that the slope of the curve is
2.12
2.1 Y = 1.40 X – 2.78
R² = 0.94
2.08
Vs-experimental (km/s)
2.06
2.04
2.02
2
1.98
1.96
1.94
3.38 3.4 3.42 3.44 3.46 3.48 3.5 3.52
Vp-experimental (km/s)
Figure 2.10 Plot of experimental shear wave velocity against compressional wave velocity.
38 Reservoir Characterization
2.08
Y = 0.29 X + 0.94
2.06
R² = 0.94
2.04
Vs-estimate (km/s)
2.02
1.98
1.96
1.94
3.55 3.6 3.65 3.7 3.75 3.8 3.85 3.9 3.95 4 4.05
Vp-estimate (km/s)
Figure 2.11 Plot of estimated shear wave velocity against compressional wave velocity.
4000
3500
Velocity (m/s)
3000
2500
2000
1500
1500 2000 2500 3000 3500 4000 4500 5000
Effective pressure (psi)
Vs-experimental (m/s) Vp-experimental (m/s)
Vs-estimate (m/s) Vp-estimate (m/s)
25
Y = –3.53 X + 72.42
23
R² = 0.11
K-estimate (MPa)
21
19
17
15
14.4 14.5 14.6 14.7 14.8 14.9 15
K-experimental (MPa )
Figure 2.13 Plot of Laboratory vs. estimated Bulk modulus (K) of rock sample.
The fluid used in this study has a patchy saturation and has created
anisotropic, heterogeneous environments. Besides that, different effec-
tive pressures that applied on a frame stone, impressed the results of
Gassmann - Greenberg – Castagna equations.
10
9.4
9.2
8.8
8.6
8.6 8.8 9 9.2 9.4 9.6 9.8 10 10.2
G-experimental (MPa)
26
Y = 1.08 X – 0.99
25.5
R² = 0.97
25
E-estimate (MPa)
24.5
24
23.5
23
22.5
22
21.5 22 22.5 23 23.5 24 24.5 25
E-experimental (MPa)
But based on the values in Figure 2.15, the experimental and the esti-
mated values of Young’s modulus are very close to one another and very
comparable (%97).
2.5 Conclusions
In this paper we evaluated the closeness of the estimated elastic wave veloc-
ities and elastic modulus of a rock sample via Gassmann - Greenberg -
Castagna formula to those obtained from laboratory measurements at
reservoir pressure.
2.6 Acknowledgment
Authors extended their appreciation to Petroleum Engineering
Department, Curtin University of Technology, Australia that provided
the authors the opportunity to utilize the laboratory core flooding system.
Also, the authors thank and appreciate very much Prof. Vamegh Rasouli,
Dr. Amin Nabipour, Dr. Mohammad Sarmadi and Dr. Mohsen Ghasemi
for their help in conducting these experiments. Finally, the authors are
very grateful and extend their deepest appreciation to the respected faculty
of Petroleum Engineering and Geophysics departments, Curtin University
of Technology that cooperated in the design, manufacturing and installa-
tion of the transducers.
References
1. A. J. Engel and G. R. Bashford, A new method for shear wave speed estima-
tion in shear wave elastography. IEEE T. Ultrason. Ferr. 62(12), 2016–2114
(2015).
2. J. K. Jang, K. Kondo, T. Namita, M. Yamakava, and T. Shiina, Comparison
of techniques for estimating shear-wave velocity in arterial wall using
shear-wave elastography - FEM and phantom study, in IEEE International
Ultrasonic Symposium, Taipei, Taiwan, p. 10, (2015).
3. J. S. L. Heureux and M. Long, Correlations between shear wave velocity and
geotechnical parameters in Norwegian clays, in 17th Nordic Geotechnical
Meeting Challenges in Nordic Geotechnic, Reykjavik, Iceland, 299–308 (2016).
4. B. Widarsono and P. M. Wong, Estimation of rock dynamic elastic property
profiles through a combination of soft computing, acoustic velocity model-
ing, and laboratory dynamic test on core samples, SPE Asia Pacific Oil and
Gas Conference, 68712 (2001).
5. G. R. Pickett, Acoustic character logs and their application in formation eval-
uation. J. Petrol. Technol. 15, 650–667 (1963).
6. P. Milholand, M. H. Manghnani, S. O. Schlanger, and G. H. Sutton,
Geoacoustic modeling of deep-sea carbonate sediments. J. Acoust. Soc. Am.
68, 1351–1360 (1980).
7. S. N. Domenico, Rock lithology and porosity determination from shear and
compressional velocity. Geophysics 49, 1188–1195 (1984).
8. L. Thomsen, Weak elastic anisotropy. Geophys 51, 1654–1966 (1986).
9. D. Han, Empirical relationships among seismic velocity, effective pressure,
porosity and clay content in sandstone. Geophysics 54, 82–89 (1989).
10. M. Krief, J. Garat, J. Stellingwerf, and J. Venter, A petrophysical interpreta-
tion using the velocities of P and S waves (full wave from sonic). Log Anal,
31, 35–369 (1990).
44 Reservoir Characterization
27. J. P. Dvorkin, and S. Alkhater, Pore fluid and porosity mapping from seismic,
First Break, 22, 53–57 (2004).
28. I. Takahashi, Rock physics as a quantitative tool for seismic reservoir charac-
terization, INPEX Corporation, p. 175 (2005).
29. P. Avseth, T. Mukerji, and G. Mavko, Quantitative Seismic Interpretation,
Cambridge University Press, Cambridge (2005).
3
Anomaly Detection within
Homogenous Geologic Area
Simon Katz1*, Fred Aminzadeh2, George Chilingar2 and Leonid Khilyuk1
California
Abstract
This paper describes the development and investigation of a methodology and
multi-step algorithms for the detection of geologic anomalies. The algorithms
presented in this paper belong to the group of machine learning one-class clas-
sification techniques. The developed methodology includes the following steps:
(a) compiling a single-class training set that includes data recorded in a known
homogeneous area, (b) construction of universal anomaly detection classifiers
(AD classifiers) and high-resolution classifiers specifically designed for detection
of an anomaly of a certain type, (c) construction of adaptive AD classifiers, and
(d) testing a hypothesis about an anomaly type using data from the part of the
anomaly detected by the AD classifier and from the training set.
Three basis AD classifiers were suggested and tested for identification of anom-
alies: (a) distance-from-the-center of the training set, (b) sparsity of neighbors
from the training set in the vicinity of a tested record, and (c) divergence from
the center of the training set. Distance-from-the-center of the training set and
sparsity of neighbors are universal and may be used for detection of anomalies
with unknown properties and for optimization of adaptive classifiers. Divergence
is a specialized classifier designed for detection of anomalies with known proper-
ties. New adaptive classifiers presented in the paper may be used for detection of
anomalies of unknown type with efficiency superior to universal classifiers. The
performed tests with the developed algorithms demonstrate the high efficiency of
the developed one-class classifiers in the detection of gas-filled sands as anoma-
lies even when moderate size training sets were utilized. A hypothesis test about
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (47–68) © 2022
Scrivener Publishing LLC
47
48 Reservoir Characterization
anomaly type had statistically significant results in the wide range of test set sizes
when an adaptive classifier was used for anomaly detection.
3.1 Introduction
An anomaly detection problem arises in geological research and reser-
voir characterization when, for example, there is a need to locate gas-filled
sands in the vicinity of the area with identified brine-filled sands or shales,
Katz et al., [8]. Other examples of anomaly detection problems include
finding the location of an overpressure zone using data obtained from an
area with normal gas pressure Dvorkin et al., [6], Gurevich et al., [7] and
finding the location of fractured carbonates filled with gas, Chilingar et al.,
[5], Maity and Aminzadeh [12].
Detection of anomalous behavior in seismic and/or well log data, which
can be attributed to the presence of hydrocarbon, has been the subject
of many research activities. Seismic attributes introduced by Taner [16]
have been widely used to detect seismic anomalies. Subsequently, many
methods have been used to combine such attributes, either through clus-
tering, Aminzadeh and Chaterjee [10], or neural networks Aminzadeh
et al., [13]. Then, the meta attributes concept was introduced which com-
bines machine and human intelligence for a targeted anomaly detection
Aminzadeh [11−15]. For a comprehensive overview of seismic attribute
methodologies see Chopra and Murfurt [17].
The methodology for anomaly detection (AD) presented in this paper
is different from multiclass classification where a training set is formed as
a union of several subsets, and each subset contains records from a single
class. Records from all classes are mixed in the training set. It is also differ-
ent from the classical methodology for outlier detection. Outlier detection
methodology is aimed at analysis of a dataset that may contain both reg-
ular and outlier records which are not identified prior to outlier analysis,
so there is no training set (Barnett [2], Barnett and Lewis [3], Agrawal and
Raghavan [1], Breunig et al. [4]).
In this paper writers tested new algorithms that belong to the group
of one-class classification machine learning techniques. One-class classi-
fication is a comparatively new area of machine learning research (see for
example Tax and Duin [18], Muñoz-Marí [19], Marti et al. [24]). Machine
Anomaly Detection within Homogenous Geologic Area 49
dist (Y , Center ) =
1
M ∑( y m − ctr ,m )2 (3.4)
m =1
ctr ,m =∑1
K k =1
y k ,m where yk,m is the m-th coordinate of the
k-th record in the training set, K is total number of records
in the training set.
2. Nearest neighbors sparsity:
spars(Y ) =
1
L ∑dist(Y ,neighbor ) l (3.5)
l =1
div(Y , Center ) =
1
M ∑a m
∗
(ctr ,m − ym ) (3.6)
m =1
aggr (Y ) = ∑y m ∗ sm (3.7)
m =1
0.5
0.0
–0.5
0 20 40 60
Index
Figure 3.1 Divergence values for records in training and test set. The horizontal dashed
line is the classification cutoff for the expected false discovery rate of 15%. The test set
contains 30 regular records and 25 anomaly records from gas-filled sands. Each record
(Vp/Vs and Posson’s ratio).
54 Reservoir Characterization
0.8 0.8
Expected FDR
Expected FDR
0.6 0.6
0.4 0.4
0.2 0.2
0.0 0.0
Figure 3.2 Expected versus posterior false discovery rate for two sizes of the training set.
Classifier: distance from the center of the training set. Coordinates of a record: Poisson’s
Ratio and Vp/Vs.
Anomaly Detection within Homogenous Geologic Area 55
Table 3.1 Mean values of false and true discovery rates for three classifiers in
detection of gas-filled sands anomaly. Size of the training set: 20 records.
Mean of
Expected false Mean of posterior posterior true
AD discovery rate false discovery discovery rate
method (%) rate (%) (%)
Sparsity 20 20.44 60.85
Distance 20 20.44 71.32
Divergence 20 20.49 85.71
Divergence AUC
200
150
100
50
0
0.5 0.6 0.7 0.8 0.9
Sparcity AUC
40
Count
20
0
0.5 0.6 0.7 0.8 0.9
Distance AUC
60
40
20
0
0.5 0.6 0.7 0.8 0.9
AUC
Figure 3.3 Histograms of area under posterior ROC Curve (AUC) for three anomaly
detection classifiers. Size of each training set is 5 records.
training set contained five records. The total number of randomly gener-
ated pairs of training and test sets was 1000. The intersections of contin-
uous vertical black lines with x-axis at Figure 3.3 shows the median value
of the AUC. Dashed lines indicate lower and upper quantiles of the AUC
distribution calculated for quantile probabilities Plow=0.1 and Pupper=0.9.
According to the Figure 3.3, the divergence classifier is characterized by
the narrowest distribution of AUC values and the largest AUC median.
Distribution of this classifier is symmetric. Distributions of distance and
sparsity classifiers are asymmetric with long tails in the direction of smaller
values. As a result, lower quantiles for the distribution of distance and spar-
sity AUC are shifted towards smaller AUC values.
Two of the three AD classifiers with AUC shown in Figure 3.3 do not
rely on the use of information about properties of petrophysical parameters
within potential anomalies. This is their advantage since they may be used
for detection of any type of anomaly with unknown properties. Although
they underperform compared to the divergence classifiers that rely on the
use of known anomaly properties, they still can produce median AUC val-
ues as high as 0.75.
Figure 3.4 shows posterior AUC histograms for sparsity classifiers cal-
culated for two sizes of the training sets. Continuous vertical lines are
median values, intersection of doted lines with x-axis are the levels Plow=0.1
Anomaly Detection within Homogenous Geologic Area 57
Size of training set: 25
40
Frequency
20
0
0.5 0.6 0.7 0.8 0.9
40
20
0
0.5 0.6 0.7 0.8 0.9
AUC
Figure 3.4 AUC histograms and quantile regions calculated from 1000 pairs of training-
test sets. Size of the training sets: 5 and 25 records. Sparsity classifier.
0.90
0.85
Median of posterior AUC
0.80
0.75
0.70
0.65
Divergence Distance
Sparsity
0.60
5 10 15 20 25
Size of the training set
Figure 3.5 Median of posterior AUC for three AD classifiers as a function of the size of
the training set.
and Pupper=0.9. According to Figure 3.4, the AUC quantile region in the case
of 25 records training sets is 40% narrower than the quantile region in the
case of the training sets of 5 records.
The median of posterior AUC values and the width of the AUC quantile
region for three AD classifiers are shown in Figures 3.5 and 3.6. According
58 Reservoir Characterization
0.2
Quantile width of posterior
AUC distribution
0.1
0.0
Divergence Distance
–0.1
Sparsity
5 10 15 20 25
Size of the training set
Figure 3.6 Quantile width of AUC distribution calculated on anomaly detection results
from 1000 randomly generated pairs of training-test sets.
to Figure 3.5, the median AUC of the divergence is higher than median
AUC of other classifiers with its values stable around 0.9. Importantly, it
is as high as 0.9 even for small training sets containing only five records.
The sparcity classifier has the lowest AUC median which may be as small
as 0.70.
According to Figure 3.6, divergence has the narrowest AUC quantile
region. AUC quantile regions of two other AD classifiers are significantly
wider than that of divergence. The width of AUC quantile regions for these
classifiers decreases with the increasing size of the training sets. It is still
about three times as wide as the AUC quantile width of divergence for the
size of training set 20, 25 records.
where sm; 1 ≤ m ≤ M are weights and aggr() is defined by Eq. 3.7, classifier
Anomaly Records are records identified by a universal classifier as anomaly,
trainSetRecords are records from the training set.
To find coefficients sm that maximize efficiency criterion (3.8) we used a
multidimensional grid search. In the grid search, the coefficients sm in Eq.
3.7 take on a discrete set of values in the following region:
−1 ≤ sm ≤ 1 (3.9)
0.5
0.0
–0.5
0 20 40 60
Index
Figure 3.7 Sparsity values on the records of the training and test sets. Horizontal dashed
line - anomaly detection cutoff producing an expected false discovery rate of 20%.
200
0
0.55 0.60 0.65 0.70 0.75 0.80 0.85
True discovery rate
Expected
250
false discovery rate: 5%
100
0
0.55 0.60 0.65 0.70 0.75 0.80 0.85
True discovery rate
Figure 3.8 Anomaly detection. Histograms of posterior true discovery rate (TDR) for two
values of expected false discovery rate. AD method: aggregated. 20 regular records in each
training set.
the training set are randomly selected out of a set of 50 records. The test
set contains 30 regular and 25 anomaly records. Anomaly records are from
gas-filled sands. Regular ones are from brine-filled sand or shale. The
assigned value of the expected false discovery rate was 20%. Records clas-
sified as a potential anomaly include 13 actual anomaly records and 2 reg-
ular records. Thus the posterior true discovery rate is very moderate - 52%.
Anomaly Detection within Homogenous Geologic Area 61
median (ϕ j (detectedAnomalyPart ))
J
anomalyIndicator = ∑ ηj ∗
median (ϕ j (trainSet ))
j =1
(3.11)
62 Reservoir Characterization
80
40
0
0.4 0.6 0.8 1.0 1.2
(a)
80
40
0
0.4 0.6 0.8 1.0 1.2
(b)
50
20
0
0.4 0.6 0.8 1.0 1.2
(c)
3.8 Conclusion
New algorithms and a methodology for machine learning one-class anom-
aly detection are presented. Detection of gas-filled sandstones, detection of
abnormal pressure zones, and detection of gas-filled fractured carbonates
are examples of geologic anomaly detection problems.
Three groups of AD classifiers were developed and tested for anomaly
detection:
References
1. R. Agrawal, and P. Raghavan, A Linear Method for Deviation Detection in
Large Databases Arning A. KDD-96, 164–169 (1996).
66 Reservoir Characterization
20. L. Bregman, The relaxation method of finding the common points of convex
sets and its application to the solution of problems in convex programming.
Computational Mathematics and Mathematical Physics 7(3), 200–217 (1967).
21. A. Banerjee, S. Merugu, I. Dhillon, and J. Ghosh, Clustering with Bregman
Divergences. J. of Machine Learning Research 6, 1705–174 (2005).
22. W. Ostander, Plane-wave refection coefficients for gas sands at nonnormal
angles of incidence. Geophysics 49(10), 1637–1648 (1984).
23. P. Jain, C. Jambhekar, and P. Pandey, Identification of gas using Vp/Vs vis-
a-vis Poisson’s ratio. 9-th Biennial Int. Conf. and Exposition on Petroleum
Geophysics, Haiderabad (2012).
24. L. Marti, N. Sanchez-Pi, L. Molina, and A. Garcia, Anomaly detection
based on sensor data in petroleum industry applications. Sensors 15(2),
2774–2797 (2015).
4
Characterization of Carbonate
Source-Derived Hydrocarbons Using
Advanced Geochemical Technologies
Hossein Alimi *
Abstract
Various crude oil classifications have been proposed by geochemists and petro-
leum refiners. Geologists and geochemists are more interested in identifying and
characterizing the crude oils, to relate them to source rocks and to measure their
grade of evolution using advanced geochemical technologies. Carbonate source
rocks studied are organically very rich, containing oil-prone kerogen capable of
generating significant quantities of hydrocarbons.
Tertiary and Cretaceous oils studied are of carbonate origin and classified as
aromatic-intermediate class. Most of the Tertiary oils studied are severely biode-
graded while those from Cretaceous reservoirs are non-degraded.
Almost all the oils investigated show the following biomarker parameters, char-
acteristic of oils derived from carbonate source rocks deposited in a marine envi-
ronment under reducing conditions:
*Email: [email protected]
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (69–80) © 2022
Scrivener Publishing LLC
69
70 Reservoir Characterization
4.1 Introduction
For obtaining a fundamental knowledge of the genetic history of crude oils
two types of geochemical methods should be used.
percentage of oil earlier than more humic organic matter of shales. The
unique Carbonate-Evaporite depositional and diagenetic environments,
without infux of major terrigenous and humic substances, produce
algal-sapropelic organic facies that yield sulfur-rich petroleum at lower
temperatures than humic-type shales. The worldwide presence of small to
very large, apparently immature, to marginally mature, non-biodegraded,
heavy oil/bitumen/asphalts deposits, rich in resins and asphaltenes, are oil
sourced from carbonate source rocks.
This study will provide an overview of the geochemical characteristics
of carbonate-derived oils and source rocks collected from different wells
and fields.
As with shales, the source potential of carbonate rocks depends pri-
marily upon the organic facies rather than the mineral matrix. Where
the depositional and early diagenetic environment is highly oxygen-
ated, the total-organic-carbon (TOC) content is low. The remaining
kerogen is highly oxygenated, with a negligible generative capacity for
hydrocarbons.
The early anoxic diagenetic depositional environment can result in
the deposition of organic-rich, fine-grained carbonate sediments that are
excellent potential source rocks [3]. Although they constitute a small per-
centage of all carbonate rocks, organic-rich, fine-grained carbonate rocks
are widespread in both time and space and are the probable source of
30-40% or more of the petroleum reserves of the world.
Gas-prone organic facies are rare in carbonate rocks, because they are
usually dominated by humic organic matter deposited in a dominantly
clay matrix. However, gas-prone organic facies may occur in carbonate
rocks as results of turbidite deposition [4] or by a mixture of kerogen
types II and III.
Oils derived from carbonate rocks are often richer in cyclic hydrocar-
bons and sulfur containing compounds than oils derived from shales,
owing to the lack of terrestrial-plant waxes in their source organic matter.
1000
Well A Well B
900 l
Type I
800
Hydrogen index (HI, mg HC/g TOC)
700
II
600 Type II
500
400
300
200
III
Type III
100
IV
Type IV
0
0 10 20 30 40 50 60 70 80 90 100
Oxygen index (OI, mg CO2/g TOC)
1.0
Immature Oil window Dry gas window
0.6
Stained or
0.5 contaminated
0.4
0.3
0.1
Low level conversion Overmature
0.0
400 425 450 475 500
Maturity (based on Tmax, °C)
NC11 NC11
NC12 NC12
IP13 IP13
IP14 IP14
NC13 NC13
IP15 IP15
NC14 NC14
IP16 IP16
NC15 NC15
NC16 NC16
IP18 IP18
IP19 NC17 IP19 NC17
PHEN PHEN
76 Reservoir Characterization
NC19 NC19
NC20 NC20
NC22 NC22
Phytane
NC24 NC24
NC25 NC25
NC26 NC26
NC27 NC27
NC28 NC28
NC29 NC29
NC30 NC30
NC31 NC31
Pr/Ph = 0.49
NC32 NC32
NC33 NC33
NC34 NC34
NC35 NC35
Pr/Ph = 1.44
NC36 NC36
NC37 NC37
NC38 NC38
NC39 NC39
NC40 NC40
NC41 NC41
G1090570.D
G2091217.D
Characterization of Carbonate Source-Derived Hydrocarbons 77
Ph
Pr
Figure 4.6 GC- Fingerprint of Tertiary Oil No. 3 which is severely biodegraded.
Figure 4.7 GC- Fingerprint of Cretaceous Oil No. 3 which is not biodegraded.
Tertiary oil # 1
C30-Hopane
C29-Hopane
marllacustrine sulfate-rich
Ter. oil4
3.5 Zone 2 Lacustrine sulfate-poor
mixed shaje-carbonate Ter. oil5
3.0 Crt. oil1
Zone 3 Marine shale and other
lacustrine Crt. oil2
2.5 Crt. oil3
Zone 4 Fluvial/deltaic
Crt. oil4
2.0
Zone 1B Crt. oil5
1.5 Crt. oil6
Crt. oil7
1.0 Crt. oil8
Crt. oil9
0.5 Zone 2 Zone 3 Zone 4
Crt. SRExt
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
Pristane / phytane
References
1. B.P. Tissot and D.H. Welte, Petroleum Formation and Occurrence, Springer-
Verlag, New York (1984).
2. K.E. Peters, C.C. Walters, and J.M. Moldowan, The Biomarker Guide Volume
1 and 2 Biomarkers and Isotopes in Petroleum Exploration and Earth History,
pp. 492, Cambridge University Press, United Kingdom (2005).
3. GJ. Demaison and G.T. Moore, Anoxic environments and oil source bed gen-
esis. AAPG Bulletin 64, 1179-1209 (1980).
4. S.E. Calvert, Oceanographic controls on the accumulation of organic matter
in marine sediments, in Marine Petroleum Source Rocks, J. Brooks and A.J.
Fleet (Eds.), pp. 137-151, Blackwell, London (1987).
5. K.E. Peters, Guidelines for evaluating petroleum source rocks using pro-
gramed pyrolysis. AAPG Bulletin 70, 318-329 (1986).
6. D.K. Baskin, Comparison between H/C and Rock-Eval hydrogen index as
indicator of organic matter quality, in The Monterey Formation: From Rocks
to Molecules, CM. Isaacs and J.Rullkoetter (Eds.), pp. 230-240, Colombia
University Press, New York (2001).
7. D.H. Welte and D. Waples, Ueber die Bevorzugung gradzahliger n-Alkane in
Sedimentgesteinen. Naturwissenschaften 60, 516-517 (1973).
8. R. Alexander, R. Kagi, and G.W. Woodhouse, Geochemical correlation of
Windalia oil and extracts of Winning Group (Cretaceous) potential source
rocks, Barrow Subbasin, Western Australia. AAPG Bulletin 65, 235-250 (1981).
9. H. Ten Haven, J.W. de Leeuw, J. Rullkoetter, and J.S. Sinninghe damste,
Restricted utility of the pristane/phytane ratio as paleoenvironmental indi-
cator. Nature 330, 641-643 (1987).
10. W.B. Hughes, A. G. Holba, and L.I.P. Dzou, The ratios of Dibenzothiophene to
Phenanthrene and Pristane to Phytane as indicators of depositional environ-
ment and lithology of petroleum source rocks. Geochimica et Cosmochimica
Acta 59, 3581-3598 (1995).
5
Strategies in High-Data-Rate
MWD Mud Pulse Telemetry
Yinao Su1, Limin Sheng1, Lin Li1, Hailong Bian1, Rong Shi1, Xiaoying Zhuang2
and Wilson Chin2*
1
China National Petroleum Corporation (CNPC), Beijing, China
2
Stratamagnetic Software, LLC, Houston, Texas, United States
Abstract
Low data rates, typically 1-3 bits/sec or less, are the norm in “mud pulse”
Measurement-While-Drilling (MWD) systems. For example, “sirens” offering
high carrier frequencies produce low-amplitude signals subject to severe atten-
uation; and strong signals from “positive pulsers” are slow since large forces are
required to overcome mud inertia. Moreover, reflections, high pump noise, ero-
sion and rapid power consumption affect all pulsers. This paper describes an inte-
grated systems approach to high-data-rate telemetry. Source strength is enhanced
using downhole constructive interference, and surface signal processing elim-
inates downward reflections and pump noise through “directional filters,” both
drawing on wave-based methods. Hydraulic properties associated with torque,
erosion, aerodynamic stability and turbine performance, and subtleties found in
large-scale acoustic wave interactions, are studied using dynamic similarity. In
particular, short and long wind tunnels are introduced with significantly reduced
test times, cost and required labor. Experimental facilities, prototypes and signal
processing methods are presented in detail.
Keywords: Echo cancellation, mud pump noise, mud siren, MWD, noise
removal
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (81–134) © 2022
Scrivener Publishing LLC
81
82 Reservoir Characterization
5.1 Summary
Measurement-While-Drilling (MWD) systems presently employing mud
pulse telemetry transmit no faster than 1-3 bits/sec from deep wells con-
taining highly attenuative mud. The reasons – “positive pulsers” create
strong signals but large axial flow forces impede fast reciprocation, while
“mud sirens” provide high data rates but are severely lacking in signal
strength. China National Petroleum Corporation research in MWD telem-
etry focuses on improved formation evaluation and drilling safety in deep
exploration wells. A high-data-rate system providing 10 bits/sec and opera-
ble up to 30,000 ft is described, which creates strong source signals by using
downhole constructive wave interference in two novel ways. First, teleme-
try schemes, frequencies and pulser locations in the MWD drill collar are
selected for positive wave phasing, and second, sirens-in-series are used to
create additive signals without incurring power and erosion penalties. Also,
the positions normally occupied by pulsers and turbines are reversed. A
systems design approach is undertaken, e.g., strong source signals are aug-
mented with new multiple-transducer surface signal processing methods to
remove mudpump noise and signal reflections at both pump and desurger,
and mud, bottomhole assembly and drill pipe properties, to the extent pos-
sible in practice, are controlled to reduce attenuation. Special scaling meth-
ods based on dynamic similarity are developed to extrapolate wind tunnel
results to real muds flowing at any downhole speed. We also describe the
results of detailed acoustic modeling in realistic drilling telemetry chan-
nels, and introduce by way of photographs, CNPC’s “short wind tunnel” for
signal strength, torque, erosion and jamming testing, “very long wind tun-
nel” (over 1,000 feet) for telemetry evaluation, new siren concept prototype
hardware and also typical acoustic test results. Movies demonstrating new
test capabilities are available upon request.
This is easier said than done. For example, in the early days of well log-
ging, formation evaluation dictated that gas or oil-filled rocks have higher
resistivities than water-filled rocks. Interpretation was simple. But through
the years, “low-resistivity pay” has become a worldwide phenomenon; with
lower oil prices driving the re-exploration of mature fields, new methods of
interpreting resistivity readings have proliferated.
Answers depend on many factors. It is known, for instance, that prop-
erties isotropic at small scales may be anisotropic macroscopically. Many
logging tools lack the resolution needed to resolve values for individual
thin beds of sand and shale, often taking averages that conceal useful reser-
voir information. Ideally, instruments would sample properties at multiple
scales and return their findings to on-site geologists. Similar arguments
apply to formation tester, acoustic, nuclear and NMR logs – more data,
especially if available during drilling, is always helpful.
The bottleneck in this quest for increasing information – again, aimed at
enhancing sustainability by reducing the likelihood of bypassing produc-
tive zones – is a direct consequence of low data transfer rates supported
by present MWD technology. In a digital age where “megabytes per sec-
ond” define the norm, data rates during drilling, typically just 1-3 bits/
sec or less, are limited by high mud attenuation, poor signal strength at
the source, and highly reverberant and distortive environments, all leading
to decreases in signal-to-noise ratio. Moreover, problems associated with
high levels of sand erosion and excessive power demands add to already
difficult engineering challenges.
Little effort has been undertaken in the past thirty years to approach
MWD mud pulse telemetry as a science based on acoustics and fluid-
dynamical principles. The present authors have approached this research
systematically over the past decade and results are reported in their recent
book, i.e., Chin, Su, Sheng, Li, Bian, and Shi [2]. This paper presents an over-
view of the new methodologies plus a major update on “direction-based”
(as opposed to frequency-oriented) signal processing techniques for sur-
face enhancement.
5.1.2 Introduction
The petroleum industry has long acknowledged the need for high-data-
rate MWD mud pulse telemetry in oil and gas exploration. This need is
driven by several demand factors: high density logging data collected by
more and more sensors, drilling safety for modern managed pressure drill-
ing and real-time decision-making, and management of economic risk by
enabling more accurate formation evaluation information.
84 Reservoir Characterization
Yet, despite three decades of industry experience, data rates are no bet-
ter than they were at the inception of mud pulse technology. To be sure,
major strides in reliability and other incremental improvements have been
made. But siren data rates are still 3 bits/sec in shallower wells and positive
pulser rates still perform at a dismal 1 bit/sec or less. Recent claims for data
rates exceeding tens of bits/sec are usually offered without detailed basis or
description, e.g., the types of mud used and the corresponding hole depths
are rarely quoted.
From a business perspective, there is little incentive for existing oil ser-
vice companies to improve the technology. They monopolize the logging
industry, maintain millions of dollars in tool inventory, and understandably
prefer the status quo. Then again, high data rates are not easily achieved.
Subtleties abound. Quadrupling a 3 bits/sec signal under a 12 Hz carrier
wave, as we will find, involves much more than running a 48 Hz carrier
with all else unchanged. Moreover, there exist valid theoretical consider-
ations (via Joukowski’s classic formula) that limit the ultimate signal pos-
sible from sirens. Innovative mechanical designs for positive pulsers have
been proposed by others and tested. Some offer extremely strong signals,
although they are not agile enough for high data rates. Unfortunately, the
lack of complementary telemetry schemes and surface signal processing
methods renders them hostage to strong reverberations and signal distor-
tions at desurgers.
One might surmise that good “back of the envelope” planning, from
a systems engineering perspective underscoring the importance of both
downhole and surface components, is all that is needed, at least in a first
pass. Acoustic modeling in itself, while not trivial, is a well-developed
science in many engineering applications. For example, highly refined
theoretical and numerical models are available for industrial ultrasonics,
telephonic voice filtering, medical imaging, underwater sonar for subma-
rine detection, sonic boom analysis for aircraft signature minimization,
and so on; several of them deal with complicated three-dimensional, short-
wave interactions in anisotropic media.
By contrast, MWD mud pulse telemetry can be completely described by
a single partial differential equation, in particular, the classical wave equa-
tion for long wave acoustics. This is the same equation used, in elementary
calculus and physics, to model simple organ pipe resonances and has been
the subject of research reaching back to the 1700s. Why few MWD design-
ers use wave equation models analytically or experimentally, by means of
wind tunnel analogies implied by the identical forms of the underlying
equations, is easily answered: there are no physical analogies that have
motivated scientists to even consider models that bear any resemblance
Strategies in High-Data-Rate MWD Mud Pulse Telemetry 85
haunts the intended upgoing signal. But unlike a shadow that simply fol-
lows its owner, the use of “phase-shift-keying” (PSK) introduces a random
element that complicates signal processing: depending on phase, the upgo-
ing and downgoing signals can constructively or destructively interfere.
Modeling of such interactions is not difficult in principle since the linearity
of the governing equation permits simple superposition methods.
However, it is now important to model the source itself: it must create
antisymmetric pressure signals and, at the same time, allow up and down-
going waves to transparently pass through it and interfere. It is also nec-
essary to emphasize that wave refraction and reflection methods for very
high frequencies (associated with very short wavelengths) are inapplicable.
The solution, it turns out, lies in the use of mathematical forcing functions,
an application well developed in earthquake engineering and nuclear test
detection where seismic waves created by local anomalies travel in multiple
directions around the globe only to return and interfere with newer waves.
Wave propagation subtleties are also found at the surface at the stand-
pipe. We have noted that (at least) two sets of signals can be created down-
hole for a single position-modulated valve action (multiple signals and
MWD drill collar reverberations are actually found when area mismatches
with the drill pipe are large). These travel to the surface past the stand-
pipe transducers. They reflect at the mud-pump and at the desurgers. For
high frequency, low amplitude signals (e.g., those due to existing sirens),
desurgers serve their intended purpose and the internal bladders “do not
have enough time” to distort signals. On the other hand, for low frequency,
high amplitude signals (e.g., positive pulsers), the effects can be disastrous:
a simple square wave can stretch, change in shape and literally become
unrecognizable.
Thus, robust signal processing methods are important. However, most
of the schemes in the patent literature amount to no more than common
sense recipes that are actually dangerous if implemented. These often sug-
gest “subtracting this, delaying that, adding the two” to create a type of
stacked waveform that hopefully improves signal-to-noise ratio. The dan-
ger lies not in the philosophy but in the lack of scientific rigor: true filter-
ing schemes must be designed around the wave equation and its reflection
properties, but few MWD schemes ever are.
Moreover, existing practices demonstrate a lack of understanding with
respect to basic wave reflection properties. For example, the mud pump is
generally viewed with fear and respect because it is a source of significant
noise. It turns out that, with properly designed multiple-transducer signal
processing methods, piston induced pressure oscillations can be almost
completely removed even if the exact form of their signatures is not known.
Strategies in High-Data-Rate MWD Mud Pulse Telemetry 87
In addition, theory indicates that a MWD signal will double near a piston
interface, which leads to a doubling of the signal-to-noise ratio. This, in
fact, has been verified experimentally in the field, a result that has prompted
improved strategies for surface transducer placement.
Drillpipe Mud
5.0" OD, 2.8" ID
MWD Collar 22 ft
8.0" OD, 4.5" ID Pulser
Mud Motor
8.0" OD, 3.5" ID 25 ft
Rotor, 2.5" dia.
Bit-Box/Nozzles
3.5" ID 2 ft
12.25"
Figure 5.2a Example MWD collar used for siren frequency and source placement
optimization analysis.
0.960
0.879
0.798
0.717
0.636 15 20
0.555 10 2
5 4
0.475 0
0.394 6
0.313 8
1.0
0.232 10
0.151 0.8
0.070 12
0.6 1.0
0.4 0.8
0.2 0.6
2
0.4
4
0.2
6
8 20
Frequency (Hz) 15
10
10
12 5
0 Source Position (Ft)
1.000
0.910
0.820
0.730 Frequency (Hz)
0.640 0 10 20 30 40
0.560 0 5 10 15 20 Source 50
0.460 0 20
0 Position (Ft) 15
0.320 F (ms) 10 10 20
0.280
0.190 (Hz) 20 20 10 15
0.100 30 30 5 10
0.010 40
40 0 5
50 50 1.0
1.0 1.0 0
1.0
0.5
0.5 0.5
0.5
0.0
0.0 0.0 0 10 20 0.0
0 5 10 15 20 30 40 50
Source Position (Ft)
2.00
1.82 OPTIMUM
1.64 Frequency (Hz) 20
00
80 1
1.46 15
1.28
1.10 60 10
0.91 80 1
00 40 5
0.73 Source Position 0 60 2.0 20
0.55 0 20 4 0 0
0.30 (Ft) 2 0 1.5
0.19 15 1.0 2.0 2.0
0.00 10
5 0.5
0 1.5 1.5
0.0
1.0 1.0
2.0
20 0.5
1.5 5 0.5
1 0.0
1.0 10 0.0
0
10
0.5
20
5 80
0.0 Source Position (Ft) 60
0 20 40 60 80 1000 0
40
Frequency (Hz)
20
0 0
frequency (left axis) and source position from the bottom (right axis). For
low frequencies less than 2 Hz, red zones indicate that optimal wave ampli-
tudes are always found whatever the source location. But at the 12 Hz used
in present siren designs, source positioning is crucial: the wrong location
can mean poor signal generation and, as can be seen, even “good locations”
can be bad. These calculations are repeated for upper limits of 50 Hz and
100 Hz in Figures 5.2c and 5.2d. In these diagrams, red means optimal
frequency-position pairs for hardware design and strong signal strength
entering the drillpipe.
That present drilling telemetry channels support much higher data rates
than siren operations now suggest, e.g., carrier waves exceeding 50 Hz, is
confirmed by independent research at Presco [4]. In our designs, we select
92 Reservoir Characterization
1
Pressure (psi)
0
100 200 300 400 500
–1 Time
–2
–3
Figure 5.3a Three step pulse recovery in noisy environment (pressure, vertical; time,
horizontal).
Strategies in High-Data-Rate MWD Mud Pulse Telemetry 93
Pressure (psi) 1
0
100 200 300 400 500
–1 Time
–2
–3
is shown in Figures 5.3a and 5.3b. Mudpump generated noise can be almost
completely removed. Experimental validations are given later.
A second method for f, leading to a time-delay difference equation, was
also designed and does not bear the (small Dz and Dt) limitations of deriv-
ative models. Updated results for both models in which the effects of trans-
ducer separation and sampling time are evaluated under typical surface
standpipe constraints are presented in detail later.
Untapered Rotor
Upstream Stator
Straight Stator Tapered Rotor
Figure 5.4a Early 1980s “stable closed’ siren (left) and improved 1990s “stable-opened”
downstream rotor design.
Gavignet, Bradbury and Quetier [8] who used the method to study flows
beneath drill bits nozzles. This counter-intuitive (but correct) approach to
modeling drilling muds provides a strategically important alternative to
traditional testing and reduces the time and cost of developing new MWD
systems. Several oil service company wind tunnel systems have since been
designed and built by Chin.
The CNPC MWD wind tunnel test facility in Beijing consists of two
components, a “short flow loop” where principal flow properties and tool
characteristics are measured, and a “long loop” (driven by the flow in the
short wind tunnel) designed for telemetry concept testing, signal process-
ing and noise removal algorithm evaluation. Field testing procedures and
software algorithms for tool properties and surface processing are first
developed and tested in wind tunnel applications and then moved effort-
lessly to the field for evaluation in real mud flows. This provides higher
efficiency than with “mud loop only” approaches.
Our “short wind tunnel,” housed at an off-campus site, is shown in
Figure 5.5a. This laboratory location was selected because loud, low-
frequency signals are not conducive to office work flow. The created signals
are as loud as motorcycle noise (typically exceeding 100 db) and require
hearing protection for long duration tests. More remarkable is the fact
that internal pipe pressures are “1,000” times louder than the waves that
escape – another 1,000 arises from the ratio of mud to air density. The com-
bined ratio implies that careful and precise acoustic signal measurement
is required to accurately extrapolate those to mud conditions. Similarly,
torques acting on sirens are 1,000 times lower. In fact, air-to-mud torque
scaling is simply proportional to the dynamic head “ρU2” ratio, where U
is the oncoming speed. Thus, wind tunnel tests can be run at lower speeds
with inexpensive blowers, provided a quadratic correction is applied for
downhole flow extrapolation. The turbine, similarly designed and tested, is
not discussed here. Wind tunnel testing allows effects like high pitch angle,
wide annular clearances, strong turbulence and unsteady transients to be
modeled accurately – effects common to MWD tools that would challenge
the best computational fluid dynamics models. Details appear in [2].
In Figure 5.5a, a powerful water-cooled blower (blue) with its own power
supply pumps more or less constant flow rate air regardless of siren rotor
blockage. A sensitive flow meter records average flow rate. Flow straighten-
ers ensure uniform flow into the siren and to remove downstream swirl for
accurate differential pressure measurement. The siren test section deserves
special comment. The motion of the rotor is governed by its own electrical
controller and is able to affect position-modulated motions as required for
telemetry testing.
98 Reservoir Characterization
For the simplest schemes, only two transducers are required; three allow
redundancies important in the event of data loss or corruption. Additional
(recorded) noise associated with real rigsite effects is introduced into the
wind tunnel using low frequency speakers or woofers.
Numerous siren concepts and shapes were evaluated. Several of the
sirens shown in this paper are not practical but were purposely designed
to be impractical; a broad range of data was accumulated to enhance our
fundamental understanding of rotating flows as they affect signal, torque
and erosion.
In our work, we re-evaluated conventional four-lobe siren designs and
developed methods that incrementally improve signal strength and reduce
torque. Results reinforced the notion that the technology has reached its
performance limits. Radically different methods for signal enhancement
and minimization of resistive torque were needed.
As noted earlier, constructive wave interference provides “free” signal
amplitude without erosion or power penalties. This is cleverly implemented
100 Reservoir Characterization
in two ways. First, FSK with alternating high-low amplitudes is used. High
amplitudes are achieved by determining optimal frequencies from three-
dimensional color plots such as those in Figures 5.2b, c, d. Design param-
eters include sound speed, source position and frequency, MWD collar
design, and whenever possible, drill-pipe inner diameter and mud density.
This information is used in the waveguide model of Figure 5.2a and also in
a model for non-Newtonian attenuation applicable over the length of the
drillpipe. Low amplitudes need not be achieved by bringing the rotor to a
complete stop. If a high-amplitude is associated with 60 Hz, then a useful
low-amplitude candidate can be found at 55 Hz, as suggested by Figures
5.2b, c, d. Thus, FSK can be efficiently achieved while minimizing the
effects of mechanical inertia. Rotor torque reduction, while an objective in
Strategies in High-Data-Rate MWD Mud Pulse Telemetry 101
Figure 5.5e Evaluation of hub convergence effects on signal strength and torque.
wind tunnel analysis, is useful but need not be the main design driver in
our approach since the rotor is never brought to a complete stop.
To make constructive wave interference work, the siren is located as
close to the most significant bottom reflector, normally the drillbit, as pos-
sible (intervening waveguides, e.g., mud motors, resistivity-at-bit subs, and
so on, do support wave transmission). Thus, the siren is placed beneath
the turbine in the MWD collar, in contrast to existing designs. Tests con-
firm that long waves pass effortlessly through turbines without reflection.
Detailed waveguide analyses suggest that signal gains of 1.5-2.0 are doable.
PSK methods, again, are undesirable because they cause wave cancella-
tions and ghost signals that hinder signal processing.
Additional signal enhancement is possible using constructive interference
of a different nature, specifically multiple sirens arranged in series or in tan-
dem. If the distance between sirens is small and siren apertures are properly
phased, signals will be additive. This idea was first proposed by the last author
102 Reservoir Characterization
Figure 5.5f Flow straighteners (PVC tubing) for upstream and downstream use.
found, to our amusement, that the large firecrackers used at Chinese wed-
dings, e.g., see Figure 5.5o, provide a useful source of low-frequency, plane-
wave noise when all else is unavailable.
Conventional siren designs are built with four lobes cut along radial
lines. Rotating sirens with additional lobes would surely increase fre-
quency or data rate, but large lobe numbers are associated with much lower
Dp signals. For this reason, they are not used in designs to the authors’
knowledge. Because constructive interference now enhances our arsenal
of tools against attenuation, we have been able to reassess the use of higher
lobe numbers. Downhole and uphole telemetry concepts are easily tested
in our wind tunnels.
Wind tunnel usage enables a scale of knowledge accumulation, together
with cost, time and labor efficiencies, not previously possible. Numerous
parameters can be evaluated, first by computational models and then by
testing in air. Design parameters include lobe number, stator and rotor
106 Reservoir Characterization
Figure 5.5m Piezoelectric transducer closest to siren for constructive interference and
harmonic generation study.
thicknesses, stator-rotor gap, rotor clearance with the collar housing, rotor
taper angles, and so on.
Tests are not limited to signal strength. Torque is important, as is the
ability to pass lost circulation material (LCM) – this is assessed by intro-
ducing debris at the upstream end of the short wind tunnel and observing
Figure 5.5o Fireworks for low frequency noise generation, when all else is unavailable.
108 Reservoir Characterization
A B C D E
1
2 12.69 208 ∆P
3 13.03 303 0.04
4 18.41 398
0.03
5 27.68 583
6 36.88 798 0.02
7 50.01 993
0.01
8 55.32 1200
HZ
9 71.4 1406 0
10 83.01 1580 10 15 20 30 40 50 60 70 80 90
11 89.5 1767
12 12.47 197 ∆P
13 16.82 294 0.02
14 25.05 390 0.015
15 27.67 609
16 36.88 803 0.01
17 49.88 1004
0.005
18 55.32 1216 HZ
19 68.95 1391 0
20 78.55 1606 10 15 20 30 40 50 60 70 80 90
21 92.22 1811
2
Series
1
Pressure (psi)
1
2
0 3
100 200 300 400 500 4
−1 Time
−2
−3
2
Series
1
Pressure (psi)
1
2
0 3
100 200 300 400 500 4
−1 Time
−2
−3
2
Series
1
Pressure (psi)
1
2
0 3
100 200 300 400 500 4
−1 Time
−2
−3
model now used to study typical MWD collars, e.g., see Figure 5.2a, is
more general and does not assume any particular reflection mechanism on
an a priori basis. Detailed calculations show that, more often than not, the
drillbit (because of its nozzles) actually acts as an open reflector – attesting
Strategies in High-Data-Rate MWD Mud Pulse Telemetry 111
to the dangers of “common sense” and visual inspection. This model cre-
ates plots similar to Figures 5.2b, c, d. The wave characteristics of siren and
positive pulsers from present MWD vendors are consistent with those in
Figures 5.2b.
If the sound speed is 5,000 ft/sec, the respective distances are 15 and 30
ft. All of these distances are doable on existing rig installations. Thus, we
ask, “Is it possible to design a good MWD echo cancellation and mud
pump noise removal algorithm that works under the restrictive environ-
ment described and performs quickly with a minimum of computation?”
Methods 4-3 and 4-4 for “directional filtering,” described in detail in [2]
are based on analytical wave equation properties. The former uses a differ-
ence equation delay formulation whereas the latter is hosted by a differential
equation for the upgoing MWD signal. The algorithms are completely dif-
ferent, but both fully eliminate propagating waves that travel in directions
opposite to the upgoing MWD signal. These can contain signal reflections
at positive displacement and/or centrifugal mud pumps, distorted reflec-
tions at the desurger, together with large-amplitude mud pump noise with
or without frequencies near those in the upgoing signal.
Information on downgoing waveform shape or amplitude is not neces-
sary. Application of these models is subject to requirements imposed by the
Nyquist-Shannon Theorem. In addition, the methods can be augmented
with damping to lessen earlier time effects and to improve computational
stability; processing times are almost instantaneous for the computations
with Intel i5 class chipsets. In practice, such “directional filters” are used
together with frequency, wavelet, white noise and other filters for robust
signal processor design. In this paper, the two methods are evaluated in
detail using synthetic datasets to examine their ability to remove all noise
traveling opposite to the upcoming signal. Both algorithms appear to work
well with large transducer separations and coarse time sampling – the
use of the two schemes together with multiple pressure transducer data
would offer redundancy that is likely to further minimize common syn-
chronization loss at the surface. The methods are also computationally effi-
cient, requiring few “multiplies and divides” compared to a typical FFT
calculation.
5.3.2 Theory
The ideas behind Method 4-3 are developed from wave equation properties.
The complete pressure disturbance is taken as the sum of two waves travel-
ing in opposite directions. Let “f ” denote the incident wave traveling from
downhole. Then, “g” will denote reflections of any type at the mudpump
(positive displacement pistons and centrifugal pumps are both allowed),
together with reflections at the desurger with any type of shape distortion
permitted, plus the mudpump noise itself. In general, we can write
Strategies in High-Data-Rate MWD Mud Pulse Telemetry 113
Subtraction yields
or
solves for f(t) given the pressure values on the right and is solved by our
2XDCR*.FOR code.
Method 4-3 is extremely powerful because it eliminates any and all
functions g(t + x/c), that is, all waves traveling in a direction opposite to the
upgoing wave. Thus, g(t + x/c) may apply to mudpump noise, reflections of
the upgoing signal at the mudpump, and reflections of the upgoing signal
at a desurger, regardless of distortion or phase delay, reflections from the
rotary hose connections, and so on. The functional form of the downgoing
waves need not be known and can be arbitrary. This is not to say that all
downward moving noise sources are removed. For example, fluid turbu-
lence noise traveling downward with the drilling fluid is not acoustic noise;
it will not be removed and may degrade performance. Additional noise
sources and filters would be used together with Method 4-3. The order
in which filters are applied will affect the outcome of any signal process-
ing, and it is this uncertainty that provides the greatest challenge in sig-
nal processor design. The model in Equation 5.7 can be solved exactly in
closed analytical form and is implemented in our 2XDCR*FOR software
series. We note that this method is similar to the two-transducer delay-line
approach of Foster and Patton [10] who first solved a model analogous to
Equation 5.7 for MWD applications using approximate frequency domain
analysis. Other ad hoc derivations have since appeared, which are solved
by undisclosed methods, but to the authors’ knowledge, none have proven
successful and there have been no other publicly available validations of
delay approaches except that presented here.
Method 4-4 is similarly developed from wave equation properties but
follows a strategy different from the delay equation approach. Our deriva-
tion at first follows from Chin [5], which gave solutions for the derivative
of the signal only. The present method, which includes a robust integrator
to handle sharp pressure pulses, substantially changes the earlier work. In
Method 4-3, we used time delayed signals for which there was no restric-
tion on time delay size. For Method 4-4, we invoke time and space deriv-
atives; thus sampling times should at least be small compared to a period
and transducer separations should be small compared to a wavelength.
As in Method 4-3, the representation of pressure as an undamped upgo-
ing and downgoing wave is still very general (attenuation between close
transducer locations is minimal), and all waves in the downward direction
are removed with no information required about the mudpump, the desur-
ger or the rotary hose. Note that “c” is the mud sound speed at the surface
and should be measured separately. In our derivation, expressions for time
and space derivatives of p(x, t) are formed, from which the downgoing
Strategies in High-Data-Rate MWD Mud Pulse Telemetry 115
wave “g” is explicitly eliminated, leaving the desired upgoing “f.” The steps
shown are straightforward and need not be explained.
high accuracy. The required processing in space and time is inferred from
the use of finite difference formulas in approximating the derivatives shown
and numerous such computational molecules are available in the numeri-
cal analysis literature. The illustrative calculations used below assume two
transducers and pressures stored at two time levels. Higher order process-
ing does not add substantially to computational or storage requirements.
5.3.3 Calculations
Method 4-3 uses a difference equation time delay equation with damping
while Method 4-4 employs a differential equation approach. Both methods
(1) require at least two (piezoelectric) transducers mounted on the surface
standpipe, (2) operate with minimal information on waveform and noise
properties, and (3) robustly function in the less-than-ideal environment
prevalent on typical drilling rigs. The surface speed of sound is required
and is easily obtained by clocking the transit time of a sharp pulse traveling
between the two transducers. The results below are reported together with
software references cited in [2]. For all the calculations given, we duplicate
relevant Fortran source code describing the assumed signals and noise in
order to clearly document the test methodology employed.
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
0.1 0.2 0.3 0.4 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
0.1 0.2 0.3 0.4 0.5 0.6 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
0.05 0.10 0.15 0.20 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
0.05 0.10 0.15 0.20 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
0.1 0.2 0.3 0.4 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
0.1 0.2 0.3 0.4 0.5 0.6 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 4
−1 Time (sec)
−2
−3
while blue shows a successfully extracted signal (using only green data)
which is almost identical to the black. Note that high-data-rate telemetry
requires closer transducer separations along with finer time samples than
in more conventional applications. As shown in our source code, the FSK
122 Reservoir Characterization
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
0.05 0.10 0.15 0.20 0.25 0.30 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
0.1 0.2 0.3 0.4 0.5 0.6 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
0.1 0.2 0.3 0.4 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1 1
Pressure (psi)
2
3
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
Pressure (psi)
1 1
2
0 3
0.1 0.2 0.3 0.4 0.5 0.6 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1 1
Pressure (psi)
2
0 3
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
100 200 300 400 500 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
100 200 300 400 500 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
100 200 300 400 500 4
−1 Time (sec)
−2
−3
Signal Deconvolution
2
Series
1
Pressure (psi)
1
2
0 3
100 200 300 400 500 4
−1 Time (sec)
−2
−3
Signal Deconvolution
3
2
Series
Pressure (psi)
1 1
2
0 3
100 200 300 400 500 4
−1 Time (sec)
−2
−3
frequencies used are 20 and 40 Hz. The Series B results assume single
non-sinusoidal pulses.
Run C-1
Internal MWD upgoing (psi) signal available as
P(x, t) = + 5.000 {H(x- 150.000-ct) - H(x- 400.000-ct)}
+ 10.000 {H(x- 600.000-ct) - H(x- 1000.000-ct)
+ 15.000 {H(x- 1400.000-ct) - H(x- 1700.000-ct)}
Units: ft, sec, f/s, psi ...
Assume canned MWD signal? Y/N: Y
Downward propagating noise (psi) assumed as
N(x, t) = Amplitude * cos {2╥ f (t + x/c)} ...
o Enter noise freq “f” (hz): 15
o Type noise amplitude (psi): 30
o Enter sound speed c (ft/s): 5000
130 Reservoir Characterization
Run C-2
Internal MWD upgoing (psi) signal available as
P(x, t) = + 5.000 {H(x- 150.000-ct) - H(x- 400.000-ct)}
+ 10.000 {H(x- 600.000-ct) - H(x- 1000.000-ct)}
+ 15.000 {H(x- 1400.000-ct) - H(x- 1700.000-ct)}
Units: ft, sec, f/s, psi ...
Assume canned MWD signal? Y/N: Y
Downward propagating noise (psi) assumed as
N(x, t) = Amplitude * cos {2╥ f (t + x/c)} ...
o Enter noise freq “f” (hz): 15
o Type noise amplitude (psi): 30
o Enter sound speed c (ft/s): 5000
o Mean transducer x-val (ft): 1750
o Transducer separation (ft): 10
Run C-3
Internal MWD upgoing (psi) signal available as
P(x, t) = + 5.000 {H(x- 150.000-ct) - H(x- 400.000-ct)}
+ 10.000 {H(x- 600.000-ct) - H(x- 1000.000-ct)}
+ 15.000 {H(x- 1400.000-ct) - H(x- 1700.000-ct)}
Units: ft, sec, f/s, psi ...
Assume canned MWD signal? Y/N: Y
Downward propagating noise (psi) assumed as
N(x, t) = Amplitude * cos {2╥ f (t + x/c)} ...
o Enter noise freq “f” (hz): 15
o Type noise amplitude (psi): 30
o Enter sound speed c (ft/s): 3000
o Mean transducer x-val (ft): 1750
o Transducer separation (ft): 30
Run C-4
Internal MWD upgoing (psi) signal available as
P(x, t) = + 5.000 {H(x- 150.000-ct) - H(x- 400.000-ct)}
+ 10.000 {H(x- 600.000-ct) - H(x- 1000.000-ct)}
+ 15.000 {H(x- 1400.000-ct) - H(x- 1700.000-ct)}
Units: ft, sec, f/s, psi ...
Assume canned MWD signal? Y/N: Y
Downward propagating noise (psi) assumed as
Strategies in High-Data-Rate MWD Mud Pulse Telemetry 131
Run C-5
Internal MWD upgoing (psi) signal available as
P(x, t) = + 5.000 {H(x- 150.000-ct) - H(x- 400.000-ct)}
+ 10.000 {H(x- 600.000-ct) - H(x- 1000.000-ct)}
+ 15.000 {H(x- 1400.000-ct) - H(x- 1700.000-ct)}
Units: ft, sec, f/s, psi ...
Assume canned MWD signal? Y/N: n
Number of rectangular pulses: 7
5.4 Conclusions
We have summarized our strategy for high-data-rate mud pulse telemetry
and means for developing the technology. Our target objective of 10 bits/
sec at 30,000 feet appears to be doable. The signal amplification approach
used, together with new surface signal processing techniques, plus the use
of specially designed tools that are integrated with mud and drillpipe prop-
erties, provide a systems oriented process that optimizes data transmission.
Needless to say, we have acquired much in our testing program, and we are
continually learning from our mistakes and developing new methods to
improve the technology.
Strategies in High-Data-Rate MWD Mud Pulse Telemetry 133
Acknowledgments
The project “Downhole High-Data-Rate Continuous Wave Mud Pulse
Telemetry” was sponsored by China National Petroleum Corporation’s
Technology Management Division under Contract 2008C-2101-1. The
authors thank the management of CNPC for permission to publish
this research and for its support and interest throughout the effort. We
also thank the University of Petroleum (East China), whose College of
Electromechanical Engineering provided test facilities for research exper-
iments, and Professor Jun Fang, Professor Xueshi Gao and Dr. Peng Jia,
who diligently contributed their expertise during numerous wind tunnel
test sessions.
References
1. A. Boyd, H. Darling, J. Tabanou, B. Davis, B. Lyon, C. Flaum, J. Klein, R.
Sneider, A. Sibbit, and J. Singer, The lowdown on low-resistivity pay. Schlumb.
Oilfield Rev. Autumn, 4–18, (1995).
2. W.C. Chin, Y. Su, L. Sheng, L. Li, H. Bian, and R. Shi, MWD Signal Analysis,
Optimization and Design, John Wiley & Sons, New Jersey (2014).
3. B.J. Patton, W. Gravley, J.K. Godbey, J.H. Sexton, D.E. Hawk, V.R. Slover,
and J.W. Harrell, Development and successful testing of a continuous-wave,
logging-while-drilling telemetry system. J. Petrol. Technol. 1215–1221 (1977).
4. Presco, Inc., Measurement while drilling, http://www.prescoinc.com/science/
drill-ing.htm (2011).
5. W.C. Chin, Multiple transducer MWD surface signal processing, US Patent
5969638, assigned to Halliburton (October 19, 1999).
6. W.C. Chin, MWD siren pulser fluid mechanics. Petrophysics 45. 363–379,
(2004).
7. W.C. Chin and J. Trevino, Pressure pulse generator, US Patent 4785300,
assigned to Schlumberger (November 15, 1988).
8. A.A. Gavignet, L.J. Bradbury, and F.P. Quetier, Flow distribution in a roller
jet bit determined from hot-wire anemometry measurements. SPE Dril. Eng.
19–26, (1987).
9. W.C. Chin, Measurement-while-drilling system and method, US Patent
5583827, assigned to Halliburton (December 10, 1996).
10. M.R. Foster and B.J. Patton, Apparatus for improving signal-to-noise ratio
in logging-while-drilling system, US Patent 3742443 assigned to Mobil Oil
Corporation (June 26, 1973).
6
Detection of Geologic Anomalies with
Monte Carlo Clustering Assemblies
Simon Katz1*, Fred Aminzadeh2, George Chilingar2, Leonid Khilyuk2
and Matin Lockpour1
2
Petroleum Engineering Program, School of Engineering, University of
Southern California, Los Angeles, CA
Abstract
Authors present new clustering-based algorithms for detection of geologic anom-
alies and results of their testing on the data containing anomalies of two types:
(a) high permeability anomaly with regular records containing smaller perme-
ability values, and (b) gas-filled sand anomaly with regular records containing
data from brine-filled sands. Results of algorithms testing, presented in the paper,
demonstrate high stability of anomaly detection with false discovery rate below
20% and with the true discovery rate exceeding 73%.
6.1 Introduction
The goal of this paper is to present new clustering assembly algorithms for
detection of geologic anomalies of various types. These algorithms may be
utilized as one of the steps in location of oil and gas reservoirs, overpres-
sure zones, and other geologic anomalies. List of publications on detection
of geologic anomalies includes finding the location of fractured carbonates
filled with gas [4], finding the location of an overpressure zone [5], and
one-class methodology for anomaly detection within the homogeneous
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (135–150) © 2022
Scrivener Publishing LLC
135
136 Reservoir Characterization
this usually does not happen, and both the train and the test sets may
be broken into several clusters. Another issue, that complicates anom-
aly detection clustering methodology, is instability of clustering process,
where minor change in the clustered data may lead to significant change
in the number of clusters and in their size. This is illustrated by Figure
6.1 that shows histograms of the number of clusters and their sizes in
repeated clustering of the dataset with records collected from brine-filled
sands areas. Clustering was performed on the same dataset with repet-
edly randomized reordering of the clustered records and removing five
records from the dataset. One can observe, that the dataset ‘brine’ is inho-
mogeneous containing clusters of different size, which number varies in
the wide range. Figure 6.1 also shows, that there are clusters as small as
one or two records. Therefore, using small cluster size as indicator of
geologic anomaly is problematic. To overcome the problem of clustering
instability and questional relations between cluster size and its potentional
anomaly, authors developed anomaly detection algorithms, that rely on the
use of clustering assemblies containing large number of individual cluster
sets. Each cluster set is generated by clustering the randomized test set, that
contains records, some of which labeled as regular and others - unlabeled
records. Clustering assembly is utilized to formulate criteria for identifica-
tion of anomalous cluster sets, clusters, and anomalous records.
Due to instability of individual clusters, anomaly detection, that rely
on the use of individual cluster set, produce unreliable results. This is
illustrated by Figure 6.2, that shows three clusters and ellipses, drawn
around clusters with confidence level 0.95 (R function dataEllipse).
Clustering was done on randomized test set, that includes anomalous
100
Frequency
Frequency
80
50
40
0
2 3 4 5 6 2 4 6 8 10 12 14
Figure 6.1 Histograms of the number of clusters and the number of records in individual
clusters (cluster size) in the dataset containing records from brine-filled sands.
138 Reservoir Characterization
0.40
Anomaly
Regular
0.30
Grain.size
Regular
0.20
0.10
5 10 15 20 25
Porosity
Figure 6.2 Clustering of a randomized test set that includes regular and anomalous
records.
and regular permeability data. Two of the clusters are regular and one is
anomalous. One can observe, that cluster ellipses overlap each other and
poorly separated.
210
BRINE GAS
Number of presence
in different clusters
190
170
150
0 10 20 30 40 50
Records
test set includes 15 randomly selected records from gas set and the same
number of records from the brine set. Number of Monte Carlo runs is 300.
One can observe, that the number of appearances for any individual
record exceeds 160. So large number of appearances of individual records
in different clusters of the clustering assembly opens the way for building
stable and reliable procedures for identification of anomalous records.
n.unlabeled( j, r ) (6.1)
Irreg ( j, r ) =
n.cluster( j, r )
According to Eq. 6.3, index of anomaly presence in the cluster set satis-
fies the following constraints:
If irregularity index is the same for all clusters in the cluster set, then:
On the other hand, if there is at least one cluster, that contains only
labeled records from the train set and at least one cluster containing only
unlabeled records, then:
J (m )
rAnom(m) =
1
J (m) ∑ Irreg (m, j,r ) (6.8)
j =1
where m is the index of the record, r is the index of the cluster containing
record with the index m at Monte Carlo run with index j, J(m) is the total
number of appearances of record with index m in the clustering assembly.
Similarly, anomaly index of the whole clustering assembly is defined as:
anS =
1
J ∑anClset( j) (6.9)
j =1
where tr is the threshold, defined using prior false discovery rate (section
6.6).
142 Reservoir Characterization
Irregularity index
8
6
4
2
0
Anomaly index of individual clusters
Percent of total
8
6
4
2
0
Claster sets anomaly index
8
6
4
2
0
0.0 0.2 0.4 0.6 0.8 1.0
Values of anomaly parameters
As shown in the Table 6.1, the prior false discovery rates for anomalous
cluster sets and individual clusters in the cluster set drop below 0.01 at the
threshold equal to 0.35.
144 Reservoir Characterization
12
10
8
6
4
2
0
Irregularity index
12
10
8
6
4
2
0
0.0 0.2 0.4 0.6 0.8 1.0
Histograms of three anomaly indexes
Anomaly index
Anomalous permeability
1.2
of records
0.6
0.0
Records
Anomaly Index
Regular permeability
0.8
of records
0.4
0.0
Records
400
800
400
600
300
300
Frequency
Frequency
Frequency
400
200
200
100
200
100
0
0
0.0 0.4 0.8 0.0 0.4 0.8 0.0 0.4 0.8
cl.set.anom cl.anom irr.prior
250
Cluster set Cluster Irregularit y
250
anomaly anomaly
300
200
250
200
200
150
Frequency
Frequency
Frequency
150
150
100
100
100
50
50
50
0
Brine
0.6
Anomaly index
0.4
of records
0.2
Gas
0.0
0 10 20 30 40 50
Records
Figure 6.9 Values of anomaly index of individual records in gas-sand and brine-sand
datasets.
Table 6.3 Thresholds, true, and false discovery rates of anomalous records
using three independently generated clustering assemblies. 300 Monte
Carlo runs.
Quantile
probabilities Thresholds False discoveries True discoveries
1 2 3 1 2 3 1 2 3
0.6 0.32 0.32 0.34 0.32 0.36 0.36 0.8 0.84 0.96
0.7 0.35 0.34 0.36 0.24 0.2 0.32 0.8 0.8 0.8
0.8 0.38 0.36 0.38 0.2 0.2 0.2 0.8 0.8 0.8
0.9 0.38 0.38 0.39 0.04 0.04 0.04 0.8 0.8 0.8
0.95 0.38 0.38 0.39 0.04 0.04 0.04 0.8 0.8 0.8
According to Table 6.3, values of true and false discovery rates obtained
using three independently generated clustering assemblies show minor
differences. For all three clustering assemblies, true discovery rates not
smaller 0.8, with false discovery rates as low as 0.04.
Detection of Geologic Anomalies with Monte Carlo 149
6.10 Notations
Gas set - dataset that contains records from gas-filled sands, brine set
- dataset that contains records from brine-filled sands.
TD and FD - true and false discovery rates.
j is the index of the cluster set formed via clustering randomized test
set, r is the index of the cluster in the cluster set with index j. Thus,
each cluster is indexed by pair of indexes (r, j).
Irreg(j, r) - irregularity index of the cluster with index r within cluster
set with the index j.
n.unlabeled(j, r) and n.cluster(j, r) are the numbers of unlabeled and
the total number of records in the cluster.
AnClset(j) - index of anomaly presence in a cluster set sCl(j).
AnC(j, r) - anomaly index of individual cluster.
rAnom(m) anomaly index of individual record.
J(m) is the total number of appearances of the record with index m in
the clustering assembly.
TD(tr), FD(tr) - posterior true and false discovery rates, fr-threshold
in anomaly detection rules.
6.11 Conclusions
• Authors present new algorithms for detection of geologic anom-
alies and results of their testing. Key element of the developed
algorithms is construction of multiple randomized test sets, and
construction of multiple cluster sets, that form clustering assem-
bly. The clustering assembly is used for calculation of irregularity
index of individual records and anomaly indeces for each cluster
set, each cluster, and individual records in the test and train sets.
• The algorithms were tested on the data with anomalies of two
types: (a) high permeability anomaly with regular records con-
taining smaller permeability values, and (b) gas-filled sand anom-
aly with regular records containing data from brine-filled sands.
Results of algorithms testing, presented in the paper, demonstrate
high stability of anomaly detection with true discovery rate higher
than 73% with false discovery rates equal or even lower than 0.2
for both anomaly types.
150 Reservoir Characterization
References
1. Aase, N., Bjorkum, P., and Nadeau, P., 1996, The effect of grain-coating
microquartz on preservation of reservoir porosity. Bull. Am. Assoc. Petrol.
Geologists 80, (1): 1654-1673.
2. Aminzadeh, F. and Chatterjee, S., 1984, Applications of cluster analysis in
Exploration Seismology, Geoexploration, 23: 147-159.
3. Charrad M, Ghazzali N., Boiteau V., Niknafs A., 2014. NbClust: An R
Package for determining the relevant number of clusters in a data set. Journal
of Statistical Software, 61 (6): 1-36.
4. Chilingar G., Mazzulo S., Rieke, H., 1992. Carbonate Reservoir
Characterization: A Geologic-engineering Analysis. Elsevier, 639 pp.
5. Dvorkin J, Mavko G., Nur A., 1999, Overpressure detection from com-
pressional and shear-wave data. Geophysical Research Letters, 26, (22):
3417–3420.
6. Gurevich A., Chilingar G, and Aminzadeh F., 1994. Origin of the formation
fluid pressure distribution and ways to improving pressure prediction meth-
ods. J. Pet. Sci. Eng., 1: 67-77.
7. Katz S., Aminzadeh F., Chilingar G., Khilyuk L., 2015. Anomaly detection
within homogeneous geologic area. J. of Sustainable Energy Engineering, 3,
(2): 169-186.
8. Lian Duan, Lida Xu, Ying Liu., 2008. Cluster-based outlier detection. Annals
of Operations Research, 168, (1): 151–168
9. Ramos, A. and Castagna, P., 2001. Useful approximations for converted-wave
AVO. Geophysics, 66: 1721–1734.
10. Zimek, A., Campello, R., Sander, J.. 2014. Ensembles for unsupervised outlier
detection. ACM SIGKDD Explorations Newsletter. 15: 11–22.
11. Zengyou He, Xiaofei Xu, Shengchun Deng. 2003. Discovering cluster-based
local outliers. Pattern Recognition Letters, 24(9-10):1641-1650.
7
Dissimilarity Analysis of Petrophysical
Parameters as Gas-Sand Predictors
Simon Katz1*, George Chilingar2, Fred Aminzadeh2 and Leonid Khilyuk1
Abstract
Multiple petrophysical parameters, whose values may be used as indicators of gas
presence in sands and sandstones, are analyzed here. Bulk and shear moduli, Lame
parameter λ, Vp/Vs, and Poisson’s ratio are among parameters subjected to com-
parative analysis as potential gas indicators. Rock density ρ, product ρ*λ and ratio
of Lame Parameter λ to shear modulus were also tested as potential gas indicators.
Additionally, two one-dimensional aggregated parameters constructed as combi-
nations of several parameters and multidimensional parameters were tested using
machine learning techniques and ROC curve analysis.
Ranking parameters and evaluation of their efficiency was based on two meth-
ods: (1) Dissimilarity-based method independent of classification technique and
(2) ROC curve analysis specific to individual classification methodology. Both
types of methods work in terms of true and false discovery rates. The parameter
built as a combination of five other parameters (Lame parameter λ, Vp to Vs ratio,
velocities of compressional and shear waves, and rock density) get high rankings
using both methods. This parameter demonstrates a higher true discovery rate
and a limited false discovery rate in identification of gas-sands versus brine-sand
or shale. The parameters Poisson’s ratio, Vp/Vs, bulk over shear modulus, Lame
parameter λ over shear modulus also rank highly.
Cross validation was used to control over classification with undue increase
of estimate of true discovery rate. Classification with most efficient parameters of
gas-sand vs. brine-sand or shale leads to the true discovery rate exceeding 80%
and a false discovery rate not exceeding 20%.
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (151–168) © 2022
Scrivener Publishing LLC
151
152 Reservoir Characterization
7.1 Introduction
Evaluation of petrophysical parameters as classifiers of lithology and fluid
content is presented in multiple publications. Wide attention was given to
Vp/Vs ratio as a possible lithology classifier. Garotta (1987), replace with
Ramos and Castagna (2001), Margrave et al. (1998) and Stewart demon-
strated that changes in lithology can lead to a change in the Vp/Vs. Picket
(1963) determined Vp/Vs values for sandstones, dolomites and limestones.
Goodway (1997, 2001) suggested that the parameters derived from Lame
parameters (λ * ρ, μ* p and λ/μ) may be used for identification of gas-sands.
Ostrander (1984) studied the effect of changing Poisson’s Ratio on reflec-
tion coefficients of seismic waves.
The goal of this paper is development and testing of methodology for
lithologic and fluid content classification based on the use of multiple pet-
rophysical parameters. The authors performed comparative analysis of
efficiency of multiple petrophysical parameters as potential indicators of
changing lithology and gas saturation. Advancement in this direction was
achieved with construction of the dissimilarity measures and analysis of
aggregated parameters built as combinations of several individual parame-
ters. In addition, evaluation of the false and true discovery rates with ROC
curve analysis improves the reliability of estimation of efficiency of gas
indicators for both individual and aggregated parameters.
4
K+ µ
Vp = 3 = λ + 2µ ; V = µ (7.1)
s
ρ ρ ρ
Dissimilarity Analysis of Petrophysical Parameters 153
µ = ρ ∗ Vs2 (7.2)
4 (7.3)
K = ρ ∗ Vp2 − Vs2
3
Vp2 4
K/µ= − (7.5)
Vs2 3
ρ ∗ (Vp2 − 2 ∗ Vs 2 ) Vp2
λ/µ= = 2 −2 (7.6)
ρ ∗ Vs2 Vs
δ=
(Vp / Vs ) − 2
2
(7.7)
2 ∗ (Vp / Vs ) − 1
2
λ*ρ (7.8)
μ*ρ (7.9)
K = λ + 2 * μ / 3 (7.10)
2.5
Gas vs Shale & Brine
Gas vs Brine
Gas vs Shale
2.0
Shale vs. Brine
1.5
Dissimilarity
1.0
0.5
–0.5 0.0
2 4 6 8 10 12 14
1.5
0.0
3 4 5 6 7 8 9
Group_A
8
4
0
3 4 5 6 7 8 9
Group_A
Figure 7.2 Histograms of the values of the Group_A parameter calculated for gas-sand
and brine-sand or shale.
4
Frequency
2
0
3 4 5 6 7 8 9
Group_A
4
2
0
3 4 5 6 7 8 9
Group_A
Figure 7.3 Histograms of the Group_A parameter calculated for brine-sand and shale.
analysis shows a true discovery rate as a function of the rate of false discov-
ery for a given classification rule. Area under ROC curve (AUC) charac-
terizes performance of a classification procedure. The larger the AUC the
better is the performance of classification algorithm. Another advantage of
ROC curve analysis is its ability to adjust classification rule to guarantee an
appropriate level of false discovery rate.
We used the leave one out cross validation technique to avoid producing
over optimistic results. This technique is performed in four steps:
a. Remove a record from the input data set and optimize pre-
dicting model onthe remaining records.
b. Make prediction for removed record with optimized pre-
dictive model.
c. Return the record back to the data set.
d. Repeat steps a to c with another record.
e. Go through all records in the data set repeating steps a-d.
1.0
0.8
True Discovery Rate
0.6
0.4
Group_A
Poisson’s ratio
0.2
Group_B
Vp/Vs
Figure 7.4 ROC curves for four high performing parameters. Method: LDA.
Table 7.2 AUC values for four classification methods and 14 predictor
parameters.
Parameter Mean
rank Parameter RF GLM LDA KNN AUC
1 Group A 0.92 0.89 0.89 0.91 0.90
2 Group B 0.81 0.89 0.89 0.84 0.86
3 Poisson’s ratio 0.78 0.88 0.88 0.84 0.85
4 Vp/Vs 0.78 0.88 0.87 0.84 0.84
5 λ 0.86 0.83 0.84 0.84 0.84
6 λ/μ 0.78 0.88 0.85 0.85 0.84
7 K/μ 0.78 0.88 0.85 0.85 0.84
8 ρ* λ 0.83 0.83 0.83 0.79 0.82
9 K 0.67 0.70 0.70 0.74 0.70
10 ρ 0.46 0.67 0.67 0.56 0.59
11 Vs 0.58 0.50 0.50 0.49 0.52
12 Vp 0.55 0.49 0.49 0.53 0.52
13 μ 0.36 0.41 0.41 0.34 0.38
14 ρ*μ 0.24 0.28 0.28 0.28 0.27
2 4 6 8 10 12 14
Parameter’s Rank
Figure 7.5 Mean area under ROC curve for 14 parameters. Area under Rock Curve
averaged over four machine learning techniques. Horizontal axis shows parameters’ ranks
given in the Table 7.2.
Dissimilarity Analysis of Petrophysical Parameters 163
Gas-Sand
Brine-Sand or Shale
–0.1
3 4 5 6 7 8 9
Group_A
Figure 7.6 Group_A vs. Poisson’s ratio cross section of gas sand vs. brine sand or shale.
164 Reservoir Characterization
20
15
Lambda
10
5
Gas-Sand
0
Brine-Sand or Shale
Figure 7.7 Vp/Vs versus λ cross section of gas-sand vs. brine-sand or shale.
0.8
True Dsaicovery Rate
0.6
0.4
0.2
LDA
KNN
Figure 7.8 ROC Curves for KNN and LDA classification techniques for 2D predictor
(Group_A, Poisson’s ratio).
discovery rate with limited rate of false discovery is the result of high sepa-
ration of the values of this predictor shown in the Figure 7.6.
7.8 Conclusions
1. Fourteen petrophysical parameters were analyzed as poten-
tial predictors of gas presence in sand formations. Vp to
Dissimilarity Analysis of Petrophysical Parameters 165
References
1. Aminzadeh F., Dasgupta S., 2013. Geophysics for Petroleum Engineering.
Elsevier, 282 pp.
2. Buryakovsky, L., Chilingar G., Rieke H., and Shin S., 2012. Fundamentals of
the Petrophysics of Oil and Gas Reservoirs. Wiley. 314 pp.
3. Garotta, R.J. and Granger, P.Y., 1987, Comparison of responses of compres-
sional and converted waves on a gas sand. 57th Ann. Internat. Mtg., Soc.
Expl. Geophys., Expanded Abstracts; 627-630.
4. Goodway, W., Chen, T., and Downton, J., 1997. Improved AVO fluid detec-
tion and lithology discrimination using Lamé petrophysical parameters;
“Lambda-Rho”, “Mu-Rho”, & “Lambda/Mu fluid stack”, from P and S inver-
sions. from P and S inversions: 67th Ann. Internat. Mtg., Soc. Expl. Geophys.,
Expanded Abstracts; 183-186.
5. Goodway W. 2001. AVO and Lame’ constants for rock parameterization and
fluid detection. Recorder, 26; 39-60.
6. Dumitrescu C., 2009. Case study of a heavy oil reservoir interpretation using
Vp/Vs ratio and other seismic parameters. SEG Annual Meeting, October
25-30, Houston, Texas.
7. Dvorkin, J. 2008. Can gas sand have a large Poisson’s ratio?. Geophysics
Volume 73 (Issue 2).
8. Everett, M., 2013. Near-Surface Applied Geophysics., Cambridge University
Press. 400pp.
Dissimilarity Analysis of Petrophysical Parameters 167
9. Gardner, G.H.F., Garner, L.W., and Greogory, A.R., 1974, Formation velocity
and density – The diagnostic basics for stratigraphic traps: Geophysics, 39;
770-780.
10. Margrave, G. F.,Lawton, D. C. and Stewart, R. R., 1998,. Interpreting channel
sands with 3C-3D seismic data: The leading edge, 17 (04); 509-513.
11. Murphy R.J. and Monteiro S.T., 2012. Evaluating classification techniques
for mapping vertical geology using field-based hyperspectral sensors;
Geoscience and Remote Sensing, IEEE Transactions. 50, (3066–3080)
(Issue: 8).
12. Ostrander W., 1984. Plane-wave reflection coefficients for gas sands at non
normal angles of incidence. Geophysics, 49; 10 (1637–1648).
13. Pickett, G.R., 1963. Acoustic character logs and their applications in forma-
tion evaluation. Journal of Petrol. Technol., 15, 659-667.
14. Ramos A., and Castagna J., 2001. Useful approximations for converted-wave
AVO. Geophysics, 66 (6). 1721–1734.
15. Sing T, et al., ROCR: visualizing classifier performance in R. Bioinformatics.
2005, Oxford Journals V. 21(20), 3940-3941.
8
Use of Type Curve for Analyzing
Non-Newtonian Fluid Flow Tests
Distorted by Wellbore Storage Effects
Fahd Siddiqui* and Mohamed Y. Soliman
Abstract
Objective: This study reconsiders the problem of wellbore storage and skin dis-
tortion for the transient flow of non-Newtonian power law fluids through porous
media injection tests. A better method of pressure analysis data from such tests is
derived which corrects some mathematical inconsistencies present in the litera-
ture. Methodology: Because the problem being considered results in non-linear
boundary conditions, Laplace transform and other solution devices cannot be
used to obtain analytical solutions. Hence the finite element method was chosen
to solve the problem numerically. This study uses the finite element method to
generate type curves for the partial differential equation resulting from the flow of
non-Newtonian power law fluids through porous media and the associated nonlin-
ear boundary condition that accounts for the wellbore storage distortion and skin
effects. Results and conclusions: The mathematical reasons for the inaccuracies in
the previously presented solutions are described. A corrected step by step analysis
procedure for obtaining the formation properties, skin factor and the associated
wellbore storage constant is described using type curves. The application is aug-
mented by one real field test data, and three simulated test data examples.
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (169–194) © 2022
Scrivener Publishing LLC
169
170 Reservoir Characterization
8.1 Introduction
Pressure data analysis is routinely performed on injection and production
tests to determine the reservoir properties such as permeability, and well-
bore properties such as skin and wellbore storage. This problem becomes
complicated if injected fluid is a non-Newtonian power law liquid. Correct
knowledge and evaluation of these properties is essential to develop the
petroleum reservoirs in the most efficient manner and to sustain an opti-
mum recovery of energy resources.
The flow of the non-Newtonian fluids has been successfully described
by previous researchers (Odeh & Yang, 1979; Ikoku & Ramey, 1979;
Siddiqui, Soliman, & House, 2014). Other authors (Ikoku & Ramey, 1980;
Vongvuthipornchai & Raghavan, 1987; Olarewaju, 1992; Igbokoyi & Tiab,
2007) have also described the wellbore storage and skin distortion.
The problem of wellbore storage and skin distortion is being reconsid-
ered here because the subject needs an effective mathematical treatment
to generate correct solutions. Previously presented solutions have mathe-
matical inconsistencies that must be addressed. The mathematical reasons
for those inaccuracies in the previously presented solutions are described
below.
1
∂p ∂p n
CD wD − D =1 (8.1)
∂t D ∂rD rD =1
1
∂p n
pwD = pD − s D (8.2)
∂rD rD =1
1E + 2
pwD and tD.dpwD/dtD
1E + 1
1E + 0
1E + 2
pwD and tD.dpwD/dtD
1E + 1
1E + 0
coefficient in that manner, then they had to consider placing the ‘n’ in
the denominator of the boundary condition (equation 8.16 in the refer-
ence) to ensure algebraic compatibility. Admittedly the authors assumed
a steady state viscosity profile in the boundary condition which yields
a good agreement only for n>0.8. For other values of n, the deviation
is ‘large’ (Vongvuthipornchai & Raghavan, 1987) in the transition zone
from storage dominated to radial flow. This is mainly because for values
of n closer to unity, the non-linearity is diminished. They showed that by
assuming in Equation 8.3, their solution is identical to the Laplace space
solution in Equation 8.4.
1
−1
µ ∗ ∂ pD n
= (8.3)
µa ∂rD
2 2
K 1− n z + s zK 2 z
1
3− n 3 − n
3− n 3 − n
pwD =
z 2 2 2
zK 2 z + C D z K 1− n z + s zK 2 z
3−n
3− n 3 − n
3− n 3 − n
3− n
(8.4)
∂ pwD ∂ pD
CD − =1 (8.5)
∂t D ∂rD rD =1
∂p
pwD = pD − s D (8.6)
∂rD rD =1
Even though the inversion of Equation 8.4 retains certain features of the
solution (for example, unit slope line and long-time solutions are correct),
the transition between the two flow regimes would be incorrect (Figure 8.1
and 8.2). Since both the studies were aiming to generate type curves, the
Analyzing Non-Newtonian Fluid Tests 173
8.2 Objective
The objective of this study is to use numerical methods to solve the prob-
lem of non-Newtonian fluid flow through porous media in a correct way
and devise a step by step analysis methodology using type curves for esti-
mating formation permeability, wellbore storage and skin factors from
injection pressure test data. The problem is defined in the following sec-
tions as a partial differential equation (Equation 8.11) in dimensionless
form along with the associated boundary conditions (Equations 8.1 and
8.2). The resulting solution is presented as type curves that can be used
by the step by step analysis methodology which gives reliable formation
properties because it removes the mathematical inconsistencies described
above.
pwf − pi
pwD = n 1−n (8.7)
q rw
2π h λeff
174 Reservoir Characterization
t
tD = n −1 (8.8)
q rw3−n
nϕ ct
2π h λeff
C
CD = (8.9)
2π nhφct rw2
k
λeff = (8.10)
µeff
Odeh and Yang (Odeh & Yang, 1979) and Ikoku and Ramey (Ikoku &
Ramey, Transient Flow of Non-Newtonian Power-Law Fluids in Porous
Media, 1979) independently developed similar partial differential equa-
tions and proposed similar analysis techniques for practical well test anal-
ysis. Equation 8.11 is the partial differential equation proposed by Ikoku
and Ramey:
∂ 2 pD n ∂ pD ∂p
2 + = rD1−n D (8.11)
∂rD rD ∂rD ∂t D
This study uses the equation above along with proper boundary condi-
tions to generate the correct solutions (in form of type curves) for analyz-
ing tests distorted by wellbore storage.
∂ pD
∂r = −1 (8.12)
D rD =1
2
K 1−n z
3− n 3 − n
pwD ( z ) = (8.13)
2
z 3/2 K 2 z
3 − n
3− n
tD
pwD = (8.14)
CD
After the wellbore storage effects have diminished completely, the gra-
dient of pressure across the sand face would become negative unity - hence
the inner boundary condition would reduce to Equations 8.12. Therefore,
the solution would reduce to Equation 8.13 after the wellbore storage
effects have diminished completely.
1 K 0 ( z ) + s z K1 ( z )
pwD ( z ) = (8.15)
K1 ( z ) + CD z K 0 ( z ) + s z K1 ( z )
3/2
z
This verifies the presented type curves against known existing solutions.
dpwD dp
= t D wD = t D pD ′ (8.16)
d ln t D dt D
3m − 1
n= (8.17)
m −1
Based on this value, the nearest type curve can be selected (from
Appendix B: Type Curve Charts for Various Power Law Indices). If, how-
ever long time data are not available, the value of n obtained from labora-
tory measurements can be used with caution. If a satisfactory match is not
obtained with the laboratory value, a different value of n should be used
178 Reservoir Characterization
2.4
2.3
p' = 58.11 t0.1648
R2 = 1
2.2
2.1
2.5 2.75 3 3.25 3.5
Log [t (mins)]
to get a better match on type curve. This is because the in-situ value of n
might be different from the one obtained under laboratory conditions.
q t
CD = (8.18)
2π nhφct rw pwf − pi USL
2
If a unit slope line is not present, then C (hence CD) can be calculated
from well-bore properties, but it may not be representative of the test con-
ditions. In any case the CD=0 curve can also be used to analyze the data free
from wellbore storage distortion, without calculating CD.
shifting along the pre-determined CD USL (if it exists) and matching the
derivative first. Once this is accomplished, the nearest skin curve can be
selected from the (pwf-pi) and pwD match. Interpolation or graphical esti-
mation of skin is also recommended. Selecting the nearest skin value is
recommended only if the curves are undistinguishable.
n
q 1−n pwD
λeff = r (8.19)
2π h w pwf − pi
MP
n −1
q tD
λeff = nϕ ct rw3−n (8.20)
2π h t MP
n
H 3n + 1 1−n
µeff = 3 (150 kϕ ) 2 (8.21)
12 n
2
n
H 3n + 1 1−n 1+n
k = λeff 3 (150kϕ ) 2 (8.22)
12 n
1E + 2
pwD and tD.dpwD/dtD
1E + 1
resulting match. Table 8.2 summarizes the results from three analyses.
Because of the differences in the transition zone (as described above)
the resulting CD values are different. However, since we used the correct
(non-linear) boundary condition, the CD value presented in this study
is correct.
pwD 5
pwf − pi = 1.69 × 10 (8.23)
MP
rw 6.09E-02 m
h 15.24 m
q 3.68E-04 m3/s
H 0.0314 Pa.sn
pwD 4
pwf − pi = 9.48 × 10
MP
184 Reservoir Characterization
1E + 2
1E + 1
pwD 5
pwf − pi = 9.40 × 10
MP
Figure 8.6 and Table 8.6 show the match and results respectively.
1E + 2
1E + 1
1E + 0
1E – 1
1E + 0 1E + 1 1E + 2 1E + 3 1E + 4 1E + 5 1E + 6
tD
1E + 2
pwD and tD.dpwD/dtD
1E + 1
1E + 0
1E – 1
1E + 1 1E + 2 1E + 3 1E + 4 1E + 5 1E + 6 1E + 7
tD
CD 10 1.04E+01
s 10.0 10.0
188 Reservoir Characterization
pwD 5
pwf − pi = 3.48 × 10
MP
Figure 8.7 and Table 8.8 show the match and results respectively.
8.7 Conclusion
The problem was reconsidered for wellbore storage effects on the flow of
non-Newtonian power law fluids through porous media to correct some of
the inaccuracies present in the literature. A new analysis methodology was
developed, based on type curves, which gives reliable results and its appli-
cation was demonstrated for a few cases. We stress that this new method
be used for analyzing power-law fluid flow pressure tests to obtain accurate
results.
Nomenclature
ct Total compressibility, Pa-1
C Wellbore storage coefficient, m3/Pa
CD Dimensionless wellbore storage coefficient
h Formation thickness, m
H Consistency (power law parameter), Pa.sn
k Permeability, m2 [1 md=9.86923x10-15 m2]
Kv Modified Bessel function of the second kind of order v
n Flow behavior index (power law parameter)
pD Dimensionless pressure drop, non-Newtonian fluid
pi Initial Pressure, Pa
pwD Dimensionless pressure drop at wellbore, non-Newtonian fluid
pwD’ Dimensionless wellbore pressure log-derivative, non-Newtonian fluid
pwf Flowing wellbore pressure, Pa [1 psi=6894.76 Pa]
q Flow rate, m3/s
rD Dimensionless radial distance
rw Wellbore radius, m
s van Everdingen-Hurst skin factor
tD Dimensionless time, non-Newtonian
USL Unit Slope Line
z Laplace parameter
Analyzing Non-Newtonian Fluid Tests 189
References
R.B. Bird, W.E. Stewart, and E.N. Lightfoot, Transport Phenomena, John Wiley &
Sons Inc, New York City (1960).
A.O. Igbokoyi and D. Tiab, New type curves for the analysis of pressure transient
data dominated by skin and Wellbore Storage—Non-Newtonian fluid. SPE
(2007, January).
C. Ikoku, Transient Flow of Non-Newtonian Fluids in Porous Media, Stanford
University (1978).
C. Ikoku, Practical application of non-Newtonian transient flow analysis. SPE
(1979).
C. Ikoku and H.J. Ramey, Transient flow of non-Newtonian power-law fluids in
porous media. SPE Journal (1979).
C. Ikoku and H.J. Ramey, Well bore storage and skin effects during the transient
flow of non-Newtonian power-law fluids in porous media. SPE Journal
(1980).
A.S. Odeh and H.T. Yang, Flow of Non-Newtonian power-law fluids through
porous media. SPE Journal (1979, June).
J.S. Olarewaju, A reservoir model of Non-Newtonian fluid flow. SPE (1992).
J.G. Savins, Non-Newtonian flow through porous media. Industrial and Engineering
Chemistry, 18–47 (1969, October).
F. Siddiqui, M.Y. Soliman, and W. House, A new methodology for analyzing Non-
Newtonian fluid flow tests. J. Petrol. Sci. Eng. 173–179 (2014, December).
S. Vongvuthipornchai and R. Raghavan, Well test analysis of data dominated
by storage and skin: Non-Newtonian power-law fluids. SPE Formation
Evaluation (1987, December).
∫
p(z) = L ( p(t) ) = e ez p(t)dt
0
190 Reservoir Characterization
1
∂ pwD ∂ pD n
L CD − = L(1)
∂tD ∂tD rD =1
1
∂ pwD ∂ pD n
L CD − L = L(1)
∂tD ∂rD rD =1
1
∂ pwD ∂ pD n
CD L − L = L(1)
∂t D ∂rD r =1
D
1
dpD n 1
CD zpw ( z ) − =
drD rD =1 z
This is because
1
1
∂ pD n ∂ pD n
L ≠ L
∂rD rD =1 ∂rD rD =1
Since
1
1
∞ 1
n ∞ ∂p n dp n
1
∂ pD n ∂ pD
L
∫
= e
∫
− zt zt D D
dt ≠ e dt =
∂rD rD =1 ∂rD rD =1 ∂rD drD rD =1
0
0
Analyzing Non-Newtonian Fluid Tests 191
1E + 2
s = 30
pwD and tD.dpwD/dtD
s = 10
1E + 1
s=5
s=0
1E + 0
C = 10
D CD = 102 CD = 103 CD = 104
1E – 1
1E + 0 1E + 1 1E + 2 1E + 3 1E + 4 1E + 5 1E + 6 1E + 7 1E + 8 1E + 9
tD
1E + 2
s = 30
pwD and tD.dpwD/dtD
s = 10
1E + 1
s=5
s=0
1E + 0
C = 10
D D C = 102 C = 103 C = 104
D D
1E – 1
1E + 0 1E + 1 1E + 2 1E + 3 1E + 4 1E + 5 1E + 6 1E + 7 1E + 8 1E + 9
tD
Figure B2 Pressure and pressure derivative type curve for n=0.4.
192 Reservoir Characterization
1E + 2
s = 30
pwD and tD.dpwD/dtD
s = 10
1E + 1
s=5
s=0
1E + 0
C = 10
D DC = 102 C = 103 C = 104
D D
1E – 1
1E + 0 1E + 1 1E + 2 1E + 3 1E + 4 1E + 5 1E + 6 1E + 7 1E + 8 1E + 9
tD
Figure B3 Pressure and pressure derivative type curve for n=0.5.
1E + 2
s = 30
pwD and tD.dpwD/dtD
s = 10
1E + 1
s=5
s=0
1E + 0
C = 10
D DC = 102 C = 103 C = 104
D D
1E – 1
1E + 0 1E + 1 1E + 2 1E + 3 1E + 4 1E + 5 1E + 6 1E + 7 1E + 8 1E + 9
tD
Figure B4 Pressure and pressure derivative type curve for n=0.6.
Analyzing Non-Newtonian Fluid Tests 193
1E + 2
s = 30
s = 10
pwD and tD.dpwD/dtD
1E + 1
s=5
s=0
1E + 0
Figure B5 Pressure and pressure derivative type curve for n=0.7.
s = 30
s = 10
1E + 1
pwD and tD.dpwD/dtD
s=5
s=0
1E + 0
Figure B6 Pressure and pressure derivative type curve for n=0.8.
194 Reservoir Characterization
s = 30
s = 10
1E + 1
pwD and tD.dpwD/dtD
s=5
s=0
1E + 0
Figure B7 Pressure and pressure derivative type curve for n=0.9.
s = 30
s = 10
1E + 1
pwD and tD.dpwD/dtD
s=5
s=0
1E + 0
Figure B8 Pressure and pressure derivative type curve for n=1.0.
Part 3
RESERVOIR PERMEABILITY
DETECTION
9
Permeability Prediction Using
Machine Learning, Exponential,
Multiplicative, and Hybrid Models
Simon Katz1*, Fred Aminzadeh2, George Chilingar2 and M. Lackpour2
1
Russian Academy of Natural Sciences, US Branch, Los Angeles, California
2
Petroleum Engineering Program, School of Engineering,
University of Southern California
Abstract
The authors present a unified methodology for permeability prediction with non-
linear multiplicative, exponential, and hybrid multiplicative-exponential nonlin-
ear models. Due to logarithmic transform of these models they may be used for
prediction with both linear regression and various machine learning methods. It
was demonstrated that enhancement of prediction accuracy is achieved with new
two-level adjustable committee machines. The new prediction methodology was
tested on data from sandstone and carbonate reservoirs with similar pattern of
improvement of prediction accuracy due to using nonlinear prediction models
and two-level committee machines.
9.1 Introduction
Modeling of relations between permeability and other petrophysical
parameters, such as porosity or grain size, leads usually to construction of
nonlinear permeability models. Results presented in multiple publications
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (197–216) © 2022
Scrivener Publishing LLC
197
198 Reservoir Characterization
k = a * E(z1,..,zm) (9.5)
k = b * M(x1,..,xn) (9.6)
The goal of this paper is to design and test permeability forecast meth-
odology characterized by two seemingly controversial features: (a) techni-
cally, forecast relies on the models with linear structure, and (b) results of
forecast reflect complex nonlinear relationships between permeability and
predictor parameters.
According to Eq. (9.2), hybrid model is transformed into additive model
of the form (1) by log transform of both sides of Eq. 9.2:
log(M) = q1 * s1 + … + qn * sN (9.8)
200 Reservoir Characterization
log(E) = α1 * z1 + … + αm * zM (9.9)
N M
H = δ0 + ∑δ ∗ s + ∑δ
r r p ∗zp (9.10)
r =1 p =1
H = δT + q1 * s1 + … + qn * sn (9.11)
H = δE + α1 * z1 + … + αm * zm (9.12)
According to Eqs. 9.10, 9.11, and 9.12, all three nonlinear models are
transformed into linear additive models with dependent variable - loga-
rithm of permeability. Forecast methodology presented in this paper is
based on this type of logarithmic transforms. It includes three basic steps:
(a) logarithmic transform of nonlinear permeability model into linear one,
(b) forecast of values of log(permeability), and (c) transform of forecast
values of log(permeability) into forecasts of permeability via exponential
transform. Additional elements of our permeability forecast methodology
are construction and use of two-level committee machines, basis function
expansion, and exhaustive search for optimum subset of predictors.
yn = aO + ∑a x (n) + ∑b q (n);
r r l l ql = θ l ( x l1 (n),, x lm (n))
r l
(9.13)
where xr(n) are base predictors, n is index of the record, xls, 1 ≥ ls ≤ lm, are
parameters from a subset of base parameters, θl are nonlinear functions
of recorded parameters, ql = θl(x11(n), ... , xlm(n)) are newly constructed
expanded base predictors. Basis function expansion allows efficient use
of linear regression, when relations between response and predictors are
nonlinear. Negative side of the use of basis function expansion may be in
a large number of constructed predictors which will result in overfitting
and instability of the forecast. To avoid potential overfitting and forecast
instability the authors combine basis function expansion with exhaustive
search for optimum subset of predictors [9] (R package ‘leaps’). Exhaustive
search presented in this package analyses possible subsets of predictor
variables and finds best subsets of predictors for approximation. The unre-
solved issue with this approach is that a subset of predictors, that guaran-
tees excellent approximation, may not produce equally accurate forecasts.
So, the authors use exhaustive search for selection candidate subsets of pre-
dictors, and then evaluate accuracy of the forecast with selected predictors
with enhanced Monte Carlo cross validation [8].
Equations 9.14 and 9.15 present models resulting from exhaustive search
methodology for additive, Eq. 9.14, multiplicative, Eq. 9.15, for three pre-
dictors in a subset.
where ϕ is porosity, g is grain size. One can observe, that large part of
selected predictors are nonlinear functions of recorded basis predictors.
12000 coefOutlier=1
Individual predictions
coefOutlier=2
8000 coefOutlier=3
4000
0
0 500 1000 1500 2000 2500 3000
Index
Figure 9.1 Plot of sorted values of individual permeability forecasts generated with
hybrid model. Dashed horizontal lines are drawn at three values of parameter coefOutlier.
Permeability Prediction 203
Table 9.1 Percent of identified outliers among individual forecasts for five values
of coefOutlier. Number of predictor variables in models of four type: 7.
Models
Cutoff
coefOutlier (mD) Additive Exponential Miltiplicative Hybrid
1 2620 0 1.26 4.01 2.04
2 5240 0 1.19 3.10 1.94
3 7860 0 0.03 1.39 1.94
4 10480 0 0.00 0.95 0.75
5 13100 0 0.00 0.07 0.71
Y(n) =
1
M (n) ∑ F(n,m) (9.17)
m
204 Reservoir Characterization
N N
Y (n) =
1
∑
W (n) m=1
F (n, m) ∗ w(n, m); W (n) = ∑w(n,m) + δ 2
m =1
(9.22)
Permeability Prediction 205
1; F (n, m) ≤ cutoff
w(n, m) = (9.23)
mp / F (n, m); F (n, m) > cutoff
1; F (n, m) ≤ cutoff
w(n, m) = (9.24)
0; F (n, m) > cutof
If for some records the number of outliers equals to the number of indi-
vidual forecasts, then, according to Eq. 9.22 and 9.24, first level committee
machine produces forecast value equal to zero, which is equivalent to not
producing permeability forecast. Zero forecasts are dealt with by the sec-
ond level committee machine. Its simplest form is given by Eq. 9.25:
where n index of the record in the test set, Y1(n) and Y2(n) forecasts by
first level committee machines defined by Eqs. 9.22 and 9.24 and produced
with two different permeability models. It is also assumed, that Y1(n) is
produced either with multiplicative, exponential, or hybrid model, so that
zero values among set of values of Y1(n) are indicators of outliers. Weight
function ω(n) in Eq. 9.25 is defined as:
1; Y (n) > 0
w(n) = 1 (9.26)
0; Y1 (n) = 0
206 Reservoir Characterization
Level 2
Level 1
Figure 9.2 Diagram of the structure of the second level committee machine. Weights are
to deal with outliers. CM is manifestation of machine learning through a set of individual
forecasts, each operating independently, outputs are combined through a gate keeper.
It follows from Eqs. 9.25 and 9.26, that output of the second level com-
mittee machine is of the form:
Let S1 and S2 are two sets of indexes n at which forecasts Y1(n) and
Y2(n) have outliers, indicated by zero values. It follows from Eqs. 9.25 to
9.27, that the number of outliers at the output of second level committee
machine will be equal to the size of intersection of S1 and S2. Therefore, if
the size of S2 is smaller than the size of S1, then the number of outliers in
P(n) will be smaller than in the Y1(n).
Diagram shown at Figure 9.2 illustrates mechanism of forming second
level committee machine.
N M
RMSE =
1
N ∑ ∑(k(n) − F(n,m))
1
M
2
(9.28)
N =1 m =1
N
MAB =
1
N ∑ k(n) − Y (n) (9.29)
N =1
N M
MINST =
1
N ∑ ∑(Y (n) − F(n,m))
1
M
2
(9.30)
N =1 m =1
where k(n) is actual permeability, n -index of the record in the studied data
set, m is index of individual forecast F(n,m) for the records with index n.
According to Eq. 9.29, parameter MAB may also be interpreted as mean
absolute error of the forecast by first level committee machine.
Plots of mean absolute bias of permeability forecast done with first level
committee machines and four forecast models are shown in the Figure 9.3.
Bias was calculated for nonoutlier records with cutoff equal to max(k)*1.25.
According to Figure 9.3, bias of the forecast with additive models is higher
than bias of the forecasts with nonlinear models. On the other hand, out-
put of the first level committee machine produced with additive model
has the smallest number of outliers. According to Table 9.2, the number of
detected outliers tend to decrease with increase of the cutoff, while bias of
200 Additive
Mean absolute bias (mD)
“Exponential
Multiplicative
160
Hybrid
120
80
1 2 3 4 5 6 7
Number of predictions
Figure 9.3 Mean absolute bias of permeability forecast with additive multiplicative,
exponential, and hybrid models. Sandstone dataset.
208 Reservoir Characterization
Table 9.2 Number of outliers at the output of the first level committee
machine. Sandstone dataset.
Permeability models
coefOutlier Additive Multiplicative Exponential Hybrid
1.0 1 4 2 2
1.15 0 3 2 2
1.25 0 2 2 2
1.5 0 1 2 2
2.0 0 1 1 2
Table 9.3 Bias of the forecast with four predictors for four permeability models
and different outlier cutoffs. Number of predictors: 4.
Permeability models
coefOutlier Additive Multiplicative Exponential Hybrid
1.00 132.4 98.6 117.9 109.3
1.15 132.4 117.4 117.5 106.7
1.25 133.6 127.4 112.6 110.4
1.5 133.2 155.6 110.4 109.1
2.0 133.1 151.7 154.5 138.5
the forecast, as shown in the Table 9.3, tends to increase with increase of
cutoff.
Tables 9.4 and 9.5 present estimates of RMS error and bias generated
with hybrid model. RMS error is significantly larger than the absolute bias.
Both parameters show two trends – decrease of its value with increase of the
number of predictors and increase of its value with increase of coefOutlier.
Instability of individual forecasts and advantage of the first level com-
mittee machines is illustrated by Figure 9.4. According to this Figure 9.4,
difference between actual permeability and individual forecasts and their
range of values may be as large as 700 mD, while difference between actual
permeability and output of the committee machine is significantly smaller.
Wide range of values of individual forecasts indicates their instability.
Permeability Prediction 209
Table 9.4 Absolute bias of the output of first level committee machine produced
with hybrid model.
Cutoff
Number of
predictors 1 1.25 1.5 2 100
2 134 134.8 134.7 133.9 424.4
3 113.2 141.8 145.1 143.5 239.6
4 106.8 108.4 149.8 151.9 232.1
5 104.1 103.1 103.3 158.4 265.5
6 96.2 96.8 97.5 159.4 250.5
7 106.4 104.4 104.4 157.8 248.4
Table 9.5 Root Mean Squared error (mD) for individual forecasts with hybrid
models.
coefOutlier
Number of
predictors 1 1.25 1.5 2 3 100
2 276.4 289.9 285.7 288.4 290.3 1972.4
3 259.2 251.4 299 401.1 433.7 995.9
4 258.8 251.2 279.6 367.7 525.5 928.5
5 227.5 240.9 211 242.9 634.8 1201.9
6 213.6 218 219.8 215.2 727.6 1167.2
7 235.7 250.4 240.6 245.4 489 1163.1
2500
2000
1500
1 2 3 4 5 6
Index
Figure 9.4 Individual forecasts, output of the first level committee machine, and actual
permeability in six records from sandstone data set. Number of predictors is 7, coefOutlier
is 1.25, hybrid permeability model.
Table 9.6 Mean instability of individual forecasts with hybrid model for the
records with permeability exceeding 700 mD. Sandstone dataset.
coefOutlier
Number of
predictors 1 1.25 1.5 2 3 100
1 34.4 39.8 47.2 62.9 63.1 60.2
3 75.6 98.3 84.2 122.7 108.5 172.2
5 91.8 93 102.1 94.3 131.9 192.2
7 103.1 133.1 115.4 144.1 173.1 173.6
excellent forecast results done with multiplicative model. The forecast with
the exponential model shows some differences with actual permeability,
while the forecast with additive model is significantly different from pre-
dicted permeability.
Exponential
200
100
0
1 2 3 4 5
Number of predictors
Figure 9.5 Bias of permeability forecast by the first level committee machine. Outlier
cutoff equal max(permeability)*1.25. Kuybushev dataset.
500 Exponential
300
100
0
1 2 3 4 5
Number of predictors
Figure 9.6 Bias of permeability forecast by the first level committee machine. Outlier
cutoff equals max(permeability)*1.25. Central Asia carbonate dataset.
212 Reservoir Characterization
2500
Permeability
Forecast with exponential model
Forecast with multiplicative model
2000
Forecast with additive model
Permeability (mD)
1500
1000
500
0
5 10 15
Index of the records
Figure 9.7 Permeability forecast with multiplicative, exponential, and additive models.
Kuybushev dataset.
2500 Permeability
Regular forecast values
Permeability (mD)
Replacement of outliers
1500
500
0
0 20 40 60 80 100
Index
9.9 Conclusion
The authors introduced new methodology for permeability forecast with
nonlinear permeability models of three types: multiplicative, exponen-
tial, and hybrid multiplicative-exponential. It includes three basic steps:
2-Level adjustable
committee machine
References
1. C. Fung, K. Wong, and H. Eren, Modular artificial neural network for predic-
tion of petrophysical properties from well log data. IEEE T. Instrum. Meas.
46(6), 1295–1299 (1997).
2. A. Bhatt and H. Helle, Committee neural networks for porosity and permea-
bility prediction from well logs. Geophys. Prospect. 50, 547–674 (2002).
3. M. Nikravesh and F. Aminzadeh, Past, present and future intelligent res-
ervoir characterization trends (editors’ view points). J. Petrol. Sci. Eng. 31,
67–79 (2001).
4. Z. Huang, J. Shimeld, M. Williamson, and J. Katsube, Permeability predic-
tion with artificial neural network modeling in the Venture gas Feld, offshore
eastern. Geophysics 61(2), 422–436 (1996).
5. T. Hastie, R. Tibshirani, and J. Friedman, The Elements of Statistical Learning:
Data Mining, Inference and Prediction, 2nd edition. Springer-Verlag, New
York, p. 745 (2009).
6. N. Aase, P. Bjorkum, and P. Nadeau, The effect of grain-coating microquartz
on preservation of reservoir porosity. Am. Assoc. Petr. Geol. B. 80(10), 1654–
1673 (1996).
7. G. Chilingarian, J. Chang, K. Bagrintseva, Empirical expression of permea-
bility in terms of porosity, specific surface area, and residual water saturation
of carbonate rocks. J. Petrol. Sci. Eng. 4(4), 317–322 (2006).
8. S. Katz, F. Aminzadeh, W. Long, G. Chilingar, and M. Lackpour, Rock per-
meability forecasts using machine learning and Monte Carlo committee
machines. J. Sustain. Energy Eng. 4(2), 182–200 (2016).
9. P. Cortez, Data Mining with Neural Networks and Support Vector Machines
Using the R/rminer Tool. Chapter in Advances in Data Mining. Applications
and Theoretical Aspects. Volume 6171 of the series Lecture Notes in
Computer Science, 572–583 (2010).
10. F. Aminzadeh and P. de Groot, Neural Networks and Soft Computing
Techniques, with Applications in the Oil Industry, EAGE Publishing, p. 164
(2006).
11. R. Gholami, A. Shahraki, and A. Paghaleh, Support vector regression for pre-
diction of gas reservoirs permeability. J. Min. Environ. 2(1), 41–52 (2011).
10
Geological and Geophysical Criteria
for Identifying Zones of High Gas
Permeability of Coals (Using the
Example of Kuzbass CBM Deposits)
A.G. Pogosyan *
Abstract
The article reviews the results of an analysis of changes in the petrophysical prop-
erties of coals and the dynamic conditions existing during their formation. The
combined effect of physical properties and dynamic parameters on the perme-
ability of coalbeds is evaluated; their relation to the distribution of flow rates in
producing wells is determined. The feasibility of downhole multiwave VSP obser-
vations in CBM fields is substantiated.
10.1 Introduction
Geological exploration is increasingly focused on the non-traditional
resources, such as shale gas, shale oil, and coalbed methane. Thus, reserves
and resources of these sites will be a priority as key exploration areas, both
on land and offshore.
In recent years, developing coalbed methane resources has been a pri-
ority in Russia. Among the coalbeds methane (CBM) basins of Russia,
*Email: [email protected]
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (217–230) © 2022
Scrivener Publishing LLC
217
218 Reservoir Characterization
Kuzbass is the largest coalbed methane basin in the world ready for large-
scale production of methane. The basin’s recoverable methane resources
are estimated at 13 trillion m3. These methane resources occur at depths
of 1800–2000 m [1]. The first Russian pilot production field for extracting
methane gas from coalbeds was launched in 2010 at the Taldinskaya area
of the Kuzbass.
Current practice of pilot development of coalbed methane at the
Taldinskaya area in Kuzbass shows that even at high concentrations of
methane resources (resources in the area are 95 billion m3, with an aerial
density of 3 billion m3/km2), methane production may be uneconomic
because of low gas recovery.
Whereas, if the gas contained in sandstones escapes to the surface due
to reservoir pressure, special channels (fractures) must be created for its
movement within coalbeds. This approach is required because coalbed
properties differ significantly from the corresponding properties of tradi-
tional sandstone and carbonate gas reservoirs.
In addition to artificially created channels, gas recovery values are also
affected by the original natural gas permeability of coal due to fracturing.
Natural gas permeability of coalbeds is the most important property nec-
essary for commercial production of methane. Without initial natural gas
permeability, and without the ability to increase the gas recovery from
coalbeds, there is no methane production potential—despite the huge
methane resources in coalbeds.
As a result, the criteria characterizing coalbed permeability should be
considered the most important of all criteria for assessing the prospects
for developing of coal and gas production. The development of geological
and geophysical criteria for identifying the zones of high gas permeability
of coalbeds and their changes within the area is the most urgent geological
task at all stages of CBM field development, making the application of var-
ious geophysical methods a priority.
The important issue is not only to define the geological and geophysical
gas permeability evaluation criteria, but also to develop an effective meth-
odological basis for studying the physical properties of thin coalbeds in
interwell space.
The physical and geological factors affecting the distribution of well flow
rates determined from an analysis of physical and mechanical properties
of coals and dynamic conditions of their occurrence are examined below
using the example of the Taldinskoye Field of the Kuznetsk Basin, com-
pared with the results of production tests of coal/methane wells.
Geological and Geophysical Criteria 219
v s2 (3 ∗2p −4 ∗ v s2 )
E = ρ∗ (10.1)
Vp 2 − Vs 2
220 Reservoir Characterization
- shear modulus:
G = ρ * Vs2 (10.2)
4
K = ρ ∗ (Vp 2 − Vs 2 ) (10.3)
3
16
E (GPa)
14
12
10
8
6
2
0
66 NN 65 64 63 62 61 60-59 58 57 56 55? 54 NN 52a 52 51 50 48 45 44 43
(a) Index of coalbed layer
G (GPa)
7
6
5
4
3
2
1
0
66 NN 65 64 63 62 61 60-59 58 57 56 55? 54 NN 52a 52 51 50 48 45 44 43
(b) Index of coalbed layer
K (GPa-1)
0,5
The compressibility factor of high rate coalbeds
0,4 has relatively low values (0.1-0.2 GPa-1)
0,3
0,2
0,1
0
66 NN 65 64 63 62 61 60-59 58 57 56 55? 54 NN 52a 52 51 50 48 45 44 43
(c) Index of coalbed layer
Coalbed parameter value High rate wells Low rate
layers exposed to hydraulic fracturing well 2 well 5 well 9
Low rate coalbeds well 4 well 6 well 10
High rate coalbeds well 7 well 8
21 40
44 43-42
19 ? 43
48-45 44
39 40
45
17 High rate wells
48 43
39 well 2
15 50 44
well 4
48 45 51
13 50
44
52 well 7
45 52 51
48
53 51
52 Low rate wells
11 50 54
52 52 52 well 5
9 56
50
52
5 well 6
54 5 54
56
well 8
7 57
60-59 58 well 9
Index of coalbed layer
5 well 10
The calculated dynamic values of the lateral thrust (λ) of coalbeds vary from
0.1 to 0.8 of the vertical overburden pressure, with high rate wells differing
from the low rate wells by high lateral thrust values of 0.5–0.8 (Figure 10.3).
Such a significant variation in λ values in rocks of the same lithotype is a better
reflection of the internal macroscopic structural features of the structure of
coalbeds; in particular, the nature and orientation of exogenous fracturing.
Thus, a comparison of data on the geostatic and lateral thrust of over-
burden pressure with the flow rates (Figures 10.2, 10.3, 10.4) and the dis-
tribution of volume compressibility factor K (Figure 10.1 c) reveals the
correlation of the “external” rock and dynamic and “internal” physical
properties of coalbeds with gas permeability.
Therefore, the effectiveness of hydraulic fracturing depends on the com-
pleteness of geological information on dynamic rock conditions and the
structure of the borehole environment, combined with the physical and
0.9
Lateral thrust
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Low rate wells
0 Index of coalbed layer
66 NN 65 64 63 62 61 60-59 58 57 56 55? 54 NN 52a 52 51 50 48 45 44 43
7000
flow rate, thousand m3/day
M 1:50 000
well 6
S N
6000
well 4
5000
well 7
well 10
well 8
well 5
4000
well 9
3000
well 2
2000
1000
0
well 4 well 2 well 7 well 5 well 8 well 6 well 10 well 9
450 450
Bed 52 a Bed 52 a
500 500
Bed 51 Bed 51
550 550
Bed 50 Bed 50
600 600
650 650
700 700
750 750
800 800
H M H M
Interlayered sandstones and fine-grained siltstones Bed 60-59 Index of coalbed layer
(b) PP
(c) 16
73
H
16 SP
51HPP
PP
SP SP
10
20 30 SP
0 20
62HPP 20
0 10
Well 14
-60 0 10
30
10
CMS_2
0
20 SP
0
0
10
SP 12 1
15
51HPP
84
H
0
SP SP
PP
62HPP
17 13
73
H SP
PP
73HPP
Reflected PP wave tracing lines Well 2 EDM parameters (GPa)
Increased coalbed
Small amplitude disjuctive faults VSP shotpoints compressibility area
Figure 10.5 Results of experimental studies using the multiwave VSP technique:
(a) structure of coalbeds in the borehole environment in the wave field of longitudinal
PP and transverse PS waves (b) identification of small amplitude faults (c) distribution
of bulk modulus (inverse of compressibility factor) in the borehole environment for the
target coalbed.
The opportunity of studying the wave field at interior points of the coal
deposit made it possible to eliminate the effect of surface interference
waves, thus improving the data quality while also widening the energy
spectrum of wave fields by adding new waves of different nature and other
seismic processes to the traditional composition of useful waves. These
processes reflect the nonlinear properties of deposits, particularly its emis-
sion activity due to internal micro-structural heterogeneity.
The appearance of methods based on the construction of fundamen-
tally new physical models of real media, taking into account the seismic
non-linearity of oil and gas reservoirs, is characteristic of the development
228 Reservoir Characterization
of modern geophysics [9−13]. The tasks associated with the study of geo-
dynamic objects include fractured gas-saturated waveguide zones, rhe-
ologically weakened zones and media with a complex stress-strain state.
Thus, in terms of its studied properties, the CBM reservoir of Kuzbass has
all of these characteristics.
The presence of microstructural irregularities in coalbeds, e.g., those
significantly larger than atomic size but still small on the elastic wavelength
scale, such as microfractures, may lead to abnormal shows of nonlinear
medium properties, significantly increasing the intensity of nonlinear
acoustic parameters and substantially changing the qualitative nature
of the nonlinearity (the appearance of a distinct frequency or amplitude
dependence) itself.
Furthermore, the linear acoustic characteristics of the medium may stay
almost constant. Thus, the “structural sensitivity” (at the microscopic level)
of the medium’s nonlinear properties is significantly higher than that of lin-
ear elasticity parameters [14]. Desorption processes form mechanical force
fields affecting the coalbed, and the cumulative effect of these forces, leads
to acoustic radiation (UV MS) of sources of increased gas dynamic activity.
Obviously, studies of weak nonlinear processes in the medium are pos-
sible only with high-quality geophysical data registration. Downhole seis-
mic technology successfully tested in a CBM section can provide a reliable
basis for research on these processes in a seismic wave field.
10.4 Conclusions
Geological and geophysical criteria influencing the formation of high gas
permeability zones in coals are established and their relationship with the
distribution of flow rates of pilot CBM wells is physically justified.
The author provides technical solutions for increasing the efficiency
of the studies of the physical properties of coalbeds and broadening the
understanding of the distribution of permeability of CBM reservoirs.
Acknowledgement
The writer is greatly indebted to academician George V. Chilingarian for
his invaluable help.
Geological and Geophysical Criteria 229
References
1. A.M. Karasevich, V.T. Khryukin, B.M. Zimakov, G.N. Matvienko, S.S.
Zolotih, V.G. Natura, and T.S. Popova, Kuzbass Basin—The Largest Resource
Base of Commercial Methane Production from Coal-beds, pp. 64, Mining
Science Academy, Moscow. (2001).
2. A.T. Ayruni, M.A. Iofias, and L.M. Zenkovich, The Scientific Basis for
Determining the Gas Permeability of Coal-beds under Varying Filtration
Parameters, pp. 52, Research Institute of Comprehensive Exploitation of
Mineral Resources, Academy Sciences USSR, Moscow. (1982).
3. Y.N. Malyshev, K.N. Trubetskoy, and A.T. Ayruni, Fundamentally Applied
Methods of Solving the Problem of Coal-bed Methane, pp. 519, Mining Science
Academy, Moscow. (2000).
4. Halliburton, Chapter 2, in Coal-bed Methane: Principles and Practices,
pp. 126, R. E. Rogers, K. Ramurthy, G. Rodvelt, M. Mullen (Eds.), Oktibbeha
Publishing Co., LLC, Starkville, MS 39759, ISBN 978-0-9794084-1-0 (2008).
5. L.V. Baltoiu, B.K. Warren, and T.A. Natras, State of the art in coalbed meth-
ane drilling fluids. Drilling Completion 23, 250–257 (2008).
6. G.N. Boganik and I.I. Gurvich, Seismic Surveying, pp. 754, AIS, Tver. (2006).
7. D. P. Zemtsova and A. G. Pogosyan, The possibility of using multiwave seis-
mic surveys to assess the elastic deformation modules of CBM sections,
Abstracts of the XIth Scientifi c Conference and Exhibition “Geomodel-2009”,
Gelendzhik, EAGE, Volume 145, 145 (2009).
8. Tang, H. Seismic prospecting technique for coal bed methane accumulating
area. Procedia Earth Planet. Sci. 3, 224–230 (2011).
9. O.L. Kuznetsov, I.A. Chirkin, Yu.A. Kuryanov, G.V. Rogotsky, and V.P.
Dyblenko, Experimental studies, in Seismoacoustics of Porous and Fractured
Geological Media: Volume 2, O.L. Kuznetsov, (Ed.), pp. 320, State Research
Center of VNIIGeosystems Research Institute, Moscow. (2004).
10. E.I. Galperin, Vertical Seismic Profiling, pp. 344, Nedra, Moscow. (1982).
11. S.E. Richardson, D.C. Lawton, and G.F. Margrave, Seismic applications in
coalbed methane exploration and development, 01 (2003).
12. M. Li, B. Jiang, S. Lin, F. Lan, and J. Wang, Structural controls on coalbed
methane reservoirs in Faer coal mine, Southwest China. J. Earth Sci. 24, 437–
448 (2013).
13. D.P. Zemtsova, A.A. Nikitin, and A.G. Pogosyan, Study of fractured zones of
a CBM section of Kuzbass basin based on seismic emission of geodynamic
noise, Bulletin of Higher Educational Institutions, Geology and Exploration,
Moscow, No. 3, 51–55 (2009).
14. V.Y. Zaytsev, N.V. Pronchatov-Rubtsov, and S.N. Gurbatov, “Non-classical”
Structure-induced Acoustic Nonlinearity: Experiments and Models, pp. 223,
Novgorod, Nizhny. (2007).
11
Rock Permeability Forecasts Using
Machine Learning and Monte
Carlo Committee Machines
Simon Katz1, Fred Aminzadeh2, Wennan Long2*, George Chilingar2
and Matin Lackpour1
Los Angeles, California
2
Petroleum Engineering Program, School of Engineering,
University of Southern California
Abstract
We developed new concepts of extended Monte Carlo cross validation and Monte
Carlo committee machines. We subsequently used those concepts to predict per-
meability by linear regression and machine learning methods such as Neural
Networks, Support Vector machines, and Regression Tree. Among the parameters
we calculated using extended Monte Carlo cross validation are: root-mean squared
error of individual forecasts, forecast bias, correlation between forecast and actual
permeability, and forecast instability as a measure of sensitivity to perturbations
of the training set. Output of Monte Carlo committee machines is constructed
as the average of machine learning outputs generated from multiple versions of
perturbed training sets. We observed that Monte Carlo committee machines pro-
duced high stability forecasts, while individual machine learning forecasts (e.g. a
single ANN) were characterized by lower stability. Higher accuracy forecasts were
achieved when we applied machine learning methods and linear regression using
permeability models that included quantitative and categorical predictors and
second-order interactions among the predictors.
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (231–252) © 2022
Scrivener Publishing LLC
231
232 Reservoir Characterization
11.1 Introduction
Current research in modeling and forecasting rock permeability includes
a variety of methods and models. The structure of permeability models
is defined by several factors which include a set of petrophysical param-
eters utilized as predictors, heterogeneity of lithology in the studied area,
a model that defines relations between permeability and predictors, and
methods used for model construction. The means used for rock perme-
ability forecasts are linear, log-linear regression, and machine learning
methods such as neural networks and support vector machines. An
excellent analysis of correlations amongst permeability, porosity, confin-
ing pressure, cementation, and grain size was done by AlHomadhi [1].
Results presented in this paper indicate the importance of factors other
than porosity in reliable permeability forecast. Log-linear regression
model for permeability with porosity, specific surface area, and irreduc-
ible fluid saturation were developed and analyzed by Chilingarian et al.
[2] (See Addendum). The model was tested on data from several carbon-
ate reservoir rock areas in the former USSR. One specific feature of the
regression models utilized in that paper is the presence of second-order
interactions. The inclusion of irreducible fluid saturation in the set of
predictors led to a high correlation between actual and predicted perme-
ability values. Generally, the advantage of linear regression permeability
models is in their interpretability. Their weak point is their rigid struc-
ture. More flexible but more difficult to interpret are machine learning
techniques, such as neural networks [3–6] and support vector machines
[5, 7] used for permeability forecasts. Additional enhancement of the
efficiency of machine learning permeability forecasts might be done with
committee machines [4, 8–11]. Efficiency of forecast is estimated with
different versions of cross validation. Cross validation version ‘leave one
out’ was used by Gholami [7] to validate the permeability model built
with support vector machines. Monte Carlo cross validation was pro-
posed in Qing’s paper [12] and used for estimating the number of com-
ponents in the calibration model. General review of machine learning
methods, such as neural networks and soft computing for reservoir char-
acterization, was presented by Nikravesh and Aminzadeh [13].
Authors of this paper concentrated on development and testing of two
algorithms: (a) extended version of Monte Carlo cross validation and (b)
algorithm for Monte Carlo committee machines. Although we will not
delve into the details of the ANN committee machines, we will include a
brief description from Aminzadeh and de Groot [11] in Appendix 2.
Rock Permeability Forecasts Using ML 233
Classic methods of cross validation, such as leave one out or k-fold have
estimated single parameter - forecast error. The extended version of Monte
Carlo cross validation is aimed at detailed analysis of the forecast and
includes estimation of additional parameters such as forecast bias, fore-
cast instability, correlation between forecast and target parameters, and
comparative analysis of accuracy of individual and committee machine
forecasts. Developed methods and algorithms have wide applications for
forecasts of different types of reservoir characterization. They may be uti-
lized, for example, for analysis and forecast of reservoir porosity distribu-
tion and in the forecast of trends in hydrocarbon production [14].
In this paper, the analyses of forecasts were obtained using a dataset
[15] published as open source at the pubs.usgs.gov website. That data-
set includes data from Jurassic sandstones collected from 15 wells in the
Norwegian sector of the North Sea in the depth range of 3300–4250 m.
Each record in the dataset contains a permeability value and respective
values of quantitative and categorical parameters. The total number of
records with no missing values is 99. Quantitative parameters in the data-
set are porosity, grain size and burial depth. Categorical variables in this
dataset qualitatively characterize content of the following mineralogical
components: microquartz, macroquartz, clay, and carbonates. They have
the following discreet values: 0 - not observed, 1 - very minor, 2 - minor,
3 - medium, 4 - much, 5 - very much.
Authors of this paper use R rminer package [5] to run machine learning
methods and R function lm() for linear regression.
1 N
Y (k ) = Σ F (k , m) (11.1)
N m=1
where k is the index of the record in the analyzed dataset, m is the index of
a randomly formed training set that was utilized to produce a forecast F(k,
m). N is the total number of randomly generated training sets that are used
to produce individual forecasts for a record with index k.
Mean absolute bias (mabF) and mean absolute error (maErF) of an
individual forecast for the set of records S and Bias of individual forecasts
(bF(k)), absolute bias averaged over a set of records S, mabF(S), and averaged
absolute error of individual forecasts, mabErF(S), are defined as:
1
bF (k ) = P(k ) − Y(k ); mabF (S ) = ∑ bF (k ) ;
N (S ) k ⊂ S
1 (11.2)
mabErF (S ) = ∑ P(k ) − F (k , m)
N (S ) k ⊂ S
Root mean squared error of the forecast by the committee machine for the
group of records is shown as:
Rock Permeability Forecasts Using ML 235
0.5
1 N (S)
RMSE(S ) ∑ ( P(k ) − Y (k ))2 (11.3)
N (S ) k ⊂ S
1 R
YR (k ) = ∑ Y (k , r ) (11.4)
R r =1
1
bCM (k ) = YR (k ) − P( K ); mabCM (S ) = Σ bCM (11.5)
N (S) k⊂ S
1 M
F (k ) = ∗ Σ Y (k ) − F (k , m) (11.6)
M − 1 m=1
1 R
instICM (k ) = ∗ Σ YR (k ) − Y (k , r ) (11.7)
R r =1
n.record(Train) (11.8)
prInd = 1 −
n.records(dataSet )
where n.records (dataSet) and n.records (Train) are respectively the number
of records in the analyzed data set that includes all available records with
known value of predicted parameter and in the randomly formed training
sets, Train in Eq. 11.8 is a randomly formed train set. This parameter satis-
fies the following constraints: 1> prInd>= 0, and for stable, not perturbed
training sets prInd = 0.
Major application of the MC committee machines forecast with a fixed
test set with records that do not include predicted parameter, so that per-
turbations are done only to the train set. In that case the goal of the Monte
Carlo committee machine is to decrease instability of the forecast of pre-
dicted parameters.
Number of forecasts
120
Forecasts number
80
40
0
0 20 60 100
Record index
One can observe that the number of individual forecasts is rather stable
and, in this example, does not go below 74.
Parameters of distribution of the random number of individual fore-
casts, calculated for four MC cross validation cycles, are shown in Table 11.1.
Each cycle had 1000 cross validation runs. As seen, the estimated parame-
ters of the distribution are similar across all four cycles. Minimum number
of forecasts within four cycles varies from 75 to 81. Lower quantiles vary
within a range of 93 to 96. Upper quantiles are in the range of 107–108.
Therefore, if a number of Monte Carlo runs in the MC cycle is not less
than 1000, one may expect that about 75 forecasts may be used to evaluate
errors, bias and instability of the output of the committee machine and of
individual forecasts.
1500
Model 1
Model 3
500
Bias
–500
–1500
Figure 11.2 Bias of the forecast with permeability Models 1 and 3. Bias is calculated for
permeability values in the range of 1100-2650 mD. Horizontal scale – permeability.
Model1
1500
0
0 20 40 60 80 100
Model3
1500
0
0 20 40 60 80 100
Figure 11.3 Ordered values of permeability and their forecasts with permeability
Models 1 and 3. Continuous line - ordered permeability, small dots - individual
forecasts. Vertical axis - permeability in mD horizontal scale - index of the record.
predicted as 1500 mD. A forecast with Model 3 is much more accurate and
has smaller bias in a full permeability range.
Table 11.3 Mean absolute bias of the forecasts by Monte Carlo committee
machines with six machine learning methods. Mean absolute bias is calculated
for forecasts with nine permeability models for each forecast method.
Methods
Models PLSR KNN SVM NN PCR RT
1 231.9 153.0 169.5 152.1 230.3 118.8
2 164.0 154.5 166.5 153.7 164.0 121.1
3 146.0 139.7 143.6 126.5 146.7 124.7
4 217.6 142.7 142.2 127.0 215.9 125.1
5 221.7 157.3 146.3 133.3 221.7 121.9
6 214.8 147.7 155.7 143.5 215.0 123.5
7 205.4 151.6 164.4 137.0 204.3 120.5
8 144.2 133.4 141.4 133.4 143.2 125.4
9 143.1 145.5 142.7 148.5 142.2 123.3
Rock Permeability Forecasts Using ML 243
SVM
1000
–2000
RT
1000
–2000
Figure 11.4 Forecast bias as a function of permeability values for two machine learning
methods with permeability model (Model 1).
forecasts is the largest for neural network. Hence, more individual fore-
casts may be necessary to build neural network or regression tree commit-
tee machine forecasts with low instability.
Smaller regression tree bias compared to the bias of support vector
machines is illustrated by Figure 11.4. As seen here, bias of forecast by the
committee machines relying on support vector machines is systematically
negative at permeability values larger than 1500 mD. Therefore, a support
244 Reservoir Characterization
Permeability
CM forecast
Individual
Permeability (md)
2000 0 4000
1 2 3 4 5 6 7 8
Record index
Individual
Committee, N = 5
200 400 600 800
Committee, N = 50
Committee, N = 100
STD
0
11.9 Conclusions
New methods of extended Monte Carlo cross validation and Monte Carlo
committee machines are introduced in this paper and tested with rock
permeability forecasts. Extended Monte Carlo cross validation is designed
for detailed analysis of bias and instability of the forecasts, forecast error,
and correlation between forecast and forecasted parameter. Monte Carlo
committee machines are expected to enhance forecast stability, which is
improved with an increase of a number of individual forecasts that form
the committee machine. Forecast bias is a basic characteristic of forecast
accuracy. It cannot be diminished.
Comparative analysis of the accuracy of individual permeability fore-
casts and forecasts by Monte Carlo committee machines relying on linear
regression, and several machine learning methods, was performed using
extended Monte Carlo cross validation protocol. A list of analyzed machine
learning methods includes neural networks, k-nearest neighbor, support
vector machine, principal component regression, partial least squared,
and regression tree. Committee machines relying on neural network and
regression tree outperformed linear regression and are characterized by
smaller forecast bias. The highest accuracy, smallest bias and highest cor-
relation between forecast and forecasted permeability was obtained with
regression tree.
Nomenclature
• MC cross validation = Monte Carlo cross validation, MC
committee machine - Monte Carlo committee machine, MC
run - Monte Carlo run.
• F(k, m) = Individual forecast for record with index k with
model built using randomly formed train set with index m.
• mabF(S) = Mean absolute bias of individual forecast.
• bCM(k) = Mean absolute bias of the forecast by the commit-
tee machine for the value of the forecasted parameter in the
record with index k.
• instIF(k) = Instability index for individual forecasts.
248 Reservoir Characterization
word ‘effective’ properly we will achieve more accurate results when assess-
ing the correlation between porosity and permeability.
In addition, specific surface area (per unit of pore volume), which is
a measure of the degree of fracturing, must be considered when evaluat-
ing the relationship between porosity and permeability. Fractures do not
contribute much to porosity but they do substantially increase the per-
meability. A few near-perfect correlations were obtained by Chilingarian,
Bagrintseva and Chang [2] for several carbonate reservoirs by adding two
additional variables: irreducible fluid saturation (Swr) and specific surface
area (Ss):
Where SS = Specific Surface Area (per unit of pore volume), Swr = irreduc-
ible water saturation and ϕ = fractional porosity. The role of the insoluble
residue (IR) content is also being investigated by the writer in determining
the relationship between porosity and permeability.
y1
Expert 1
w1
y2
Expert 2
w2
Integrator
x y
Input Final output
yN
Exert N
wN
Figure A2.2 Multilayer Perceptron (a) without and (b) with short cut connections and
passive input neurons (c) Completely Connected Perceptron (CCP) with active input
neurons (from Fruhwirth and Steinlechner, 2005).
committee chair. The role of the global expert is to make rulings or judg-
ments pertaining to the local experts, assign a significance factor or weight
to their respective outputs, and to determine what role each should play in
the final outcome of the combined network. The number of outputs of the
gating network is the same as the number of individual networks that are
combined. Figures 11.1–11.6 is an example of a network.
Dynamic committee machines are capable of handling more compli-
cated problems. Examples of applications of committee machines in the oil
industry include those in the inversion of induction of log data Zhang and
Rock Permeability Forecasts Using ML 251
Poulton [17] and prediction of shear wave logs from sonic and other suites
of logs Fruhwirth and Steinlechner [18].
Fruhwirth and Steinlechner [18] used a Completely Connected
Perceptron (CCP) with active input neurons, a common MLP as well as
an MLP with short cuts for log prediction application. They used the net-
work to predict shear wave logs from a suite of other logs in a well. The
training was done using an existing shear wave and other suite of logs from
another well. The training started using solely the CCP architecture for 10
network generations starting without any hidden units (Figures 11.1–11.6).
This operation is equivalent to a multi-linear regression in many attribute
analysis applications except that the nodes in CCP can account for any
non-linear relationship between the known (in this case the known log
suites) versus the unknown (the shear wave log).
The final CCP had 9 hidden units. In each of these generations 20 differ-
ent and randomly initialized networks in parallel were trained, which can
be considered another type of realization of a modular neural network. The
goal was to prevent getting trapped too much in local error minima.
*This Appendix is adopted from Aminzadeh and de Groot [11].
References
1. E.S. AlHomadhi, New correlations of permeability and porosity versus con-
fining pressure, cementation, and grain size and new quantitatively correla-
tion relates permeability to porosity. Arabian Journal of Geosciences 7(7),
2871–2879 (2014).
2. G. Chilingarian, J. Chang, and K. Bagrintseva, Empirical expression of per-
meability in terms of porosity, specific surface area, and residual water satu-
ration of carbonate rocks. J. Petrol. Sci. Eng. 4(4), 317–322 (2006).
3. C.C. Fung, K.W. Wong, and H. Eren, Determination of a generalised BPNN
using SOM data-splitting and early stopping validation approach, Proceedings
of Eighth Australian Conference on Neural Network (ACNN’97), pp. 129–133
(1997).
4. B. Alpann and H.B. Helle, Committee neural networks for porosity and per-
meability prediction from well logs. Geophysical Prospecting. 50(6), 547–
674 (2002).
5. P. Cortez and P. Perner, Data mining with Neural Networks and Support
Vector Machines using the R/rminer Tool. Chapter in Advances in Data
Mining Applications and Theoretical Aspects, Volume 6171 of the series
Lecture Notes in Computer Science, 572–583 (2010).
252 Reservoir Characterization
Abstract
As offshore hydrocarbon development in the Gulf of Mexico has moved into
deeper waters and more technically challenging subsurface environments, the
tools to evaluate and reduce risks and potential impacts of drilling continue to
evolve. Science-based decision-making, risk reduction, and identification of tech-
nology gaps are key to the responsible development of extreme offshore hydro-
carbon resources. This paper specifically focuses on providing a review of data
and information related to the subsurface petroleum system for the U.S. Gulf of
Mexico. This information is vital to understanding the current state of knowl-
edge about the subsurface geology and hydrocarbon system for this region, and
for quantifying and assessing knowledge gaps and uncertainty. This review paper
summarizes relevant peer-reviewed and open-source publications, as well as pub-
licly available databases, focusing on regions associated with deepwater (>500’
water depth) and ultra-deepwater (>5000’ water depth) settings.
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (255–268) © 2022
Scrivener Publishing LLC
255
256 Reservoir Characterization
Introduction
The Gulf of Mexico (GOM) basin is a petroleum province of global and
domestic economic importance. The United States’ Bureau of Ocean
Energy Management (BOEM) estimates that in the federal offshore GOM,
undiscovered technically recoverable resources total 87.5 billion barrels
of oil equivalent [1], and hydrocarbon companies are moving into deeper
water and riskier plays in search of the profitable prospects needed to
supply global demand that is projected to grow to 27 million incremental
barrels per day by 2020 [2]. Uncertain subsurface conditions in explor-
atory wells, already one of the riskiest phases of hydrocarbon development
[3, 4], are exacerbated by extreme environments that include water depths
of up to 10,000 feet, total well depths nearly 30,000 feet below the mudline,
pressures of almost 30,000 psia (pounds per square inch, absolute), and
temperatures over 300 degrees Fahrenheit [5]. In addition, recent events,
including impacts from natural events (e.g. Hurricanes Katrina and Rita)
and anthropogenic events (e.g. Deepwater Horizon oil spill), have high-
lighted gaps in our ability to predict risks and effectively prevent delete-
rious outcomes throughout the lifecycle of hydrocarbon development in
extreme offshore settings [3].
To reduce uncertainty, identify knowledge and technology gaps, support
risk assessments, and improve flow rate estimates in the GOM region, a
sufficient understanding of the subsurface is required for a variety of stake-
holders. While proprietary data and information are available to specific
entities, they are typically associated with restrictions for use and access.
However, information from published literature and publicly available data
sources can be integrated to support good comprehension of the petroleum
system across the GOM, and when appropriate, offer insights into field- or
site-specific areas, as well. Information from published literature includes
site-specific reservoir descriptions and basin-wide syntheses, studies of
basin structure and salt, analyses of the petroleum system, and reservoir
fluids found within hundreds of references. Publicly available databases
containing information such as subsurface attributes are less abundant
and typically focus on field or well specific scale information. However,
BOEM does release a well-populated GOM-wide public database of res-
ervoir sands characteristics that contains field-specific information and
production statistics [6]. While information about the subsurface GOM
petroleum system ranges in quantity, accessibility, and resolution, there is a
foundation from existing studies that can be utilized to reduce uncertainty
and evaluate for spatial and temporal trends at the field to regional scale.
The Gulf of Mexico Petroleum System 257
sands and shales within the basin itself [7, 11, 17, 20–22]. In some places,
Cenozoic basin deposits accumulated to more than 10,000 meters thick
[17]. Deposition in the Pliocene and Pleistocene is also characterized by
thick, interbedded terrigenous deposits, which themselves are more than
3000 meters thick in some areas [17, 20].
The result of this large and rapid deposition of sediment was destabili-
zation and gravitational detachment of the sediment fill on the underly-
ing shale and salt layers, in addition to mobilization of the autochthonous
Louann salt [17]. Large detachment fault systems parallel the current basin
shoreline in Texas and Louisiana, including Paleocene and Eocene fault
systems under the Texas coastal plain and Oligocene and Miocene detach-
ments on the offshore Texas-Louisiana shelf [12] (Figure 12.1). Periods of
extension caused coeval compression at the toe of the continental slope,
creating provinces of compressional salt-cored anticlines in the western
and central basin (the Perdido and Mississippi Fan fold belts) (Figure 12.1).
MFFB
Folds
Fault Zones
PDFB Salt
Bathymetry (ft)
Value
0
-13300
0 1530 60 90 120
Miles
Figure 12.1 Map of regional fault zones, folds, and salt occurrences in the northern
GOM. Regional fault zones are down-to-the-basin. PDFB: Perdido Fan Fold Belt, MFFB:
Mississippi Fan Fold Belt. Data supporting this map are publicly available for download
from the USGS Energy Data Finder [23]. Bathymetry data from [24].
The Gulf of Mexico Petroleum System 259
Petroleum System
Classic elements of a petroleum system include the geologic and hydro-
logic components and processes necessary to generate and store hydrocar-
bons, including a hydrocarbon source, migration pathway, reservoir rock,
trap and seal, and appropriate timing [25, 26]. For the GOM, information
about the basin’s petroleum system and sub-systems can be found in many
published articles and volumes, including the Atlas of Northern Gulf of
Mexico Gas and Oil Reservoirs [8, 9] and the AAPG Bulletin theme issue
Gulf of Mexico Petroleum Systems [27]. Updated literature with descrip-
tions of recent plays are presented in publications such as a 2012 study of
traps and reservoirs in the central northern Gulf of Mexico by Weimer and
Bouroullec [7] and a review of the properties of the deep water Wilcox
formation by Oletu et al. [5]. Reports from BOEM update the estimated
hydrocarbon reserves annually [28, 29], and their 2012 Outer Continental
Shelf Assessment summarizes the current knowledge of the northern Gulf ’s
plays, including their geology and estimated hydrocarbon resources and
reserves [1, 30].
For publicly available information pertinent to reservoir properties and
hydrocarbon properties, one of the most extensive available resources is
BOEM’s Atlas of Gulf of Mexico Gas and Oil Sands Data [31], which is pub-
licly downloadable and updated annually. This expansive data set presents
data aggregated and averaged to the sand level from non-releasable, res-
ervoir-specific data sources. This includes production statistics, estimated
reserves, age, physical and reservoir properties of the sands, and properties
of the hydrocarbons produced.
Reservoir Geology
Deepwater and ultra-deepwater reservoirs are dominantly associated with
turbidite deposits, particularly with the channel and levee systems and
fan/lobe sheet sand mass-transport deposit components of these turbidite
260 Reservoir Characterization
systems [1, 7, 32, 33]. The GOM is a notably prolific basin in that hydrocar-
bon plays span the vertical and lateral extent of the system. Plays occur ver-
tically throughout the stratigraphic column, and laterally across the basin,
the multitude of established large and small fields illustrate the breadth
of the system as well [11, 34]. Miocene-age reservoirs, many of which
are associated with deepwater and ultra-deepwater plays, have been the
most productive reservoir system, producing more than 23 billion barrels
of oil equivalent as of 2009 [1]. In addition, notable deepwater and ultra-
deepwater hydrocarbon plays in the GOM are also found within Pliocene
and Pleistocene sediments. In the future, the Paleocene, Eocene, and
Oligocene sediments that comprise Lower Tertiary plays on the continen-
tal slope are expected to contain the most undiscovered, technically recov-
erable resources, estimated at nearly 18 billion barrels of oil equivalent [1].
The Lower Tertiary Wilcox trend has produced at least a dozen discoveries
in the last decade [32, 35] (Figure 12.2).
Reservoir quality and type varies largely as a function of depositional
and structural environment, but can also be a function of secondary alter-
ation and diagenetic processes. Turbidite sands associated with Miocene to
Miocene Trend
Bathymetry (ft)
value
0
-13300
015 30 60 90 120
Miles
Figure 12.2 The general location of identified deepwater hydrocarbon trends in the
GOM. Based on data from [31, 35]. Bathymetry from [24].
The Gulf of Mexico Petroleum System 261
Hydrocarbons
Hydrocarbons are produced from almost 1,300 fields in the federal offshore
GOM, 209 of which are in the deepwater and ultra-deepwater. According
to BOEM, more than half of the hydrocarbon reserves remaining in the
GOM are in the deepwater and ultra-deepwater reservoirs. As of 2011,
21.91 billion barrels of oil and 192.4 trillion cubic feet of gas have been
produced [29]. The northern GOM is more prone to gas than oil, though
many fields produce both [29, 33, 36]. Oil and gas compositions, as well as
associated fluids and gases (e.g. CO2 or H2S) are important as indicators for
reservoir connectivity, age, source, and timing of fluid migration [37–41].
Average gas-oil ratios, oil and gas gravities, and production estimates are
available in public data from BOEM [6, 31].
Hydrocarbon sources in the GOM have been heavily influenced by the
timing of sediment deposition. Lower sediment influx to the basin in the
Mesozoic Era, followed by the rapid, voluminous deposition in the Cenozoic
Era, has left much of the modern continental slope over-pressured and with a
depressed temperature gradient, leaving Mesozoic rocks on the slope within
the necessary temperature regimes to generate hydrocarbons [42]. Analyses
of regional hydrocarbon sources can be found in [33, 36, 37, 43].
The primary sources throughout the GOM basin are the carbonates
and shales deposited in the Upper Jurassic, Upper Cretaceous, and early
Cenozoic Eras [33, 37, 42]. Hood et al.’s analysis of GOM hydrocarbon sys-
tems interprets the source rocks supplying the most of the deepwater GOM
reservoirs to be Upper Jurassic sediments of Tithonian age, and links sea-
floor seeps in the ultra-deepwater to Oxfordian sources [37]. Reservoirs
in the western offshore basin and the Texas-Louisiana continental shelf
are charged with hydrocarbons from Paleocene and Eocene source rocks.
Hood et al. [37] subdivide these systems further by original composition.
262 Reservoir Characterization
Data from both wells and seeps were used in Hood et al.’s [37] analysis
to map the locations and maturity of the sources and hydrocarbons. GOM
deepwater and ultra-deepwater reservoirs tend to be less mature than res-
ervoirs with similar sources closer to the basin rim [31, 37]. Detailed chem-
ical analyses of seeps and reservoir fluids can be used to provide insights
into hydrocarbon sources and migration pathways [37–41]. Research
and data on the presence and indications of seeps throughout the Gulf of
Mexico is also available from BOEM [41, 44].
Conclusions
Knowledge of the GOM geology and petroleum system, particularly from
publicly available sources, is continuously evolving. When used appropri-
ately, these information sources offer industry, regulators, and scientists
opportunities to study the system, identify knowledge and technology gaps,
and reduce risks related to exploration and development of the hydrocar-
bon system to mutual benefit. Challenges to drilling and production are
only going to continue to grow as operations move farther offshore [2, 4]
and to increasing sub-seafloor total depths. Insights into the subsurface
from myriad publicly available resources, including publications and data
sets, are key to reducing uncertainty in models and therefore risks [3], and
helping to ensure responsible and enduring access to domestic hydrocarbon
resources. Direct detection and characterization of the subsurface is costly
and limited to locations where sediment cores, sidewall cores, and proper-
ties from wellbore devices offer in situ measurements. The vast majority of
the data and information available for use in interpreting and predicting
264 Reservoir Characterization
employees, nor URS Energy & Construction, Inc., nor any of their employ-
ees, makes any warranty, expressed or implied, or assumes any legal liability
or responsibility for the accuracy, completeness, or usefulness of any infor-
mation, apparatus, product, or process disclosed, or represents that its use
would not infringe privately owned rights. Reference herein to any specific
commercial product, process, or service by trade name, trademark, manu-
facturer, or otherwise, does not necessarily constitute or imply its endorse-
ment, recommendation, or favoring by the United States Government or
any agency thereof. The views and opinions of authors expressed herein do
not necessarily state or reflect those of the United States Government or
any agency thereof.
References
1. R. Klazynski, E. Klocek, L. Nixon, A. Petty, and P. Post, Assessment of
Technically Recoverable Hydrocarbon Resources of the Gulf of Mexico
Outer Continental Shelf as of January 1, 2009. U.S. Bureau of Ocean Energy
Management Gulf of Mexico OCS Regional Office, Office of Resource
Evaluation, New Orleans, LA. (2012).
2. H. Elshahawi, Deepwater exploration and production in the Gulf of Mexico-
challenges and opportunities. Petrophysics 55, 81–87 (2012).
3. K. Rose, F. Aminzadeh, L. Sim., R. Ghanem, C. Disenhof, J. Bauer, M.
Mark-Moser, C. Thimmisetty, N. Jabbari, and A. Khodabakhshnejad, Risks
and impact assessment for deepwater and ultra-deepwater Gulf of Mexico
resources, Offshore Technical Conference, 25364-MS (2014).
4. D. Izon, E.P. Danenberger, and M. Mayes, Absence of fatalities in blowouts
encouraging in MMS study of OCS incidents 1992–2006. Well Control July/
Aug (2007).
5. J. Oletu, U. Prasad, A. Ghadimipour, S. Sakowski, K. Vassilellis, B. Graham,
N. Park, and L. Li, Gulf of Mexico wilcox play property trend review, Offshore
Technical Conference, 24186-MS (2013).
6. Bureau of Ocean Energy Management Data Center, www.data.boem.gov
(2014).
7. P. Weimer and R. Bouroullec, Petroleum Geology of the Mississippi Canyon,
Atwater Valley, Western Desoto Canyon, and Western Lloyd Areas, Northern
Deep Gulf of Mexico: Traps, reservoirs, and their timing. 32nd GCSSEPM
Research Conference Proceedings. 32, 110–132 (2012).
8. S.J. Seni, T.F. Hentz, W.R. Kaiser, and E.G.J. Wermund, (Eds.), Miocene and
Older Reservoirs: Atlas of Northern Gulf of Mexico Gas and Oil Reservoirs,
Volume 1, Bureau of Economic Geology, The University of Texas at Austin.
(1997).
266 Reservoir Characterization
9. T.F. Hentz, S.J. Seni, and E.G.J. Wermund, (Eds.), Pliocene and Pleistocene
Reservoirs: Atlas of Northern Gulf of Mexico Gas and Oil Reservoirs, Volume
2, Bureau of Economic Geology, The University of Texas at Austin. (1997).
10. A. Salvador (Ed.), The Gulf of Mexico Basin. The Geology of North America.
Geological Society of America, Boulder, Colorado. (1991).
11. W.E. Galloway, Depositional evolution of the Gulf of Mexico sedimentary
basin, in The Sedimentary Basins of the United States and Canada, A.D. Miall
(Ed.), pp. 505–549, Elsevier. (2008).
12. F.A. Diegel, J.F. Karlo, D.C. Schuster, R.C. Shoup, and P.R. Tauvers, Cenozoic
structural evolution and tectono-stratigraphic framework of the Northern
Gulf Coast continental margin, in Salt Tectonics: A Global Perspective, M.P.A.
Jackson, D.G. Roberts, and S. Snelson (Eds.), 109–151, AAPG Mem. 65
(1995).
13. S.H. Hall, The role of autochthonous salt inflation and deflation in the north-
ern Gulf of Mexico, Mar. Pet. Geol. 19, 649–682 (2002).
14. M.R. Hudec, M.P.A. Jackson, and F.J. Peel, Influence of deep Louann struc-
ture on the evolution of the Northern Gulf of Mexico. AAPG Bull. 97, 1711–
1735 (2013).
15. T.P. Dooley, M.P.A. Jackson, and M.R. Hudec, Coeval extension and short-
ening above and below salt canopies on an uplifted, continental margin:
Application to the Northern Gulf of Mexico. AAPG Bull. 97, 1737–1764
(2013).
16. A. Salvador, Triassic-Jurassic, in The Gulf of Mexico Basin: The Geology
of North America, A. Salvador (Ed.), pp. 131–180, Geological Society of
America, Boulder, Colorado. (1991).
17. A. Salvador, Origin and development of the Gulf of Mexico basin, in The Gulf
of Mexico Basin: The Geology of North America, A. Salvador (Ed.), Geological
Society of America, Boulder, Colorado. (1991).
18. E.J. McFarlan, and L.S. Menes, Lower cretaceous, in The Gulf of Mexico Basin:
The Geology of North America, A. Salvador, (Ed.), pp. 181–204, Geological
Society of America, Boulder, Colorado. (1991).
19. N.F. Sohl, R.E. Martínez, P. Salmerón-Ureña, and F. Soto-Jaramillo, Upper
Cretaceous, in The Gulf of Mexico Basin: The Geology of North America,
A. Salvador, (Ed.), pp. 205–244. Geological Society of America, Boulder,
Colorado. (1991).
20. J.M. Coleman, H.H. Roberts, and W.R. Bryant, Late quaternary sedimenta-
tion, in The Gulf of Mexico Basin: The Geology of North America, A. Salvador,
(Ed.), pp. 325–352, Geological Society of America, Boulder, Colorado.
(1991).
21. A. McDonnell, R.G. Loucks, and W.E. Galloway, Paleocene to eocene
deep-water slope canyons, Western Gulf of Mexico: Further insights for
the provenance of deep-water offshore Wilcox Group plays. AAPG Bull. 92,
1169–1189 (2008).
The Gulf of Mexico Petroleum System 267
22. W.E. Galloway, D.G. Bebout, W.L. Fisher, J.B.J. Dunlap, R. Cabrera-Castro,
J.E. Lugo-Rivera, and T.M. Scott, Cenozoic, in The Gulf of Mexico Basin:
The Geology of North America, A. Salvador, (Ed.), pp. 245–324, Geological
Society of America, Boulder, Colorado. (1991).
23. United States Geological Survey, USGS Energy Data Finder, http://certmap-
per.cr.usgs. gov/geoportal/catalog/main/home.page (2014).
24. NOAA National Geophysical Data Center, GEODAS Grid Translator,
ETOPO11-minute Global Relief data, http://www.ngdc.noaa.gov/mgg/gdas/
gd_designagrid.html (2012).
25. L. Magoon, The petroleum system—A classification scheme for research,
exploration, and resource assessment. U.S. Geol. Surv. Bull. 1870, 2–15
(1988).
26. L.B. Magoon and Z.C. Valin, Overview of petroleum system case studies.
AAPG Mem. 60, 329–338 (1994).
27. N. Hurley and P. Weimer, Gulf of Mexico petroleum systems. AAPG Bull. 82,
865–1112 (1998).
28. Bureau of Ocean Energy Management, Reserves Inventory Program- Gulf
of Mexico (GOM) OCS Region, http://www.boem.gov/Reserves-Inventory-
Program-Gulf-of-Mexico-OCS-Region (2014).
29. E.G. Kazanis, D.M. Maclay, and N.K. Shepard, Estimated Oil and Gas
Reserves, Gulf of Mexico OCS Region, December 31, 2011. OCS Report
2014-656, U.S. Bureau of Ocean Energy Management Office of Resource
Evaluation, New Orleans, La. (2014).
30. Bureau of Ocean Energy Management, Resource Assessment- Gulf of
Mexico OCS Region, http://www.boem.gov/Oil-and-Gas-Energy-Program/
Resource-Evaluation/ Resource-Assessment/RA-Gulf.aspx (2014).
31. Bureau of Ocean Energy Management, Atlas of Gulf of Mexico Gas and Oil
Sands Data, http://data.boem.gov/homepg/data_center/gandg/gandg.asp
(2014).
32. D. Meyer, L. Zarra, and J. Yun, From BAHA to Jack, evolution of the lower
tertiary wilcox trend in the deepwater Gulf of Mexico. Sedimen. Rec. 5, 4–9
(2007).
33. R. Nehring, Oil and gas resources, in The Gulf of Mexico Basin: The Geology
of North America, A. Salvador, (Ed.), pp. 445–494, Geological Society of
America, Boulder, Colorado. (1991).
34. B.J. Bascle, L.D. Nixon, and K.M. Ross, Atlas of Gulf of Mexico Gas and
Oil Sands as of January 1, 1999, OCS Report MMS 2001-086, U.S. Minerals
Management Service, New Orleans, LA. f(2001).
35. Halliburton, Deepwater Gulf of Mexico, 2000–2008 Deepwater Discoveries,
http://www.halliburton.com/public/solutions/contents/Deep_Water/
related_docs/GOM_ DWMap.pdf (2008).
36. A.C.J. Huffman, P.D. Warwick, and W.I. Finch, Energy resources of the
Northern Gulf of Mexico basin, in Gulf of Mexico Origin, Waters, and Biota:
268 Reservoir Characterization
Volume 3, Geology, N.A. Buster and C.W. Holmes, (Eds.), pp. 229–245, Texas
A&M University Press, College Station, Texas. (2011).
37. K.C. Hood, L.M. Wenger, O.P. Gross, and S.C. Harrison, Hydrocarbon sys-
tems analysis of the Northern Gulf of Mexico: delineation of hydrocarbon
migration pathways using seeps and seismic imaging, in Surface Exploration
Case Histories: Applications of Geochemistry, Magnetics, and Remote Sensing,
D. Schumacher and L.A. LeSchack (Eds.), pp. 25–40, AAPG (2002).
38. I.R. MacDonald, J.F. Reilly, Jr., S.D. Best, R. Venkataramaiah, F. Sassen, N.L.
Guinasso, Jr., and J. Amos, Remote sensing inventory of active oil seeps
and chemosynthetic communities in the Northern Gulf of Mexico, in M66:
Hydrocarbon Migration and its Near-Surface Expression, D. Schumacher and
M.A. Abrams (Eds.), pp. 27–37, AAPG Memoir (1996).
39. O.C. Mullins, J.Y. Zuo, K. Wang, P.S. Hammond, R. De Santos, H. Dumont,
V.K. Mishra, L. Chen, A.E. Pomerantz, C.L. Dong, H. Elshahawi, and D.J.
Seifert, The dynamics of reservoir fluids and their substantial systematic
variations. Petrophysics 55, 96–112 (2014).
40. B.N. Orcutt, S.B. Joye, S. Kleindienst, K. Knittel, A. Ramette, A. Reitz, V.
Samarkin, T. Treude, and A. Boetius, Impact of natural oil and higher hydro-
carbons on microbial diversity, distribution, and activity in Gulf of Mexico
cold-seep sediments. Deep-Sea Research II 57, 2008–2021 (2010).
41. W. Shedd, R. Boswell, M. Frye, P. Godfriaux, and K. Kramer, Occurrence
and nature of “bottom simulating reflectors” in the Northern Gulf of Mexico.
Mar. Pet. Geol. 34, 31–40 (2012).
42. J.A. Nunn and R. Sassen, Framework of hydrocarbon generation and migra-
tion, Gulf of Mexico Continental Slope. AAPG Bull. 70, 1187–1187 (1986).
43. T. Matava, A petroleum systems study of the Northern Gulf of Mexico. Lead.
Edge. 25, 478–482 (2006).
44. Bureau of Ocean Energy Management, Seismic Water Bottom Anomalies
Map Gallery, http://www.boem.gov/Seismic-Water-Bottom-Anomalies-Map-
Gallery (2014).
45. NOAA National Geophysical Data Center, Marine Trackline Geophysical
Data, http:// www.ngdc.noaa.gov/mgg/geodas/trackline.html (2014).
46. P. Weimer, J.R. Crews, R.S. Crow, and P. Varnai, Atlas of petroleum fields and
discoveris, Northern Green Canyon, Ewing Bank, and Southern Ship Shoal
and South Timbalier Areas (Offshore Louisiana), Northern Gulf of Mexico.
AAPG Bull. 82, 878–917 (1998).
47. M.R. Hudec, I.O. Norton, M.P.A. Jackson, and F.J. Peel, Jurassic evolution of
the Gulf of Mexico salt basin. AAPG Bull. 97, 1683–1710 (2013).
13
Forecast and Uncertainty Analysis
of Production Decline Trends with
Bootstrap and Monte Carlo Modeling
Simon Katz1*, George Chilingar2 and Leonid Khilyuk1
2
Petroleum Engineering Program, School of Engineering,
University of Southern California
Abstract
A new multi-step procedure was developed and tested for the analysis and forecast
of production decline curves. It includes multidimensional grid search over values
of nonlinear least squares errors, nonlinear least squares approximation, Monte
Carlo and block bootstrap simulation of production trends.
Several decline curve models were tested in grid search and iterative minimiza-
tion: SEPD, extended Hyperbolic, Duong and the Power-Exponential. Grid search
and iterative minimization worked equally well on all tested models.
Multidimensional grid search finds a starting point for nonlinear iterative pro-
cess. Then, iterative minimization finds parameters of an optimum approximating
model. The approximating model is used for the forecasting of production trends.
Block bootstrap and new Monte Carlo simulation methods produce third-level
aggregated forecasts of production rate. In addition, these two methods character-
ize the range of possible values of predicted production (uncertainty range).
Comparative analysis of Monte Carlo and block bootstrap simulation indicates
that these methods are characterized by different widths of the uncertainty regions
and by a certain mutual shift of predicted production curves. Thus, the joint use of
both techniques may result in a more reliable forecast of future production.
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (269–288) © 2022
Scrivener Publishing LLC
269
270 Reservoir Characterization
13.1 Introduction
Production decline curve analysis uses empirical decline curve models for
forecasting hydrocarbon production trends. Construction of the produc-
tion decline models is based on the use of production data in individual
wells or in a group of wells. Widely used methods of the decline curve anal-
ysis and forecast of production often rely on a graphical approach and may
include slope analysis of certain modified curves and estimates of decline
curve value at zero time. Another approach is minimization of approxi-
mation error via nonlinear regression. Darwis et al. [1] and Towler et al.
[2] used a robust linear regression technique for the prediction of decline
curve production. Nonlinear regression techniques seem more flexible and
usable with models of different type in a unified manner.
There are two important issues in the nonlinear regression: (a) selec-
tion of a starting point for stable and converging iterative minimization of
approximation error and (b) probabilistic analysis of uncertainty of final
results due to the effect of random fluctuations in input data. The first issue
is resolved if the starting point is close enough to the minimum of the
approximation error. To locate the starting point of this kind we developed
and utilized multidimensional grid search.
The probabilistic analysis of decline curves is discussed and developed
in a number of publications. A priori assumptions about the distribution of
a model’s parameters were introduced in Lin et al. [3]. These assumptions
were then used for analysis of production decline uncertainty. Jochen et al.
[4] applied the bootstrap method to develop stochastic reserves estima-
tion. Cheng et al. [5] applied a modified bootstrap method for analysis of
future production of an oilfield, using the Arps hyperbolic model.
Probabilistic analysis done in this paper relies on the unified approach
of Monte Carlo and block bootstrap methods that generates multiple pre-
dicted decline curves.
We developed and tested the multi-step forecasting and probabilistic
analysis on several decline curves models defined by Eqs. 13.1 to 13.4:
q0
Modified Hyperbolic model: u(t ) = (13.1)
(1 + g ∗ t )c
a
Duong model: u(t ) = q1 ∗ t −m ∗ exp ∗ (t å −m − 1) (13.3)
1− m
Autocorrelation
1.00
Autocorrelation
0.75
0.50
0.25
0.00
-5 0 5
Decline curve
Production (bbl/day)
80
60
40
20
20 40 60
Months
Autocorrelation
Decline curve
Figure 13.1 SEPD decline curve and autocorrelation of its random component.
Forecast and Uncertainty Analysis of Production Trends 273
a( x ) =
1
J ∑W ∗(Model(t , X ) − u(t ))
j j j
2
(13.6)
j
where: j is the time index, ωj are positive weighs, X is the vector of model
parameters, X = (x1,x2,..,xn). u(tj) are values of the decline curve at time
moments tj, model (tj, X) is the continuous nonlinear or linear function of
time and vector of parameters X. In the following sections of this paper,
model (tj, X) is defined by one of the Eqs. 13.1 to 13.4. Vector parameter
of the approximating model X is defined as one that minimizes weighted
residual of Eq. 13.6. The multi-step minimization procedure includes
multi-dimensional grid search followed by iterative minimization of the
approximation error (Eq. 13.6) and random generation of multiple decline
curves via Monte Carlo or bootstrap methods.
The main advantage of the grid search is its flexibility. The grid search
works with any approximating model of any complexity. It always results in
a location of a global minimum on a grid. Although the grid minimum does
not coincide with the actual global minimum, it gets closer to the actual
global minimum when the distance among grid nodes decreases. Another
advantage of the grid search is that it does not need a starting point. The
range for model parameters and distances between grid nodes are the only
parameters that should be pre-defined. Unfortunately, a decrease in the
distance among the grid nodes leads to an increase of the number of nodes
and, respectively, the computer time necessary for performing the grid
search. Thus, a certain balance is necessary. The authors envision the role
of the grid search as an instrument for finding the location of a starting
point for iterative methods of minimization of nonlinear least squares.
Efficiency of the grid search for several approximating models is illus-
trated in Figure 13.2. It shows four curves approximating SEPD decline
curve distorted by random noise. Dots in the Figure 13.2 show this approx-
imated curve. The approximating curves were obtained via minimization
of the approximating error (Eq. 13.7). Resulting approximating curves
shown as continuous lines are constructed with SEPD, Power Exponential,
Hyperbolic, and Duong models. According to Figure 13.2, approxima-
tion with all four models results with approximating curves close to the
15
14
Production (bpl/day)
13
12
11
10
9
0 10 20 30 40 50 60 70
Months
Figure 13.2 Approximation of the SEPD curve via multidimensional grid search. Four
approximating models.
Forecast and Uncertainty Analysis of Production Trends 275
Approximate curve
500
Sepd
Hyperbolic
Duong
Production (bb/day)
Power. Exp
300
100
0 10 20 30 40 50 60 70
Months
Figure 13.3 Approximated decline curve and four starting decline curves with
parameters different from parameters of approximated model. Model for approximated
curve - Hyperbolic. Four starting approximated curves are constructed using Hyperbolic,
Power. Exp, SEPD, and Duong models.
276 Reservoir Characterization
Approximated curve
200
Sedm
Hyperbolic
Duong
Power. Exp
Production (bb/day)
150
100
50
0 10 20 30 40 50 60 70
Months
model and each of the starting curves. Approximation results are shown in
the Figure 13.4. All four approximating models lead to excellent approx-
imation, and four approximating curves for four approximating models
are practically indistinguishable. This indicates potential flexibility in the
selection of approximating models for real data applications where actual
the model for a decline curve is uncertain. It also indicates potential prob-
lems since different models can produce equally good approximations and
at the same time may generate different forecasts.
The problem with the Levenberg-Marquardt type minimization is that
for certain starting parameters approximation fails. Thus, the selection
of starting parameters is important. A two-step procedure that includes
search on the multidimensional grid and iterative Levenberg-Marquardt
minimization is developed and reviewed in the following sections. It over-
comes the starting-point problem and avoids troublesome search for the
set of starting parameters.
10
Approximation error Grid search
8 Grid search +
Iterative minimization
6
4
2
Figure 13.5 Approximation errors for grid search followed by iterative minimization.
step is the grid search. Results of the grid search are used as a starting point
for Levenberg-Marquardt iterative minimization.
Figure 13.5 illustrates a decrease of the approximation error by the
iterative Levenberg-Marquardt procedure that follows the grid search.
According to Figure 13.5, approximation errors are systematically lower
after the second step approximation compared to the grid search errors at
the first step.
Iterative minimization of least squares produces an approximating
curve with parameters Xapprox. These parameters are used to extrapolate the
approximation model and to construct a forecasted decline curve with val-
ues for time t outside the approximation segment:
where: u0(t) is the decline curve produced by the model used in the grid
search and iterative approximation.
sections for statistically justified forecasting of the decline curves and esti-
mation of the range of forecasted values. We analyze here two techniques
for joint forecasting and analysis of forecasted uncertainty:
uk (t ) = u(t , X k ) (13.10)
declMeanR(t ) =
1
K ∑u(t , X )k (13.11)
k =1
and
where mean and median are calculated for a set of values u(t, Xk) at the
fixed time t.
The aggregated forecasts of Eq. 13.11 and 13.12 are different from each
other if distributions of randomly simulated values u(t, Xj) are not sym-
metric which is often the case for Monte Carlo simulated decline curves.
Forecast and Uncertainty Analysis of Production Trends 279
åoutside R1 (13.14)
lower upper
Quantile(P=0.1)
Quantile(P=0.9)
Median
1000
Count
500
8 10 12 14
Production(bbl/day)
Three columns of the Table 13.1 give estimates of the parameters of the
SEPD curve, standard deviation of the estimates, and P-values for esti-
mated parameters. In this example the P-values are extremely small and
standard deviations are much smaller than the estimates of the respective
parameters. This indicates that the estimate of the parameters of approxi-
mating model is highly reliable.
Table 13.2 shows the example of estimated covariance matrix of the
coefficients of SEPD model.
Examples of several simulated vectors of parameters for SEPD model
derived from estimated covariance matrix are given in the Table 13.3
The first 10 rows in this table show 10 Monte Carlo simulated vector
coefficients produced using data illustrated by the Tables 13.1 and 13.2.
Mean_10,000 in the Table 13.3 is the average over 10,000 generated coef-
ficients. Deterministic is the actual vector of coefficients shown in the first
column of the Table 13.1. Table 13.3 illustrates that randomly generated
coefficients fluctuate around its deterministic value.
Table 13.1 Example of the estimated parameters of the SEPD-type decline curve.
Figure 13.7 shows fifty Monte Carlo generated curves with parameters
derived from estimates of coefficient of approximated curve and covari-
ance matrix of estimated parameters. One can observe asymmetry in the
distribution of simulated curves and presence of outlier curves.
Figure 13.8 shows uncertainty regions defined by the Eq. 13.13 for
two pairs of quantiles calculated for a set of 10,000 Monte Carlo simu-
lated decline curves. Quantiles were calculated for a range of production
months of 100-500. Inasmuch as the uncertainty range for the quantile
pair is wider, the prediction of possible production values is less certain.
On the other hand, estimated probability for predicted decline curve to
be within uncertainty range limits is higher for the quantile pair (P=0.05,
0.95). The black line shifted to the upper quantile bound is the actual pro-
duction decline curve. According to Figure 13.8, the actual decline curve
is within the uncertainty region for both quantile pairs (p=0.05, 0.95) and
(P=0.10, 0.90) although it is close to the upper boundary of the quantile
range (P=0.10, 0.90).
Forecast and Uncertainty Analysis of Production Trends 283
50
40
Production (bbl/day)
30
20
10
40
Daily production (bbl)
30
20
10
Figure 13.8 Two quantile ranges for Monte Carlo generated production curves.
Approximation segment - 1-100 months. Prediction range - 100-500 months. Both,
approximated decline curve and approximating model are SEPD.
284 Reservoir Characterization
Nb
declMeanBT (t ) =
1
Nb ∑u(t , X )b (13.16)
b =1
and
40
Daily production (bbl)
30
20
10
Figure 13.9 Bootstrap quantile uncertainty regions for SEPD approximating curve.
Approximated curve is generated using SEPD model and disturbed by random noise.
and block bootstrap simulation are both close to the center of the quantile
region. They may be treated as predicted values of the declined production.
Table 13.4 Forecasted mean and median production values derived from Monte
Carlo and bootstrap methods.
Forecasted mean production Forecasted median production
Months Monte Carlo Bootstrap Monte Carlo Bootstrap
100 30.783 24.883 30.357 24.849
150 21.467 18.726 21.012 18.67
200 16.132 15.096 15.674 14.985
250 12.686 12.672 12.234 12.521
300 10.29 10.928 9.854 10.747
350 8.539 9.608 8.119 9.389
400 7.212 8.572 6.814 8.329
Bootstrap quantiles
Monte carlo quantiles
40
Oil production (bbl/day)
20 30
10
Figure 13.10 Upper and lower quantiles for Monte Carlo and bootstrap generated
decline curves.
Forecast and Uncertainty Analysis of Production Trends 287
Figure 13.10 shows the upper and the lower quantiles for the two sets
of Monte Carlo and bootstrap generated decline curves. The total num-
ber of decline curves in both data sets is 2500. The upper and lower
quantiles were calculated for probabilities 0.9 and 0.1. The area between
upper and lower quantiles shows forecast uncertainty area. Probability
for the decline curve to be outside the uncertainty area in both data sets
is 0.2.
There is significant similarity between two uncertainty areas. They over-
lap, although Monte Carlo generated uncertainty area is systematically
narrower and slightly shifted towards larger production values. The most
reliable strategy would be to use results of both methods of decline curves
simulation for analysis of uncertainty and forecast.
13.12 Conclusions
Conclusions can be summarized as follows:
New multi-step production trend forecast and uncertainty analysis is
presented in this paper. It includes multi-dimensional grid search as the
first step followed by iterative minimization of nonlinear least squares.
Estimated parameters of the approximating model are used then for Monte
Carlo or bootstrap simulation of multiple decline curves, calculation of
aggregated forecast values, and analysis of uncertainty of forecasted pro-
duction trends. Parameters of the approximating model are used for the
deterministic decline curve forecast.
The following decline curve models are tested and their forecast effi-
ciency is evaluated: Power-Exponential, SEPD, Duong, and Hyperbolic.
Both grid search and two-step iterative minimization proved to be flexible
and able to work with any model with similar efficiency. In some situations
however, due to inappropriate selection of a starting point, iterative min-
imization may be unstable. In order to produce stable results the iterative
minimization needs a grid search as a first step.
Three types of forecast of production decline curves are designed and
analyzed in this paper:
References
1. S. Darwis, B.N. Ruchjana, and A.K. Permadi, Robust decline curve analysis.
J. Indones. Math. Soc. (MIHMI), 15(2): 105–111 (2009).
2. B.F. Towler, and S. Bansal, Hyperbolic decline-curve analysis using linear
regression. J. Petrol. Sci. and Engr., 8 (4): 257–268 (1993).
3. Z-Sh. Lin, C-H. Chen, G.V. Chilingar, H.H. Rieke, and S. Pal, Uncertainty
analysis of production decline data. Energy Sources, 27: 463–483 (2005).
4. V. Jochen, S. Holditch, and J. Spivey, Probabilistic reserves estimation using
decline curve analysis with the bootstrap method. SPE Annual Technical
Conference and Exhibition, 6–9 Oct., Denver, Colorado (1996).
5. Y. Cheng, Y. Wang, D. McVay, and J. Lee, Practical application of a probabilis-
tic approach to estimate reserves using production decline data. SPE Annual
Technical Conference and Exhibition, 9–12 Oct., Dallas, Texas (2010).
6. J.J. Arps. Analysis of decline curves. Trans. AIME, 160: 228–247 (1945).
Agarwal et al., Analyzing well production data. SPE Reservoir Eval. & Eng.,
2(5): (1999).8. D.M. Bates and D.G. Watts, Nonlinear Regression Analysis and
Its Applications, Wiley: 359 pp. (1988).
7. M. Khulud, K.M. Rahuma, H. Mohamed, N. Hissein, and S. Giuma,
Prediction of reservoir performance applying decline curve analysis. Int. J.
Chem. Eng. Appl., 4 (2): 74–77 (2013).
8. G.A. Watson, (Ed.), The Levenberg-Marquardt algorithm: implementation
and theory, in Lecture Notes in Mathematics: Numerical Analysis, Springer-
Verlag: Berlin, pp. 105–116 (1978).
14
Oil and Gas Company Production,
Reserves, and Valuation
Mark J. Kaiser *
Abstract
The sustainability of an oil and gas company depends upon its ability to replace
reserves at a faster pace than its production rate. The primary determinants
of the value of an oil and gas company are its cash flows and earnings, which
are dependent upon the quantity and quality of the hydrocarbons it produces,
along with commodity sales prices, production potential, which is described by
its reserves, reserves replacement rate, and its inventory of capital assets, equip-
ment, infrastructure, and acreage. The purpose of this paper is to establish the
relationship between the primary factors that influence value for a cross-section of
oil and gas companies for the year ending 2010. We construct regression models
for majors and a random sample of North American independents according to
production, reserves, technology application, and geographic diversification. We
show that reserves and production are strong indicators of market capitalization,
and for independents, production and total assets are better proxies of company
value than reserves. Multinational independents are valued higher than domestic
producers and companies producing primarily from conventional assets exhibit a
modest price premium to unconventional producers. We infer the effective cap-
italization for a sample of private companies and the National Oil Companies of
OPEC and compare model-predicted market caps for companies domiciled out-
side North America.
*Email: [email protected]
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (289–336) © 2022
Scrivener Publishing LLC
289
290 Reservoir Characterization
14.1 Introduction
The value of an oil and gas company and/or property is intended to reflect
the worth of the company and/or property on the open market. According
to the 2005 International Valuation Standards, worth is defined as “the
value of property to a particular investor, or class of investors, for iden-
tified investment objectives.” The market value of a property is defined as
the “estimated amount for which a property should exchange on the date
of valuation between a willing buyer and a willing seller in an arms-length
transaction after proper marketing wherein the parties had each acted
knowledgeably, prudently, and without compulsion… reflecting the col-
lective perceptions and actions of a market …” [1].
The primary value of any company is derived from its cash flows and
earnings, which are dependent upon the quantity and quality of the
product that it provides, along with the sales price. Oil and gas compa-
nies derive their earnings from producing commodities that serve other
businesses and consumers in the economy. Production is derived from
reserves and the inventory of capital assets, production equipment, infra-
structure, and acreage. Reserves lie below the surface and have not yet
been produced but are economically and technically viable to extract. In
North America, the United States Securities and Exchange Commission
(SEC), the Ontario Security Commission (OSC), Toronto Stock Exchange
(TSE), and Canadian Security Administration (CSA) provide guidelines
on resource classifications and company requirements to list on their
stock exchanges.
Any member of society with enough money can buy shares of a public
company, but a private company has only a few owners whose shares are
not offered to the public. To estimate the value of a public company, there
are four basic valuation techniques commonly employed—book value of
assets, discounted cash flow, price earnings multiple, and market value—
which can vary considerably depending on the assumptions applied [2, 3].
For a private company, the first three methods are not an option because
detailed financial information is not publicly released.
The sustainability of oil and gas production in modern society is a cen-
tral theme of economists and policy makers alike, not to mention a concern
among the general public and government agencies, but unfortunately, the
basic premises are frequently misunderstood. Sustainable oil and gas pro-
duction is measured by the reserves to production ratio, R/P, reported by
public companies on their annual statements, and collected on a country
Oil and Gas Company Production, Reserves, and Valuation 291
basis and reported by BP’s Annual Statistical Report and related surveys.
The world R/P ratio for oil has held reasonably steady at 40 for the past 40
years, while for gas, R/P has varied around 60. The exact value for the ratio
is unknowable, of course, as countries with the most significant reserves do
not follow third-party accounting systems for reporting. Most public com-
panies have R/P ratios less than 15–20, dictated in large part by the trade-
offs between capital investment and rate of return thresholds involved in
building reserves. In recent years, R/P ratios for oil and gas companies have
increased because of significant new unconventional resource discoveries
and developments.
The purpose of this paper is to describe the primary factors that impact
the value of an oil and gas company and establish the relationships that
exist between market capitalization, reserves, production, and assets
for a cross-section of public companies. We fix the time of assessment
on December 31, 2010 to coincide with the release of production and
reserves data. By fixing the time of assessment, we eliminate the impact
of commodity price volatility on market capitalization, but the model
results are necessarily linked to a specific point in time1. Information on
†
1
†
This is not a significant drawback, however, since it is relatively easy to incorporate time
into the analysis with additional data collection.
292 Reservoir Characterization
14.2 Reserves
14.2.1 Proved Reserves
The primary assets of oil and gas companies are their entitlements to future
production from reserves. Proved reserves are defined as the estimated2 ‡
‡
2
The physical attributes of the asset class, located miles under the ground in rocks with
variable properties and uncertain boundaries, relying on indirect measurements that are
expensive to perform, in environments that range from benign to harsh, in both stable
and politically risky countries, with contract terms that are time dependent, means that
reserves estimates and deliverability are always uncertain. Because future production is
subject to variable production rates, unknown prices and cost, and is impacted by regula-
tory and fiscal uncertainty, the value of reserves is also uncertain.
Oil and Gas Company Production, Reserves, and Valuation 293
is the most risky and least certain reserves class. Companies that produce a
majority of their production from unconventional formations tend to have
a large percent of unconventional PUD reserves.
§
3
Society of Petroleum Engineers, World Petroleum Congress, Society of Petroleum
Evaluation Engineers, American Association of Petroleum Geologist.
294 Reservoir Characterization
14.3 Production
Production is the causal result of reserves and an important measure of
performance because it determines gross revenue, and when combined
with costs, the cash flow and profitability of a property. Production data
for publicly traded companies are compiled at various levels of aggrega-
tion and frequency depending on the state/country of operation. Wells,
reservoirs, properties, fields, and projects are typical evaluation units and
production is frequently reported to regulatory authorities on a monthly
basis. Wells are the preferred unit for performance analysis and fields are
the highest level for which reserves are estimated.
Operators produce at rates to maximize return on investment, but dif-
ferences arise in how oil and gas is produced depending on ownership,
product type, location, and prices. National Oil Companies have broader
constraints and obligations than public corporations and OPEC NOCs are
subject to restrictions on production via a quota system.
Oil and Gas Company Production, Reserves, and Valuation 295
stock exchange listing. IOCs pursue activities to grow their stock price
and are primarily concerned about production, cash flows, and booking
reserves. Several IOCs are integrated across the supply chain in the sense
that they not only explore for, develop, and produce oil and gas, but also
refine, transport, and market their petroleum products. Exxon, Shell, BP,
Chevron, Total, ConocoPhillips and Eni are integrated IOCs commonly
referred to as the majors or supermajors. Exxon was the largest gas pro-
ducer in the U.S. in 2011 and the world’s largest publicly traded oil com-
pany. Eni is publicly traded and also government owned. In 2012, Conoco
Phillips split into separate E&P and refining businesses and is no longer
classified as an integrated IOC.
¶
4
Estimated state ownership in 2010 denoted in parenthesis.
298 Reservoir Characterization
are employed which change over time. Some companies, such as BP and
Chevron, balance their production and refinery capacity, while Exxon and
Shell have a much higher refinery capacity than their production business
can provide, which require they purchase additional crude for their refin-
ery runs. Major exporting countries and NOC production capacity usually
far exceeds domestic refining capacity.
for large companies may translate into a value premium relative to small
companies.
bons have been found, acquired, and developed. Direct lifting costs are
total production spending minus production taxes. Total lifting costs are
the sum of direct lifting costs and production taxes. Production taxes typ-
ically rise and fall with changes in the prices of oil and natural gas, and
lifting cost varies by region and time. In 2009, worldwide total lifting costs
for U.S. IOCs were approximately $10/boe [20].
Each oil and gas deposit has its own extraction, processing, and transpor-
tation cost. Extraction cost depends on the size of deposit, reservoir conti-
nuity, location, fluid flow characteristics, and other factors. Processing cost
depends on the type and mixture of oil and gas, while transportation cost
is incurred to deliver the product to market and is usually a small part of
the total cost. Production cost is often a small part of the commodity price
over most of the life cycle of a field, but when deposits deplete, the cost of
production rises as a deposit approaches its economic limit. Companies
that are listed on U.S. exchange markets are required to file data on capital
expenditures, and for PDP reserves, the per-unit costs of extraction, but
because cost are aggregated over regional basins, the ability to infer useful
information from reported data is constrained.
**
5
Heat-equivalent reserves are computed using 6,000 cf = 1 boe.
300 Reservoir Characterization
likely to be valued higher than companies with a high finding cost because
of its impact on project profitability and perception of market participants.
14.4.7 Assets
Assets represent capital invested. The scale of assets required to produce
oil and gas is enormous, especially for complex and large field develop-
ments in harsh environments. A company’s balance sheet summarizes its
financial position at a point in time and is a list of all its assets, liabilities,
and equity. Total assets are the total fixed assets of the firm reduced by
the accumulated depreciation of those assets as they age. The book value
of assets reflects the historical costs rather than current market value or
replacement cost. Assets typically include surface facilities, equipment and
infrastructure, processing plants, offshore structures, refineries, loading
facilities, pipelines, property, and buildings. Capital equipment is often
custom-designed and permanently installed at specific fields and have little
or no alternative uses.
††
For example, Argentina’s largest oil company, YPF SA, majority owned by Spain’s Repsol
6
YPF, was the target of a recent government takeover. Prior to the takeover, many of YPF’s
concessions were revoked, sending its market capitalization to $9 billion in April 2012,
from $17.5 billion in 2011 [21]. Some industry insiders theorize the Argentine govern-
ment encouraged provincial leaders to revoke YPF’s oil concessions to reduce its mar-
ket capitalization to lower the amount the government would have to pay in the event of
nationalization.
Oil and Gas Company Production, Reserves, and Valuation 303
14.5.2 Variables
A summary of the model variables and data sources are provided in Table
14.1. By fixing the time of assessment on December 31, 2010, we eliminate
the impact of oil and gas price on company valuation, and match the pro-
duction and reserves reporting of companies.
The market capitalization of a publicly traded corporation is the num-
ber of shares issued and outstanding, multiplied by the per share price at a
point in time.
100000 80
10000 70
Cumulative Reserves (Bboe)
Proved Reserves (MMboe)
60
1000
50
100
40
10
30
1
20
0.1 10
0.01 0
1 11 21 31 41 51 61 71 81 91 101 111 121 131
Companies
Figure 14.1 Independents sampled from the 2010 OGJ150 are shown as dark lines.
304 Reservoir Characterization
‡‡
7
Consolidated subsidiaries are the business entities whose voting stock is less than 50%
controlled by the company.
Oil and Gas Company Production, Reserves, and Valuation 305
14.5.5 Independents
The 70 independents sampled represented 29 large and 41 small cap com-
panies distributed across 40 oil and 30 gas producers (Table 14.3). In total,
the 70 independents held combined reserves of 30 Bboe, produced 2.3
Bboe, and had a combined market cap of $401 billion at the end of 2010.
Apache, Devon, and Anadarko are the largest oil independents with
each company’s proved reserves over 2 Bboe and more than 200 MMboe
annual production, with market caps ranging from $34 billion (Devon) to
$46 billion (Apache).
Chesapeake, EOG Resources, and Talisman Energy are the largest
North American gas producing companies with reserves greater than 8
Tcfe and market caps ranging from $16.9 billion (Chesapeake) to $23.2 bil-
lion (EOG Resources). Both Chesapeake and EOG Resources report large
proved undeveloped reserves in unconventional plays (47% and 48% of
their reserves base, respectively).
Most independents hold less than 200 MMboe reserves: independents
with more than 200 MMboe reserves are classified as large-cap companies,
and the top-tier large cap independents have more than 1 Bboe reserves.
Top-tier large cap independents include oil companies Apache, Devon,
Anadarko, Marathon Oil, and Pioneer, and gas companies Chesapeake,
EOG Resources, Talisman Energy, and Noble Energy.
The average large-cap oil producer had 1.1 Bboe reserves, 88 MMboe
production, and a market cap of $15.5 billion on December 31, 2010, while
the average small-cap oil producer had 36 MMboe reserves, 3.5 MMboe
production, and $787 million market cap. The large cap oil producers had
a slightly smaller D/E ratio (1.1 vs. 1.7) and a slightly larger R/P ratio (12
vs. 10) compared to the small producer group and a smaller PUD/R per-
centage (34% vs. 53%).
The average large-cap gas producer held 861 MMboe reserves, pro-
duced 60 MMboe, and had a market cap of $10.2 billion on December 31,
2010, while the average small-cap gas producer had 61 MMboe reserves,
4.4 MMboe production, and $776 million market cap. Gas producers were
similar in R/P and PUD/R ratios, but large producers had slightly higher
average D/E ratios (1.3 vs. 1.1).
Aggressive-growth independents (e.g., Callon and ATP Oil & Gas)
leverage their financial strength to capture business opportunities and
have D/E > 10, while more conservative companies (e.g., Evolution and
Houston American Energy) follow a “zero-debt” policy to avoid carrying
interest-bearing debt and have D/E < 0.2. The independents had a group
Oil and Gas Company Production, Reserves, and Valuation 309
average D/E ratio of 1.1 similar to the majors but with a much higher stan-
dard deviation reflecting a larger variation in financing structure.
CAP = a + b ⋅R boe + c ⋅R / P + d ⋅D / E
CAP = a+ b ⋅ R o + c ⋅R g +d ⋅ R / P + e ⋅ D / E.
where CAP is the market capitalization ($); Ro, Rg, Rboe represent the vol-
umes of oil (bbl), gas (cf) and heat-equivalent (boe) reserves, respectively;
R/P is the reserves to production ratio (yr); and D/E is the debt equity ratio.
14.6.2 Expectations
Proved reserves are expected to be a primary indicator of company value
and to be positively correlated with market capitalization at a given point in
time. Reserves represent the inventory of the company, and larger reserves
are expected to be associated with higher valuation and longer produc-
tion life. For all other things equal, high R/P ratios indicate that produc-
tion and cash flows are weighted to the future, and thus, high R/P ratios
could be negatively correlated with market value, but if large inventories
are perceived by the market as valuable, high R/P ratios could be positively
interpreted. Hence, the “expected” sign is ambiguous and the market is the
determining force in its realization.
Highly leveraged firms have higher fixed charges in the form of inter-
est payments relative to discretionary outlays, such as dividend payments.
The higher the D/E ratio, the greater the financial risk, and the lower the
expected market cap. The market cap for oil producers are expected to
dominate gas producers because of the price premium of liquid hydrocar-
bons, but obviously, many other factors also occur. Companies that are less
geologically and geographically diversified, or have significant percentage
310 Reservoir Characterization
400
Exxon
y = 0.0132x–0.3031
Market Cap ($billion)
300
R2 = 0.7524
Shell
200 Chevron
BP
Eni Total
100
ConocoPhillips
0
0 5,000 10,000 15,000 20,000 25,000 30,000
Proved Reserves (MMboe)
Figure 14.2 Market capitalization and proved reserves – integrated oil companies (2010).
400
Exxon
y = 0.2183x–57.829
Market Cap ($billion)
300
R2 = 0.7188
Shell
200 Chevron
BP
Eni Total
100
ConocoPhillips
0
400 600 800 1,000 1,200 1,400 1,600 1,800
Production (MMboe)
Figure 14.3 Market capitalization and production – integrated oil companies (2010).
Oil and Gas Company Production, Reserves, and Valuation 311
500
Exxon
400
Market Cap ($billion)
y = 1.3696x–88.798
R2 = 0.3468
300
Shell
200
Chevron BP
Total
100
ConocoPhillips Eni
0
100 150 200 250 300 350
Total Assets ($billion)
Figure 14.4 Market capitalization and total assets – integrated oil companies (2010).
50
y = 0.014x + 0.445
40 R2 = 0.9124
Market Cap ($billion)
30
y = 0.0072x + 4.142
R2 = 0.6303
20
10
0
0 500 1,000 1,500 2,000 2,500 3,000 3,500
Proved Reserves (MMboe)
Gas Oil
Small-Cap Oil Producer Group Small-Cap Gas Producer Group
Linear (Gas) Linear (Oil)
Figure 14.5 Market capitalization and proved reserves — large-cap independents (2010).
4
Market Cap ($billion)
3
y = 0.012x + 0.366
R2 = 0.3537
2
y = 0.011x + 0.0768
R2 = 0.4388
1
0
0 50 100 150 200
Proved Reserves (MMboe)
Gas Oil Linear (Gas) Linear (Oil)
Figure 14.6 Market capitalization and proved reserves — small-cap independents (2010).
Oil and Gas Company Production, Reserves, and Valuation 313
50
y = 0.8053x + 1.0413
R2 = 0.8088
40
Market Cap ($billion)
30
y = 0.8301x + 1.7071
R2 = 0.7654
20
10
0
0 10 20 30 40 50 60
Total Assets ($billion)
Gas Oil Linear (Gas) Linear (Oil)
50
y = 0.014x + 0.308
R2 = 0.9476
40
Market Cap ($billion)
30
20
y = 0.0091x + 1.419
R2 = 0.762
10
0
0 500 1,000 1,500 2,000 2,500 3,000 3,500
Proved Reserves (MMboe)
Gas Oil Linear (Gas) Linear (Oil)
50
y = 0.014x + 0.5704
R2 = 0.9424
40
Market Cap ($billion)
30
20
y = 0.0073x + 1.1748
10 R2 = 0.7119
0
0 500 1,000 1,500 2,000 2,500 3,000 3,500
Proved Reserves (MMboe)
Multinational Domestic Linear (Multinational) Linear (Domestic)
12
10
Market Cap ($billion)
y =0.016x + 0.200
8
R2 = 0.7512
y = 0.011x + 0.467
6
R2 = 0.746
4
0
0 100 200 300 400 500 600 700 800
Proved Reserves (MMboe)
Conventional Unconventional Linear (Conventional) Linear (Unconventional)
300
250
y = 0.0826x + 0.1143
Production (MMboe)
R² = 0.9614
200
150
y = 0.0692x + 1.2049
R² = 0.9206
100
50
0
0 500 1,000 1,500 2,000 2,500 3,000 3,500
Proved Reserves (MMboe)
Gas Oil Linear (Gas) Linear (Oil)
Class Product a b c d e R2
All Gas 1,872 27(6.85) 1.0(6.83) -7.3 -502 0.87
(1.61)a (-0.24) (-0.70)
Oil 555(0.84) 16(7.70) 2.1(6.84) -19(-0.81) -56(-0.53) 0.95
All 851 21(16.06) 1.2(10.86) -18(-0.92) -80(-0.70) 0.92
(1.55)
Large- Gas 8,958 22(4.61) 0.75(4.29) -42(-1.13) -3,278 0.83
Cap (3.01) (-1.88)
Oil 8,540 14(2.95) 2.2(4.00) -336 -1,658 0.92
(1.68) (-1.41) (-0.53)
All 7,702 19(9.1) 1.1(6.02) -46(-0.97) -3,849 0.88
(2.61) (-2.24)
Small- Gas 405(0.75) 50(3.62) 0.57(0.70) -18(-0.43) -53(-0.22) 0.58
Cap Oil 490(2.31) 23(2.31) 0.028 -6.1 -36(-1.32) 0.31
(0.02) (-1.02)
All 343(1.91) 26(4.53) 0.89(1.86) -5.0 -44(-1.55) 0.44
(-0.83)
t-statistics are denoted in parenthesis.
a
significant explanatory variable, and while the R/P and D/E coefficients
are both negative, they are not statistically significant.
Petrobras Brazil 10.76 11.95 12.76 878 14.5 0.7 229 124 222
CNOOC China 1.92 6.46 2.99 263 11.4 0.5 106 86 34
BG Group U.K. 0.95 11.69 2.89 236 12.3 0.9 69 53 49
Woodside Australia 0.18 6.45 1.25 69 18.0 0.8 34 46 14
Source: Companies’ 2010 Annual Report, Financial Times, Yahoo-Finance, Scottrade.
Oil and Gas Company Production, Reserves, and Valuation 323
400
Exxon
PetroChina
300
Market Cap ($billion)
Petrobras y = 0.0132x–0.3031
R2 = 0.7524
200 Chevron
Shell
BP
CNOOC Total
100
Eni ConocoPhillips
BG Group
Woodside
0
0 5,000 10,000 15,000 20,000 25,000 30,000
Proved Reserves (MMboe)
IOC State-Owned Major International Independent
Figure 14.12 Market capitalization and proved reserves – state-owned majors and other
independents (2010).
14.12 Conclusions
Constructing robust market capitalization models of oil and gas compa-
nies is subject to a number of estimation issues. In this analysis, we fixed
the time of assessment to coincide with the release of reserves and produc-
tion data, and it is clear that the model results will change over time and
are only valid on a relative basis. However, the relative positions of com-
panies and model results for most companies will not change significantly
on a year-to-year basis, unless exceptional events occur (e.g., BP and the
324 Reservoir Characterization
Macondo oil spill, merger, and acquisition activity) because finding and
developing reserves is a relatively slow and sequential process.
A large number of factors may impact capitalization of a company,
including the cost of production, earnings, management quality, and explo-
ration potential, etc. Some of these factors can be quantified but many can-
not. Factors that are not reported or disclosed limit the use of regression
models, but this limitation is relevant only if the excluded factors are sig-
nificant. In most cases, these factors are believed to be secondary, and thus
their exclusion is not expected to significantly impact the model results.
All of the major oil and gas companies in North America were enu-
merated and a large portion of independents was sampled. Large samples
reduce selection bias but do not eliminate its impact, and because all of the
measured factors are reported according to U.S. GAPP, the data sources are
believed to be reasonably consistent and accurate.
There are only a few majors and the use of multi-factor models are
over-specified. For independents, the model fits were generally robust, but
several coefficients of interest were not individually statistically significant.
As long as the coefficients are jointly significant, however, the estimating
equation can be utilized with benefit. The purpose of the regression model
is to estimate market value and is not necessarily concerned about the
direct significant of each control variable.
Because only a small number of transactions occur in the market place,
comparisons for private companies are difficult to perform, while for
the NOCs of OPEC, the values are speculative and based entirely on the
reserves volumes of the companies. We demonstrated comparisons for
GSEs and international companies listed on the U.S. stock exchange with
the model results.
References
1. International Valuation Standard, 7th Edn., International Valuation Standard
Committee, Washington, DC (2005).
2. J.B. Abrams: Quantitative Business Valuation. Wiley, New York (2010).
3. A. Damodaran: Investment Valuation. Wiley, New York (2002).
4. V.A. Van Vactor: Introduction to the Global Oil and Gas Business. PennWell,
Tulsa, OK (2010).
5. D. Chapman: Energy Resources and Energy Corporations. Cornell University
Press, Ithaca, NY (1983).
6. N. Antill and R. Arnott, Valuing Oil and Gas Companies. Woodlands Publishing
Ltd, UK (2000).
Oil and Gas Company Production, Reserves, and Valuation 325
7. A.W. Howard and A.B. Harp, Oil and gas company valuations. Bus. Val. Rev.
28, 30–35 (2009).
8. SPE/WPC/AAPG/SPEE, Petroleum Resources Management System,
http://www.spe.org/industry/docs/Petroleum_Resources_Management_
System_2007. pdf#redirected_from=/industry/reserves/prms.php (2007).
9. F. Demirmen, Reserves estimation: the challenge to the industry. Journal of
Petroleum Technology, 5: 80–88 (2007).
10. US Securities and Exchange Commission (US SEC). Modernization of the
Oil and Gas Reporting Requirements. Confirming version (proposed rule),
17 CFR Parts 210, 211, 229, and 249, Release Nos. 33-8995; 34-59192; FR-78;
File No. S7-15-08, RIN 3235-AK00, http://www.sec.gov/rules/final/2008/33-
8995.pdf (2008).
11. R. Harrel and C. Cronquist, Estimation of Primary Reserves of Crude Oil,
Natural Gas, and Condensate, Chapter 16, in Reservoir Engineering and
Petrophysics, Vol V(B), E. Holstein, (Ed.), pp. 1480–1570, SPE, Richardson,
TX (2007).
12. B.G. Dharan, Improving the relevance and reliability of oil and gas reserves
disclosures. Prepared/Written testimony for the US House Committee on
Financial Services. 108th Congress, Second Session, 21 July 2004, http://
archives.financialservices.house.gov/media/pdf/072104bd.pdf (2004).
13. J.R. Etherington, Managing your business using integrated PRMS and SEC
standards, SPE 124938. SPE Annual Technical Conference and Exhibition, 4–7
October 2009, New Orleans, Louisiana, (2009).
14. W.J. Lee, Modernization of the SEC oil and gas reserves reporting require-
ments, SPE 123793. SPE Annual Technical Conference and Exhibition, New
Orleans, Louisiana, 4–7 October 2009, (2009).
15. M.A. McLane, Reserve overlooking – the problem no one wants to talk about,
Paper SPE 68580. SPE Hydrocarbon Economics and Evaluation Symposium,
Dallas, Texas, 2–3 April 2001, (2001).
16. G.T. Olsen, W.J. Lee, and T.A. Blasingame, Reserves overbooking: the prob-
lem we’re finally going to talk about, SPE 134014. SPE Annual Technical
Conference and Exhibition, Florence, Italy, 20–22 September 2010, (2010).
17. E.D. Young and C.L. McMichael, Evaluation of fiscal terms, reporting
reserves, and financials. SPE 77510. SPE Annual Technical Conference and
Exhibition, San Antonio, TX, September 29–October 2 2001, (2002).
18. L. Denning, Nailing big oil’s moving production target. Wall Street Journal,
July 3 (2012).
19. L. Denning, Chevron battles Exxon for big oil’s crown. Wall Street Journal,
Aug 15 (2012).
20. Performance Profiles of Major Energy Producers. U.S. Energy Information
Administration, Office of Energy Statistics, Washington, DC (2009).
21. T. Turner and M. Moffett, Behind the battle for Argentina’s oil. Wall Street
Journal, May 3 (2012).
326 Reservoir Characterization
Resources
McMoRan 15 192 280 15 20 18.6 0.7 2,691 13
Exp.
Petroquest 2 175 184 28 31 6.5 1.1 467 108
Energy
Crimson 5 135 166 13 52 12.9 1.2 192 26
Exploration
Delta 2 123 134 15 8 9.1 1.0 22 64
Petroleum
(Continued)
Table A.2 Sample of independent gas producersa (2010). (Continued)
Reservesb
Rg/Ro
Liquid Gas Total Production PUD/R Market Capc (Mcf/
Company (MMbbl) (Bcf) (Bcfe) (Bcfe) (%) R/P (yr) D/Ec ($million) bbl)
Double Eagle 0.4 113 115 9 35 12.6 0.7 54 296
Pet.
Gasco Energy 0.5 40 43 4 0 9.8 0.9 42 85
CREDO 0.5 9 12 2 4 7.4 0.1 81 18
Petroleum
Mexco Energy 0.2 8 10 1 40 15.1 0.2 15 35
Royale Energy 0.02 6 6 1 12 9.1 1.2 22 337
All 2,560 67,637 82,995 5,793 45 14.3 1.3 164,471 26
a
Data collected from companies’ 2010 Annual Report, Financial Times, Yahoo-Finance, Scottrade.
b
Reported proved reserves on 2010 Annual Reports.
c
As of December 31, 2010.
Oil and Gas Company Production, Reserves, and Valuation 335
Part 5
UNCONVENTIONAL RESERVOIRS
15
An Analytical Thermal-Model for
Optimization of Gas-Drilling in
Unconventional Tight-Sand Reservoirs
Boyun Guo1*, Gao Li2 and Jinze Song1
2
Petroleum Engineering Department, Southwest Petroleum University
Abstract
Drilling wells with gas, or gas-drilling, has recently been adopted to drill uncon-
ventional oil and gas reservoirs including tight sands, gas shale and oil shale.
However, the performance of gas-drilling is very unpredictable in many areas due
to lack of optimization of drilling parameters. Because gas properties are greatly
affected by temperature, a reliable thermal model is required for gas-drilling opti-
mization. Such a model is not available and this paper fills the gap. An analytical
thermal-model was derived in this study for predicting bottom hole gas tempera-
ture under various flow conditions. The result given by the analytical model was
compared with that by an existing approximation, showing an accuracy improve-
ment of 14%. Sensitivity analysis with the new model indicates that formation
fluid influx dominates the temperature profile inside and outside the drill string,
while Joule-Thomson cooling and drill cutting’s intrusion affect temperature
only at bottom hole. Applications of the model are demonstrated in an example.
This paper provides an analytical tool to drilling engineers for optimizing their
gas-drilling operations.
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (339–362) © 2022
Scrivener Publishing LLC
339
340 Reservoir Characterization
15.1 Introduction
Drilling optimization is a process of using optimal drilling parameters
to increase rate of penetration and enhance drilling performance. Such
parameters include weight on bit, rotary speed and fluid flow rate. Drilling
optimization is particularly important in drilling unconventional oil and
gas reservoirs including tight sands, gas shale and oil shale where the rate
of penetration is usually very low. The process of drilling wells with gas,
or gas-drilling, is difficult to optimize because gas properties are greatly
affected by temperature. A reliable thermal model is required for predict-
ing bottom hole temperature in optimizing weight on bit and rotary speed.
Such a model is not available from literature.
Gas-drilling (drilling with air, nitrogen, etc.) has been traditionally used
for increasing rate of penetration (ROP) in hard rock formations [1]. This
technology has recently gained a strong momentum in unconventional
oil and gas field development in North America and other regions. Holt
et al. [2] reported that this technology reduced drilling time by 50% in
hard rock drilling in China. George et al. [3] illustrated significant cost sav-
ings achieved through the use of PDC bits with air/foam drilling. Lays and
Grayson [4] documented the performance gains realized with PDC bits in
air/foam drilling in the Appalachian Basin. Wilhide et al. [5] reported the
first rotary steerable system drilling with dry air used to improve low cost
development of an Unconventional gas reservoir. Zreik et al. [6] showed
improved surface hole air-foam drilling performance in the Karstifed
Limestone (Papua New Guinea). Pletcher et al. [7] demonstrated that the
application of air drilling improved drilling efficiency in horizontal sand-
stone wells. The first application of gas-drilling in shale gas field develop-
ment was reported by Ford et al. [8]. It reduced drilling costs and improved
efficiency of developing the Fayetteville shale gas reserves on the northern
Arkansas side of the Arkoma Basin. The advantages of drilling with air as
the circulating medium rather than mud were found significant. Air drill-
ing delivered faster ROP compared and reduced mud costs and incidence
of lost circulation. Utilizing this technology enabled the operator to drill
approximately 58% more footage at high ROP. The increased durability has
reduced the number of bits/trips required to complete the interval. Maranuk
et al. [9] reported a unique system for Underbalanced drilling using air in
the Marcellus shale. They described a drilling system and drilling param-
eters highlighting the differences between mud and air drilling. They dis-
cussed the modifications to the bottom hole assembly (BHA) to increase
reliability and drilling fluids and their effects on various pressure regimes.
Analytical Thermal-Model for Optimization of Gas-Drilling 341
They identified the major disadvantages of using air for drilling as its lim-
itation to handle fluid influx, the reduction of carrying capacity compared
to foam and other normal mud regimes, and the increased flow velocities
required to ensure adequate cuttings removal.
A few researchers have investigated the methods for predicting fluid
temperature profiles in drilling circulation systems. Among them are
Zhang et al. [10], Hasan and Kabir [11], Eppelbaum et al. [12], and Kutasov
and Eppelbaum [13]. Unfortunately, all these methods were developed
for liquid-drilling, not for gas-drilling. The only method for gas-drilling
is the numerical simulator developed by Wang et al. [14]. The paper was
published in a Chinese journal and the simulator is not accessible to the
authors. Li et al. [15] presented an approximate mathematical model for
predicting bottom hole gas temperature in gas-drilling unconventional
tight reservoirs where the drilling annulus is treated as an insulator.
A new analytical solution for predicting gas temperature profiles inside
drill string and in the annulus was derived in this study for gas-drilling,
considering the effects of formation fluid influx, Joule-Thomson cooling,
and entrained drill cuttings. The bottom hole temperature given by the
new analytical solution are found significantly higher than that given by
Li et al.’s [15] model. Results of sensitivity analyses show that formation
fluid influx can significantly increase the temperature profiles in both the
drill string and the annulus. The Joule-Thomson cooling effect lowers the
temperature in the annulus only in a short interval near the bottom hole.
The drill cuttings entrained at the bottom hole can slightly increase the
temperature profile in the annulus.
W/m- °C, respectively. The high contrast (>50) in the thermal conductivity
values makes the casing a super conductor for the heat conduction in the
radial direction. Therefore, the first assumption is valid.
Heat capacity of gas is a function of temperature and pressure [16]. In
the temperature range between 0 °C and 100 °C at atmospheric pressure,
the heat capacity of air varies between 1,005 J/kg-C and 1,009 J/kg-C, or
within 0.40%. In gas-drilling operations the gas pressure in the drill string
is in a narrow range between 7 MPa and 10 MPa. The heat capacity of
air varies between 1,016.2 J/kg-C and 1,021.6 J/kg-C, or within 0.53%, in
this pressure range [17]. Considering the extreme condition of 0 °C and
10MPa, the heat capacity of air varies between 1,005 J/kg-C and 1,021.6 J/
kg-C, or within 1.65%, which justifies the second assumption.
All gases used in gas-drilling are dilute gases in the above-mentioned
ranges of pressure and temperature. Gas density varies from 1 to 100 kg/
m3 and gas viscosity changes from 13.3 × 10–6 m2/s to 22.1 × 10–6 m2/s
[17]. The friction pressure drop in the whole circulation system is 15 MPa
at most, with the friction pressure drop inside the drill string being less
than 5 MPa over a few thousand meters of length. This low pressure drop
is not expected to generate significant amount of heat, and thus the third
assumption is valid.
When gas is injected into a drill string, the heat brought to the inside of
string is proportional to the product of fluid heat capacity C and mass flow
rate ṁp where
ṁp = ρpQp (15.1)
where ρp and Qp are gas density and volumetric flow rate inside drill string.
As the gas fows down the drill string, the rate of heat transfer through
drill string is proportional to the thermal conductivity of string Kp. When
the drilling fluid expands below the bit orifices, its temperature drops due
to Joule-Thomson cooling effect. The downstream temperature can be
expressed as [18]:
k −1
P k
Tdn = Tup dn (15.2)
P
up
where Tdn and Tup are the absolute temperatures in the downstream and
upstream of bit orifices, respectively, Pdn and Pup are the absolute pressures
in the downstream and upstream of bit orifices, respectively, and k is the
Analytical Thermal-Model for Optimization of Gas-Drilling 343
specific heat ratio of gas (k = 1.3 for gas). The temperature drop at the bit
is expressed as:
The gas receives heat from the entrained drill cuttings and formation oil
influx. Assuming all formation fluid influx occurs at bottom hole, the fluid
temperature should change at bottom hole in the annulus by
where ∆Tc and ∆Tf are temperature changes due to added drill cutting
and fluid influx, respectively. It can be shown that in the practical drilling
conditions where the rate of penetration is less than 50 m/hour, the term
∆Tc is negligible. The annular temperature at the bottom hole Tbh can be
expressed as:
where Tp is the temperature of fluid inside the drill string, Cf is the heat
capacity of the fluid influx, mf is the mass flow rate of fluid influx, and Tmax
is the geo-temperature at the bottom hole depth. Therefore the tempera-
ture change at the bottom hole can be expressed as:
The heat transfer in the annulus depends on the product of mixture heat
capacity Ca and mixture mass fow rate ṁc where
where the product of heat capacity and mass fow rate of solid cuttings Csṁs
is further expressed in two terms:
where Ch and Cr are the heat capacities of hydrocarbon and dry rock in the
cuttings, respectively. The mass flow rates of the hydrocarbon and rock in
the cuttings are respectively expressed as:
π 2
m h = Db R pϕρ h (15.9)
4
and
π 2
m r = Db R p (1− ϕ ) ρ r (15.10)
4
where Db, Rp, φ, ρh, and ρr are drill bit diameter, rate of penetration, rock
porosity, density of hydrocarbon, and density of rock, respectively. The Cf
in (15.4) is heat capacity of formation influx fluid (usually oil) and mass
flow rate of formation fluid influx is expressed as:
ṁf = ρfQf (15.11)
where ρf and Qf are density of fluid influx and flow rate of fluid influx,
respectively. As the fluid mixture flows up the annulus, the rates of heat
transfer through drill string and cement sheath are proportional to the
thermal conductivities of drill string Kp and cement sheath Kc, respectively
(the thermal conductivity of casing is assumed to be infinity compared to
that of cement sheath).
The gas temperatures inside the drill string Tp and in the annulus Ta
take the following forms respectively (derivation of solution is given in
Appendix):
AG + ABTg 0 − G( B + E )
Tp = C1 Ae r1L + C2 Ae r2 L + GL + (15.12)
AB
and
AG + ABTg 0 − EG
Ta = C1 ( A + r1 )e r1L + C2 ( A + r2 )e r2 L + GL +
AB
(15.13)
Analytical Thermal-Model for Optimization of Gas-Drilling 345
where
B + E − A + ( B + E − A)2 + 4 AB
r1 = (15.16)
2
B + E − A − ( B + E − A)2 + 4 AB (15.17)
r2 =
2
where
π dpK p
A= (15.18)
C pm pt p
π dc K c (15.19)
B=
Cam atc
π dpK p
E= (15.20)
Cam at p
Wooley [20], Marshall and Bentsen [21], Kabir et al. [22] and Hasan and
Kabir [11]. Unfortunately all these models were developed for liquid or
multi-phase flow. They are not applicable to gas flow in gas-drilling. The
analytical model for gas-drilling presented by Li et al. [15] was used for
comparison. The data used in the models are provided in Table 15.1.
Gas is the major component of air (>78%). Heat capacity of gas is a func-
tion of temperature and pressure [16]. In the temperature range between
0 °C and 100 °C at atmospheric pressure, the heat capacity of air varies
between 1,005 J/kg-C and 1,009 J/kg-C, or within 0.40%. In gas-drilling
operations the gas pressure in the drill string is in a narrow range between
7 MPa and 10 MPa. The heat capacity of gas varies between 1,016.2 J/
kg-C and 1,021.6 J/kg-C, or within 0.53%, in this pressure range [17].
Considering the extreme condition of 0 °C and 10MPa, the heat capacity
of gas varies between 1,005 J/kg-C and 1,021.6 J/kg-C, or within 1.65%.
Therefore, the heat capacity of gas was assumed to be constant in this study.
Figure 15.1 indicates that the injected gas is cooled down in the upper
section of drill string by the geothermal gradient. Gas is then heated up by
the geothermal gradient in the lower section of drill string. After arrival
in the annulus, the gas is quickly heated up by the geothermal gradient in
the lower section of the annulus. Eventually the gas is cooled down in the
upper section of the annulus by the geothermal gradient.
Also presented in Figure 15.1 is the gas temperature profile inside the
drill string given by Li et al.’s [15] analytical model. It is shown that the
bottom hole temperature given by the new analytical model is 7 °C higher
80
70
60
Temperature (C)
50
40
20 Geothermal temperature
Temperature inside drill string
10 Temperature inside drill string by Li et al. (2015)
0
0 500 1000 1500 2000 2500
Measured depth (m)
than that given by Li et al.’s [15] model. If the new model is considered to
be accurate, Li et al.’s [15] model is expected to underestimate bottom hole
temperature by 14%.
80
70
60
Temperature (C)
50
40
0
0 500 1000 1500 2000 2500
Measured depth (m)
70
60
50
Temperature (C)
40
30
70
60
50
Temperature (C)
40
30
k −1
P k
Tdn = Tup dn (15.21)
P up
where Tdn and Tup are the absolute temperatures in the downstream and
upstream of bit orifces, respectively, pdn and pup are the absolute pressures
in the downstream and upstream of bit orifces, respectively, and k is the
specific heat ratio of gas (k = 1.3 for air). Equation (15.21) gives in this case:
1.3−1
1.28 1.3
Tdn = (49 + 273.15) = 287°K = 14.2°C
2.10
The fold of increase in weight on bit (WOB) of PDC bit can be estimated
based on a simple relation derived from Glowka and Stone’s [23] thermal
model for bit cutters:
where
FWOB = fold of increase in WOB by Joule-Thomson cooling
tcr = critical temperature of PDC cutter, 350 °C
tdn,C = gas temperature in the downstream of bit after cooling, °C
tdn,N = gas temperature in the downstream of bit before cooling, °C
350 − 14.2
FWOB = = 1.12 or 12% increase.
350 − 49
Analytical Thermal-Model for Optimization of Gas-Drilling 351
15.6 Conclusions
A new closed-form analytical solution for predicting gas temperature pro-
files inside drill strings and in the annulus was derived in this study for
gas-drilling. The new solution has advantages over existing solutions in
that it can handle formation fluid influx, Joule-Thomson cooling effect,
and entrained drill cuttings. The following conclusions are drawn from this
study:
Nomenclature
Aa = cross-sectional area of annulus (m2)
Ap = cross-sectional area of drill pipe (m2)
Ca = heat capacity of fluid in the annulus (J/kg-°C)
Cf = heat capacity of formation fluid influx (J/kg-°C)
Ch = heat capacity of hydrocarbons in cuttings (J/kg-°C)
Cp = heat capacity of fluid inside drill pipe (J/kg-°C)
Cr = heat capacity of rock (J/kg-°C)
Cs = heat capacity of solid in the annulus (J/kg-°C)
Db = bit-diameter (m)
dc = inner-diameter of cement sheath (m)
Dc = outer-diameter of cement sheath (m)
dp = inner-diameter of drill pipe (m)
Dp = outer-diameter of drill pipe (m)
G = geothermal gradient (°C/m)
k = specific heat ratio of gas (dimensionless)
Kc = thermal conductivity of cement (W/m- °C)
Kp = thermal conductivity of drill pipe (W/m-°C)
L = wellbore depth along the drill string (m)
Lmax = the maximum hole depth (m)
ṁa = mass flow rate in the annulus (kg/s)
ṁf = mass flow rate of formation fluid influx (kg/s)
ṁh = mass flow rate of hydrocarbons in cuttings (kg/s)
ṁp = mass flow rate inside the drill pipe (kg/s)
ṁs = mass flow rate of solid cuttings in the annulus (kg/s)
ṁr = mass flow rate of rock (kg/s)
Pdown = absolute pressure in the downstream (psi)
Pup = absolute pressure in the upstream (psi)
Qf = formation fluid influx rate (m3/s)
Qp = fluid injection flow rate (m3/s)
RP = rate of penetration (m/s)
tc = thickness of cement sheath (m)
tp = wall thickness of drill pipe (m)
Ta = temperature of annular fluid (°C)
∆Tb = temperature change at drill bit (°C)
Tdn = absolute temperature in the downstream (°C)
Tg = geothermal temperature at depth (°C)
Tg0 = geothermal temperature at surface (°C)
Tp = temperature of fluid inside drill pipe at depth (°C)
Tp0 = temperature of fluid inside drill pipe at surface (°C)
Tup = absolute temperature in the upstream (°C)
Analytical Thermal-Model for Optimization of Gas-Drilling 353
Greeks
φ = porosity of rock (dimensionless)
ρa = fluid mixture density in the annulus (kg/m3)
ρf = density of formation fluid (kg/m3)
ρh = density of hydrocarbons in cuttings (kg/m3)
ρp = fluid density inside drill pipe (kg/m3)
ρr = density of dry rock (kg/m3)
ρs = solid density in the annulus (kg/m3)
Acknowledgements
This research was supported by the China National Natural Science
Foundation Founding No. 51274220, No. 51134004, No. 51221003,
51274045, 51274221, and No. 51334003.
References
1. W.C. Lyons, B. Guo, and F.A. Seidel, Air and Gas Drilling Manual, 2nd edition,
McGraw-Hill, New York (2001).
2. C. Holt, S.W. Nas, C. Shen, and X. Niu, Managed pressure drilling reduces
China Hard Rock Drilling by Half, Paper SPE 105490 presented at the SPE/
IADC Drilling Conference held in Amsterdam, 20–22 February, 2007.
3. B. George, E.S. Grayson, R. Lays, F.C. Felderhoff, M.L. Doster, and M.A.
Holmes, Significant cost savings achieved through the use of PDC bits in
compressed air/foam applications, Paper SPE 116118 presented at the SPE
Annual Technical Conference and Exhibition held in Colorado, Denver,
21–24 September, 2008.
4. R. Lays and E.S. Grayson, Performance gains realized with PDC bits in Air/
Foam applications in the Appalachian Basin, Paper SPE 117695 presented
at the SPE Eastern Regional/AAPG Eastern Section Joint Meeting held in
Pittsburgh, 11–15 October, 2008.
5. S. Wilhide, J. Smith, D. Doebereiner, B. Raymond, D. Weisbeck, and B.
Ziemke, First rotary steerable system drilling with dry air is used to further
improve low cost development of an unconventional gas reservoir, Paper SPE
135471 presented at the SPE Annual Technical Conference and Exhibition
held in Florence, Italy, 19–22 September, 2010.
6. A. Zreik, R.H.J. Pinkstone, and A.I. Ross, Improving surface hole air-foam
drilling performance in Karstifed Limestone (Papua New Guinea), Paper
SPE 135859 presented at the IADC/SPE Asia Pacific Drilling Technology
354 Reservoir Characterization
22. C.S. Kabir, A.R. Hasan, G.E. Kouba, and N.M. Ameen, Determining circu-
lating fluid temperature in drilling, workover, and well-control operations.
SPE Drilling & Completion 11(2), 74–79 (June 1996).
23. D.A. Glowka and C.M. Stone, Thermal response of polycrystalline diamond
compact cutters under simulated downhole conditions. SPE Journal (April
1985), 143–156 (1985).
24. P. L. Moore, Five factor that affect drilling rate. Oil Gas J. 56(40), 141–162
(1958).
Governing Equation
Figure 15.5 depicts a small element of a borehole section with a drill string
at center. Consider the heat flow inside the drill pipe during a time period
of ∆t. Heat balance is given by
where
Qp,in = heat energy brought into the drill pipe element by fluid due to
convection, J
Qp,out = heat energy carried away the drill pipe element by fluid due to
convection, J
qp = heat transfer through the drill pipe due to conduction, J
Qp,chng = change of heat energy in the fluid, J.
These terms can be further formulated as
Cement
Cement Annulus
Inside
Annulus
drill pipe
ΔL qp qa
∂Tp
q p = π d p K p∆L − ∆t (A.4)
∂r
∂ Tp
C pm p∆t (Tp , L − Tp , L +∆L ) + π d p K p∆L t = ρpC p Ap∆L∆Tp
∂ r
(A.6)
∂ Tp ρ p Ap ∂ Tp π d p K p ∂ Tp
+ = (A.8)
∂L m p ∂ t C pm p ∂ r
∂ Tp Τa − Tp
= (A.9)
∂r tp
∂ Tp ∂ Tp
+ λp + α p (Tp − Ta ) = 0 (A.10)
∂L ∂t
where
ρ p Ap
λp = (A.11)
m p
π dpK p
αp = (A.12)
C pm pt p
Consider the heat flow in the annulus during a time period of ∆t. Heat
balance is given by
where
Qa,in = heat energy brought into the drill pipe element by fuid due to
convection, J
Qa,out = heat energy carried away the drill pipe element by fuid due to
convection, J
qa = heat transfer through casing and cement due to conduction, J
Qa,chng = change of heat energy in the fuid, J.
358 Reservoir Characterization
∂T
qa = π dc K c ∆L − a ∆t (A.16)
∂r
∂ Tp
Cam a∆t (Ta,L +∆L − Ta,L ) + π d p K p∆L ∆t
∂ r
∂T
+ π da K a∆L a ∆t = ρaCa Aa∆L∆Ta
∂r
(A.18)
∂ Ta ∂T ∂ Tp ∂T
Cam a − ρaCa Aa a − π d p K p + π dc K c a = 0
∂t ∂t ∂r ∂r
(A.20)
∂ Tp Τa − Tp
= (A.21)
∂r tp
and
∂ Ta Τ g − Ta
= (A.22)
∂r tc
∂Ta ∂T
∂L ∂t
( )
− λ a a + β a Tp − Ta − α a Ta − Tg = 0
( ) (A.23)
where
ρa Aa
λa = (A.24)
m a
π dpK p
βa = (A.25)
Cam at p
π dc K c
αa = (A.26)
Cam atc
∂ Tp
∂L
(
+ α p Tp − Ta = 0 )
(A.27)
∂ Ta
∂L
( ) (
+ β a Tp − Ta − α a Ta − Tg = 0
) (A.28)
360 Reservoir Characterization
Boundary Conditions
The boundary conditions for solving (A.27) and (A.28) are expressed as
Tp = Tp0 at L = 0 (A.30)
Solution
The governing (A.27) and (A.28) subjected to the boundary conditions
(A.30) and (A.31) were solved with the method of characteristics. The
solutions take the following form:
Aa + ABb − a( B + E )
Tp = C1 Ae r1L + C2 Aer2 L + aL + (A.32)
AB
Aa + ABb − aE
Ta = C1 ( A + r1 )e r1L + C2 ( A + r2 )e r2 L + aL + (A.33)
AB
where
B + E − A + ( B + E − A)2 + 4 AB
r1 = (A.36)
2
B + E − A − ( B + E − A)2 + 4 AB
r2 = (A.37)
2
Abstract
Production of natural gas from unconventional gas-hydrate reservoirs faces sev-
eral challenges. One of them is the well control issue due to the natural gas released
from gas hydrates during well drilling. It is very important for drilling engineers
to know if the temperature of drilling fluid in the borehole is lower than the crit-
ical temperature that causes hydrate dissociation. However, there is no reliable
method to predict the fluid temperature for designing drilling hydraulics. This
paper fills the gap. An analytical model was developed in this study for predict-
ing temperature profile in drilling gas-hydrate deposits. A case study indicates a
good consistency between model-implications and field observations. Sensitivity
analyses with the model show that the bottom-hole temperature in gas-hydrate
drilling is dominated by the temperature and flow rate of the injected drilling fluid.
The temperature of drilling fluid in the annulus can become greater than the geo-
temperature at the same depth at high fluid flow rate. The Joule-Thomason cooling
effect below the drill bit nozzles rapidly diminishes in a short interval above the
bottom hole due to the heating effect of geo-thermal gradient. The rate of pene-
tration of drill bit has a negligible effect on the fluid temperature profile due to the
low percentage of heat flow contributed by the drill cuttings. This paper provides
drilling engineers a rigorous method for predicting wellbore temperature profile
during drilling gas-hydrates reservoirs.
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (363–382) © 2022
Scrivener Publishing LLC
363
364 Reservoir Characterization
16.1 Introduction
The amount of energy trapped in the naturally occurring gas hydrates has
been found in the world to be about twice the amount of energy found in all
recoverable fossil fuels. Due to the large amount of natural gas stored in the
huge deposits of gas hydrates, gas hydrates have been considered to be the
future clean energy resources [1]. Production of natural gas from gas hydrates
faces great challenges, uncertainties, and special issues [2, 3]. Challenges
encountered during drilling gas hydrates wells are complications due to the
gas released from hydrates. The in-situ gas hydrates decompose when pres-
sure is reduced and temperature is increased due to the circulation of drilling
fluid in the borehole. The release of gas from the hydrates can cause borehole
stability [4] and well control problems [5, 6]. Therefore, accurately predicting
the temperature in the wellbore is critical to drilling gas hydrates wells.
Some mathematical models are available for estimating fluid tempera-
ture during drilling conventional wells. An approximate solution that cou-
ples a pseudo-steady heat transfer in the wellbore to transient heat flow in
the formation was presented by Ramey [7]. Raymond [8] derived complete
heat transfer equations for the well-bore-reservoir system. A number of
further investigations have been conducted since then. Holmes and Swift
[9] presented a method to calculate circulating mud temperatures under
pseudo-steady flow conditions. Sump and Williams [10] used a mathemat-
ical model to predict wellbore temperature during mud circulation and
cementing operations. All of these early analytical solutions do not con-
sider the rate of penetration and heat-transport due to mixing of mud with
drill cuttings. Keller et al. [11] provided a numerical transient heat transfer
model for predicting temperature distribution in circulating mud columns.
Wooley [12] applied finite difference method to computing downhole tem-
peratures in circulation, injection, and production wells. His model predic-
tions were found in agreement with field temperature data. Following Keller
et al.’s approach Marshall and Bentsen [13] developed a computer model to
calculate the temperature distribution in a well-bore using finite difference
method with an improved solution procedure. Arnold [14] calculated tem-
perature variation in circulating wellbore fluid. Kabir et al. [15] presented
mathematical models for determining the circulating fluid temperature in
drilling, workover, and well-control operations. Results from their models
do not agree with that from Wooley’s (1980) [12] model near the bottom
hole. This is attributed to the fact that Kabir et al. made an assumption
of stationary temperature at the bottom hole, i.e., the temperature gra-
dient in the axial direction was assumed to be zero at the bottom hole.
Development of Analytical Model for Predicting Fluid Profile 365
where
Qp,in = heat energy brought into the drill pipe element by fluid due to
convection, J
Qp,out = heat energy carried away the drill pipe element by fluid due to
convection, J
qp = heat transfer through the drill pipe due to conduction, J
Qp,chng = change of heat energy in the fluid, J.
Inside
Annulus drill pipe Annulus
ΔL qp qa
∂ Tp
q p = π d p K p∆L − ∆t (16.4)
∂ r
∂ Tp
C pm p∆t (Tp , L − Tp , L +∆L ) + π d p K p∆L ∆t = ρ pC p Ap∆L∆Tp (16.6)
∂ r
Dividing all the terms of this equation by ΔLΔt yields
∂ Tp ρ p Ap ∂ Tp π d p K p ∂ Tp
+ = (16.8)
∂L m p ∂ t C pm p ∂ r
The radial-temperature gradient in the insulation layer can be formu-
lated as
∂ Tp Ta − Tp
= (16.9)
∂r tp
Substituting Eq. (16.9) into Eq. (16.8) yields
∂ Tp ∂ Tp
+ λp + a p (Tp − Ta ) = 0 (16.10)
∂L ∂t
where
ρ p Ap
λp = (16.11)
m p
368 Reservoir Characterization
π dpK p .
ap = (16.12)
C pm pt p
Consider the heat flow in the annulus during a time period of Δt. Heat
balance is given by
where
Qa,in = heat energy brought into the drill pipe element by fluid due to
convection, J
Qa,out = heat energy carried away through the drill pipe element by fluid
due to convection, J
qa = heat transfer through casing and cement due to conduction, J
Qa,chng = change of heat energy in the fluid, J.
∂T
qa = π dc K c ∆L − a ∆t (16.16)
∂r
∂ Tp ∂T
Cam a∆t (Ta,L +∆L − Ta,L ) − π d p K p∆L ∆t + π dc K c ∆L a ∆t
∂r ∂r
∂ Ta ∆T ∂ Tp ∂T
Cam a − ρaCa Aa a − π d p K p + π dc K c a = 0
∂L ∆t ∂r ∂r
(16.20)
∂ Tp Ta − Tp
= (16.21)
∂r tp
and
∂ Ta Tg − Ta (16.22)
=
∂r tc
∂Ta ∂T
− λa a + βa (Tp − Ta ) − aa (Ta − Tg ) = 0 (16.23)
∂L ∂t
where
ρa Aa
λa = (16.24)
m a
π dpK p
βa = (16.25)
Cam at p
π dc K c
aa = (16.26)
Cam atc
For steady heat flow, Eqs. (16.10) and (16.23) can be written as:
∂ Tp
+ a p (Tp − Ta ) = 0 (16.27)
∂L
∂ Ta
+ βa (Tp − Ta ) − aa (Ta − Tg ) = 0 (16.28)
∂L
Tg = Tg0 + GL (16.29)
Tp = Tp0 at L = 0 (16.30)
AG + ABTg 0 − G( B + E )
Tp = C1 Ae r1L + C2 Ae r2 L + GL + (16.32)
AB
and
AG + ABTg 0 − EG
Ta = C1 ( A + r1 )e r1L + C2 ( A + r2 )e r2 L + GL +
AB
(16.33)
Development of Analytical Model for Predicting Fluid Profile 371
where
B + E − A + ( B + E − A)2 + 4 AB (16.36)
r1 =
2
B + E − A − ( B + E − A)2 + 4 AB (16.37)
r2 =
2
where
π dpK p
A= (16.38)
C pm pt p
π dc K c
B= (16.39)
Cam atc
π dpK p .
E= (16.40)
Cam at p
where the mass flow rate inside the drill string is expressed as
.
m p = ρpQp. (16.42)
where ρp and Qp are density of fluid inside the pipe and fluid flow rate,
.
respectively The product of heat capacity and mass flow rate of solid Csm s
is further expressed in two terms:
where
π 2
m h = Db R pϕρh (16.44)
4
and
π 2
m r = Db R p (1 − ϕ )ρr (16.45)
4
where Db, Rp, ϕ, and ρr are drill bit diameter, rate of penetration, rock
porosity, and density of rock, respectively.
It is understood that the temperature change at drill bit due to Joule-
Thomason effect (Tb0) depends on fluid properties. For compressive flu-
ids the temperature in the downstream of the bit orifices (bottom hole) is
lower than the temperature in the upstream and can be predicted based on
isentropic process [21]:
k −1
P k
Tdn = Tup dn (16.46)
P up
where Tdn and Tup are the absolute temperatures in the downstream and
upstream of bit orifices, respectively, Pdn and Pup are the absolute pressures
in the down-stream and upstream of bit orifices, respectively, and k is the
specific heat ratio of fluid (k = 1.3 for air). For incompressive fluids, such as
water, the temperature in the downstream of the bit orifices (bottom hole)
has been found to be higher than the temperature in the upstream due
to the negative Joule-Thompson coefficient [22]. Because this temperature
change is fluid-dependent, the value of Tb0 in the analytical model is left
user-specified.
Development of Analytical Model for Predicting Fluid Profile 373
12
10
8
Temperature (C)
4
Temperature in the annulus
2 Geothermal temperature
Temperature inside drill string
0
0 100 200 300 400 500 600 700
Depth below sea floor (m)
12
10
Temperature (C)
4
Temperature in the annulus
2 Geothermal temperature
Temperature inside drill string
0
0 100 200 300 400 500 600 700
Depth below sea floor (m)
12
10
Temperature (C)
4
Temperature in the annulus
2 Geothermal temperature
Temperature inside drill string
0
0 100 200 300 400 500 600 700
Depth below sea floor (m)
Figure 16.4 Model-predicted temperature profiles with mud flow rate 0.05 m3/s.
12
10
8
Temperature (C)
Figure 16.5 Model-predicted temperature profiles with 3 °C-temperature drop at the drill
bit due to Joule-Thomson effect.
Figures 16.1 and 16.3 shows that the fast injection of cold mud reduces bot-
tom hole temperature. The upward flowing fluid mixture (mud and rock
cuttings with hydrates) in the annulus is heated to the geo-temperature in
a short interval above the bottom hole. Also the temperature of the upward
flowing fluid mixture in the annulus is slightly elevated due to the fast con-
vection of heat from the deeper section.
Figure 16.5 demonstrates temperature profiles calculated by Eqs. (16.32)
and (16.33), considering a 3 °C temperature drop due to Joule-Thomson
Development of Analytical Model for Predicting Fluid Profile 377
effect below the drill bit nozzles. This assumption is made for oil-base mud
with a positive Joule–Thomason coefficient. Oil-base muds are often utilized
in drilling hydrate-bearing formations to prevent hole enlargement due to
decomposition of gas hydrates. A comparison between Figures 16.1 and 16.4
shows that the Joule-Thomason effect rapidly diminishes due to the heating
effect of geo-thermal gradient in a short interval above the bottom hole.
The effect of ROP on annular temperature was investigated with Eq.
(16.33). When the ROP is doubled, its effect was found not noticeable. This
is because the heat added by the drilling cuttings is less than 0.5% of the
total heat flow at the total well depth.
16.5 Conclusions
An analytical model was developed in this study for predicting tempera-
ture profiles in drilling gas hydrates deposits. A case study with the model
was performed. Parameter sensitivity analyses were carried out with the
model. The following conclusions are drawn:
Acknowledgements
This research was supported by the China National Natural Science
Foundation Founding No. 51134004, No. 51274220, and No. 51334003.
Nomenclature
Aa = cross-sectional area of annulus open for fluid flow (m2)
Ap = cross-sectional area of drill pipe open for fluid flow (m2)
Ca = heat capacity of fluid in the annulus (J/kg- °C)
Cp = heat capacity of fluid inside drill pipe (J/kg- °C)
Cs = heat capacity of solid in the annulus (J/kg- °C)
Ch = heat capacity of gas hydrates (J/kg- °C)
Cr = heat capacity of rock (J/kg- °C)
Db = bit-diameter (m)
dc = inner-diameter of cement sheath (m)
Dc = outer-diameter of cement sheath (m)
dp = inner-diameter of drill pipe (m)
Dp = outer-diameter of drill pipe (m)
k = specific heat ratio of fluid, dimensionless
Kc = thermal conductivity of cement (W/m- °C)
Kp = thermal conductivity of drill pipe (W/m- °C)
L = wellbore depth along the drill string (m)
Lmax = the maximum hole depth (m)
.
m a = mass flow rate in the annulus (kg/s)
.
m p = mass flow rate inside the drill pipe (kg/s)
.
m s = mass flow rate of solid cuttings in the annulus (kg/s)
.
m h = mass flow rate of gas hydrates (kg/s)
.
m r = mass flow rate of rock (kg/s)
Qp = fluid flow rate (m3/s)
Development of Analytical Model for Predicting Fluid Profile 379
Greeks
ϕ = porosity of rock
ρa = fluid density in the annulus (kg/m3)
ρh = density of hydrates (kg/m3)
ρp = fluid density inside drill pipe (kg/m3)
ρr = density of dry rock (kg/m3)
ρs = solid density in the annulus (kg/m3)
References
1. Y.F Makogon, Natural gas hydrates–A promising source of energy. J. Nat.
Gas Sci. and Eng. 2, 49–59 (2010).
2. G.J. Moridis, T.S. Collett, R. Boswell, M. Kurihara, M.T. Reagan, C. Koh, and
E.D. Sloan, Toward Production from gas Hydrates: Current status, assess-
ment of resources, and simulation-based evaluation of technology and
potential. SPE J. 12(5), 745–771 (2009).
3. G.J. Moridis, T.S. Collett, M. Pooladi-Darvish, S. Hancock, C. Santamarina, R.
Boswell, T. Kneafsey, J. Rutqvist, M.B. Kowalsky, M. Reagan, E.D. Sloan, A.K.
Sum, and C.A. Koh, Challenges, uncertainties, and issues facing gas produc-
tion from Gas-Hydrate deposits. SPE Reserv. Eval. Eng. 14, 76–112 (2011).
4. T. Khabibullin, G. Falcone, and C. Teodoriu, Drilling through gas-hydrate
sediments: Managing Wellbore-Stability risks. SPE Drilling & Completion
26, 287–294 (June 2011).
5. D.M. Hannegan, R.J. Todd, D.M. Pritchard, and B. Jonasson, MPD–uniquely
applicable to methane hydrate drilling, SPE/IADC Underbalanced Technology
Conference and Exhibition, SPE/IADC 91560 (2004).
380 Reservoir Characterization
25. W.J. Winters, Physical and geotechnical properties of gas hydrate bearing
sediment from offshore India and northern cascadia margin compared
to other hydrate reservoirs, 7th International Conference on Gas Hydrates,
pp. 1–22 (2011).
26. F. Incropera and D. Dewitt, Heat Transfer, John Wiley and Sons Inc. New
York (1985).
27. J. Li, B. Guo, and B. Li, A closed form mathematical model for predicting gas
temperature in gas-drilling unconventional tight reservoirs. J. Nat. Gas Sci.
Eng. 27, 284–289 (2015).
17
Distinguishing Between Brine-
Saturated and Gas-Saturated Shaly
Formations with a Monte-Carlo
Simulation of Seismic Velocities
Simon Katz*, George Chilingar and Leonid Khilyuk
Abstract
The Monte-Carlo method was used to model seismic velocities in shaly sands
defined by a pair of linear equations with randomized coefficients. The starting
point for this research was a deterministic model with linear equations defining
compressional (Vp) and shear waves (Vs) velocities in gas or brine-filled shaly
deposits in the North Sea. We introduced two levels of randomization. At the first
level, porosity and clay content were treated as random variables. At the second
level, in addition to randomized porosity and clay content, coefficients of equa-
tions were randomized and random errors in seismic velocities were added. The
effects of errors in measured seismic velocities and random coefficients were not
separable. Therefore, we introduced a summary measurement of the randomiza-
tion effect that includes the effects of all random parameters.
The goal of the Monte Carlo simulation was to evaluate the efficiency of iden-
tification of gas-saturated deposits using seismic velocity data. We tested three
methods for the identification of gas-filled formations: k-nearest neighbor, recur-
sive partitioning, and linear discriminant analysis. The efficiency of identification
was characterized by two conditional probabilities: (1) The Probability of True
Discovery, which is defined as the probability of identification of a gas-saturated
formation as gas saturated, and (2) The Probability of False Discovery which is
defined as the false identification of a brine-saturated formation as gas saturated.
Both probabilities were evaluated as functions of the width of distribution of
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (383–398) © 2022
Scrivener Publishing LLC
383
384 Reservoir Characterization
17.1 Introduction
The goal of this paper is to analyze the efficiency of classification of gas
and brine-filled shaly formations using the Monte Carlo technique. The
economic success of shale gas in the US over the last ten years has led to
the rapid development of shale gas in the world, thus improving the sus-
tainability of worldwide energy resources.
The Monte Carlo method relies on repeated random sampling followed
by the calculation of statistical attributes defined by a set of randomized
equations. We generated up to 10000 random samples of porosity, clay
content and coefficients of model equations to calculate multiple replicas
of Vp and Vs velocities.
The results of the Monte Carlo simulations were used to evaluate the
efficiency of identifying gas-filled formations against brine-filled ones. The
identification efficiency is characterized by the following two conditional
probabilities:
data and on the uncertainty of the coefficients in the Vp, Vs models. Another
factor that affects the classification probabilities is the set of predictors used
for classification.
The Monte Carlo simulation and all statistical analysis is done with
statistical software R [4–6]. The general principles of statistical classifica-
tion techniques have been studied [1, 6, 7]. Some approaches to classifi-
cation techniques and machine learning methods in geological problems
are presented in academic research [8]. The Monte Carlo techniques were
used in seismic inversion and geological research in various publications
[9–11]. Deterministic equations that link seismic velocities with porosity,
clay content and the effect of pore fluid are discussed in studies by Schon
and Buryakovsky et al. [12, 13]. The starting point for the analysis done in
this paper was a model for Vp and Vs velocities expressed as linear functions
of porosity and clay content. These types of models were built as a result of
linear regression for the data inside a limited range of porosity variations,
De-hua Han*, A. Nur (1986).
ag,0 = 4.82; ag,1 = 5.02; ag,2 = 0.597; bg,0 = 3.26 ; bg,1 = 3.03; bg,2 = 0.892
ab,0 = 5.46; ab,1 = 6.29 ; ab,2 = 1.1; bb,0 = 3.32 ; bb,1 = 3.62; bb,2 = 0.952
We assume that equations 17.1–17.4 define the mean values of the sto-
chastic Vp and Vs velocities. For purposes of the Monte-Carlo simulation
386 Reservoir Characterization
Random components hg,1, hg,2, hb,1, and hb,2 simulate errors in the Vp, Vs
velocities estimated from seismic data and the imprecise nature of coeffi-
cients ag,0, bg,0, ab,0, bb,0.
b. Equations 17.1–17.4 with randomized coefficients in Equations 17.1–17.4.
Random variables δg,0, δg,1, μg,2, μg,0, μg,1, μg,2, δb,0, δg,1, μb,2, μb,0, μb,1, δg,0
model the imprecise nature of equations 17.1–17.4. It is important to note
that the effect of random variables δg,0, μg,0, δb,0, μg,0 is identical to the effect
of the errors in Vp, Vs estimated from seismic data.
We assume that the random variables in equations 17.9–17.12 have zero
mean, mutually independent, and their probability distributions are uni-
form in a predefined interval symmetric around zero.
Distinguishing Brine-Saturated and Gas-Saturated Formations 387
1
; x <= a (17.13)
P( x ) = 2a
0; x > a
Figures 17.1 and 17.2 show standard deviations calculated from two sets
of simulated Vp, Vs velocities as functions of half the width of the interval of
distribution (parameter a in eq. 17.13). For each value of parameter a, ten
thousand Vp(k) and Vs(k), 1<=k<=10000, samples are calculated. These
samples are used to calculate standard deviations.
According to Figures 17.1 and 17.2, the standard deviations of the Vp,
Vs and Vp/Vs ratios are strongly dependent on the variability of random
components in equations 17.9–17.12 and monotonically increase with an
increase of half the width of the distribution interval.
One can observe the strong effect of the width of distribution interval
of random components in eqs. 17.9–17.12 on the variability of Vp, Vs and
their ratio. The Vp standard deviation is somewhat higher than the stan-
dard deviation of Vs. The standard deviation of the Vp/Vs ratio is about the
same for gas-filled and brine-filled formations.
388 Reservoir Characterization
0.14
Velocity STD
0.10
std(Vp)
0.06
std(Vs)
Figure 17.1 Gas-filled formations. The standard deviations of calculated Vp, Vs serve
as unctions of the half-width of the distribution interval for random components in
eqs. 17.10–17.13. The standard deviation is calculated using 10000 Vp and Vs samples
for a range of porosity from 0.1 to 0.16 and clay content in the range of 0.3 to 0.4. Standard
deviation is given in km/sec.
0.08
STD of Vp/Vs ratio
0.04
GAS
BRINE
0.00
0.05 0.10 0.15 0.20
Half width of the distribution interval
Figure 17.2 Standard deviations of the Vp/Vs ratio for gas and brine saturated formations.
Vp and Vs velocities were calculated for 10000 samples of random components in eqs.
17.10–17.14.
4.5 4.5
4.0 4.0
Vp
Vp
3.5 3.5
GAS GAS
3.0 Brine Brine
3.0
Figure 17.3 Clusters of (Vp,Vs) velocities for gas-filled and brine-filled formations.
Two velocity clusters are perfectly separable in the case of the first level
randomization of the model. The velocity clusters partially overlap if the
models are randomized with strong random components in the coeffi-
cients of equations 17.9–17.12.
TU = V.std + a (17.14)
GAS
where Class =
BRINE
(17.16)
and the predictor component in eq. 17.15 has one of the following forms:
predictors = Vp (17.17)
predictors = Vp/Vs (17.18)
Each record in the training set has the form (class, predictors).
Estimates of the probability of true and false discoveries were calculated
as follows:
N test (G | G ))
P .true.discovery = (17.21)
N test (G )
Distinguishing Brine-Saturated and Gas-Saturated Formations 391
N test (G | B))
P . false.discovery = (17.22)
N test ( B)
where
Ntest (G|G)) - number of records in the test set with Class variable equal
GAS that is classified as GAS.
Ntest (G|B)) - number of records in the test set with class variable equal
BRINE that is classified as GAS.
Ntest (G)) - number of records in the test set with Class variable equal
GAS.
Ntest (B)) number of records in the test set with Class variable equal
BRINE.
Additional useful parameter – the probability of missed discovery is
defined as follows:
N test (G | G )) (17.23)
P .missed .discovery = 1 −
N test (G )
1.0
0.8
Probability
0.6 Predictors
Vp
Vp+Vs+Vp/Vs
Vp+Vs+Vp/Vs +Vp*Vs
0.4
Vp/Vs
0.4
0.3
Probability
0.2
Predictors
0.1 Vp
Vp+Vs+Vp/Vs
Vp+Vs+Vp/Vs +Vp*Vs
0.0 Vp/Vs
When total uncertainty is as high as 0.4, two sets of predictors (eqs. 17.19,
17.20) lead to the probability of false discovery being no higher than 0.3.
According to Figure 17.6, two predictor sets (eqs. 17.19 and 17.20)
guarantee the highest probability of true discovery at small values of total
uncertainty. For these predictors, the probability of true discovery is larger
than 0.9 in the range of total uncertainty (0, 0.1). In the case of high total
uncertainty ~ 0.4 , two predictor sets (eqs. 17.19 and 17.20) guarantee that
the probability of true discovery is around 0.7. As in the case of KNN clas-
sification, single predictors produce inferior results compared to results
with the group of predictors.
According to Figure 17.7, classification with a single predictor Vp leads
to a comparatively high probability of false discovery even when total
1.0
0.8
Probability
0.6 Predictors
Vp
Vp+Vs+Vp/Vs
Vp+Vs+Vp/Vs +Vp*Vs
0.4
Vp/Vs
0.3
Probability
0.2
Predictors
0.1 Vp
Vp+Vs+Vp/Vs
Vp+Vs+Vp/Vs +Vp*Vs
0.0 Vp/Vs
uncertainty is around zero. Three other sets of predictors give the proba-
bility of false discovery as smaller than 0.1 in a range of total uncertainty of
(0, 0.1). When total uncertainty is as high as 0.4, two sets of predictors (eqs.
17.19, 17.20) lead to a probability of false discovery of around 0.33. Vp/Vs as
a predictor gives a false discovery rate as high as 0.4.
1.0
0.8
Probability
0.6 Predictors
Vp
Vp+Vs+Vp/Vs
Vp+Vs+Vp/Vs +Vp*Vs
0.4
Vp/Vs
Figure 17.8 Linear discriminant analysis and the probability of true discovery.
Distinguishing Brine-Saturated and Gas-Saturated Formations 395
0.4
0.3
Probability
0.2
Predictors
0.1 Vp
Vp+Vs+Vp/Vs
Vp+Vs+Vp/Vs +Vp*Vs
0.0 Vp/Vs
Figure 17.9 Linear discriminant analysis and the probability of false discovery.
1.0
0.9
Probability
0.8
0.7
KNN
RPART
LDA
0.6
0.1 0.2 0.3 0.4
Total uncertainty
Figure 17.10 The probability of true discovery and the three classification techniques
Classification model: class =Vp + Vs + Vp/Vs + Vp *Vs.
396 Reservoir Characterization
0.30
0.20
Probability
0.10
KNN
0.00 RPART
LDA
17.6 Conclusions
1. The Monte-Carlo method was used to model seismic velocities
defined by a pair of linear equations with randomized coefficients.
Deterministic versions of these equations were designed by [14]
to model Vp and Vs velocities in gas or brine-filled shaly deposits
in the North Sea.
2. We introduced two levels of randomization. At the first level,
porosity and clay content were treated as random variables. At
the second level, in addition to randomized porosity and clay
content, coefficients of equations were randomized and random
errors in seismic velocities were included in the model. The effects
of errors in measured seismic velocities and random coefficients
in the equations were not separable. Therefore, we introduced a
summary measurement of randomization effects that included
the effects of all random parameters.
3. A summary attribute named ‘total uncertainty’ was constructed
to characterize the effect of random parameters and errors in
Distinguishing Brine-Saturated and Gas-Saturated Formations 397
References
1. R. Duda, E. Peter, E. Hart, and. D. Stork, Pattern Classification (2nd Edition),
pp. 637, Whiley Interscience Publication (2000).
2. L. Breiman, J. Freidman, and R. Olsen, C. Stone, Classification and Regression
Trees, p. 368, Chapman & Hall/CRC, Boca Raton (1984).
3. G.J. McLachlan, Discriminant Analysis and Statistical Pattern Recognition,
pp. 530, Wiley Interscience (2004).
398 Reservoir Characterization
Abstract
In the last decade, the unconventional resource has changed the global energy
and petroleum industry. The technological innovations of hydraulic fracturing
and horizontal drilling techniques, which require a good understanding of the
rock mechanical properties of shale, have driven the economical production of
shale reservoirs. However, the intrinsic rock-mechanical properties of shale rocks
are extremely complex; they are affected by many factors, including confining
pressure, water content, water salinity, TOC (total organic carbon), clay content,
bedding plane orientation, mineralogy, anisotropy and others. Some factors (i.e.,
water content, TOC and clay content etc.) can significantly decrease the strength
of shale rocks, and this weakening mechanism needs to be considered in hydraulic
fracture designing and wellbore stability.
This paper systematically investigates various key factors that impact the rock-
mechanical properties especially the mechanical properties of shale rocks, including
confining pressure, water content, TOC (Total organic carbon), clay content, bed-
ding plane orientation, porosity, anisotropy, and temperature effects. We experimen-
tally investigated the impact of water saturation on Barnett shale’s rock-mechanical
properties (Young’s modulus, E and Poisson’s ratio, ν and uniaxial unconfined com-
pressive strength (UCS)) using a Material Testing System (MTS-810). Experimental
results showed that: 1) water saturation has a significant impact on Young’s modulus
and UCS. Young’s modulus decreases 6.1% with an increase of a water saturation
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (399–426) © 2022
Scrivener Publishing LLC
399
400 Reservoir Characterization
(1%) of Barnett shale core corresponding to the value of dry shale’s Young’s modu-
lus; 2) Young’s modulus and UCS decrease linearly with increasing water saturation;
And 3) the relationship between water saturation and Poisson’s ratio is not obvious.
18.1 Introduction
In the last decade, the unconventional resource has changed the global
energy and petroleum industry. According to the International Energy
Agency (IEA), the volume of unconventional gas resources is estimated at
380,000 billion cubic meters (Gm3), equivalent to about 50% of global gas
resources, and shale gas accounts for the biggest share of these resources
(Unconventional Gas, 2012) [1]. In the U.S.A., Shale gas has been produced
for many years, and accounts for about 8% of total natural gas production
(Warlick, 2006) [2]. Gas shales in the U.S.A. are predicted (AEO, 2011)
to become the source of 45% of all gas production by 2035 [3], especially
considering that other fossil energy resources constitute larger threats for
climatic shift, environmental pollution and potential risks for production/
exploration. The increasing significance of shale gas plays has led to the
need for deeper understanding of shale behaviors.
Shale is a sedimentary rock formation which is generally composed
of clay, quartz and other minerals (Survey of Energy, 2010) [4]. Shale has
extremely low permeability (ranging from 10–6 md to 10–2 md) and nano-
pores in shale matrix (Liu et al., 2012) [5]. Therefore, Shale gas cannot be
produced at economical rates without stimulation, typically hydraulic frac-
turing, and horizontal wells (Tran, 2009) [6]. Some of the most critical factors
that affect hydraulic fracture propagation, wellbore stability, and the produc-
tivity of shale gas wells are the rock-mechanical properties of shale rocks
(i.e., Young’s modulus, E, Poisson’s ratio, ν, and the strength of shale rock)
(Sharma et al., 2004) [7]. Meanwhile, mechanical properties of gas shale are
affected by many factors, including confining stress, water content, TOC
(Total organic carbon), bedding plane orientation, anisotropy, temperature
and so on. Accordingly, the development of shale reservoir should consider
the effects of these factors and shale gas development nowadays becomes
one of the thorniest technical challenges faced by the petroleum industry.
bedding plane orientation, total organic content and so on. Each factor is
reviewed as follows.
Pe = Pc − nPp (18.1)
6.0
5.0
VP (km/s)
4.0
3.0
2.0
0 10 20 30 40 50
Effective pressure (MPa)
Figure 18.1 Vp versus effective pressure for sandstone samples [9] (Zhang and Bentley,
2000).
402 Reservoir Characterization
Rupture
line
Sheor
o
d
c
•
Tension 0 Compression
f 'c f 'c
Figure 18.2 Typical Mohr rupture diagram for concrete [13] (Wuerker, 1959).
stress effects and confinement can be treated together with Mohr’s theory
(Wuerker, 1959). Mohr’s theory of failure (Mohr’s rupture line) has been
found to be one of most useful tools in analyzing the stress of rocks (Chong
et al., 1984; LaI, 1999). Coates and Denoo (1981) applied Mohr’s Circle to
determine a sand’s ultimate strength. Zhang et al., (2006) and Islam et al.,
(2010) conducted shale wellbore stability analysis using Mohr’s theory.
Rocks will not fail if the stress is within the Mohr’s rupture line. Figure 18.2
shows the typical Mohr rupture diagram made by Wuerker (1959) [13–18].
18.2.2 Porosity
The influence of porosity on shale’s rock-mechanical properties has been
well investigated. Spikes and Dvorkin (2004) studied the relation between
Poisson’s ratio and porosity using synthetic seismic modeling. Based on
the simulation and well logging analysis, Poisson’s ratio increases with the
increment of porosity (see Figure 18.3) [19].
Kumar et al., (2012) studied porosity dependent elastic modulus of
shale by using nano-indentation technique. Based on their research,
Young’s modulus decreases with an increase of shale porosity (Figure
18.4). Furthermore, Kumar et al. (2012) concluded that the development of
porosity took place during the thermal maturation of organic matter, and
localized presence of porosity played a crucial role in mechanical proper-
ties of organics in shale [20, 21].
Shale Mechanical Properties Influence Factors 403
0.38 130
1.0 0.8
0.36 120
110
0.34
Poisson’s ratio
100
0.32 90
0.30 80
0.0 70
0.28
60
0.26 50
0.6 0.4 0.2
0.24 Clay content 40
0.00 0.05 0.10 0.15 0.20 0.25 0.30 GR
Porosity
Figure 18.3 Poisson’s ratio versus total porosity [19] (Spikes and Dvorkin, 2004).
100 Woodford
Ei = 67–3.8*φ Barnett
80 R2 = 0.4
Haynesville
60 Eagle ford
Ei, GPa
40
20
0
0 5 10 15
φ, %
Figure 18.4 The relationship of Young’s modulus and porosity [20] (Kumar et al., 2012).
water content will make the menisci become weaker or disappear, and
the rock will become weaker [22–25].
Van Eeckhout (1976) studied the effect of humidity on the strength and
elastic properties of a few samples obtained from coal mine shale. Their
experimental work showed that with an increase of moisture content (rel-
ative humilities of 0.48 and 100%), there was a reduction in compressive
strength and Young’s modulus, and an increase in Poisson’s ratio [26].
Lashkaripour and Passaris (1993) compiled a database with selected
values of shale rock properties and concluded that there was also a sig-
nificant negative correlation between water content and compressive
strength in coal mines shale. Uniaxial tests were widely used to study
the weakening effect of water content on shale’s rock-mechanical proper-
ties. Steiger and Leung (1989) performed uniaxial compressive strength
(UCS) tests on three categories of shale rocks under both dry and wet
conditions. The testing results demonstrated that a dryer sample (lower
water content) has a higher unconfined compressive strength. Table 18.1
lists the uniaxial compressive strengths measured with both wet and dry
samples. As Table 18.1 shows, the uniaxial compressive strengths mea-
sured on dry samples can be 2 to 10 times larger than that of the wet
samples [27, 28].
The elastic properties changed remarkably during desaturation and
resaturation tests (Pham et al., 2005). Pham et al. (2005) carried out
desaturation and resaturation uniaxial compressive tests on Bure shale to
observe the influence of water content or RH (relative humidity) on elastic
properties of shale specimens. The UCS and Young’s modulus increased
when the specimens were desaturated and decreased during resaturation
(Figure 18.5 (a) and (b)) [29].
Lin and Lai (2013) conducted laboratory tests on water satura-
tion effects on Barnett shale’s geomechanical properties using uniaxial
strength testing method.
The results showed that Young’s modulus and unconfining compressive
strength decreased linearly with increasing water saturation. Poisson’s ratio
showed insensitivity to the change of water content [30].
Table 18.1 Uniaxial compressive strength of typical shales [28] (Steiger and
Leung, 1989).
Sample condition Unconfined compressive strength (1000 psi)
Shale E Shale F Shale G
Wet 6.5 4 0.5
Dry 14 12 5
Shale Mechanical Properties Influence Factors 405
12000
11000
8000
7000
6000
5000
4000
30 40 50 60 70 80 90 100
(a) Relative humidity (%)
60
55
Compressive strength (MPa)
50
45
40
35
30
25
30 40 50 60 70 80 90 100
(b) Relative humidity (%)
Figure 18.5 (a) The variation of Young’s modulus with RH [29] (Pham et al., 2005) and
(b) The variation compressive strength with RH [29] (Pham et al., 2005).
3.5 80
VP –fast (km/s)
VS –fast (km/s)
5.5
70
5 3
60
4.5 50
2.5
4 40
30
0 2 4 6 8 10 12 0 2 4 6 8 10 12 0 2 4 6 8 10 12
TOC (%) TOC (%) TOC (%)
Figure 18.6 The relationship of Vp-fast & Vs-fast and Young’s modulus with TOC [37]
(Gupta et al., 2012).
Shale Mechanical Properties Influence Factors 407
100 Woodford
y = 58.5–3.3* TOC
R2 = 0.48 Barnett
80
Haynesville
60 Eagle Ford
Ei, GPa
40
20
0
0 5 10 15
TOC, wt%
Figure 18.7 The relationship of Young’s modulus and TOC [21] (Kumar et al., 2012).
100 Woodford
90 Barnett
y = 63–0.45* clay
80 R2 = 0.47 Haynesville
70 Ordovician
60 Eagle ford
Ei, GPa
50
40
30
20
10
0
0 20 40 60 80 100
Clay, wt%
Figure 18.8 The relationship of Young’s modulus and clay content [21] (Kumar et al.,
2012).
250
200
UCS (MPa)
150
Figure 18.9 The influence of clay content on UCS [41] (Sone and Zoback, 2011).
1.2
1
normalized to that measured
perpendicular to bedding
0.8
0.6
0.4
0.2
0
0 10 20 30 40 50 60 70 80 90
Sample orientation to bedding (0 = perpendicular, 90 = parallel)
Tournemire27 Siltshale28 Mudshale28
2 2
HRZ shale Kingak shale Pierre 1 shale
Figure 18.10 Experimental data showing strength anisotropy in shales [42] (Willson
et al., 2007).
Due to the limited number of samples, the authors claimed that the effects
of bedding plane to the strength was unclear and needed more experimen-
tal supportive results [43].
Wu and Tan (2010) measured acoustic velocity on shale samples with
various bedding planes. They found that the Young’s modulus from paral-
lel to the bedding planes was about 37% higher than that perpendicular to
the bedding planes. According to the triaxial stress experiment conducted
by Li et al., (2012), the core samples perpendicular to the bedding plane
have the highest strength, then the samples paralleling bedding plane, and
410 Reservoir Characterization
100
Middle-vertical
Middle-horizontal
90 Upper-vertical
Deviatoric strength (MPa) Upper-horizontal
80
70
60
50
40
6 7 8 9 10 11 12 13 14
Figure 18.11 Strength of horizontally and vertically cored samples from Upper and
Middle Woodford [43] (Abousleiman et al., 2008).
the strength of the samples in the bedding plane orientation 60° is lowest
(see Figure 18.12) [44, 45].
Alqahtani et al. (2013) conducted triaxial tests and ultrasonic velocity
tests to study the effects of bedding plane orientation (0°, 45° and 90°) on
Green River, Eagle Ford and Mancos shales. The results demonstrated that
the maximum compressive strength, static elastic moduli, and compressive
velocity occurred at horizontal samples (see Figure 18.13) [46].
250
a
Compressive strength, MPa
200
a
150
100
50
0
0 20 40 60 80 100
Angle of coring and bedding planes normal, deg
Figure 18.12 The strength of shale core in different direction [45] (Li et al., 2012).
Shale Mechanical Properties Influence Factors 411
25000
20000
Compressive strength (psi)
15000
Vertical
45
10000 Horizontal
5000
0
Green river Mancos Eagle Ford, La Salle
Figure 18.13 Effect of the bedding angle on compressive failure strength [46] (Alqahtani
et al., 2013).
18.2.8 Mineralogy
Mineralogy is another key factor that affects shale mechanical properties.
To examine mineral compositions effects on elastic moduli, Marouby and
Heidari (2011) used a self-consistent approximation (Voigt-Reuss-Hill’s
approximation) method to calculate shale matrix bulk (Kmat) and shear
moduli (μmat) based on the interpretation of sonic, nuclear, and resistivity
logs. Table 18.3 shows different mineral constituents of three-layer shale
formation (Cq, Ccal, Cill, Cpy and Cker respectively denotes quartz, calcite,
0 0
5 5
10 10
15 15
Depth (ft)
20 20
25 25
30 30
35 35
40 40
45 45
13 14 15 16 8 9 10
Ksat (GPa) µsat (GPa)
Figure 18.14 The bulk and shear moduli with different mineral concentrations of 3 layer
formation [47] (Marouby and Heidari, 2011).
100
y = 26.3 + 0.4* QC
80 R2 = 0.48
60
Ei, GPa
40
20
0
0 20 40 60 80 100
Quartz plus carbonate, wt%
Figure 18.15 The dependence of Young’s modulus on quartz and carbonate [20] (Kumar
et al., 2012).
illite, pyrite and kerogen concentration), and Figure 18.14 shows that dif-
ferent mineral concentrations lead to different bulk and shear moduli.
Based on the study of Marouby and Heidari (2011), mineral constituent is
a critical factor that infuences elastic properties of shale [47].
Shale Mechanical Properties Influence Factors 413
18.2.9 Anisotropy
Shale generally is a highly anisotropic laminated sedimentary rock. The
degree of anisotropy of shale depends on many factors such as kerogen con-
tent (Vernik and Landis, 1996), TOC and clay content (Kumar et al., 2012).
The larger the kerogen, TOC and clay content, the higher is the anisotropy
of the shale (Vernik and Nur, 1992; Sondergeld et al., 2000; Sondergeld and
Rai, 2011; Kumar et al., 2012 and Zhu et al., 2012) [20, 48–52].
Abousleiman et al., (2008) studied the relation of elastic strength and
anisotropy of Woodford shale samples using an inclined direct shear test-
ing device (IDSTD). According to the study of Abousleiman et al., (2008),
anisotropy influences the elastic properties of shale specimens. But the
relationship of Woodshale strength and anisotropy is not yet clear due
to the limited number of tests. Dewhurst et al., (2008) conducted triaxial
tests using North Sea Shale samples and measured ultrasonic P-wave and
S-wave. They concluded that the shale speciments exhibited slightly larger
P-wave anisotropy and moderate S-wave anisotropy [11, 43].
18.2.10 Temperature
Temperature also can influence geomechanical properties of shale. High
temperatures will lead to kerogen decomposition, thereby weakening the
shale strength. Handin and Hager (1958) performed short-time triaxial
compression tests on shale samples to study the deformation of rocks at
elevated temperature conditions. The experimental results show that an
increase of temperature at constant pressure reduces the yield stress and
raises the ultimate strength by increasing the ductility of shale. Closmann
and Bradley (1979) investigated the temperature effects on mechanical
properties of oil shale. The testing results demonstrated that with tem-
perature increasing, Young’s modulus, tensile strength and compressive
strength all decrease. Similarly, Akrad et al., (2011) used nano-indentation
technique and found out that the Young’s modulus of shale decreased
significantly when they were heated to high temperatures (~300 °F) [33,
53, 54].
414 Reservoir Characterization
Plot Poisson’s ratio vs. water saturn Figure 18.17. It can be seen that
partially saturated Barnett shale’s Poisson’s ratio fluctuates around the val-
ues of the dry samples. In other words, there is no obvious relationship
between water saturation and Poisson’s ratio.
Shale Mechanical Properties Influence Factors 415
φ Sw E ν UCS
Test # % % (10 ) psi
6
psi
1 0 3.88 0.224 4,528
2 15.5 32.2 2.4 0.287 3,159
3 15.1 20.9 3.58 0.45 3,989
4 13.8 2.4 3.64 4,193
5 14.3 29.9 3.11 0.092 2,939
6 14.0 17.8 3.4 0.311 3,433
7 15.0 7.4 3.59 3,872
8 0 3.64 4,429
9 13.9 35.1 2.87 0.21 2,965
10 14.4 33.6 2.72 0.725 2,793
11 14.3 9.3 2.73 2,903
12 14.5 11.3 3.49 4,232
13 14.6 10.9 3.52 0.278 3,960
14 15.4 22.2 3.51 0.277 4,463
15 15.3 8.4 3.37 3,444
16 16.0 8.6 3.33 0.04 3,871
17 15.2 12.8 3.73 0.109 4,077
18 15.3 24.7 3.14 0.107 2,760
19 16.3 13.2 2.84 1,906
Where φ is porosity; Sw is water saturation; E is Young’s modulus, psi; ν is
Poisson’s ratio; and UCS is uniaxial compressive strength, psi.
4000000
3800000 Young’s modulus vs. Sw
3600000
Young’s modulus (psi)
3400000
3200000
3000000
2800000 y = –22819.98x + 3758858.52
R2 = 0.69
2600000
2400000 Young’s modulus
2200000 Fitting line of Young’s modulus
2000000
0 10 20 30 40
Water saturation (%)
0.8
Poisson’s ratio vs. Sw
0.7
0.6
Poisson’s ratio
Poisson’s ratio
0.5
Fitting line of Poisson’s ratio
0.4
0.3
0.2
y = 0.0012x + 0.2166
0.1
R2 = 0.0203
0
0 10 20 30 40
Water saturation (%)
Previous work and our experimental results show that induced water
tends to decrease the strength and Young’s modulus. Shales are rich in clay
minerals, and in the presence of water, clays with expandable minerals
such as smectites will expand and make shale softer and weaker.
Shale Mechanical Properties Influence Factors 417
5000
UCS vs. Sw
4500
4000
3500
UCS (psi)
3000
2500
y = –43.252x + 4357.2
2000 UCS
Fitting line of UCS R2 = 0.7627
1500
0 10 20 30 40
Water saturation (%)
C 3,674,567.36 4,128.51
Sy 519,658.53 890.88
418 Reservoir Characterization
4000000 5000
Young’s modulus vs. Sw UCS vs. Sw
4500
Young’s modulus (psi)
UCS (psi)
3000000
3000
Young’s modulus
2500 UCS
2500000 The best line The best line
Fitting line of Young’s modulus 2000
Fitting line of UCS
2000000 1500
0 10 20 30 40 50 0 10 20 30 40 50
(a) Water saturation (%) ( b) Water saturation (%)
Figure 18.19 (a) and (b) The fitting line and the “best” line for E and UCS.
m=
∑ x y − xy i i
∑ x − (∑ x ) / N
2 (18.3)
2
i
C = y − mx (18.4)
∑ D = ∑( y − Y )
i
2
i i
2
(18.5)
Sy =
1
N −2 ∑D i
2
(18.6)
18.4 Conclusions
1.
Shale is a highly laminated anisotropic sedimentary rock. The
mechanical properties of shale are affected by many factors. After a
thorough review of key factors that affect shale’s mechanical proper-
ties, the following conclusions can be drawn.
Shale Mechanical Properties Influence Factors 419
4000000
3000000
Figure 18.20 (a) and (b) The data range analysis for E.
UCS
2500 3000 3500 4000 4500 5000
5000
UCS vs. Sw
4500
4000
UCS (psi)
3500
UCS (psi)
3000
2500
y = –43.252x + 4357.2
2000 R2 = 0.7627
1500
0 10 20 30 40 50 0 5 10 15 20 25 30 35
(a) Water saturation (%) (b) Water saturation (%)
Figure 18.21 (a) and (b) The data range analysis for UCS.
Acknowledgements
The authors are grateful to Aramco Service Company for the support and
the permission to publish this paper. We thank the Petroleum Engineering
Department of ULL for the support of this study.
References
1. Total: Unconventional Gas. Unconventional Gas Exploration & Production,
Paris, France (2012).
Shale Mechanical Properties Influence Factors 421
2. D. Warlick, Gas shale and CBM development in North America. Oil Gas
Financial Journal. 3(11), 1–5 (2006).
3. Annual Energy Outlook (AEO). National Energy Modeling System, Run Ref,
D120810C (2011).
4. Survey of Energy Resources: Focus on Shale Gas, Report of World Energy
Council, London, United Kingdom (2010).
5. C. L. Liu, R. Sharma, and A.Tutuncu, Shale Gas Resources: Energy Potential and
Associated Exploitation Challenges for Coupled Geomechanics and Transport
Characteristics, 46th U.S. Rock Mechanics/Geomechanics Symposium, Chicago,
Illinois, June 24–27, Paper No. ARMA-2012-209 (2012).
6. M. H. Tran, Geomechanics Field and Laboratory Characterization of
Woodford Shale, A thesis for the degree of Master of Science. University of
Oklahoma, Norman, Oklahoma (2009).
7. M. M. Sharma, P. B. Gadde, R. Sullivan, R. Sigal, R. Fielder, D. Copeland, L.
Griffn, and L. Weijers, Slick Water and Hybrid Fracs in the Bossier: Some
Lessons Learnt, SPE Annual Technical Conference and Exhibition, Houston,
TX, USA, September 26–29, Paper No. SPE 89876 (2004).
8. A. Tinni, C. H. Sondergeld, C. S. Rai, and H. Kouam, Effective Pressure
and Microstructure Control on Resistivity Formation Factor and Seismic
Waves Velocities, SPE Annual Technical Conference and Exhibition, Denver,
Colorado, USA, 30 October-2 November, Paper No. SPE-147432-MS
(2011).
9. J. J. Zhang and L. R. Bentley, Change of Elastic Moduli of Dry Sandstone
with Effective Pressure, SEG Annual Meeting, August 6–11, Calgary, Alberta,
Paper No. SEG-2000-1826 (2000).
10. M. N. Toksoz, C. H. Cheng, and A. Timur, Velocities of seismic waves in
porous rocks. Geophysics 41, 621–645 (1976).
11. D. N. Dewhurst, A. F. Siggins, U. Kuila, M. B. Clennell, M. D. Raven, and
H. M. Nordgard-Bolas, Elastic, Geomechanical and Petrophysical Properties
of Shales, 42nd US Rock Mechanics Symposium and 2nd U.S.-Canada Rock
Mechanics Symposium, San Francisco, USA, June 29- July 2, Paper No. ARME
08-208 (2008).
12. A. M. Al-Tahini and Y. N. Abousleiman, Pore-pressure-coeffcient anisot-
ropy measurements for intrinsic and induced anisotropy in sandstone. SPE
Reservoir Evaluation & Engineering 13(2), 265–274 (2010).
13. R. G. Wuerker, Infuence of Stress Rate and Other Factors on Strength and
Elastic Properties of Rocks, 3rd U.S. Symposium on Rock Mechanics (USRMS),
Golden, CO, April 20–22, Paper No. ARMA-59-0032 (1959).
14. K. Chong, J. Chen, G. Dana, and S. Sailor, Triaxial testing of Devonian oil
shale. J. Geotech. Engrg. 110(10), 1491–1497 (1984).
15. M. LaI, Shale Stability: Drilling Fluid Interaction and Shale Strength, SPE
Latin American and Caribbean Petroleum Engineering Conference, Caracas,
Venezuela, April 21–23, Paper No. SPE 54356 (1999).
422 Reservoir Characterization
Abstract
In order to explore the potential applications of nanotechnology in oil industry to
enhance oil recovery, numerous studies regarding nanoparticle mobility in porous
media have been conducted in the past decade. Theoretical modeling was carried
out to investigate nanoparticle retention mechanism as well. However, most of the
simulations were either limited to laboratory scale or only accounted for transport
and retention behaviors. In this paper, the effect of hydrophilic nanofuids on oil
recovery was quantified at a reservoir scale. Simulations were performed using an
in-house simulator and three essential tasks were accomplished: (1) The in-house
simulation framework consists of two critical components: a two-phase displace-
ment module (accounts for waterflooding) and a transport/retention module
(accounts for Advection-Dispersion-Retention behavior exhibited by nanoparti-
cles). Those two modules were verified with two simulation tools (Eclipse from
Schlumberger and MNM1D by Tosco et al.), respectively in this study; (2) A solu-
tion algorithm bridging the above two modules was developed and implemented
to demonstrate the effect of nanoparticle retention on improving displacing effi-
ciency; (3) Different injection scenarios simulated include continuous injection,
slug injection, and postflush, and the effects of injection time, injection rate, and
slug size on oil recovery were studied. The results discovered that when nanofuids
flooding is used after waterflooding as a tertiary recovery method, early nano-
fuids injection improves higher oil recovery, but results in more nanoparticle loss.
Higher injection rate of nanofuids could help improve the flooding efficiency, but
not the ultimate oil recovery and cause more nanoparticle loss as well. Compared
to continuous injection, slug injection with postflush saves a lot of nanoparticles
but present a similar contribution in a Nano-EOR process.
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (429–462) © 2022
Scrivener Publishing LLC
429
430 Reservoir Characterization
19.1 Introduction
Introduced by Feynman in 1959, nanotechnology was an idea to
precisely manipulate atoms and molecules and to create nanoscale
machines. Nanotechnology has shown its potential for revolution-
ary changes in oil and gas industry in the past few years. Engineered
nanoparticles have been proposed for many potential applications in
upstream oil and gas industry, especially as emulsion/foam stabilizers for
enhanced oil recovery (EOR) purpose or as nano-sensors for formation
evaluation. Zhang et al., (2009), and Espinosa et al., (2010) both have
found that surface-coated silica nanoparticles can stabilize emulsions
and foams for mobility control in EOR through experimental studies.
Functionalized nanoparticles can be injected into reservoir formations
as sensors to detect certain reservoir rock and/or fluids properties [1].
Other engineering applications in the improved oil recovery process
include injecting nanoparticle dispersion into reservoirs for improved
oil recovery [2–6]. Several studies have proved that SiO2 nanoparticles
(NPs) can alter the wettability of reservoir rock and reduce the interfa-
cial tension (IFT) between crude oil and brine phases by their adsorp-
tion onto the rock surface [7], (Onyekonwu and Ogolo 2010). Le et al.
(2009) studied synergistic blends of SiO2 nanoparticles and surfactants
for EOR in high-temperature reservoirs. They proved that some blends
have great potential in EOR application because of their thermostability
at 91°C, extremely good IFT reduction and very high oil displacement
efficiency. Hendraningrat et al., (2013) used Hydrophilic silica nanopar-
ticles for a coreflood experiment to investigate nanofuids for EOR. Their
results showed that oil recovery increased with nanofuids flooding in
high permeability sand stone core sample, but not for low permeabil-
ity one. For both environmental concern and reservoir application, it is
important to understand nanomaterial transport properties in porous
media with single- or two-phase flow.
Over the past few years, nanoparticles transport in porous media has
been studied. Both experiments and theoretical models were carried
out to evaluate the nanoparticle mobility and investigate the nanopar-
ticle retention mechanism. Systematical nanomaterial transport exper-
iments were carried out by Lecoanet et al. [8] to evaluate their mobility
Numerical Investigation of Enhanced Oil Recovery 431
The Darcy’s law and the equations of mass conservation for each phase
govern the two-phase immiscible incompressible flow in a homogeneous
porous medium. The equations are given below.
Kkrw
υw = − ∇Pw (19.1)
µw
Kkro
υo = − ∇Po (19.2)
µo
∂ Sw
φ + ∇ • υw = 0 (19.3)
∂t
434 Reservoir Characterization
∂ So
φ + ∇ • υo = 0 (19.4)
∂t
Sw + So = 1 (19.5)
where the subscripts w and o denote the wetting phase (water) and
non-wetting phase (oil), respectively, P is pressure, v is Darcy velocity vec-
tor, Sw is water saturation, So is oil saturation, kr is the relative permeability
and μ is viscosity.
The sum of the velocities of both the wetting and non-wetting phases is,
Kkrw Kk
υt = υ w + υ o = − ∇Pw − ro ∇Po (19.6)
µw µo
Based on Eqs. (19.1), (19.2) and (19.5), one may get that,
∇ • υt = 0 (19.7)
Or,
Kk Kk
∇ • rw ∇Pw + ro ∇Po = 0 (19.8)
µw µo
Pc = Po − Pw (19.9)
Po = Pw (19.10)
k k
∇ • K rw + ro ∇Pw = 0 (19.11)
µw µo
Numerical Investigation of Enhanced Oil Recovery 435
S
krw ,s = krw + (k ∗ − krw ) (19.12)
S ∗ rw,s
S
kro ,s = kro + (k ∗ − kro ) (19.13)
S ∗ ro,S
where krw,S* and kro,S* are the relative permeabilities of water and oil phases
when the surfaces per unit bulk volume of the porous media are completely
occupied by the nanoparticles, while krw,S and kro,S are the relative permea-
bilities of water and oil phases when the surfaces per unit bulk volume of
the porous media are partially occupied by the nanoparticles.
In this study, two groups of relative permeability data obtained from
experiments were used. One is for S = 0, and the other one is for S = S*. For
all the situations in between (0 < S < S*), a linear relationship was assumed.
The experimental data is based on the study of Parvazdavani et al., in 2012
as shown in Table 19.2 [15].
∂ (φ Sw Cw )
+ υw • ∇Cw = ∇ • (φ Sw Dw ∇Cw ) (19.14)
∂t
∂ (Sw Cw ) ρb ∂ S
+ =0 (19.15)
∂t φ ∂t
∂ S Swφ S
= ka 1 − Cw − kd S (19.16)
∂ t ρb Smax
Swφ S
ka 1 − Cw − kd S = 0 (19.17)
ρb Smax
Smax
S∗ =
ρ k S (19.18)
1 + b d max
φ ka Cw
Which implies that S* is always smaller than Smax when kd is not zero. In
this study, S* is denominated as the equilibrium adsorption at given values
of adsorption and desorption rates, which is also the maximum “possible”
adsorption (less than Smax).
on the results from the second step and the third step, krw,S and kro,S can be
calculated by linear interpolation.
So the whole process is a decoupled process. The operator splitting
error is acceptable according to the results of verification cases previously.
The nanoparticles transport and retention model influence flow model by
updating the relative permeability of both water and oil.
The solution for mathematical models for one time step is shown in
Figure 19.1.
The entire process is actually known as an ADRE (Advection-Dispersion-
Reaction-Equation) problem. To simplify the calculation, operator-splitting
was employed and the ADRE was therefore split into an ADE problem (Eq.
19.14) plus Reaction behaviors (Eq. 19.15-19.16), mathematically a PDE
and coupled ODEs, which are solved by ADE solver and Retention model,
respectively (as seen in the yellow box of Figure 19.1). The simultaneous
ADE and Reaction (the reaction here is referred to as retention including
adsorption and desorption of nanoparticles) are considered sequentially in
the presented modeling (ADE first and then retention). This splitting will
introduce numerical error that is why a verification case (4.3) was studied
Start
Solve ADE
(Eq. 19.14),
obtain Cw
Linear interpolation
table 19.2 and table 19.3
obtain krw, 0 and kro, S* and Solve retention model
krw, 0, kro, S* (w, n) (Eq. 19.15 & 19.16),
obtain S and Cw
Figure 19.1 Flowchart of solution for mathematical models for one time step.
Numerical Investigation of Enhanced Oil Recovery 439
to prove that the produced splitting error is negligible. ADE are solved
implicitly and the ODEs can be solved by many methods, such as Rouge-
Kutta or an iterative method as used in this study.
0.9
0.8
0.7
Oil recovery, percent
0.6
0.5
0.4
0.3
0.2 Delta t = 0.1d
0.1 Delta t = 0.05d
0
0 500 1000 1500 2000 2500
Time, d
calculating the oil recovery factor with the time step of 0.1d, which is thus
employed as the time step for the following simulations.
The residual oil saturation after water flooding is 0.270, and that after
nanofluids flooding is 0.143, proving that nanofuids flooding could
enhance oil recovery significantly.
442 Reservoir Characterization
1.0
0.9
0.8 krw-water
0.7 kro-water
krw-nano
Kro, Krw
0.6
kro-nano
0.5
0.4
0.3
0.2
0.1
0.0
0.0 0.2 0.4 0.6 0.8 1.0
Sw
Figure 19.3 Oil and water relative permeability curves (repotted from Parvazdavani et al.,
[15].
The oil recovery factors (RF) produced from two simulators presented
a great agreement, which proved the consistency of two-phase flow com-
putation by in-house simulator comparing to a well-known commercial
simulator.
0.9
0.8
Oil recovery, percent 0.7
0.6
0.5
0.4 Nano (eclipse)
0.3 Nano (this study)
Water (eclipse)
0.2
Water (this study)
0.1
0
0 500 1000 1500 2000 2500
Time, d
Figure 19.4 Comparison of oil recovery change with time after water flooding or
nanofuids flooding computed by Eclipse and in-house model.
1.0
Normalized concentration (C/C0)
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1 Tosco et al. (2009)
This study
0.0
0.0 2000.0 4000.0 6000.0
Time (sec)
19.5 Results
In this chapter, the effect of nanoparticle injection time, injection rate,
and nanoparticle slug injection were all investigated by simulation. For
all the different injection scenarios, the amount of nanoparticles trapped
in the formation and ultimate oil recovery were calculated for compar-
ison. The distributions of nanoparticles on the rock surface and in the
fluid were plotted.
444 Reservoir Characterization
1.0
The model used here is a five-spot flooding model, and the parameters
of the model are shown in Table 19.4. One injection well is in the center of
the squared reservoir and four production wells are at each corner.
The input data used for analysis is based on literature review. Oil and
water relative permeability data is listed in Table 19.2 and Table 19.3
for water flooding and nanoflooding, respectively. Other input data is
based on the experiment #65 conducted by Zhang [20]. The unit of the
data is converted to the unit used in the model and the data is shown in
Table 19.5.
0.9
0.8
120
Nanoparticle trapped, ton
100
80
60
40
20
0
300 600 900 1200 1500
Injection time, d
50 0.84 50 0.86
0.82 0.84
40 0.8 40 0.82
0.78
Water saturation
0.8
30 0.76 30
0.78
0.74
0.76
20 0.72 20
0.74
0.7
10 0.68 10 0.72
Figure 19.9 Water saturation distribution at 0.5 PV, 1.5 PV, 2.25 PV and 3 PV respectively
when the time of nanofuids injection t = 300d (the color bar represents water saturation).
50 1 1
0.99
40 0.98 0.95
0.97
30 0.96 0.9
0.95
20 T = 600d 0.94 0.85
T = 900d
0.93
10 0.92 0.8
0.91
0 0.9 0.75
50 1 0.85
0.95 0.84
40 0.9 0.82
0.85
30 0.8
0.8
0.78
0.75
20 0.76
T = 1200d 0.7 T = 1500d
0.65 0.74
10
0.6 0.72
0 0.55 0.7
0 10 20 30 40 50 0 10 20 30 40 50
Figure 19.10 Nanoparticle distribution in the reservoir at the end of the nanofuids
injection (3PV) under different nanofuids injection time (the color bar represents
normalized nanoparticle concentration).
0.9
0.8
0.7
Oil recovery, percent
0.6
0.5 q = 2000 stb/d
0.4 q = 2500 stb/d
0.3 q = 3000 stb/d
q = 5000 stb/d
0.2
q = 10000 stb/d
0.1
0
0 500 1000 1500 2000 2500
Time, d
102
101
Nanoparticle trapped, t
100
99
98
97
96
2000 2500 3000 5000 10000
Injection rate, stb/d
Figure 19.12 The amounts of nanoparticles trapped in the reservoir at the end of
injection under different injection rate.
by 2000d (3PV), but when flow rate is 2000 stb/d, there is the least amount
of nanoparticle trapped (Figure 19.12), which makes it the best flow rate.
Also, the difference between the nanoparticles loss decreased as the
injection rate increased from 2000stb/d to 10000stb/d. This is because of
the existence of nanoparticle equilibrium concentration (S*). No more
nanoparticles would be adsorbed onto the rock surface when the concen-
tration increased to S*.
0.8
0.7
Oil recovery (fraction)
0.6
0.5
0.4
t_inj = 300d
0.3
t_inj = 600d
0.2 t_inj = 900d
0.1
0
0 500 1000 1500 2000 2500
Time (d)
Figure 19.13 Oil recovery with slug injection of nanofuids under different injection time.
Based on the result, it can be seen that for slug injection of nanofuids,
early injection would lead to higher oil recovery and less nanoparticle
retention. This is actually related to the desorption property of nanopar-
ticles. When the nanofuids were first injected into the reservoir, a lot of
nanoparticles were adsorbed onto the rock surface. Then after the slug
injection of nanofuids, water is injected into the reservoir, and nanopar-
ticles are desorbed and pushed out by water flooding. So with longer
water flooding, more nanoparticles would be pushed out, resulting in less
nanoparticle loss.
The water cut changes when injection time is 300d is shown in Figure
19.14. It can be seen from the curve that water cut decreased significantly
when nanofuids were in effect, and then increased slowly and stabilized at
100%. This means that oil production increased due to nanofuids injec-
tion. Based on the plot, the water cut reached 98% by 1450d (2.2 PV), so
in field application, the injection should have been stopped for this case
if only considering oil production. For nanofuids flooding, nanoparticle
Numerical Investigation of Enhanced Oil Recovery 451
1.2 1.2
0.6 0.6
0.4 0.4
t_inj = 300d
0.2 Water cut 0.2
0 0
0 500 1000 1500 2000 2500
Time (d)
Figure 19.14 Oil recovery and Water cut against injection time (nanofuids injection time
and time length were both 300d).
loss is also a problem for economy. Longer water flooding after nanofuids
flooding might be preferred since it can help recover more nanoparticles.
0.9
0.8
Oil recovery (fraction)
0.7
0.6
0.5
0.4 100d
0.3 300d
500d
0.2
0.1
0
0 500 1000 1500 2000 2500
Time (d)
Figure 19.15 Oil recovery with slug injection of nanofuids at different slug size.
452 Reservoir Characterization
called postflush. Three different slug sizes were selected, 100d, 300d and
500d respectively. The result is shown in Figure 19.15 and Table 19.7.
In conclusion, both injection time and slug size of nanofuids flood-
ing have obvious influence on oil recovery and nanoparticle retention.
Postflush could help to recover more nanoparticles, and long time post-
flush may be needed considering economic and environmental issues.
flush flow rate, e.g. brine water flow rate, might be helpful in nanoparticle
recovery. So the flow rate ratio between water flooding and nanoflood-
ing was looked into here. Three groups of simulation were completed.
Nanofuids injection time length was 300d. The flow rate of nanoflooding
was set at 2000stb/d, and the injection rate of water post flush changed
from 2000stb/d to 4000stb/d. The result is shown in Figure 19.17, Figure
19.18 and Table 19.8.
It can be seen from Figure 19.17 and Table 19.8 that there is not much
difference in the ultimate oil recovery for the post flush processes under
three different injection ratios. But, the difference between the mass of
nanoparticle retention is significant. Higher flow rate ratio results in higher
nanoparticle recovery. However, the water flow rate can’t be too high due to
the pump capacity and water flooding cost.
According to Figure 19.18, the nanofuids concentration was still increas-
ing after the nanoflooding period. This shows that the trapped nanoparti-
cles were washing out by brine water flooding. During the flooding period,
the effluent concentration was always above 0, which means the whole
process, including the postflush, is actually nanoflooding. But the concen-
tration of the fluid is not constant, it will increase first due to nanoparticle
desorption, and then decrease due to nanoparticles being washed out of
the formation. This happens probably because after nanoflooding, there
are a lot of nanoparticles adsorbed onto the rock surface, and the following
brine flooding could help nanoparticles desorb from the substrate and get
back into the water phase due to concentration difference. The desorbed
nanoparticles would be pushed forward and adsorb onto the unsaturated
0.9
0.8
0.7
Oil recovery (fraction)
0.6
0.5
0.4 t_inj = 0d
0.3 t_inj = 100d
t_inj = 300d
0.2
t_inj = 500d
0.1 t_inj = 700d
0
0 500 1000 1500 2000 2500
Time (d)
Figure 19.16 Oil recovery curves under different length of nanofuids injection time
followed by post flush.
454 Reservoir Characterization
0.8
0.7
0.6
Oil recovery (fraction)
0.5
0.4 Qw/Qn = 1
0.1
0
0 500 1000 1500 2000 2500
Time (d)
Figure 19.17 Oil recovery under different flow rate ratio between water flooding and
nanoflooding.
0.4
0.35
Qw/Qn = 1
Dimensionaless C/C0
0.3
concentration C/C0
Qw/Qn = 1.5
0.25
Qw/Qn = 2
0.2
0.15
0.1
0.05
0
0 500 1000 1500 2000 2500
Time (d)
surface which could contribute to higher oil recovery. The whole process
is like the recycling of nanoparticles. So postflush not only help improve
the nanoparticles recovery, but also extend the nanoflooding period, thus
enhancing the oil recovery.
In conclusion, for this simulation model, the best developing strategy
would be using nanofuids flooding with water postflush as secondary oil
recovery method. Nanofuids would be injected first at 2000stb/d for 300d,
and then water is injected at 4000stb/d for the rest of the time.
8.570e-01
0.81
0.72
0.63
0.54
(a)
0.45
3.690e-01
(b)
(c)
Figure 19.20 Comparison results of (a) water saturation under water flooding,
(b) nanofuids flooding and (c) Post flush (300d nanoflooding + 700d water flooding)
under the same color cale.
19.6 Discussions
Mechanism of Postflush Enhancing Oil Recovery with
Less Nanoparticles Loss
The results on postflush may surprise the researchers. Before, postflush
was considered to be a good method to recover nanoparticles trapped in
the reservoir. Based on this study, if it’s used after a short time of nano-
flooding (0.5 PV or more), it could serve as a perfect method to enhance
oil recovery with the least nanoparticles loss (compared to nanoflooding
all the time). In order to find explanations for these results, another group
of simulations were completed. This time, nanofuids were injected into the
458 Reservoir Characterization
Nanoflooding Concentration
1.000e+00
0.75
0.5
(a) 0.25
(b)
reservoir for 100 days (0.15 PV) first, then brine water was injected up to
400 days. The nanofuids concentration distribution and nanoparticles con-
centration on the rock were both analyzed. The results are shown in Figure
19.22 and Figure 19.23.
From Figure 19.22, it can be seen that when the nanofuids is injected
for 100d, there is a nano-circle generated around the injection well, and
there is a concentration gradient inside the circle. It’s the same situation
for the nanoparticles concentration on the rock surface (Figure 19.23).
Nanoparticles adsorption onto the rock surface means oil molecules
desorption from the rock. So the 100d nanoflooding actually helps free the
oil around the injection well, inside the nano-circle area.
After nanoflooding, postflush has been conducted for 300 days. It can be
seen that the nano-circle expands and covers a bigger area. While, in the
center area, around the injection well, there are almost no nanoparticles,
indicating that water is pushing nanoparticles forward towards the produc-
tion wells. So the nano-circle becomes a nano-loop. Meanwhile, the same
amount of nanoparticles would desorb from the nano-circle area and adsorb
onto the rocks in the nano-loop area to help extract more oil. So on and so
forth, the nano-loop keeps expanding and the nanoparticles injected in the
Numerical Investigation of Enhanced Oil Recovery 459
20 1 20 0.4
Nano-circle 0.9 Nano-loop
0.35
0.8
15 15 0.3
0.7
0.6 0.25
10 0.5 10 0.2
0.4 0.15
0.3
5 5 0.1
0.2
0.1 0.05
t = 100d t = 200d
0 0 0 0
0 5 10 15 20 0 5 10 15 20
20 0.25 20 0.18
0.16
0.2
15 15 0.14
No-nano-circle 0.12
0.15
0.1
10 10
0.08
0.1
0.06
5 5
0.05 0.04
0.02
t = 300d t = 400d
0 0 0 0
0 5 10 15 20 0 5 10 15 20
first 100 days keep being pushed forward. At 400d, there are nanoparticles
all around the reservoir except the area around the injection well, which is
called no-nano-circle here. It can be predicted that, if the postflush is contin-
ued for a long time, the no-nano-circle will expand and eventually, most of
the nanoparticles injected at the first 100 days would be washed out.
So in the whole process of this EOR method, the same amount of
nanoparticles would be used again and again to extract oil through the
whole reservoir, thus reducing the amount of nanoparticles required and
enhancing the oil recovery at the same time.
20 0.03 20 0.01
Nano-loop 0.009
Nano-circle
0.025
0.008
15 15
0.007
0.02
0.006
10 0.015 10 0.005
0.004
0.01
0.003
5 5
0.002
0.005
0.001
t = 100d t = 200d
0 0 0 0
0 5 10 15 20 0 5 10 15 20
20 0.006 20 0.0045
0.004
0.005
15 15 0.0035
0.0025
10 0.003 10
0.002
0.002 0.0015
5 5
0.001
0.001
0.0005
t = 300d t = 400d
0 0 0 0
0 5 10 15 20 0 5 10 15 20
References
1. M. Prodanovic, S. Ryoo, A. R. Rahmani, R. Kuranov, C. Kotsmar, T.E.
Milner, K.P. Johnston, S.L. Bryant, and C. Huh, Effects of magnetic field on
the motion of multiphase fluids containing paramagnetic particles in porous
media. SPE 129850 presented at SPE Improved Oil Recovery Symposium,
Tulsa, OK (2010).
2. B. Ju, S. Dai, Z. Luan, T. Zhu, X. Su, and X. Qiu, A study of wettability and
permeability change caused by adsorption of nanometer structured polysili-
con on the surface of porous media. SPE 77938 presented at SPE Asia Pacific
Oil and Gas Conference and Exhibition, Melbourne, Australia (2002).
3. M.K. Chaudhury, Complex fluids: Spread the world about nanofuids. Nature
423(10), 131–132 (2003).
4. D.T. Wasan and A.D. Nikolov, Spreading of nanofuids on solids. Nature 423,
156–159 (2003).
5. P.L.J. Zitha, Smart fluids in the oilfields. Exploration & Production: The Oil &
Gas Review 66–68 (2005).
6. I. N. Evdokimov, N. Y. Eliseev, A. P. Losev, and M. A. Novikov, Emerging
petroleum-oriented nanotechnologies for reservoir engineering. SPE 102060
presented at SPE Russian Oil and Gas Technical Conference and Exhibition,
Moscow, Russia (2006).
7. B. Ju and T. Fan, Experimental study and mathematical model of nanoparti-
cle transport in porous media. Powder Tech. 192, 195–202 (2009).
8. H. Lecoanet, J. Bottero, and M. Wiesner, Laboratory assessment of the mobil-
ity of nanomaterials in porous media. Environ. Sci. Technol. 38, 5164–5169
(2004).
462 Reservoir Characterization
9. S.W. Jeong and S.D. Kim, Aggregation and transport of copper oxide
nanoparticles in porous media. J. Environ. Monit. 11, 1595–1600 (2009).
10. F. He, M. Zhang, T. Qian, and D. Zhao, Transport of carboxymethyl cellu-
lose stabilized iron nanoparticles in porous media: Column experiments and
modeling. J. Colloid Interface Sci. 334, 96–102 (2009).
11. N. Saleh, H. Kim, T. Phenrat, K. Matyjaszewski, R.D. Tilton, and G.V. Lowrey,
Ionic strength and composition affect the mobility of surface-modified Fe0
nanoparticles in water-saturated sand columns. Environ. Sci. Technol. 42,
3349–3355 (2008).
12. A. Tiraferri and R. Sethi, Enhanced transport of Zerovalent Iron nanoparti-
cles in saturated porous media by Guar Gum. J. Nanopart. Res. 11, 635–645
(2009).
13. M.J. Murphy, Experimental analysis of electrostatic and hydrodynamic forces
affecting nanoparticle retention in porous media, MS Thesis, The University
of Texas at Austin (2012).
14. H. Yu, Transport and retention of surface-modified nanoparticles in sedi-
mentary rocks. PhD Dissertation, The University of Texas at Austin (2012).
15. M. Parvazdavani, M. Masihi, and M.H. Ghazanfari, Investigation of the effect
of water based nano-particles addition on hysteresis of oil and water relative
permeability curves. SPE International Oilfeld Nanotechnology Conference,
Noordwijk (2012).
16. M.F. El-Amin, A. Salama, S. Sun, and K. Abdullah, Modeling and simula-
tion of nanoparticles transport in a two-phase flow in Porous Media. SPE
154972, SPE International Oilfield Nanotechnology Conference, Noordwijk,
The Netherlands, June 12–14 (2012).
17. Louisiana Optical Network Initiative. Last updated Tuesday, April 15, 2014,
http://www.loni.org/ (accessed November 17, 2015).
18. ParaView. www.paraview.org (accessed November 17, 2015).
19. PETSc. http://www.mcs.anl.gov/petsc (accessed November 17, 2015).
20. T. Zhang, Modeling of nanoparticle transport in porous media. PhD disser-
tation, The University of Texas at Austin, August (2012).
21. T. Tosco and R. Sethi, MNM1D: A numerical code for colloid transport in
porous media. Implementation and validation. Am. J. Env. Sci. 5, 517–525
(2009).
22. J. Kou and S. Sun, On iterative IMPES formulation for two-phase flow with
capillarity in heterogeneous porous media. Int. J. Numer. Anal. Modeling 1,
20–40 (2010).
23. S.K. Das, S.U.S. Choi, W. Yu, and T. Pradeep, Nanofuids Science and Technology,
John Wiley & Sons, Inc. Publishing, Hoboken, NJ, ISBN 0470074736 (2008).
23. M.A. Sbai and M. Azaroual, Numerical modelling of formation damage by
two-phase particulate transport processes during CO2 injection in deep het-
erogeneous porous media. Adv. Water Resour. 34, 62–82 (2011).
20
3D Seismic-Assisted CO2-EOR
Flow Simulation for the Tensleep
Formation at Teapot Dome, USA
Payam Kavousi Ghahfarokhi1*, Thomas H. Wilson1 and Alan Lee Brown2
2
Schlumberger NExT, Director of Geoscience Training for NAM, San Felipe,
Houston, USA
Abstract
The Tensleep Formation at Teapot Dome, Wyoming is a naturally fractured res-
ervoir that has been studied for CO2-EOR. A realization of fractures is generated
using a discrete fracture network (DFN) approach. The model DFN has three frac-
ture sets: a dominant, high intensity N76°W set and two lower-intensity sets that
strike N28°W and N75°E. The fracture intensity in the model DFN is extracted
from a most negative curvature dissimilarity attribute calculated from the post-
stack 3D seismic data. In addition, probable deformation bands are interpreted
from the fracture intensity seismic attribute and incorporated in the reservoir
model as permeability barriers parallel to N40°E striking S1 fault. The gener-
ated model DFN is plugged into a dual porosity compositional reservoir simu-
lator (Eclipse 300) and adjusted to obtain a production history match. A three
parameter Peng-Robinson equation of state (EOS) is calibrated against the CO2
swelling tests laboratory data to obtain accurate phase behavior for CO2-EOR in
the Tensleep reservoir. Two CO2-EOR models are evaluated. CO2 is injected in
the B1 Tensleep Sandstone from the end of the production history (December,
2005) for 3 years at an injection rate of 1 MMSCF/day. The first model includes
three horizontal injectors parallel to the main fracture set and the second model
has three horizontal injection wells perpendicular to the dominant fracture set.
Although immediate CO2 breakthrough is observed in both models, model 1
yields a higher oil recovery and a lower CO2 mole fraction at the surface. The oil
recovery improvement is less than 700 MSTB for both models. Finally, we show
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (463–486) © 2022
Scrivener Publishing LLC
463
464 Reservoir Characterization
that ultimate CO2-EOR recovered oil can vary up to 150 MSTB and depends on
how impermeable the permeability barriers are.
20.2 Introduction
Naturally fractured reservoirs (NFR) are challenging to model, since
fracture networks are often complex and incorporate fluid transfer from
matrix to the fractures. High permeability fractures typically cause major
fluid flow in fractures rather through the matrix block. However, they can
also act as flow barriers when mineralized or through infilling of fine-
grain materials in formerly open fractures. Wireline image logs, out-
crop studies, core observations, and seismic data analysis provide direct
and indirect measurements of fracture properties at different scales.
Observations made at these different scales must be integrated to pro-
vide a representative model of the reservoir DFN [1–4]. Fracture spacing,
aperture, and length distributions can be incorporated in the reservoir
model using wireline image logs, core data, and field studies [2, 5]. While
uncertainties remain, these models are consistent with observations made
at a variety of scales.
Dual porosity (DP) models have been widely used to simulate the
fluid flow between the fractures and matrix medium [6–9]. Dual porosity
3D Seismic-Assisted CO2-EOR Flow Simulation 465
models incorporate coupled fow continua: one for the matrix and the
other for the fracture network. Fluid flow through the fracture network
along with flow from the matrix into the fracture network can be modeled
with a transfer function referred to as the shape factor [7, 8, 10, 11]. More
recently, discrete fracture models (DFM) have been employed to simulate
the flow in 2D and 3D fractured media [12–21]. Fracture permeability
upscaling is necessary to model the fluid flow in fractured reservoir using
a dual porosity model. Oda [22] algorithm calculates a fracture tensor,
which only considers the geometry of the fractures (aperture, size and
orientation). Oda’s approach is often used as a calculation technique in
commercial software as a fast algorithm to calculate fracture permeability
tensors. However, this method assumes that all fractures are connected
and overestimates effective permeability [17, 18, 23–25]. Oda’s permea-
bility tensor can be corrected using a flow-based numerical method by a
linear transformation.
We use a DFN along with the corrected Oda’s method to simulate pro-
ductions in a fully compositional dual porosity simulator for the Teapot
Dome oilfield, Wyoming (Figure 20.1). The history matched reservoir
model is then used in a compositional simulator to model CO2-EOR.
CO2-EOR can increase ultimate oil recovery by 7–23% of original oil in
place [26]. CO2-EOR efficiency significantly depends on optimization of
the injection process which requires a thorough understanding of reser-
voir heterogeneities and phase behaviors [27]. Injected CO2 usually follows
less resistant flow paths such as fractures or high permeability channels
to production wells. This effect is intensified by the low viscosity of CO2
at reservoir conditions. Although optimum well location relative to reser-
voir heterogeneities increases sweep efficiency, early breakthrough cannot
be avoided in most cases, just postponed [27]. CO2 and oil have multiple
contact miscibility. It is a dynamic fluid-mixing process in which CO2 ini-
tially mixes with the oil and makes it lighter. This process continues by
vaporization of oil components into CO2 rich phase. Exchanges of com-
ponents continue until no interfaces remain between the CO2-enriched
and oil-enriched CO2 phases [26]. Although miscibility is a function of
reservoir temperature and pressure, only pressure controls the miscibility
in isothermal reservoirs [28].
Minimum miscibility pressure (MMP) refers to a pressure at which
miscibility occurs. It can be achieved by injecting CO2 at a higher pres-
sure than MMP or enriching the gas with intermediate-weight hydrocar-
bons [27]. Injected CO2 displaces oil, develops miscibility, swells oil, and
reduces oil viscosity [29]. CO2-EOR sweep efficiency can be greatly influ-
enced by well pattern, well type, injection rate, reservoir heterogeneity,
466 Reservoir Characterization
2000 ft
790000 800000
976000 976000
972000 972000
968000 968000
954000 954000
944000
953000 953000
800000
953200 953200 N
952800 952800
952400 952400
952000 952000
951000 951000
951200 951200
802000
(b)
Figure 20.1 (a) The Tensleep horizon interpreted from 3D seismic data shown in subsea
depth (ft.) (b) Contour map of cumulative oil production (bbl) from 1978 to 2004 in
Section 10 overlaid by the reservoir grid for flow simulation. The gridding size is 300 ft by
300 ft and it is approximately oriented parallel to the main fracture set (N76°W). Note that
suffix “-TPX-10” is removed from the well names for visualization.
3D Seismic-Assisted CO2-EOR Flow Simulation 467
Tenseep
2-way time (ms)
Pennsylvanian sandstone
–1000
Alco
va L Amsden formation
Tens s.
leep
Ss. Madison limestone
Mad Mississippian Regional aquifer
ison
Prec LS. and water drive
amb
rian Englewood formation
–1500 gran
ite Devonian
Fremont canyon Ss.
1 km Precambrian Granite
(a) (b)
Figure 20.2 (a) West–east seismic section across the Teapot Dome. The reverse fault
(shown by black line) has a southeast dip. (b) Stratigraphic column Precambrian-Triassic
strata. The Madison Limestone is the regional aquifer that provides the water drive for the
Tensleep reservoir, and the Goose Egg Formation is the seal (taken from Wilson et al. [2]).
3D Seismic-Assisted CO2-EOR Flow Simulation 469
that provides the water drive for the Tensleep reservoir. The Chugwater
and Permian Goose Egg formations (a mixture of shale, siltstone with
some sandstone) seal the Pennsylvanian Tensleep reservoir (Figure 20.2).
8 30
88 88
6
20
4
80 10
80
2
0
0 0 0.05 0.1 0.15 0.2 0.25 0.3
72 0 0.07 0.14 0.21 0.28 0.35 0.42 0.49
Wireline_Image_Log_Apertures
72
Core_A_Apertures
64 % 4 8 12 16 20 24 24 32 36 40 44 48 52 64
10
8
56 56
6
48 2 48
0
0 0.35 0.7 1.05 1.4 1.75 2.1 2.45
(%)
Core_B_Apertures
40 40
32 32
24 24
16 16
8 8
0 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Aperture (mm)
Figure 20.3 Apertures are log-normally distributed in the wireline image logs, Core A,
and Core B. Note apertures measured in Core A and Core B are under no overburden
pressure.
3D Seismic-Assisted CO2-EOR Flow Simulation 471
% 2 4 6 8 10 12 14 16 18 20 22 24 26 % 24 4 6 8 10 12 14 16 18 20 22 24 26 28
9 13
8 A Sand 12
B Dolomit e
11
7 10
6 9
8
5
7
4 6
5
3
4
2 3
2
1
1
0 0
0.4 0.6 0.8 1 0.4 0.6 0.8 1
% 2 4 6 8 10 12 14 16 18 20 22 24 26 28 % 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
6.5
11
6
10 B Sand Aquifer
9 5
8 4.5
7 4
6 3.5
5 3
2.5
4
2
3
1.5
2 1
1 0.5
0 0
0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Figure 20.4 Distributions of the fracture intensity attribute for each zone in the grid.
Note that the B Sand fracture intensity histogram consists of the B1 Sand and B2 Sands.
The aquifer includes C Dolomite and C sand.
Kavousi and Wilson [3] qualitatively related the fracture intensity attri-
bute to the production history of the field. Wells located on NW seismic
discontinuities are in general more productive than wells located on NE
trending discontinuities. NW discontinuities (Figure 20.5) are interpreted
as possible open fracture zones which mainly include N76°W fracture set.
Wilson et al. [2] showed that the dominant fracture set (N76°W) is roughly
parallel to the present-day maximum horizontal compressive stress (Shmax)
inferred from drilling induced fractures. They also interpreted permeabil-
ity barriers roughly parallel to the S1 fault.
In this study, NE seismic discontinuities (Figure 20.5) are interpreted as
possible low-permeability deformation bands. Chiaramonte [28] showed
that a 3300 psi reservoir pressure increase is necessary to make the S1 fault
472 Reservoir Characterization
SH
ma
x
(a) (b)
0 1000 2000 3000 ft
N Low high
Figure 20.5 (a) Possible deformation bands are shown by red arrows, note that they are
parallel to sub-parallel to the S1 fault. (b) The fracture intensity attribute introduced by
Kavousi and Wilson.2 Wells located on the NW discontinuities (yellow arrows) are more
productive than wells on the NE discontinuities (red arrows).
72-TpX-10
62-TpX-10
52-1-TpX-10
53-TpX-10
73-TpX-10
63-TpX-10
43-TpX-10
43-2-TpX-10
54-TpX-10
74-CMX-10-WD
44-1-TpX-10 75-TpX-10
55-TpX-10
76-TpX-10
46-TpX-10
56-TpX-10
i
67-1-X-10
Figure 20.6 The permeability barriers are roughly parallel the S1 fault. The grids j
direction is almost parallel to the dominant fracture set (N76°W). The grid block size is
300 ft by 300 ft. Note Kxx = Ki and Kyy = Kj.
3D Seismic-Assisted CO2-EOR Flow Simulation 473
X-axis
800000 802000 804000 806000
958000 958000
950000 950000
Figure 20.7 Locations of 8 Wells with core data used in Chiaramonte (2009) reservoir
modeling study are shown on Tensleep structure. The black polygon over the A Sand
horizon shows areal extent of high oil production zones. Subsea depth and coordinates are
in feet.
The 8 component EOS parameters are tuned to fit the swelling test data
from well 72-TPX-10 oil sample. The resulting EOS properly outputs the
oil sample response to the swelling tests (Figure 20.8). The swelling tests
show that oil responds to CO2 injection with volume increase and viscos-
ity reduction. We modeled the liquid viscosity at each swelling test step
and compared that with laboratory results (Figure 20.9). The 8-component
EOS simulates the liquid viscosity reductions with a reasonable error at
each swelling test step.
Table 20.2 Matrix porosity (ft3/ft3), permeability (mD.), and average cell thickness (ft.) for the
Tensleep static model (modified from Chiaramonte [28]).
Horizontal Vertical Number of Average cell
Zone Porosity% permeability permeability the layers thickness
A Sand 4.74 3.69 1.11 2 23
B Dolomite 3.03 0.21 0.10 1 26.29
B1 Sand 10.54 29.31 6.14 1 19
B2 Sand 6.95 3.45 1.05 2 26.25
Aquifer 20 10 2.53 8 50
3D Seismic-Assisted CO2-EOR Flow Simulation 475
476 Reservoir Characterization
CO2
Swelling Calc.-After regression
3000.0 1.50
2800.0 1.40
2600.0 1.30
2400.0 1.20
Saturation pressure (psia)
2200.0 1.10
2000.0 1.00
1800.0 0.90
Swelling factor
1600.0 0.80
1400.0 0.70
1200.0 0.60
1000.0 0.50
800.0 0.40
600.0 0.30
400.0 0.20
200.0 0.10
.0 0.00
0.000 0.100 0.200 0.300 0.400 0.500 0.600 0.700 0.800
Composition (mol fraction)
Legend
Psat S.F. Exp.Psat Exp.S.F.
Figure 20.8 Saturation pressure and swelling factor calculated in the swelling
experiments and model EOS are plotted against mole fraction CO2 added to the oil sample
from well 72-TPX-10 at 190.4°F. The saturation pressure increases as higher mole fraction
of CO2 is added to the fluid sample.
4 70
3.5
60
3
50
Solvant CO2 (model%
)
Fluid viscosity (cp)
2.5
40
2
30
1.5
20
1
0.5 10
0 0
0 500 1000 1500 2000 2500 3000
Bubblepoint pressure (psia)
Figure 20.9 Fluid viscosity is plotted vs. bubble point pressure at each swelling test step.
The modeled viscosity reduction curve (orange line) is close to the swelling test viscosity
results (blue line).
3D Seismic-Assisted CO2-EOR Flow Simulation 479
20.8 CO2-EOR
The 8-component EOS is used in a compositional simulator (Eclipse 300)
to model CO2-EOR in the Tensleep reservoir at Teapot Dome, Wyoming.
Two models are designed for CO2-EOR using three horizontal wells within
72
63 73
43 43-2 Water saturation
54 1.0000
0.9500
0.9000
0.8500
44 75 0.8000
55 0.7500
0.7000
76 0.6500
0.6000
0.5500
0.5000
56
67
Figure 20.10 Streamline analysis on 03-28-2004 for water saturation around producing
wells. The streamline simulator assigns the streamlines to the center of the grids where
wells are located. The model mainly produced water. Well 56-TPX-10 is the most oil
prolific well in this field. Note how streamlines change around permeability barriers.
480 Reservoir Characterization
X-axis X-axis
802000 802000
Y-axis
Y-axis
Y-axis
802000 802000
X-axis X-axis
(a) (b)
Figure 20.11 (a) Model 1 has injectors (I4, I5, and I6) parallel to the main fracture set
(N76°W). (b) In model 2, Injectors (I7, I8, and I9) are perpendicular to the dominant
fracture set. Each grid block is 300 ft by 300 ft.
3D Seismic-Assisted CO2-EOR Flow Simulation 481
Field
Jan 2006 Jul 2006 Jan 2007 Jul 2007 Jan 2008 Jul 2008 Jan 2009
8000
400 600 800 1000 1200 1400
6000
Gas flowrate (MSCF/d)
2000
m1 oil production
200
m2 oil production
0
0
Jan 2006 Jul 2006 Jan 2007 Jul 2007 Jan 2008 Jul 2008 Jan 2009
Figure 20.12 A high mole fraction of CO2 is observed at the producers beginning almost
immediately after injection in model 2. Model 1 has a higher oil production and lower
gas production rates during the injection at 1000 Mscf/day. Both models suffer from early
breakthrough because of the fractures which act as easy flow conduits for CO2 to reach the
producers while bypassing the matrix blocks.
640 MSTB and 580 MSTB more, respectively, than the case with no CO2
injection (Figure 20.13).
We undertook a sensitivity analysis on the fault transmissibility mul-
tiplier (MULTFLT) for the CO2-EOR models. We consider two cases: all
permeability barriers are completely sealing (MULTFLT=0), and all per-
meability barriers are completely leaky (MULTFLT=1). Results show that
recovered oil varies up to 150 MSTB and depends on MULTFLT values.
The higher the MULTFLT, the higher the recovered oil (Figure 20.14).
The model studies show that early CO2 breakthrough limits sequestra-
tion effectiveness. Wilson et al. [2] noted that Tensleep fracture lengths
are much less than the spacing between laterals with the majority of frac-
ture lengths between 5 m and 15 m. Their outcrop studies also indicated
considerable variation of fracture intensity within Tensleep sequences
exposed in the Fremont Canyon area. The lower sequences were generally
less intensely fractured than those in the upper Tensleep. The LIDAR mea-
surements of Zahm and Hennings ([38], their Figure 20.8) indicate that
sequence bound fracture intensities are somewhat less in the lower lime-
stone and dolostone dominated sequences exposed in the Alcova anticline.
Sequence bound fracture intensities also appear to form vertical zones of
alternating lower and higher intensity that cut through the Tensleep [38]
482 Reservoir Characterization
2.4E+06
2.2E+06
2.2E+06
Oil production rate (STB/d)
2E+06
2E+06
1.8E+06
1.8E+06
Jan 2006 Jul 2006 Jan 2007 Jul 2007 Jan 2008 Jul 2008 Jan 2009
Date
Figure 20.13 The oil production for Models 1 and 2 are compared to an oil production
case with no CO2 injection. The recovery improvement of 640 MSTB (black curve) is
achieved in Model1 compared to no CO2 injection case (Blue curve).
2.2E+06 2.4E+06
2.4E+06
Liquid production volume (STB)
2.2E+06
2E+06
2E+06
1.8E+06
1.8E+06
Jan 2006 Jul 2006 Jan 2007 Jul 2007 Jan 2008 Jul 2008 Jan 2009
Date
Model 2 _FLT1
Figure 20.14 Increasing fault multiplier results in higher oil production for both CO2-
EOR models.
(Figure 20.8). Wilson et al. [2] also observed alternating vertical zones of
high and low fracture intensity in Fremont Canyon Tensleep exposures.
Seismic discontinuities observed in the Teapot Dome 3D seismic data set
are interpreted to represent zones of increased fracture intensity. Estimated
3D Seismic-Assisted CO2-EOR Flow Simulation 483
widths of these zones are less than twice the seismic bin spacing or approx-
imately 60 m with lengths that vary from about 200 m to 1000 m. Wilson
et al. [2] suggested that perforation into these zones should be avoided to
reduce breakthrough time. We also suggest that wells land in the lower,
less intensely fractured, Tensleep. Completion in the lower Tensleep away
from seismic discontinuities may reduce breakthrough time and increase
sequestration volume.
20.9 Conclusions
• Streamlined simulation shows that the dominant (N76°W)
fracture set has significant influence on flow orientation.
Two other low intensity sets do not significantly control flow
directions except for the highest producers when they allow
the well to access larger reservoir volume.
• CO2 tends to emerge in the producers with the start of injec-
tion for both models: at a higher rate in model 2 where the
injectors are perpendicular to the dominant fracture set
(N76°W) and at a lower rate in model 1 where the injectors
are oriented parallel to the dominant fracture set.
• Horizontal wells parallel to the main anisotropy axis
resulted in a higher sweep efficiency in model 1 compared
to model 2.
• A threshold pressure of 3300 psi for reservoir sealing was
assumed based on work of Chiaramonte [28], however, at
an injection rate of 1000 MSCF/day, this pressure was never
reached in either model. Thus, the sealing properties of the
S1 fault and other permeability barriers is not compromised
during CO2-EOR.
• We suggest that wells be completed in the lower, less intensely
fractured, Tensleep and that perforations avoid zones of seis-
mic discontinuity to reduce breakthrough time and increase
sequestration volume.
Acknowledgement
I appreciate Rocky Mountain Oil Testing Center for providing me with
3D seismic and log data for Teapot Dome. The use of Eclipse and Petrel
from Schlumberger, and CMG from Computer Modelling Group is greatly
484 Reservoir Characterization
appreciated along with the insights on how to use the software from Dr.
Alan Lee Brown from Schlumberger and adjunct Professor at West Virginia
University; these software resources made the analysis presented in this
paper possible. I also thank Vicki Stamp for providing me with PVT report
of oil samples. Production data were sorted and formatted for Petrel by
Valerie Smith from West Virginia University.
References
1. A. Ouenes, T.C. Anderson, D. Klepacki, A. Bachir, D. Boukhelf, G.C.
Robinson, M. Holmes, B.J. Black, and V.M. Stamp, Integrated characteriza-
tion and simulation of the fractured Tensleep Reservoir at Teapot Dome for
CO2 injection design. SPE Western Regional Meeting (2010).
2. T.H. Wilson, V. Smith, and A. Brown, Developing a model discrete fracture
network, drilling, and enhanced oil recovery strategy in an unconventional
naturally fractured reservoir using integrated field, image log, and three-
dimensional seismic data. AAPG Bulletin. 99, 735–762 (2015).
3. P. Kavousi Ghahfarokhi and T.H. Wilson, Fracture intensity attribute for the
Tensleep reservoir at Teapot Dome, Wyoming, USA. Interpretation. 3, SZ41–
SZ48 (2015).
4. P.K. Ghahfarokhi, Reservoir Characterization and Flow Simulation for CO2-
EOR in the Tensleep Formation Using Discrete Fracture Networks, Teapot
Dome, Wyoming. Diss. West Virginia University (2016).
5. S.P. Cooper, Deformation Within a Basement-Cored Anticline: Teapot Dome,
Wyoming, Citeseer (2000).
6. G. Barenblatt, I.P. ZHELTOV, and I. KOCHINA, Basic concepts in the theory
of seepage of homogeneous liquids in fissured rocks [strata]. J. Appl. Math.
Mech. 24, 1286-1303 (1960).
7. J. Warren and P.J. Root, The behavior of naturally fractured reservoirs. Soc.
Petrol. Eng. J. 3, 245–255 (1963).
8. H. Kazemi, L. Merrill Jr., K. Porterfeld, and P. Zeman, Numerical simula-
tion of water-oil flow in naturally fractured reservoirs. Soc. Petrol. Eng. J. 16,
317–326 (1976).
9. L.K. Thomas, T.N. Dixon, and R.G. Pierson, Fractured reservoir simulation.
Soc. Petrol. Eng. J. 23, 42–54 (1983).
10. G. Penuela, F. Civan, R. Hughes, and M. Wiggins, Time-dependent shape
factors for interporosity flow in naturally fractured gas-condensate reser-
voirs. SPE Gas Technology Symposium (2002).
11. P. Sarma and K. Aziz, New Transfer Functions for Simulation of Naturally
Fractured Reservoirs with Dual Porosity Models. SPE J. 11, 328–340 (2006).
12. M. Karimi-Fard and A. Firoozabadi, Numerical simulation of water injec-
tion in fractured media using the discrete-fracture model and the Galerkin
method. SPE Reserv. Eval. Eng. 6, 117–126 (2003).
3D Seismic-Assisted CO2-EOR Flow Simulation 485
Abstract
This chapter gives a brief overview of reservoir characterization, with a focus
on how the emerging artificial intelligence, machine learning and data ana-
lytics techniques can improve the ability to do reservoir characterization (RC).
Specifically, the ability to use AI to integrate different data types with different
Scale, Uncertainty, Resolution, and Environment, that we refer to it as the SURE
challenge is discussed. We show how AI offers a natural toolbox for reservoir prop-
erty estimation, and their uncertainties. Machine/Deep Learning-based methods
perform much like a human brain. They can receive variety of data from many
different sources with drastically different characteristics, and undertake neces-
sary evaluations and perception-based measures, and eventually make the right
decisions and/or solve complicated problems. Human intelligence (engineers,
geoscentists) will always have a superior performance with qualitative data than
computers that are better in dealing with quantitative data. Several examples on
how effective human-machine interfaces to create hybrid solutions to address dif-
ferent reservoir characterization problems are provided.
Keywords: Reservoir characterization, big data artificial intelligence (AI),
support vector machines, data analytics, neural networks, fuzzy logic
*Email: [email protected]
Fred Aminzadeh (ed.) Reservoir Characterization: Fundamentals and Applications, (489–524) © 2022
Scrivener Publishing LLC
489
490 Reservoir Characterization
for optimizing field development and operation, and for reservoir valuation.
Dynamic reservoir characterization allows us to understand and predict the
changes in reservoir properties and to monitor its performance as hydrocar-
bons are produced from the reservoir. Characterization is also critical when
stimulating the reservoir for enhancing production. This is accomplished by
the analysis of data from a combination of different sources, to extract addi-
tional information about the in situ conditions of the reservoir, including
the formation temperature, pressure, and the properties of the oil, gas, and
brine. Other reservoir properties that can affect measured data are density,
hydrocarbon viscosity, stresses, and fractures.
As shown in Figure 21.1, an integrated reservoir characterization starts
with collecting data from geological, petrophysical, seismic and engineer-
ing data. A multi-disciplinary data analysis process creates a model which
includes characterizing the architecture, lithologies, facies, fluid composi-
tions, rock-fluid interactions, the geometry of the flow units and physical
rock properties such as porosities and permeabilities of flow layers. Three
properties are related to the pore space: porosity—the fraction of the entire
volume part occupied by pores, cracks and fractures; internal surface: the
magnitude of the surface of pores as related to the rock mass pore volume
and controls interface-effects at the boundary grain; pore fluid, permeabil-
ity: the ability to flow fluid through rock pores. Porosity and specific inter-
nal surface are scalar properties, permeability is a tensor. Given different
w1
w2
Time, days
Uncertainty Analysis
Multiple Disciplinary
Analysis
Rate
(a) Well-separated clusters. Each point is (b) Center-based clusters. Each point is
closer to all of the points in its cluster than closer to the center of its cluster than to
to any point in another cluster. the center of any other cluster.
(c) Contiguity-based clusters. Each point is (d) Density-based clusters. Clusters are
closer to at least one point in its cluster than regions of high density separated by
to any point in another cluster. regions of low density.
38.83 2-1
0
1
38.82
2-2 2-4
2-3 -1
38.81
Depth
(1m)
Latitude [deg]
-2
38.8
3 -3
38.79
-4
38.78
-122.84 -122.83 -122.82 -122.81 -122.8 -122.79 -122.78 -122.77
Longitude [deg] -5
Figure 21.3 Fuzzy clustering to monitor fluid movement in a geothermal reservoir [5].
Machine Learning in Reservoir Characterization 495
496 Reservoir Characterization
Original
D Training data
Step 1:
Create Multiple D1 D2 Dt-1 Dt
Data Sets
Step 2:
Build Multiple C1 C2 Ct-1 Ct
Classifiers
Step 3:
Combine C*
Classifiers
Each neuron in a layer has a weight and has directed connection to the
neuron(s) in the next layer. The inputs are multiplied by the weight and fed
Machine Learning in Reservoir Characterization 499
into the activation function. The output of the activation function cascades
to the next neuron; eventually the output layer generates a prediction. In
the backpropagation, the weights are updated as per the original connec-
tions between the neurons, but in the opposite direction, back from the
output layer.
To give the network an ability to learn non-trivial solutions, non-linear
activation functions are important. There are many activation functions –
both linear and non-linear; following are a few examples of the non-linear
functions:
Weights
Constant 1
W0
X1 Weights Sum
W1
Out
Inputs Wn-1
Xn-1 Step Function
Wn
Xn
(a)
Input #1
Input #2
Output
Input #3
Input #4
(b)
Figure 21.6 (a) Perceptron learning process [11]. (b) Multi layer perceptron architecture
[22].
500 Reservoir Characterization
There are several architectures of the ANN which are popular and widely
deployed to solve challenges that are difficult to solve with traditional NN
structure. In this chapter, we have restricted ourselves to Convolutional
Neural Network (CNN), Recurrent Neural Networks (RNN), Auto Encoders
(AE) and Generative Adversarial Networks (GAN). Although the reviewed
literature makes use of variants of the architectures discussed earlier.
CNNs are very effective at tasks involving pattern recognition in images
(multi-band images). The primary benefit of a CNN is that they success-
fully capture spatial and temporal dependencies in an image (or sequence).
Practically CNNs are the driving force behind most computer vision tasks –
semantic segmentation, instance segmentation, panoptic segmentation.
As the name suggests, the CNN use convolution between a kernel and the
Convolution Pooling Convolution Pooling Fully Fully Output
+ReLU +ReLU Connected Connected predictions
dog (0.01)
Cat (0.01)
Boat (0.94)
Bird (0.94)
X1 X1
‹
X2 X2
‹
X3 X3
hw,b(x)
‹
X4 X4
‹
X5 X5
+1
‹
X6 X6
Layer L2 Layer L3
+1
Layer L1
Discriminator
loss
Real images Sample
Discriminator
Random input
Denerator
loss
Generator Sample
0.1
Topographic
Topographic 16.5
Wetness Index 7.2
(TWI)
Surface
Change in 69.2
sedimentary 0
thickness
Lithologic Sedimentary subsurface
Basement surface 81.0
curvature 0
Crystalline basement
Horizontal derivative of
total magnetic intensity
Figure 21.12 Proxy dataset maps aligned to data types (right) and concept layers (left)
from Justman et al. [27].
506 Reservoir Characterization
Evaluation
resistivity, neutron density, etc. They note that both BiLSTM and Inception
convolutional autoencoder reached human-level performance in the blind
test, with accuracy of nearly 90%.
The traditional procedure used for estimating permeability and poros-
ity, such as the experimental method and statistical methods used to
determine these properties of the reservoir, is very expensive and imprac-
tical due to the complexity of realistic conditions. Anifowose et al. [17],
Priezzhev et al. [25], Otchere et al. [14], and Saikia and Baruah [46]
studied and demonstrated the use of AI for predicting reservoir prop-
erties using machine learning techniques. Saikia and Baruah [46] used
an ensemble method, named StackNet to predict well log data as output.
Otchere et al. [14] performed a comparative study on the logging data as
the target variable but used different supervised learning models—ANN,
Support Vector Machines (SVM) and Relevant Vector Machine—and
determined SVM worked better than ANN in data scarce environment.
They also commented on achieving improved performance by hybridiz-
ing multiple algorithms and the clear benefit that had on enhanced reser-
voir characterization.
He et al. [23] developed and tested a framework to predict sedimentary
facies of uncored areas with high accuracy using Multi-Layer Perceptron
(MLP) trained on data characterizing cored sampled petrophysical logs.
Imamverdiyev and Sukhostat [26] proposed a new CNN architecture with
one dimensional kernels to identify lithological facies from petrophysical
logs. Alizadeh et al. [2] used hierarchical methods to cluster Total Organic
Column (TOC) according to their geochemical composition and used
NNs to relate Logging While Drilling data to source rock. This allowed
the authors to draw burial thermal history diagrams and identify potential
zones for hydrocarbon production.
In shale plays, fracture network and fracture identification is a critical
part of the reservoir characterization process. Tian and Daigle [57] used
single shot object detection method on SEM images to identification and
characterization microfractures in shales (carbonate-rich shale and a sili-
ceous shale). With this approach they were able to identify abundance,
obtained statistics of length and areal porosities of the fractures as well.
Suhag et al. [50] evaluated the performance of NNs for forecasting shale
oil production using data such as petrophysical logs, past-production, and
well interventions – stimulation. This analysis focused on the Bakken,
North Dakota, formation. Through a selection of physical properties from
different sources, they built an NN model that fits with the production
data in wells that have a diverse production history. Their work has shown
508 Reservoir Characterization
Chen et al. [9] studied the techniques mentioned above to map organic
carbon content in topsoil. They noted accuracy gains in using hybrid and
machine learning models – ANNK, SVR, respectively, with ANNK per-
forming the best.
Present subsurface predictions often rely upon disparate and limited
a priori information. Even regions with concentrated subsurface explo-
ration still face uncertainties that can obstruct safe and efficient explora-
tion of the subsurface. Uncertainty may be reduced, even for areas with
little or no subsurface measurements, using methodical, science-driven
geologic knowledge and data. Rose et al. [44] identified this and devel-
oped a hybrid spatiotemporal statistical-geologic approach, subsurface
trend analysis (STA), that provides improved understanding of subsur-
face systems. The STA method assumes that the present-day subsurface is
not random, but is a product of its history, which is a sum of its system-
atic processes. With even limited data and geologic knowledge, the STA
method can be used to methodically improve prediction of subsurface
properties. To demonstrate and validate the improved prediction poten-
tial of the STA method, it was applied in an analysis of the northern Gulf
of Mexico. This evaluation was prepared using only existing, publicly
available well data and geologic literature. Using the STA method, this
information was used to predict subsurface trends for in situ pressure,
in situ temperature, porosity, and permeability. The results of this STA-
based analysis were validated against new reservoir data. STA-driven
results were also contrasted with previous studies. Both indicated that
STA predictions were an improvement over other methods. Overall, their
510 Reservoir Characterization
Alabama
Mississippi
Florida
Louisiana
Texas
N
0 25 50 100 150 200
Miles
Texas Texas
0 5 10 20 30 40 0 5 10 20 30 40
Miles Miles
(a) (c)
0
R2 = 0.9284 R2 = 0.9999 R2 = 0.9318
Depth (TVD ft)
5000
10000
0 5000 10,000 0 5000 10,000 0 5000 10,000
Initial pressure (psi) Initial pressure (psi) Initial pressure (psi)
(b)
Figure 21.14 Example of domain boundary revision and iteration using an example from
central GoM. Rose et al. [44].
Florida
Louisiana
Texas
Validation error
0.4
0.2
Figure 21.15 Subsurface Trend Analysis (STA) results. Data-based prediction vs.
knowledge-data-driven predictions [44].
512 Reservoir Characterization
CMOST-AI Thickness
Residual oil
Latin Hypercube Design saturation to CO2 stored
water flood
Residual oil
Training & Testing
saturation to
gas flood DDM
N samples Injection rate
Producer BHP
CO2 retained
Porosity
Blind validation DDM
Training Simulation using
CMG-GEM
Input layer Hidden layer Output layer
Prediction real fields
Artificial Neural Network ROZs using DDM
Hand-shaking
Objective functions Reservoir simulation and ANN
(Cumulative CO2 storage,
retained and Oil Output collection End
Production)
a
First Second
Data Data Stage Stage
Validation Analysis Screening Screening
relevant EOR data. Data analysis and validations are the other stages of the
screening flow.
Giro et al. [19] introduce a methodology that uses Naïve Bayes Classifier
and selects EOR materials for specific reservoir conditions. They used
this model with physical and chemical representations of injection fluids,
including EOR materials, with reservoir-specific information on lithology,
porosity, permeability, as well as oil, water and salt conditions to provide
recommendation for EOR “cocktail” for injection fluids.
Suhag et al. [51] and Temizel et al. [56] worked on data-driven mod-
els with application that focused on determining sensitivity of controlling
parameters and controlling agents. This work provides a better under-
standing of the influence of surfactant adsorption and thus, a number of
chemicals to be used in an efficient manner. Optimum values for control-
lable decisions were determined by coupling a commercial optimization
software with the reservoir simulator. This work aids in screening and
deciding operation parameters.
21.5 Conclusion
Reservoir characterization is a critical step for understanding the sub-
surface. It is crucial for making optimal decisions to recover resources
from hydrocarbon reservoirs, geothermal assets, and ground water from
aquifers. Given the characterization process involves integration of multi-
disciplinary processes, computational intelligence tools are ideal for this.
Keeping this in mind, this chapter introduces various machine learning
techniques – supervised and unsupervised methods, deep learning archi-
tectures, and many other soft computing techniques and their applications
to reservoir characterization and enhanced oil recovery.
This chapter begins with a detailed overview of some popular machine
learning and deep learning techniques that have been used for various
subtasks of reservoir characterization. Broadly, the chapter breaks down
different subtasks of reservoir characterization—structural model devel-
opment, sedimentary modeling, petrophysical modeling, and dynamic
modeling. Similarly for the enhanced oil recovery section, the chapter
focuses on EOR performance and economics, and EOR screening. The
goal is to introduce the rationale behind a method and list some relevant
literature that might assist readers with their own research problem. With
regards to the deep learning methods, the goal has been to introduce the
fundamental architecture, different activation functions and the process
of selecting them for a specific problem statement. With deep learning
518 Reservoir Characterization
Acknowledgement
The author acknowledges Anuj Suhag’s contributions in reviewing and
adding some of the refences to a few case histories.
References
1. Ahmadi, M. A. and Pournik, M. (2016), A predictive model of chemical-
flooding for enhanced oil recovery purposes: Application of least square sup-
port vector machine, Petroleum, Volume 2, Issue 2, 2016, pp. 177-182, ISSN
2405-6561, https://doi.org/10.1016/j.petlm.2015.10.002.
2. Alizadeh, B., Najjari, S., & Kadkhodaie-Ilkhchi, A. (2012). Artificial neural
network modeling and cluster analysis for organic facies and burial history
estimation using well log data: A case study of the South Pars Gas Field,
Persian Gulf, Iran. Computers & Geosciences, 45, 261-269.
3. Aminzadeh, F., Temizel, C. and Y. Hajizadeh., 2022, Intelligent Data Analysis
for Exploration and Production, Wiley & Sons.
4. Aminzadeh, F., 2021, Reservoir Characterization: Combining Machine
Intelligence with Human Intelligence, E&P Plus, April 2021, Vol. 96 Issue 4,
E&P Plus, Hart Energy.
5. Aminzadeh, F., 2015, Characterizing Fractures in Geysers Geothermal
Field by Micro-seismic Data, Using Soft Computing, Fractals, and Shear
Wave Anisotropy, Topic Area 23: Fracture Characterization Technology
DE-FOA-0000075-23, Final Report.
6. Aminzadeh, F. and Dasgupta, S., 2013. Geophysics for Petroleum Engineers,
Elsevier.
7. Brunsdon, C., Fotheringham, S., & Charlton, M. (1998). Geographically
weighted regression. Journal of the Royal Statistical Society: Series D (The
Statistician), 47(3), 431-443.
8. Bruyelle, J., & Guérillot, D. (2019, October). Proxy Model Based on Artificial
Intelligence Technique for History Matching-Application to Brugge Field. In
SPE Gas & Oil Technology Showcase and Conference. Society of Petroleum
Engineers.
9. Chen L, Ren C, Li L, Wang Y, Zhang B, & Wang Z, A Comparative Assessment
of Geostatistical, Machine Learning, and Hybrid Approaches for Mapping
Topsoil Organic Carbon Content. ISPRS International Journal of Geo-
Information. 2019; 8(4):174.
Machine Learning in Reservoir Characterization 519
10. Chopra, S., & Marfurt, K. J. (2018). Seismic facies classification using
some unsupervised machine-learning methods. Society of Exploration
Geophysicists.
11. Complete Guide to Artificial Neural Network Concepts & Models. (n.d.).
MissingLink.Ai. Retrieved February 16, 2021, from https://missinglink.ai/
guides/neural-network-concepts/complete-guide-artificial-neural-networks/.
12. Cortes, C., and Vapnik, V. Support-vector networks. Mach Learn 20, 273–
297 (1995). https://doi.org/10.1007/BF00994018.
13. Da Silva, L. M., Avansi, G. D., & Schiozer, D. J. (2020). Support Vector
Regression for Petroleum Reservoir Production Forecast Considering
Geostatistical Realizations. SPE Reservoir Evaluation & Engineering.
14. Daniel Asante Otchere, Tarek Omar Arbi Ganat, Raoof Gholami, & Syahrir
Ridha, Application of supervised machine learning paradigms in the predic-
tion of petroleum reservoir properties: Comparative analysis of ANN and
SVM models, Journal of Petroleum Science and Engineering.
15. Demyanov, V., Kanevsky, M., Chernov, S., Savelieva, E., & Timonin, V.
(1998). Neural network residual kriging application for climatic data. Journal
of Geographic Information and Decision Analysis, 2(2), 215-232.
16. Enning Wang, and Jeff Nealon; Applying machine learning to 3D seismic
image denoising and enhancement. Interpretation; 7 (3): SE131–SE139. doi:
https://doi.org/10.1190/INT-2018-0224.1.
17. Fatai Anifowose, Suli Adeniye, Abdulazeez Abdulraheem, & Abdullatif
Al-Shuhail, Integrating seismic and log data for improved petroleum res-
ervoir properties estimation using non-linear feature-selection based
hybrid computational intelligence models, Journal of Petroleum Science and
Engineering, Volume 145, 2016, pp. 230-237,ISSN 0920 4105,https://doi.
org/10.1016/j.petrol.2016.05.019.(http://www.sciencedirect.com/science/
article/pii/S0920410516301814).
18. Gharbi, R. B. (2000). An expert system for selecting and designing EOR pro-
cesses. Journal of Petroleum Science and Engineering, 27(1-2), 33-47.
19. Giro, Ronaldo, Lima Filho, Silas Pereira, Neumann Barros Ferreira, Rodrigo,
Engel, Michael, and Mathias Bernhard Steiner. “Artificial Intelligence-
Based Screening of Enhanced Oil Recovery Materials for Reservoir-Specific
Applications.” Paper presented at the Offshore Technology Conference Brasil,
Rio de Janeiro, Brazil, October 2019. doi: https://doi.org/10.4043/29754-MS.
20. Guillen-Rondon, P., Cobos, C., Larrazabal, G., & Diz, A. (2019, August 28).
Machine learning: A deep learning approach for seismic structural evalua-
tion. Society of Exploration Geophysicists.
21. Harris, P., Fotheringham, A. S., Crespo, R., & Charlton, M. (2010). The use
of geographically weighted regression for spatial prediction: an evaluation of
models using simulated data sets. Mathematical Geosciences, 42(6), 657-680.
22. Hassan, Hassan & Negm, Abdelazim & Zahran, Mohamed & Saavedra,
Oliver. (2015). Assessment of artificial neural network for bathymetry
520 Reservoir Characterization
34. Martin, E., Wills, P., Hohl, D., & Lopez, J. L. (2017, February). Using machine
learning to predict production at a Peace River thermal EOR site. In SPE
Reservoir Simulation Conference. Society of Petroleum Engineers.
35. Moreno, J. E., Gurpinar, O. M., Liu, Y., Al-Kinani, A., & Cakir, N. (2014,
December). EOR advisor system: a comprehensive approach to EOR selec-
tion. In International Petroleum Technology Conference. International
Petroleum Technology Conference.
36. Mudunuru, M.K., O’Malley, D., Srinivasan, S., Hyman, J., Sweeney, M.,
Frash, L., Carey, B., Gross, M., Welch, N., Karra, S., Vesselinov, V., Kang, Q.,
Xu, H., Pawar, R., Carr, T., Li, L., Guthrie, G.D., & Viswanathan, H. (2020).
Physics-Informed Machine Learning for Real-time Reservoir Management.
AAAI Spring Symposium: MLPS.
37. Muhammad Saiful Islam, Using deep learning based methods to classify salt
bodies in seismic images, Journal of Applied Geophysics, Volume 178, 2020,
104054, ISSN 0926-9851, https://doi.org/10.1016/j.jappgeo.2020.104054.
38. Müller, M., Macklin, M., Chentanez, N., Jeschke, S., & Kim, T. Y. (2020,
December). Detailed rigid body simulation with extended position based
dynamics. In Computer Graphics Forum (Vol. 39, No. 8, pp. 101-112).
39. Navrátil Jiří, King Alan, Rios Jesus, Kollias Georgios, Torrado Ruben, Codas
Andrés, Accelerating Physics-Based Simulations Using End-to-End Neural
Network Proxies: An Application in Oil Reservoir Modeling, Frontiers in
Big Data Vol. 2, 2019. URL=https://www.frontiersin.org/article/10.3389/
fdata.2019.00033, DOI=10.3389/fdata.2019.00033/
40. Ojukwu, C., Smith, K., Kadkhodayan, N., Leung, M., & Baldwin, K. (2020,
August 11). Reservoir Characterization, Machine Learning and Big Data –
An Offshore California Case Study. Society of Petroleum Engineers.
doi:10.2118/203642-MS/
41. Overview of GAN Structure | Generative Adversarial Networks. (n.d.). Google
Developers. Retrieved February 16, 2021, from https://developers.google.
com/machine-learning/gan/gan_structure/
42. Pedregosa, F., Varoquaux, G., Gramfort, A, Michel, V., Thirion, B., Grisel, O.,
Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos,
A., Cournapeau, D., Brucher, M., Perrot M., and Duchesnay, E., 2011, Scikit-
learn: Machine Learning in Python., Journal of Machine Learning Research
12 (2011) 2825-2830.
43. PetroWiki. (2016, January 18). Basic elements of a reservoir characterization study.
https://petrowiki.spe.org/Basic_elements_of_a_reservoir_characterization_
study.
44. Rose, K. K., Bauer, J. R., & Mark-Moser, M. (2020). A systematic, science-driven
approach for predicting subsurface properties. Interpretation, 8(1),
T167-T181.
45. Rumelhart, D., Hinton, G. & Williams, R. Learning representations
by back-propagating errors. Nature 323, 533–536 (1986). https://doi.
org/10.1038/323533a0.
522 Reservoir Characterization
46. Saikia, P., & Baruah, R. (2019). Investigating Stacked Ensemble Model for Oil
Reservoir Characterisation. 2019 IEEE International Conference on Systems,
Man and Cybernetics (SMC), 13-20.
47. SHAHKARAMI, A. MOHAGHEGH, S. Applications of smart proxies for
subsurface modeling, Petroleum Exploration and Development, Volume 47,
Issue 2, 2020, pp. 400–412, ISSN 1876-3804.
48. Sitharam, T. G., Samui, P., & Anbazhagan, P. (2008). Spatial variability of rock
depth in Bangalore using geostatistical, neural network and support vector
machine models. Geotechnical and Geological Engineering, 26(5), 503-517.
49. Stanford University & MIT University. (n.d.). CS 230 - Recurrent Neural
Networks Cheatsheet. Stanford.Edu. Retrieved February 16, 2021, from
https://stanford.edu/%7Eshervine/teaching/cs-230/cheatsheet-recurrent-
neural-networks
50. Suhag, A, Ranjith, R, and Aminzadeh, F. Comparison of Shale Oil Production
Forecasting using Empirical Methods and Artificial Neural Networks. SPE
Annual Technical Conference and Exhibition, San Antonio, Texas, USA,
October 2017.
51. Suhag, Anuj, Ranjith, Rahul, Balaji, Karthik, Peksaglam, Zumra, Malik,
Vidhi, Zhang, Ming, Biopharm, Frontida, Putra, Dike, Energy, Rafflesia,
Wijaya, Zein, Dhannoon, Diyar, Temizel, Cenk, and Fred Aminzadeh.
“Optimization of Steamflooding Heavy Oil Reservoirs.” Paper presented
at the SPE Western Regional Meeting, Bakersfield, California, April 2017.
doi: https://doi.org/10.2118/185653-MS.
52. Tan PN, Steinbach M, Kumar V (2005) Introduction to data mining.
Addison-Wesley Longman Publishing Co., Inc, Boston.
53. Tapoglou, E., Karatzas, G. P., Trichakis, I. C., & Varouchakis, E. A. (2014).
A spatio-temporal hybrid neural network-Kriging model for groundwater
level simulation. Journal of Hydrology, 519, 3193-3203.
54. Tarrahi, M., Afra, S., & Surovets, I. (2015, October). A Novel Automated
and Probabilistic EOR Screening Method to Integrate Theoretical Screening
Criteria and Real Field EOR Practices Using Machine Learning Algorithms.
In SPE Russian Petroleum Technology Conference. Society of Petroleum
Engineers.
55. Tembely, M., AlSumaiti, A., & Alameri, W. (2020). A deep learning perspec-
tive on predicting permeability in porous media from network modeling to
direct simulation. Computational Geosciences, 24, 1541-1556.
56. Temizel, C., Tiwari, A., Kirmaci, H., Aktas, S., Ranjith, R., Zhu, Y., Tahir, S.,
Aminzadeh, F., Yegin, C. (2016-January). “Improved optimization through
procedures as pseudo objective functions in nonlinear optimization of oil
recovery with next-generation reservoir simulators,” Proc. - SPE Annu. Tech.
Conf. Exhib., vol. 2016.
57. Tian, X., & Daigle, H. (2018). Machine-learning-based object detection in
images for reservoir characterization: A case study of fracture detection in
shales. Geophysics, 37, 435-442.
Machine Learning in Reservoir Characterization 523
58. Tokpanov, Y., Smith, J., Ma, Z., Deng, L., Benhallam, W., Salehi, A., … Castineira,
D. (2020, October 21). Deep-Learning-Based Automated Stratigraphic
Correlation. Society of Petroleum Engineers. doi:10.2118/201459-MS.
59. Unsupervised Feature Learning and Deep Learning Tutorial. (n.d.). Stanford.
Edu. Retrieved February 16, 2021, from http://ufldl.stanford.edu/tutorial/
unsupervised/Autoencoders/.
60. Vo Thanh, H., Sugai, Y. & Sasaki, K. Application of artificial neural network
for predicting the performance of CO2 enhanced oil recovery and storage
in residual oil zones. Sci Rep 10, 18204 (2020). https://doi.org/10.1038/
s41598-020-73931-2.
61. Wei Xiong, Xu Ji, Yue Ma, Yuxiang Wang, Nasher M. AlBinHassan, Mustafa
N. Ali, and Yi Luo, (2018), “Seismic fault detection with convolutional neural
network,” Geophysics 83: O97-O103.
62. Wingo, P., Justman, D., Creason, G., Jones, K., Bauer, J., Rose, K., SIMPA,
2019-03-29. https://edx.netl.doe.gov/dataset/simpa-tool.
63. Yenwongfai, H.D., Mondol, N., Lecomte, I., Faleide, J.I., & Leutscher, J. (2019).
Integrating facies‐based Bayesian inversion and supervised machine learn-
ing for petro‐facies characterization in the Snadd Formation of the Goliat
Field, south‐western Barents Sea. Geophysical Prospecting, 67, 1020–1039.
64. Zhao, X, Popa, A.S., Ershaghi, I, Aminzadeh, F., Li, Y., and Cassidy, S., 2019,
Reservoir Geostatistical Estimation of Imprecise Information Using Fuzzy
Kriging Approach, Accepted for SPE-190051- Reservoir Evaluation &
Engineering.
65. Zhong, Z., Sun, A. Y., & Wu, X. (2020). Inversion of Time‐Lapse Seismic
Reservoir Monitoring Data Using CycleGAN: A Deep Learning‐Based
Approach for Estimating Dynamic Reservoir Property Changes. Journal of
Geophysical Research: Solid Earth, 125(3), e2019JB018408.
Index
525
526 Index
Analytical model for predicting fluid p-values for hypothesis tests, 63t,
profile. see Wellbore temperature 64
profile during drilling quality characteristics of, 52
gas-hydrates reservoirs sparsity, 51, 55, 56–58, 59–60, 61,
Analytical thermal-model for 63f, 64
optimization of gas-drilling, trainSetRecords, 59
340–351 universal classifiers, 58–61
introduction, 340–341 identification rule, 50
mathematical model, 341–345 methodology, 49–50
model applications, 349–351 overview, 48–49
model comparison, 346–348 performance
sensitivity analysis, 348 one-class anomaly detection
Anifowose, F., 507 technique, 53
Anisotropy, 37, 406–407, 413 posterior quality characteristics,
Anomalous cluster sets, prior FD rates 52–55
for, 142–143 prior quality characteristics,
Anomalous records, 52–55
absence of, 142 ROC-curve analysis, 52
efficiency and reliability of posterior analysis of efficiency of,
identification, 145 144–146
prior and posterior TD and FD rates ROC curve analysis, 55–58
for, 142 training and test sets, 49, 55, 59, 60
in randomized test sets, 144f Anomaly index(es),
Anomaly cutoff, 53 cluster, 140–141, 144–146
calculation of, 52 gas and brine datasets, 146–148
defined, 50 individual records, 141, 144–146
functions of, 52 regular records, 145
Anomaly detection (AD), Anomaly indicator,
bootstrap bootstrap based tests of anomaly
based tests, 61–64 type hypothesis, 61, 62–63
for statistical analysis of, 52 percent of values of, 63t
classifiers for three types of classifiers, 63f,
adaptive aggregated, 52, 63f, 64 64
aggregated, optimization of, Apache, 297, 308
58–61 ArcGIS Desktop 10.6, 506, 511
Anomaly records, 59, 60 Area under ROC curve (AUC)
AUC for, 55–58 analysis, 159, 160
construction of, 49, 50 posterior, 55–58, 61t
distance, 50–51, 55, 56–58, 61, values, ranking parameters
63f, 64 according to, 161–162
divergence, 51, 53–55, 56–58, 61 Arnold, F. C., 364
histograms of bootstrap generated Arps hyperbolic model, 270
values, 63f, 64 Artificial intelligence (AI), 491
Index 527
Inclined direct shear testing device Iraq National Oil Company (INOC), 320
(IDSTD), 413 Irregularity index, of individual
Incompressibility, of reservoir rock, clusters in cluster set, 139–141
14 Islam, M. A., 402
Independent oil and gas companies,
307t, 308–309, 327t–335t Jeong, S. W., 431
Individual clusters, irregularity index Jochen, V., 270
of, 139–141 Joukowski formula, 87
Individual forecasts, Joule-Thomson cooling effect, 341–
enhancement of stability of MC 342, 348, 350, 365, 372, 375–377
CMs forecast via increase of Ju, B., 431, 432
number, 246–247 Justman, D., 505
instability of, analysis, 244–246
by linear regression and machine Kabir, C. S., 341, 347, 364–365
methods, 239–240 Kang, Q., 508
with ML methods, 242–244 Karstifed Limestone (Papua New
parameters of distribution of Guinea), 340
number of, 237–238 Kavousi, P., 467, 469, 471
Individual records, KazMunaiGas (Kazakhstan), 296
anomaly indexes of, 141 Keller, H. H., 346, 364
prior FD rates for anomalous cluster Kernel functions (KF), 492
sets, clusters and, 142–143 Kerogen, 71, 407, 413
Inhomogeneity of training and test conversion and maturity, 72, 75f
sets, 136–138 type, 72, 74f
Instability, Kim, T. Y., 431
of clustering process, 136–138 K-means clustering, 493
of individual forecasts, analysis, K-nearest neighbor (KNN), 391–392
244–246 AUC values, 161, 162
Instability index, individual forecasts with, 242–244
of individual forecasts, 246 true and false discovery rates with,
for MC CM forecast, 235 163
Integrated oil company reserves, 306t Kumar, M., 402, 407, 413
Integration, Kutasov, I. M., 341, 365
conventional seismic, 15f Kuwait Oil Company (KOC), 320
data Kuybyshev, Along-Volga Region,
challenges, 7–10 Russia, 249
geophysical, 6, 20 Kuzbass (Russia), CBM resources,
seismic and geologic data, 7–10 217–218, 221, 226, 228
3D seismic interpretation, 17
information from seismic data, 18 Lai, B. T., 404
reservoir structure, 5f Langmuir-like model, 436
Intel i5 class chipsets, 112 Laplace space solution, 172
International oil companies, 295–296, Laplace transform, 170, 189–190
300, 305, 310–311 Laramide Orogeny, 257
Index 535
Energy Storage 2nd Edition, by Ralph Zito and Haleh Ardibili, ISBN
9781119083597. A revision of the groundbreaking study of methods for
storing energy on a massive scale to be used in wind, solar, and other
renewable energy systems. NOW AVAILABLE!
Nuclear Power: Policies, Practices, and the Future, by Darryl Siemer, ISBN
9781119657781. Written from an engineer’s perspective, this is a treatise
on the state of nuclear power today, its benefits, and its future, focusing on
both policy and technological issues. NOW AVAILABLE!