IWC (Intelligent Well)
IWC (Intelligent Well)
IWC (Intelligent Well)
TESE DE DOUTORADO
Thesis presented to the Programa de Ps-Graduao em
Engenharia Mecnica of the Departamento de Engenharia
Mecnica, PUCRio as partial fulfillment of the
requirements for the degree of Doutor em Engenharia
Mecnica.
Advisor: Prof. Marcio da Silveira Carvalho
Co-Advisor: Prof. Arthur Martins Barbosa Braga
Rio de Janeiro
August 2012
Bibliographic Data
Silva Junior, Manoel Feliciano da
Intelligent well transient temperature signal reconstruction /
Manoel Feliciano da Silva Junior ; advisor: Marcio da Silveira
Carvalho ; co-advisor: Arthur Martins Barbosa Braga. 2012.
162 f. : il. (color.) ; 30 cm
Tese (doutorado) Pontifcia Universidade Catlica do Rio de
Janeiro, Departamento de Engenharia Mecnica, 2012.
Inclui bibliografia
1. Engenharia Mecnica Teses. 2. Intelligent well. 3. Numeric
flow simulation. 4. Finite elements. 5. Data analysis. 6. Robust
principal component analysis. 7. Diffusion maps. I. Carvalho, Marcio
da Silveira. II. Braga, Arthur Martins Barbosa. III. Pontifcia
Universidade Catlica do Rio de Janeiro. Departamento de
Informtica. III. Ttulo.
CDD: 621
Acknowledgements
It has been a very long journey since the work on this thesis started. Finally, after
all the efforts it became true.
First of all, I would like to thank all my family for backing me up along the way.
My wife Luciane and my two sons Manoel and Gustavo were of big importance
and created a great environment for me along these years. Without them I would
not have the strength to peacefully follow my path in order to make this
achievement.
To my advisors Marcio Carvalho and Arthur Braga from PUC-Rio and David
Davies from Heriot-Watt I would like to express all my gratitude for the everyday
consideration and encouragement for the realization of this work. Their
contribution and patience were immense.
I also want to give special thanks to Dr. Khafiz M Muradov who has also
provided an important support and contribution for this thesis.
Further, I would like to send thanks out to all my colleagues from PUC-RIO, in
particular to Andr T. Machado, for providing a great environment and productive
discussions.
Last but certainly not least, I am very thankful for all the support given from
Petrobras which made everything possible.
Abstract
Da Silva Junior, Manoel Feliciano; advisor: Carvalho, Marcio da Silveira;
co-advisor: Braga, Arthur Martins Barbosa. Intelligent Well Transient
Temperature Signal Reconstruction. Rio de Janeiro, 2012. 162p. D.Sc.
Thesis Departamento de Engenharia Mecnica, Pontifcia Universidade
Catlica do Rio de Janeiro.
Intelligent Well (IW) technology has built-up several years production
experience. Numerous publications have described how remote flow control and
monitoring capabilities can lead to fewer interventions, a reduced well count and
improved reservoir management. Despite the maturity of IW equipment, the
concept of the integrated IW as a key element in the Digital Oil Field still not
fully developed. Permanent monitoring systems in this framework play an
important role as source of the necessary information about actual production
system aiming model calibration and uncertainty minimization. However, each
extra permanently installed sensor increases the wells installation complexity and
operational risk. A well-founded understanding of what data is actually needed
and what analysis techniques are available to extract the required information are
key factors for the success of the IW project. This work proposes a new
framework to real-time data analysis through centralizing pre-processing. A
numeric IW transient temperature model is developed, tested and validated to
generate synthetic data. It was chosen without loss off generality as a
representative application to test and validate the cleansing and feature
extraction algorithms developed. The results achieved are compared with the state
of the art ones showing advantages regarding efficiency and potential to capture
mutual influence among processes.
Keywords
Intelligent Well; Numeric Flow Simulation; Finite Elements; Data Analysis;
Robust Principal Component Analysis; Diffusion Maps.
Resumo
Da Silva Junior, Manoel Feliciano; orientador: Carvalho, Marcio da
Silveira; co-orientador: Braga, Arthur Martins Barbosa. Reconstruo de
sinais transientes de temperatura em poos inteligentes. Rio de Janeiro,
2012. 162p. Tese de Doutorado Departamento de Engenharia Mecnica,
Pontifcia Universidade Catlica do Rio de Janeiro.
A tecnologia de poos inteligentes j possui muitos anos de experincia de
campo. Inmeras publicaes tem descrito como o controle de fluxo remoto e os
sistemas de monitorao podem diminuir o nmero de intervenes, o nmero de
poos e aumentar a eficincia do gerenciamento de reservatrios. Apesar da
maturidade dos equipamentos de completao o conceito de poo inteligente
integrado como um elemento chave do Digital Oil Field ainda no est
completmente desenvolvido. Sistemas permanentes de monitorao nesse
contexto tem um papel fundamental como fonte da informao a respeito do
sistema de produo real visando calibrao de modelos e minimizao de
incerteza. Entretanto, cada sensor adicional representa aumento de complexidade
e de risco operacional. Um entendimento fundamentado do que realmente
necessrio, dos tipos de sensores aplicveis e quais tcnicas de anlises esto
disponveis para extrair as informaes necessrias so pontos chave para o
sucesso do projeto de um poo inteligente. Este trabalho prope uma nova forma
de tratar os dados em tempo real de poos inteligentes atravs da centralizao do
pr-processamento dos dados. Um modelo poo inteligente numrico para
temperatura em regime transiente foi desenvolvido, testado e validado com a
inteno de gerar dados sintticos. A aplicao foi escolhida sem perda de
generalidade como um exemplo representativo para validao dos algortmos de
limpeza e extrao de caractersticas desenvolvidos. Os resultados mostraram
aumento da eficincia quando comparados com o estado da arte e um potencial
para capturar a influncia mtua entre os processos de produo.
Palavras Chave
Poo Inteligente; Simulao de escoamento; Elementos Finitos; Anlise de
Dados; Anlise de Componentes Principais Robusta; Mapas de Difuso.
Contents
1. Introduction
20
1.1. Scope
21
1.2. Contributions
22
1.3. Outline
22
2. Intelligent Wells
23
25
2.2. Packers
25
26
27
29
29
30
31
3. IW Monitoring Systems
33
34
34
34
35
35
36
36
37
38
39
42
43
44
46
50
52
54
55
63
68
69
71
71
72
73
74
75
76
76
76
79
80
82
87
87
88
89
91
92
94
96
98
6. Data Analysis
102
108
108
112
112
117
123
124
124
125
127
128
131
131
134
134
138
138
140
143
143
143
144
Bibliography
145
Appendix A
151
Appendix B
153
Appendix C
155
Appendix D
159
Symbols
S
Saturation
Rs
T
u
Temperature
K JT
Cp
Joule-Thomson coefficient
ugm
k
kr
p
q
qT
A
P
Area
Velocity vector
Permeability tensor
Relative permeability
Pressure
Volumetric flow
Density
Potential
Porosity
U
H
KT
Thermal flow
Compressibility
Internal energy
Enthalpy
Thermal conductivity
Stress tensor
Relative mobility
Standard gravity
U wb
fF
Radius
Viscosity (dynamic)
R
yp
Te
xp
gT
Geothermal gradient
wall
Co
Distribution coefficient
D
Dh
Diameter
Surface tension
Hydraulic diameter
Fp
Pipe roughness
Rm
Re
U af
h lat, jj'
U ta
I
Pr
Inclination factor
Gr
Grashof number
Ra
Raleigh number
Cd
Discharge coefficient
Prandt number
Subscript
w
o
s
z
r
p
Water component/phase
Oil phase component/phase
Rock
z direction
r direction
Phase
m
I
f
t
a
Mixture
Inflow
Fluid
Tubing
Annulus
Christmas Tree
DAS
DM
Diffusion Maps
DTS
DTSS
DVS
EM
Electromagnetic Method
ESP
FBG
GLV
GOR
Gas-Oil Ratio
GVF
ICV
IPR
IW
Intelligent Well
IWC
IWCS
IWIS
IWSCS
JIP
MAD
MD
Measured Depth
MEMS
MIMOSA
maintenance standards
MPFM
Multi-Phase Flowmeter
MTTF
NIST
NP
Non-Polynomial Time
NTP
OBC
OLEDB
OPC
PCP
PDG
PES
PRODML
PVT
Pressure-Volume-Temperature Analysis
RED
RESQML
RMAD
RMSE
ROV
RPCA
RTD
SCAD
SCADA
SCM
SCRAMS
SHD
SOI
Silicon on Insulator
SONAR
SoS
Speed of Sound
SVD
TCP/IP
TDM
TH
Tubing Hanger
TSR
TTA
TVD
VSP
WC
Water Cut
WCT
WDM
WH
Wellhead
WITSML
Figures
24
27
27
28
28
32
38
40
43
44
46
46
46
48
53
56
56
57
59
64
65
66
66
67
68
69
71
75
77
77
83
87
Figure 5.2 - Temperature and pressure in a vertical well with 500 bpd
of oil production in the tubing and in the annulus.
88
90
91
92
93
93
94
94
95
95
96
96
97
97
98
98
99
99
102
103
105
108
109
Figure 7.3 - Noise and drift of a real PDG (the green signal is the
clean one).
110
111
113
113
114
115
116
Figure 7.10 - Wavelet filter performance for the hard threshold (red is
the clean signal).
117
Figure 7.11 - Wavelet filter performance for the soft threshold (red is
the clean signal).
117
118
120
121
122
122
123
126
127
128
129
132
132
133
134
136
138
151
152
155
159
Figure D.2 - Control volume of the reservoir mass balance and flow
equations.
159
161
Tables
42
61
62
74
75
136
137
139
141
142
The dictionary is the only place where success comes before work.
Vince Lombardi
1
Introduction
21
the different disciplines without important simplifications that result in losing the
incremental value of the integrated modeling and optimization framework
(Hudson, Alves et al. 2011). A new data analysis approach with a centralized preprocessing is proposed aiming to design a robust permanent monitoring
architecture to deliver the full Added Value of the IW technology. It also allows
feature extraction that can potentially improve the quality of the post-processing
and interpretation predictions. This new approach has not yet been fully explored,
even though a successful application can be found in offshore production and
maintenance operations (Friedemann, Varma et al. 2008).
A numeric non-isothermal dynamic well-reservoir model is developed and
validated as an IW representative application to generate synthetic data and
illustrate the applicability of this new data analysis approach. It combines the
reservoir temperature model from (Sui, Zhu et al. 2008) and the well temperature
model from (Muradov and Davies 2008) including the necessary modifications for
the two phase flow formulation and multilayer reservoir.
The signal reconstruction algorithms can be used with single point,
multipoint, and distributed/quasi-distributed temperature monitoring systems. The
idea is identifying the impact of the temporal and spatial resolution as well as
noise and outlier in the new data analysis framework proposed. Two algorithms
are used: robust principal component analysis (Cands, Li et al. 2011) for
denoising and outlier removal and DM (Diffusion Maps) (Coifman and Lafon
2006) for feature extraction. The results are compared with the ones from (Olsen
2011) showing in which cases there were improvements. Compression and time
synchronization are discussed but they are suggested as future work.
1.1
Scope
The goal of this thesis is to develop and study a new data analysis
framework with centralized pre-processing. The study was limited to a
temperature transient application as a representative example without loss of
generality.
22
1.2
Contributions
The contributions of this thesis can be summarized as:
1.3
Outline
The organization of this thesis can be summarized as follows:
Chapter 2 reviews intelligent completion technology and introduces the concept
of integrated intelligent completion.
Chapter 3 reviews permanent monitoring systems and introduces a new IW
monitoring system design framework. The benefit of dealing with data quality
control is presented.
Chapter 4 describes the numerical non-isothermal dynamic well-reservoir model
for multi-layer reservoir and vertical wells equipped with inflow control valves
and its practical implementation.
Chapter 5 validates the model described in chapter 4 and analyses temperature
transients caused by step-like change of downhole flow control devices.
Chapter 6 reviews data analysis and interpretation concepts and proposes a new
data analysis framework with centralized pre-processing.
Chapter 7 reviews theoretical background for denoising, outlier removal and
feature extraction based on robust principal component analysis and DM.
Chapter 8 validates and compares the denoising and feature extraction algorithms
with the state of the art ones.
We conclude our research with a summary of our findings and a discussion for
future work in Chapter 9.
2
Intelligent Wells
24
25
2.2
Packers
IW packers are used to provide hydraulic isolation of zones which can be different
reservoirs or different layers of the same reservoir allowing selective control; the
uppermost completion packer is also responsible for anchoring the tubing and
providing the first safety barrier for the annulus, the same basic functions of the
conventional production packer. There are also packers for isolation purposes that
have only isolating material with no anchoring which reduces the force needed for
unsetting them and allows a large number of isolation intervals. The IW packers
also have feedthroughs to bypass control lines; typically an IW packer presents
from four to nine feedthroughs. During the setting procedures, the IW packer must
not have any relative movement of its components so it does not transmit any
tension to the control lines.
26
2.3
Inflow Control Valve
Inflow control valves are responsible for allowing selective control of
production or injection. They can be actuated hydraulically, electrically or through
a combination of both (multiplexed). Even though the control of ICVs (Inflow
Control Valves) has been changed towards only hydraulic systems due to its
reliability and cost, nowadays there is a trend back to electrical and downhole
multiplexed systems. Mainly due to reliability improvement in downhole
electronics and number of feedthroughs restrictions in tubing hanger and WCT
(Wet Christmas Tree) for subsea production. The most common control type is
still hydraulically actuated, which uses one opening control line for each valve
and a common close control line for the system to reduce the number of hydraulic
control lines installed. In electrically actuated valves, an electrical motor is
responsible for the sleeve movement. The motors are actuated using one single
electrical control line for all motors that supplies also the addressing information,
which is decoded in each valve. The electric/hydraulic systems use an electric
control line for multiplexing and one or more hydraulic control lines to provide
power to move the sleeve. The valves can also be classified according to the flow
control they provide as on-off, multiposition and infinitely variable. The on-off
valves only provide the selectivity by allowing or not the flow. The multiposition
valves provide several steps of choking, and are designed accordingly to the flow
rate expected on the well. They can use an index system to restrict the course and
provide the choking or an external device that provides a very controlled volume
of hydraulic fluid in each shifting. The infinitely variable valves are more
complex and require feedback on their position to adjust the correct choking.
There are also different geometries for the valve orifices, but the most common
are circular and variable area slots, which are used in on-off/multiposition and
infinitely variable respectively. It is worth to mention that inflow control valves
not only control flow from annulus to tubing but also control flow from tubing to
tubing using its shrouded version. The main parameters on specifying IW valves
are: number of choking positions, choking profile, flow range, maximum pressure
and maximum differential pressure.
27
2.4
Lines, connectors, feedthrough connectors and clamps
IW requires control lines to remotely control its components and receive
data from the sensors. These lines have to be protected against shocks and
chemical attacks during intervention procedures. Industry standard lines are
generally used to offer mechanical resistance and a polymeric material to provide
chemical protection. To improve the support resistance, armor cables can also be
used alongside the lines, forming a flatpack. In subsea IW applications the
electrical control lines are limited to single wire or twisted pair due to tubing
hanger wet-mate feedthrough connector system. It is recommended to use the
twisted pair electrical control line due to its noise characteristics and
improvements in telemetry to handle multiple sensors at faster data rate. Fiber
optic applications have at least one fiber available (typically three) and there are
mechanical restrictions in the TH/WCT connector system. Subsea fiber optic
applications up to the writing of this thesis are not available due to TH/WCT wetmate connector system and downhole wet-mate connector.
Control lines protectors clamps are used to hold the lines to the
production tubing and to offer extra protection to it near the tubing couplings,
where the diameter of the tubing is increased.
28
Figure 2.5 - TH/WCT wet-mate feedthrough connector system for vertical and horizontal
WCTs.
29
2.5
Wet Disconnection Tool
IW is used integrated with other completion tools such as artificial lift and
sand control. Typically when there are reliability issues between upper and lower
completion a specific tool named WDT (Wet Disconnection Tool) is necessary. It
enables upper completion to be disconnected from lower completion and
reconnected afterwards for maintenance purposes. An artificial lift using a high
performance ESP (Electric Submersible Pump) which has lower MTBF (Mean
Time Between Failure) is a representative example of the reliability issues
mentioned. Basically it is a TSR (Tubing Separation Tool) equipped with control
lines wet-mate connectors available to hydraulic lines and electric lines but still in
development to fiber optic lines. Advances, however, have been made do avoid
orientation requirement and expand the number of channels being available.
Based on the connection principle it can be classified as discrete or concentric as
can be seen in Figure 2.6. The former uses conventional wet-mate connectors and
the latter and a manufacturer specific design.
2.6
Multi-point Chemical Injection
A typical downhole chemical injection system consists of a chemical control
line and a chemical injection mandrel with a valve installed as part of the
production tubing string. Usually a check valve is also installed at the point of
injection to prevent flow from the production tubing to the injection flow path. A
high-pressure pump, capable of overcoming the tubing pressure, is installed on the
surface to pump fluid into the downhole point of injection. They are used for
scaling control as water cut or condensate water increases in oil wells and gas
30
2.7
Intelligent Well Control System
Subsea IW is controlled and monitored by a production SCM (Subsea
Control Module) installed on a WCT or a manifold. The SCM provides one or
more hydraulic control lines to operate the downhole safety valves and inflow
control valves as well as electrical/optical control lines to the downhole sensors in
the completion. Umbilical and in-field jumpers supply normally low/high
hydraulic pressure, electrical power and communications to SCM which come
from topside. This configuration is presented in the IWIS (Intelligent Well
Interface Standardization) addendum of the ISO 132628-6 as option 1
configuration 1. The first impression of this is a fully integrated system, less
complex with a reduced cost. To date this is difficult to accomplish due to the lack
of a full compatible set of interfaces (mechanical, electric, protocols, hydraulic
power, etc.) leading to time consuming and costly custom solutions with impacted
reliability. In mature fields, upgrades to install IW in this way are even worse. In
addition, IW systems are becoming more complex and adding new sensors and
functions to face the integrated framework challenges, which mean continuous
changes. The recommended solution for subsea IWCS (Intelligent Well Control
System) to achieve reliable solutions is the use of IWIS option 3 configuration 2
as detailed in IWIS Recommended Practice A2. This kind of solution allows
production control and IWCS tests to be done independently and faster
introduction of new technologies with reduced risk. At the operation phase, it also
makes the management of responsibilities clear. Besides IW interfaces
standardization there are two more efforts in this direction (standardization) that is
31
(Subsea
Fiber
Optic
Monitoring).
Improvements
in
subsea
2.8
Intelligent Well Surface Control System
IW surface control system is generally integrated to the field process
network through an application layer instead of a data layer. This means that a
proprietary SCADA (Supervisory Control and Data Acquisition System) is being
used by the IW supplier and might be a problem to integrate with the existing one
in the production unit. Due to the variety of suppliers and equipment involved, the
connectivity between them should be addressed at the project phase. Open
protocol specification should be adopted to simplify the project by reducing the
need of additional protocol translation. Among open protocols that have been used
in IWSCS (Intelligent Well Surface Control System) we should highlight OPC
(OLE for Process Control Connectivity) and TCP/IP(Transmission Control
Protocol/Internet Protocol) running over industrial Ethernet as the most accepted
and aligned with the standardization efforts mentioned earlier. Therefore, the
improved interoperability can provide the same information and services even
when a change of supplier is needed. In the integrated framework scenario high
levels of instrumentation led to demand for improvements in the timing and
format of the provided data which is important for all optimization processes. The
result was an effort of several companies for creating a high level protocol called
PRODML (Production Markup Language) that comes to try solving the data
exchange problem through standardization. Other solutions based on OLEDB
(Object Linking and Embedding Database) and historian servers are also available
but PRODML framework seems to be stronger and supported by almost all major
operators and service companies. In Figure 2.7, is possible to see its scope which
includes: drilling, completion and interventions (WITSML Wellsite information
Transfer Standard Markup Language); maintenance (MIMOSA Non-profit trade
association dedicated to operations and maintenance standards), SCADA and
reservoir characterization (RESQML Reservoir Characterization Markup
Language).
32
The real-time data comes from the IW monitoring system through the
IWSCS becoming available to the applications or to the centralized pre-processing
proposed in this thesis and discussed in the chapter 6. The following chapter
reviews the IW monitoring system introducing a new design framework.
33
3
IW Monitoring Systems
34
3.1
Electronic Sensors
3.1.1
Pressure and Temperature
Electronic sensors have been used for more than 20 years in PDGs for
measuring pressure and temperature. Quartz Resonator technology completely
dominates the single point sensor market. It has been responsible for increasing
pressure and temperature sensor reliability since the early nineties (Eck
1999/2000). MEMS (Micro Electro-Mechanical Systems) and the associated SOI
(Silicon on Insulator) technology has been responsible for the recent advances in
sensor reliability when installed in a High Pressure and High Temperature
(HPHT) environment. MEMS also have shown promise in the development of
sensors for measuring new physical quantities. Typical pressure and temperature
sensor data is shown in Table 3.2.
The mechanical protection, the type of telemetry employed and the available
number of transducers for each sensor depend on the manufacturer and the price.
Commercial sensors are typically capable of having up to 4 sensors on the same
electric control line (1/4 inch tubing with one twisted pair).
3.1.2
Quasi-Distributed Temperature
Increasing the number of sensor systems makes the installation more
complex and time consuming, hence less suited for downhole installation due to
the increasing level of operational risk. New multiplex techniques have allowed
the installation of a larger number of smaller temperature sensors on the same
cable without the need for multiple connectors and mandrels. For example, up to
48 temperature measurement sensors on a single cable using high resolution, RTD
(Resistance Temperature Detectors) are now commercially available (Gambhir,
Shrivastav et al. 2008).
Typical sensor data is shown in Table 3.2.
35
3.1.3
Single Phase and Two-Phase Flowmeter
Differential pressure flowmeters based on the Venturi principle are
commercially available for IW applications. They depend on the differential
pressure, a quantity easily measured by two pressure sensors. Flowmeters based
on the Venturi principle are best suited to single-phase measurements (e.g. water
injection). Their measurement uncertainty is < 1.5% (of the average) with a
rangeability of 4 to 1 (ratio of the maximum and minimum flow rates that can be
accurately measured). Their application to production wells is more restricted
since the fluids flow regime and the erosion velocity in the Venturis throat can
both compromise the measurement.
By contrast, an inverted Venturi is suited to high flowrates and has been
commercialized as a fullbore flowmeter (Ong, Aymond et al. 2007). It provides
the volumetric flow rate in the tubing together with temperature and pressure
measurements in the annulus and tubing. Some manufacturers provide calculated
measurements, such as WC and density, by including a densitometer or a
gradiometer. They can achieve an uncertainty of <5%, depending on the PVT
(Pressure-Volume-Temperature Analysis) data uncertainty (Kiryushkina, Sikandar
et al. 2011). An insertable Venturi has an operational advantage if low cost
interventions can be carried out by wireline, making it only a practical option for
conventional wells with dry trees.
3.1.4
Streaming Potential
A common logging tool used to characterize reservoir properties is the
spontaneous potential log. The streaming potential (one of the component of the
spontaneous potential) is strongly influenced by the resistivity of the fluids and
the differential pressure resulting from fluid flow through the porous media
towards the well. Laboratory results and numerical experiments have shown that
one can measure strong signals (up to 100 mV) in production wells when low
salinity formation water is present (Jackson, Vinogradov et al. 2011). It should be
possible to interpret those measurements to detect reservoir fluid saturation
changes, e.g. due to encroaching water, at distances of tens or even hundreds of
meters from the well (Jaafar, Jackson et al. 2009). Closed-loop reservoir
36
It is
3.1.6
Permanent Downhole Seismic
Improvements in time-lapse 3D seismic technology have remarkably
improved the quality of reservoir images. Borehole and surface seismic can define
geometric boundaries with reasonable accuracy, but not the fluid content, at long
distances (Strack 2010). The latter is best measured by the electromagnetic
technique. The downhole seismic measurements are performed by time-lapse VSP
(Vertical Seismic Profile) for borehole-to-borehole seismic imaging and by
surface and OBC (Ocean Bottom Cable) seismic calibration for permanent or
temporary deployment. During the production phase it can be used to passively
monitor rock fracturing from both injection and production activity. A typical
multilevel, downhole seismic monitoring systems has up to 30 nodes with 4
channels/node (3c geophones + 1 hydrophone for multiples correction) operating
37
with a sampling rate varying from 4 to 0.25 ms. Correct coupling with the
formation is essential for this type of system (Wilson, Floch et al. 2008).
MEMS accelerometers are now replacing the older electromagnetic
geophones due to their higher dynamic range, broadband linear phase/amplitude
response, small size and weight; though they still need improvement to minimize
noise at low frequencies (Mougenot, Cherepovskiy et al. 2011). The specification
of a downhole seismic accelerometer should consider the noise floor, linearity,
dynamic range and (multi-component) cross-axis sensitivity. Resonant frequency
should also be included since the sensor's upper useful frequency is a fraction of
its resonant frequency. It thus determines the sensitivity and displacement per g of
acceleration.
Typical 3c geophone data and the equation required to relate the geophones
electrical and mechanical data are given in Table 3.2.
Permanent downhole seismic has not been installed in an IW completion todate. The extra cost and complexity of equipping an IW with permanent downhole
seismic needs to be justified by the Added Value of, for example, enabling
proactive control of the reservoir. Current technical developments in the
combination of seismic and EM methods may make this feasible.
3.2
Fiber Optic Sensors
Fiber optic sensors have been applied to permanent downhole monitoring
systems since the mid-nineties (Zisk 2005). Initially only DTS was adopted by
operators for temporary well temperature profile logging. It was only at the end of
the nineties with the development of FBG (Fiber Bragg Grating) technology that
fiber optic sensing started to be used in permanent downhole monitoring systems
because FBG gave greater reliability and had the potential to measure a variety of
physical quantities in the same cable. Single point pressure and temperature
sensors, flowmeters and 3C geophones are now all commercially available at a
similar level of operational complexity and cost as their electronic counterpart, but
with a claimed higher level of reliability. Fiber optics great advantage for sensing
is its distributed capability, even though electronic quasi-distributed sensors can
now incorporate a larger number of sensors on the same cable (>45) which raise
questions about comparisons of those systems.
38
The technological gap for fiber optic sensing remains a reliable tubing
hanger wet-mate connector for subsea deepwater applications and a downhole
wet-mate connector for completions with multi trip installation or upper
completion equipment that needs to be changed regularly. For example, high rate
electric submersible pumps which have to be replaced at a frequency greater than
the remainder of the completion.
The next sections describe the fiber optic sensors available, their
metrological specifications and the basic principles of the technology that has
been exploited for permanent downhole monitoring: FBG, Raman and Brillouin
backscattering, coherent Rayleigh backscattering and distributed polarimetry.
3.2.1
Pressure and Temperature, Quasi-Distributed Temperature, and
Quasi-Distributed Strain Sensors
The fiber optic alternative to electronic single point and quasi-distributed
sensors is based on FBG technology. An FBG is produced by forming a periodic
structure with a permanent change in the refraction index along the propagation
axis of the fiber optic (Culshaw and Kersey 2008). The frequency reflected by this
periodic structure depends on the optical period of the grating, which is itself a
function of temperature and strain (Figure 3.1). The change in the optical path
length is of the order of 1 ppm/C (CT) and 10 ppm/ (Cs). Downhole
application of this technology faces the normal challenge of packaging and long
term stability.
39
reliability issues with their electronic counterpart in this harsh application with
levels of vibration under producing conditions.
Typical fiber optic pressure and temperature sensor data can be seen in
Table 3.2.
Quasi-distributed sensors based on FBG technology are also available for
temperature and strain measurements. The FBG sensors can be easily multiplexed
by WDM (Wavelength Division Multiplex). TDM (Time Division Multiplex) may
be included when the number of sensors increases up to a hundred, but with a loss
of accuracy and resolution. Current solutions include both approaches: WDM for
temperature measurement and WDM + TDM for strain measurement. The main
limitation regarding the number of sensors is the number of control lines,
connectors and fibers that can be handled simultaneously without making the
installation too complex and risky.
Permanent monitoring of deformations experienced by well tubular, casing
and screens identifies the loads due to reservoir compaction or changes in the
geological conditions, e.g. due to a squeezing salt layer as experienced by wells
completed in pre-salt reservoirs. Strain measurements allow early prediction of
catastrophic failure situations. Corrective action may then be taken to avoid the
complete loss of the well. The axial strain, the radius of curvature of bending and
crushing may be measured and, after signal processing, identification of the
deformation mode: axial compressing, bending, ovalization, shearing, pressuring
and/or thermal expansion can be computed (Rambow, Dria et al. 2010). This
system is currently in the field trial stage.
Data for both types of quasi-distributed sensors (temperature and strain plus
temperature) are shown in Table 3.2.
3.2.2
Single Phase and Two-Phase Flowmeter
A fullbore, permanent, downhole flowmeter must operate over a wide range
of pressures, temperatures, mixtures of fluids and flow regimes. Traditional
multiphase flowmeters are not suited for this environment. Only one downhole
flowmeter, with limitations related to minimum velocity and maximum acoustic
noise, is currently being marketed. This flowmeter simultaneously measures the
volumetric flow rate and the fluids speed of sound using passive SONAR (Sound
40
Navigation and Ranging) technology (Gysling and Loose 2003). The time
averaged flow profile in turbulent flow is measured by each element of an array of
sensors. This value is then used to identify and track spatially coherent structures
(eddies) that are travelling with the flow. The signals from each element of the
array are processed to deconvolve its frequency (f) and the wavelength (or length
scale component, ). The convection velocity can be then determined from the
signals spectrum by Equation 3.1:
(3.1)
Where
U
k
f
The volumetric flow rate is calculated from this velocity after a calibration
procedure. Figure 3.2 shows a plot of radian frequency () against wavenumber
(K). The dotted line in this plot corresponds to the convection (or bulk) velocity of
the mixture.
The acoustic waves travelling in the fluid are also measured. They propagate
both with and against the flow direction as a one dimensional acoustic wave
(Figure 3.2), representing the same coherent structures convecting along the axis
41
of the tubing. The velocity of the mixture and the SoS (Speed of Sound) can be
determined from those velocities, see Eq. 2 and Eq. 3 (Unalmis, Johansen et al.
2010):
c c
2
c c
SoS
2
cm
(3.2)
(3.3)
1.84SoS
.
D
The volumetric fractions for single phase or liquid mixture are obtained
from Eq. 4:
cm ycl1 (1 y)cl 2
(3.4)
Where
y - water fraction or water cut;
cl1, cl2 - constants obtained from the calibration procedure.
The speed of sound is proportional to the square root of the ratio of the
compressibility and the density for multiphase flow. The mixture compressibility
and density follow a similar relationship in which they are equal to the
volumetrically averaged properties of the individual components (Eq. 5 and Eq.
6):
m x g (1 x) l
(3.5)
1
x
1 x
2
pm cm pg cg pl cl2
(3.6)
Where
x - represents the gas volume fraction;
cl , cg - constants obtained from the calibration procedure.
This technology actually measures the dynamic strain in the tubing caused
by pressure fluctuation due to the turbulent flow. This passive technique is only
suitable for relatively high rate wells.
42
The acoustic noise generated by the IW valve can also be used to confirm
the valves opening positions. The majority of multi-position valves currently
used do not have position sensors with the valve position being derived from
previous information on valve movements stored in the control unit. Operational
experience has shown that it is not unusual for knowledge of the valve position to
be uncertain. It can only be confirmed by closing the valve completely, with
consequent loss of production.
Other diagnostics could be added to improve the condition monitoring of
downhole equipment. As far as we are aware this is not even being trialed.
Typical sensor data is shown in Table 3.2.
3.2.3
Permanent Downhole Seismic Sensors
The first fiber optic seismic sensors for permanent downhole measurements
were installed in southwestern France for a walkway VSP in 2002 (Knudsen,
Havsgard et al. 2003). It was a six level, tri-component (3C) accelerometer system
used for monitoring a gas storage field. Neither electronic nor fiber optic
permanent downhole seismic sensors have been trialed for installation in an IW
to-date. The advantage of the fiber optic accelerometers is their improved
metrological data (dynamic range, distortion, noise floor and bandwidth). In
addition, fiber optic accelerometers bring operational advantages for OBC seismic
(no batteries or heavy copper cables). The result is high resolution seismic
imaging and better dynamic reservoir characterization (4D) when one adds up
seismic acquisition reproducibility, acquisition frequency flexibility and improved
metrological parameters of the sensors. Typical data for downhole seismic sensors
are shown in Table 3.2. Table 3.1 is a simple comparison of the three seismic
sensor technologies available. This table illustrates the important differences
regarding noise and bandwidth between the two new technologies which result in
improvements in noise of seismic deeper traces and in trace resolution.
Electromagnetic
MEMS
Fiber Optic
Noise floor
< 1000 ng/Hz
< 700 ng/Hz
< 100 ng/Hz
Bandwith
10-240 Hz
0-800 Hz
1-1.4 kHz
Dynamic range
90 dB
120 dB
130dB
Distortion
-70 dB
< -90 dB
< -90 dB
Table 3.1 - Seismic sensors comparison by technology.
43
3.2.4
Distributed Temperature and Distributed Temperature and Strain
Sensors
Distributed downhole sensors gave the missing spatial component to the
reservoir and production engineers. The first distributed sensor for downhole
applications was based on Raman backscattering, the DTS. It is also available
today DTSS (Distributed Temperature and Strain Sensing) based on Brillouin
backscattering. Both methods are based on changes in the spectral content as a
result of nonlinear interactions of the light propagating through the fiber optic. It
uses the optical domain reflectometry technique which sends a series of optical
pulses into the fiber and extracts the scattered light from the same end of the fiber
(Figure 3.3). Raman backscattering measures the intensities of the Raman signal
at equal energy differences in opposite shifted directions, producing a ratio which
depends only on temperature (Culshaw and Kersey 2008). The backscattered light
is coupled, due to stimulated Brillouin backscattering, with an acoustic wave
which has half of the wavelength of the incoming light. The measured frequency
shift is the frequency of the acoustic wave which depends on the density ratio and,
therefore, on the temperature and strain (Culshaw and Kersey 2008).
44
one has to adjust the spatial and time resolution to meet the specifications of the
system being designed using a chart which plots the data as a function of the
above parameters. Figure 3.4 is such a performance chart for the hypothetical
system. It illustrates that a distributed measurement system based on optical
domain reflectometry needs more than one parameter, the distance and sample
rate for this system, to define its metrology performance.
3.2.5
Distributed Acoustic/Vibration Sensors
Acoustic signals are widely measured & interpreted in upstream Oil and Gas
production activities. They are used for condition monitoring, failure prediction
and diagnosis of the cause of failure in rotating machinery. They have become an
important component of the predictive maintenance strategy. In addition, acoustic
signals can be correlated to flow rate, the position of a downhole valve, the correct
functioning of a gas lift valve, etc.
45
46
47
48
49
f I E, M
f
*
E ,M * E, M
2
I
E2, M *
(3.7)
Where
f
* E, M *
E ,M
2
E , M * is the uncertainty from the measurements and
50
Te Twf
dTwf dz
C p K f (t ) RwbU
2 KRwbU
(3.8)
Another example is the relation between the measured acoustic noise level
(>600 Hz) N*600 and the single-phase flowrate q and cross-sectional area A
(McKinley and Bower 1979):
*
N 600
C
q3
A2
(3.9)
Note that the relative uncertainty of this method is relatively low (I ~ 0.2).
The time resolution required to reliably identify the desired event depends
on the event characteristic time scale. A good selection of characteristic time
scales is provided by (Nyhavn, Vassenden et al. 2000).
Optimizing the placement of a single point sensor can be done by the use of
Experimental Design. The wellbore and reservoir are modeled for different flow
conditions to provide (noisy) input data for back-calculation of phase flow rates
with the gauges placed at a number of locations. Multiple completion designs may
be evaluated and ranked according to their confidence level. Examples are
available from (Naevdal, Vefring et al. 2001).
3.3.2
Reliability
The reliability of a system or component is defined as its ability to perform
a required task under the design conditions over a specified period of time
(Rausand and Hyland 2004). It is generally presented as either a probability of
failure or a probability of availability. Long-term reliability was the first
barrier to IW technology adoption. Even today, the design engineer must include
the suppliers quality assurance and reliability management program when
51
52
53
equipment
failure,
etc.
Subsequent
well
54
3.5
Application Areas
A summary including measurement relevance to different application areas
is presented in Table 3.3. Note that instead of sensors there are measurements to
the application areas relation due to the fact that a combination of different
sensors can provide similar sets of measurements. As always, some of the standalone measurements are impossible to detect certain events, measurement
combinations however help. The application areas have been grouped into:
Conditional monitoring;
Well performance;
Well stimulation;
Flow assurance;
Advanced completions;
Reservoir characterization.
Published studies which have a similar intent of this chapter are commented
below:
(Nyhavn, Vassenden et al. 2000), who provided a very good insight on what
information is needed for effective IW control, ranked the available
measurements and sensors according to their availability (at that time),
reliability, and simplicity to the application area. They have analyzed the
major problems of IW management complex and computationally
demanding model based IW control, associated information needed,
measurement and event time scales and subsequently large data storage, and
processing and analysis power needed. Introduction of the advanced data
processing methods multivariate analysis, time-lapse processing and softsensing - were viewed to be of particular importance in the years to come.
Recognition that much IW value derives from the ability to handle surprises
and promptly take action based on the high quality information from the
downhole monitoring systems has been justified over the last decade.
55
3.6
IW Monitoring System Design Example
Let us consider a synthetic abstract example based on the PUNQS3
reservoir model; a simplified version of a real reservoir that is publically
available. It has previously been shown by (Birchenko, Demyanov et al. 2008;
Grebenkin and Davies 2010) to be a suitable candidate for IW. We will use the
optimal production strategy as developed by these authors. It consists of one,
horizontal, 4 zone IW producer and two vertical injection wells to support
reservoir pressure and dispose the produced gas. Figure 3.11 (middle) shows the
reservoir top view, the horizontal IW production well and the two injectors. The
production limits are liquid rate 600 sm3/d and bottom hole pressure > 120 bar.
A reactive strategy (after breakthrough) is planned for controlling the IW,
the ICVs are operated based on zonal WC (Water Cut) and GOR (Gas-Oil Ratio)
(Grebenkin and Davies 2010):
ai
i
Qliq
i
Qoil
Bo (GORi Rs) Bg
1 BwWC i
(3.10)
The optimization strategy uses the normalized value ai as the choke criterion
for each ICV trying to improve oil production while keeping the same liquid rate.
The ICV of the zone i is closed (or choked) iteratively starting always from the
zone with highest WC. The control decision made is confirmed if an oil
production improvement is achieved at each time step. A commercial reservoir
simulator (Eclipse 100) with multi-segment capability was used in the simulations
resulting an additional oil production of 0.17 MMsm3 (i.e. about $64 million
assuming oil price of $60/bbl.) for the worst case scenario or reservoir uncertainty
56
(Grebenkin and Davies 2010). All associated IW completion and operation costs
should be compared against this additional revenue.
Figure 3.11 - Structure (left), well locations (middle) and permeability distribution (right)
for the PUNQS3 model (Grebenkin and Davies 2010).
57
The well design consisted of a 9 5/8 casing above the reservoir with a 7
liner installed across the 8 /12 openhole reservoir section with 4 production
tubing, isolation packers, 3 multi-position inflow control valves and sensors
(Figure 3.12). The total depth of the well is 4500 m with GLV (Gas Lift Valve)
located at 1560 m (all measured depths). This is not an optimal design for high
rate wells or long horizontal wells due to maximum rate restriction but this is not
the case.
The reservoir data and well design suggest that the monitoring system
should be able to monitor zonal WC, zonal pressure and the performance of the
gas lift system. The required resolution of the zonal WC measurement was
identified as being the smallest change in WC that would lead to a control
decision being made by the equation 3.10. The optimal oil production for the 4
zones are: 80.6, 141.1, 15 and 100 sm3/d after valve 1 and 4 being choked which
corresponds to water cut changes of: -2, 5, 10 and -10% respectively and 4.6% of
improvement in oil production. They were found using the WC optimization
strategy on commercial reservoir flow simulation (Eclipse TM 100). Figure 3.13
shows the simulation results (marked lines) and the zonal WC values (solid lines).
The smallest change in WC between two control actions occurred for Zone 1 in
1976. It corresponds to 2% of the absolute value which would need to be
measured at zonal liquid rates ranging from 50 to 250 sm3/d.
WCT at ICVs
90
80
70
WCT, %
60
50
40
30
20
10
0
1/1/67
1/1/69
2/1/71
100% open ICV1 WCT %
100% open ICV2 WCT %
Optimal ICV1 WCT %
Optimal ICV2 WCT %
2/1/73
Date
3/1/75
3/1/77
4/1/79
100% open ICV4 WCT %
100% open ICV3 WCT %
Optimal ICV4 WCT %
Optimal ICV3 WCT %
58
integrated production model (reservoir, well, chokes, and pipeline models) and an
error minimization procedure between calculated and measured data. It has at
least 10% of model uncertainty (depending on the commercial software used) plus
less than 1% of measurement uncertainty which using equation 3.7 results in
10.05% of total uncertainty. Analyzing the second alternative, the most important
decision is the appropriate choice of the flowmeter. The low zonal flow rate
implies that SONAR or Inverted Venturi based flowmeter is not an option due to
their minimum flow rate limits. An insertable Venturi based flowmeter could be
used but it has a diameter restriction as a drawback which leads to a time
consuming and risky operation every time an intervention is needed below the
flowmeter. The uncertainty in this case is 2.5% due only to the measurement
uncertainty. The third alternative is a surface MPFM which has typically 5% of
WC measurement uncertainty for GVF (Gas Void Fraction) lower than 60%
which is the case for this example.
The alternatives for the zonal pressure and gas lift performance are:
59
(3.11)
Where
p
i p, T g , i 1, 2.
z i
Figure 3.14 - Temperature in lower completion showing two and all valves open.
60
Sensitivity (S)
1.32 V/in/s
Natural frequency f 1
0 2
10 Hz
0.57
7.8 g
2066
2400 ohms
R
s2
V s S s
U s
2
Rt s 2 s b0 Rt S k
mRt m
Where
V(s)
U(s)
m
k
0
Voltage
Velocity
Moving mass
Elastic constant
Open circuit damping
Shunt resistence
Total resistence
61
Application
Artificial lift
(operating GLV, ESP
performance, etc.)
Well/Pipeline Integrity
(casing leak, flow behind
casing, packer isolation,
etc.)
Well/Pipeline Structural
integrity (reservoir
compaction, formation
movement, corrosion,
etc)
Distributed
T
Influx identification
Producing zone/layer
identification
Perforated intervals
identification
Acidized intervals
identification
Well cleanup
Hydraulic fracture
(height/length/location
identification)
Flow
Assurance
Sand production
Cross-flow between
zones/layers
ICD/ICV/AICD
performance monitoring
Well Stimulation
Reservoir
Characterization
Hydrates formation
ICV position
Structural features
(faults, folds, etc)
Boundaries
Saturation Profiles
Well test
62
EM
Injection or production
flow rate profiling
Advanced
Completions
Well Performance
Type
Condition Monitoring
4
Numeric Non-Isothermal Dynamic Well-Reservoir Model
Several researchers have published results showing the need of wellreservoir coupled simulation. (Vicente, Sarica et al. 2001) showed in his
isothermal model that the traditional approach of decoupling wellbore flow from
reservoir flow in horizontal wells do not capture the interaction between them at
early times. (Grubert, Wan et al. 2009) showed that the impact of completion
design in the long-term well performance is not correctly achieved without two
tier coupling between well and reservoir. (Alberts, Belfroid et al. 2007) analyzed
the dynamic behavior in the well and reservoir identifying the time and space
scales at which the well-reservoir coupling becomes important. In Figure 4.1, a
time-space map from that work is reproduced in order to show the spatial and
temporal relationship among common production processes. It is possible, for
example, to see that well clean-up and reservoir transients have strong coupling
and then should not be analyzed separately (coincide completely). This simple
idea can illustrate when one should consider well-reservoir coupling.
Multilayer reservoir well testing models have been developed and used
successfully during the past two decades. The advances in downhole monitoring
systems in recent years have motivated new testing and analysis techniques. Using
transient temperature and pressure data (Sui, Zhu et al. 2008) developed a
numerical simulator for qualitative analysis of changes in permeability and skinfactor in a multilayered vertical well. They indicated that trasient temperature can
be more informative than pressure due to its sensitivity to damage radius and
permeability. (Duru and Horne 2010) have found out that properties such as
porosity and saturation could be estimated in an inverse problem from pressure
and temperature transient analysis which are not available from conventional
pressure transient analysis tools. (Valiullin, Ramazanov et al. 2009) highlighted
the importance of TTA (Temperature Transient Analysis) especially when the
barothermal effect (transient fluid compression/expansion at early times) takes
place for well test analysis.
64
TTA is not a mature technique and all possibilities have not been explored
yet. A basic analysis workflow is available for sandface temperature of a single
layer for a horizontal well (Muradov and Davies 2011). It assumes that downhole
pressure and temperature measurements and PVT data are available. Its
applicability depends on the observation of the initial temperature change period.
In addition, when the formation and fluid thermal properties are known then TTA
is sufficient to estimate fluid and reservoir flow properties. Otherwise, the multilayer problem is more complex due to every downstream temperature response
being affected by the upstream ones. Two major issues should be overcome to
solve this problem: transient heat losses and transient flow rates. Transient heat
losses depend on the rate of the fluid heat loss due to the difference between the
temperature of this fluid and the formation. Assuming the zonal temperatures are
stable before and shortly after the transients than a HTC (Heat Transfer
Coefficient) between annulus and formation can be defined and used. Transient
zonal flow rates can be assumed constant during the transient period because their
fluctuations have only a second order impact on the temperature signals (Muradov
and Davies 2012).
Figure 4.2 shows pressure and temperature behavior of a producing
horizontal well through six sensors placed in different positions. It shows clearly
distinct temperature change and practically no change in pressure. In addition,
temperature transients cannot be seen from all the sensors which indicate that they
are affected by upstream flowing zones.
65
Figure 4.2 - Sensors positions and their response in a producing horizontal well (Valiullin,
Ramazanov et al. 2009)
66
Figure 4.3 - Zonal IW transient pressure response (Muradov and Davies 2012).
Figure 4.4 - Zonal IW transient temperature response (Muradov and Davies 2012).
The pressure and temperature responses of the toe zone as shown in Figure
4.3 and Figure 4.4 is presented in Figure 4.5 for verification purposes only. They
are the sandface pure response as stated and discussed previously.
Analytical and numerical studies towards a thermal multiphase flow model
in pipes and wellbore have been published in the literature for a long time.
(Ramey Jr. 1962) presented the first single phase model which provides
temperature as a function of time and depth. (Alves, Alhanati et al. 1992) added
the two phase flow and well inclination in the model. (Hasan and Kabir 2002)
have generalized the model to gas, oil and two phase flow taking into account
Joule-Thomson effect and kinetic energy for the wellbore/formation system.
(Yoshioka, Zhu et al. 2005) presented a multiphase steady state temperature and
67
pressure model along the wellbore in horizontal wells. (Muradov and Davies
2009) proposed a general multiphase steady state well temperature model which
includes simultaneous fluid flow and heat transfer not only in the annulus and
tubing but also in the formation through more rigorous varying heat transfer
coefficients. It is at least to date the most complete model for IWs. The model was
derived from the total energy and mass balance equations assuming known
pressure distribution.
270
260
255
100.4
250
245
240
100.3
235
230
Sandface Temperature, C
265
100.5
225
220
22/08/04
23/08/04
100.2
24/08/04
Time
Figure 4.5 - Pressure and temperature transient sandface pure response in a horizontal
IW.
68
69
flow is assumed at the well head but without storage coefficients or skin effect
being considered (the model is limited to the reservoir region).
The model studied is two dimensional for the reservoir and one dimensional
for the well. Reservoir is considered homogeneous, anisotropic (khkv) and twophase (oil/water or gas/liquid) saturated with the initial pressure above the bubble
point. The well is vertical and drift-flux multiphase model (Appendix B) is
applied through correlations for the gas/liquid case (Ghajar and Tang 2010) where
the initial rates are given. In liquid/liquid case it simplifies to homogenous
multiphase model. Mass balance equations are derived using weighted properties
over volume fractions for the annulus and tubing. Pressure drop model is applied
with corrections due to inflow (Ouyang and Aziz 1998). Energy balance neglects
kinetic energy and viscous shear energy terms (Yoshioka, Zhu et al. 2005) but
heat transfer for annulus, tubing and near wellbore region are taken into account
(Muradov and Davies 2008).
4.1.1
Reservoir Layer Model (derived in appendix D)
Assumptions made:
70
Immiscible fluids;
Well segment length is small when compared with the total length of
the well;
dH C p dT
1 T dp
ct
p
t k p t gz
t
(4.1)
Where
ct c f j c j S j
t j j
t j j
General energy balance equation yields:
p t
T
p
p
t T
c f p f C pf T
t
t
t
C p t k p t gz T
1 T t k p t gz p kT T
Where
kT j S j k j (1 )k f
p t
j j S j C pj (1 ) f C pf
C
p
j S j C pj
t j S j j
j j
(4.2)
71
4.1.2
Well Segment Model
Assumptions made:
Phase fractions (yl) are given for liquids and comes from (Ghajar and
Tang 2010) correlations for gas/liquids (xg) for given rates;
4.1.2.1
Mass balance
Annulus:
w a
z
w ' I
Where
Tubing:
(4.3)
w t
z
72
(4.4)
Where
2
pa ma uma
2 f u2
Fa ma ma ma gsin
z
z
Dha
(4.5)
Tubing:
2
pt mt umt
2 f u2
Ft mt mt mt gsin
z
z
Dt
(4.6)
Where the average properties (for the annulus and tubing) are:
m j j y j
um
yu
y
j
The frictional force per unit of volume between a single phase fluid and the
wall is evaluated through the friction factor (fFt and fFa) which is calculated by
using classic friction factor correlations for pipe flow (Appendix A). Chen
correlation is used for this purposes as recommended in (Hasan and Kabir 2002).
It is not a strong effect in vertical wells but should be considered for high rate
wells.
For a gas-liquid flow in a vertical pipe the superficial gas velocity (cross
sectional area averaged velocity) is expressed as a sum of the distribution
coefficient (Co) and the gas drift velocity (ugm). It captures not only the
mechanisms of higher concentration of gas near the center of the pipe (Coum) but
also the tendency of buoyancy (ugm). This explains why gas inside a vertical pipe
moves faster than liquid.
ug Coum ugm
73
um usg usl 1 yl ug yl ul
Drift-flux model (Appendix B) is used through correlations from (Ghajar
and Tang 2010) to obtain Co, ugm and yl from given rates.
4.1.2.3
Temperature model
The temperature model (Muradov and Davies 2008) is derived from total
energy balance and mass balance equations. It was simplified here to fit the
assumptions made. Continuous inflow from the formation, heat transfer tubingannulus and annulus-formation, and one phase change is accounted for in the
model presented in (4.7) and (4.8). The third RHS term indicated by index I in
(4.7) represents the radial mass inflow per unit of length (w). The last RHS terms
in (4.7) and (4.8) accounts for phase change from j to j and Joule-Thomson effect
where Hlat,jj is the latent heat to change from the phase j to j and kJT is the JouleThomson
k JT
coefficient
which
is:
k JT
1
T 1
C p
for
liquid
and
T z
for gas. It should be noted that there is only one phase
C p z T p
dissolved in other and Rs is the gas-oil solubility ratio from black oil model. All
coefficients in brackets are average coefficients for annulus and tubing. Heat
transfer mechanisms among tubing, annulus and formation are accounted for
using heat transfer coefficients (Uta and Uaf).
Annulus:
wC
p a
Ta
2 Rci 1 U af T f Ta 2 RtiU ta Ta Tt w ' C p TI Ta
I
z
aj R
s
wa gsin wC p k JT wa , j ' H lat , jj '
a
aj ' a
Tubing:
p
a
z
Ta
(4.7)
wC
p t
74
Tt
2 RtiU ta Ta Tt wt gsin
z
tj R
s
wC p k JT w t , j ' H lat , jj '
t
tj ' t
Tt
p
t
z
(4.8)
4.1.3
Initial and Boundary Conditions
Two sets of initial and boundary conditions are defined. One set is used for
the first layer and another for any intermediary layer. The first layer is assumed to
be the deepest one experiencing uniform flow without tubing but with or without
a valve. The intermediary one represents a conventional IW segment. The
presence of a valve is taken into account as a different boundary condition for the
next layer (Table 4.1 and 4.2). Crossflow and heat transfer between layers are also
considered exceptions for the first and last layers (Figure 4.6). Transients can be
created by the wellhead choke or by the ICVs. Total flow is kept constant (the
sum of all layer flows) if the transient is created by the ICVs. Equation 4.14 or
4.15 are used to account for the flowrate change. In this case there will be flowrate
changes in the other zones to keep the total flowrate constant.
Layer 1
Initial Conditions
Boundary Conditions
T1t |( r Re , z ) Te1 gT z
T1 t 0 |( r , z ) Te1 gT z
Reservoir
T1 t
t
|( r Rci , z ) U wbf
T1t Twbt
r
t
t
T
T
kT2 2 |( r , z 0) kT1 1 |( r , z L1 )
z
z
T1t |( r , z 0) Te1
kT1
p1t |( r Re , z ) pe1 t1 gz
p1 t 0 |( r , z ) pe1 t1 gz
kh p1t
q1
|
r ( r Rci , z ) 2 Rci L1
kv p2 t
k p t
|( r , z 0) v 1 |( r , z L1 )
z
2
1 z
p1t |( r , z 0) pe1
Twb
Well
( t 0)
|( z ) Te1 gT z
Twb (t ) |( z 0) Te1
( t 0)
|( z ) pe1 m gz
pwb (t ) |( z 0) pe1
|( z ) 0
uwb (t ) |( z 0) 0
pwb
uwb
( t 0)
Layer j
Initial Conditions
75
Boundary Conditions
T jt |( r Re , z ) Te j gT z
t 0
Tj
T j t
kT j
|( r , z ) Te j gT z
t
|( r Rci , z ) U wbf
T j t Tajt
j
kT j1
T j 1
|( r , z 0) kT j
T j
|( r , z L j )
TN |( r , z LN ) TeN gT LN
Reservoir
pjt |( r Re , z ) pe j t j gz
pjt 0 |( r , z ) pe j t j gz
t
qj
kh p j
|( r Rci , z )
r
2 Rci L j
t
t
kv p j 1
k p j
|( r , z 0) v
|( r , z L j )
j 1 z
j z
wC
p t
j 1
Ta j
( t 0)
|( z ) Te j gT z
Tt j
( t 0)
|( z ) Te j gT z
Ta j
(t )
Tt j1 ( t ) |( z j1 0) wC p Tt j t |( z j L j )
|( z j 0) T j |( r Rci , z j 0)
Tt j1 |( z j1 0)
t
tj
wC
p t
j
Tt j |( z j L j ) wC p
t
pa j ( t 0) |( z ) pe j m j gz
pt j (t 0) | z pe j m j gz
pa j ( t ) |( z j 0) p jt |( r Rci , z j 0)
ua j (t 0) | z 0
ut j (t 0) | z 0
ua j (t ) |( z j 0) 0
aj
wC wC
p t
j
Well
aj
or
t
|( z j L j ) T t
p a
j
ptjt1 |( z j1 0) ptjt |( z j L j ) or
ptjt1 |( z j1 0) pa tj |( z j L j ) p t
wt
t
j 1
|( z j 1 0) w t |( z j L j )
j
4.2
Heat transfer coefficients
A well-reservoir segment is shown in Figure 4.9 with its radial layers:
tubing, annulus, cement and formation. The heat transfer mechanisms among
them can be described by conduction, convection and radiation. The overall heat
transfer coefficients for the radial geometry are based on the ones proposed by
Muradov (Muradov and Davies 2008). They are divided in tubing-annulus and
annulus formation heat transfer coefficients (Appendix C).
76
4.2.1
Heat transfer coefficient between annulus and tubing
Convection, conduction and radiation due to the fluids in the annulus (the
latter was neglected here) and tubing as well as the convection and conduction
through the tubing wall are taken into account:
U ta
Rti
1
ht Ktubing
1
R R
ln to ti
Rti Rto ha
(4.9)
U af
R
1
ci
ha K casing
1
R
R
Rci
ln co
ln wb
Rci K cement Rco
(4.10)
4.3
Inflow control valve
An inflow control valve is a downhole valve which the basic mechanical
project is similar to a sliding sleeve valve but with remote control. It reminds a
cage style trim control valve which provides a balanced plug and a sliding ringtype seal. There are two kinds of flow area profiles available: oblong (continuous
flow area) or circular (discrete flow area) as shown in Figure 4.10. The flow area
profiles can be exposed to the flow completely (closed-open) or partially (multiposition or continuous). In the multi-position case, there are typically 2 to 6
intermediary positions with different areas. Both uses equal percentage as flow
control strategy. Due to an industry standard the max valve area (full open) is
equal to an equivalent tubing area, e. g. a 3 valve has its full open area equals
77
to a 3 tubing area (for same tubing weight). It should be noted that the valve
flow area profile always depends on the application, in other words, on the fluids
involved, pressure, temperature and expected flow rate. A simplified approach for
ICV flow area design based on nodal analysis and IPR (Inflow Performance
Relationship) can found in (Konopczynski and Ajayi 2004).
The flow control valve model used in this development is achieved using
Bernoulli and mass balance equations for an ideal incompressible fluid using the
geometry shown in Figure 4.11:
78
A2
wideal
2p
2
(4.11)
A2
1
A1
We obtain the approximation equation for the real incompressible fluid flow
correcting for viscous losses and vena contracta through the valve discharge
coefficient (Cd):
Cd
wreal
wideal
(4.12)
wreal
Cd YFa
2
A2 2p
A
(4.13)
1 2
A
1
When the geometric factor is included Cd (4.13) becomes flow coefficient
(C):
wreal CYFa A2
(4.14)
2p
All factors are obtained experimentally and it is assumed Re > 10000 and
wreal CA2YFa
2p l
1 y
0.5
0.95
0.02
(4.15)
Both equations (4.14 and 4.15) are presented in (Oliveira, Passos et al.
2009) and they are based on ISO-5167 (Measurement of fluid flow by means of
pressure differential devices inserted in circular cross-section conduits running
full).
Hydraulic connection between the annulus and tubing is made by an ICV.
The flow area of the valve is the same for the tubing when it is in a fully opened
position as a standard industry design. The pressure in the tubing at the valve
position can be derived from (4.14) neglecting compression and thermal
expansion effects:
pt ,d pa ,u
p
79
a A12ua 2
2Cd 2 A2 2
a A12ua 2
(4.16)
2Cd 2 A2 2
The annulus density and velocity used in this equation can be modeled as
single phase or by averaging over volume fraction for two-phase (liquid/liquid).
Two phase (gas/liquid) can be derived from (4.15) using phase fraction from
(Ghajar and Tang 2010) but not neglecting compression and thermal expansion
effects.
The temperature change in the ICV can be modeled through the isenthalpic
model:
H
1 T dp 0
H
dH
dp C p dT
dT
T p
p T
T 1 dp K dp
dT
(4.17)
JT
C p
From (4.18) the average two-phase temperature change in the valve can be
calculated as:
t ,d
Ta ,u
wC K p
wC
p
pa ,u K JT pt ,d pa ,u
JT
t ,d
T K JT p
(4.18)
wC
p t ,u
Tt ,u wC p
wC
T
wC
p t ,u
a ,u
a ,u
(4.19)
p a ,u
The equation (4.19) is used as boundary condition for layer j when layer j-1
is equipped with an ICV. It is assumed that there is no phase change within the
valve trim and only liquids or slighted compressible fluids are flowing through at
below critical velocities.
4.4
Numerical Solution
Variational equations are derived from the differential equations of the wellreservoir coupled model and then discretized in time by implicit Euler and in
80
4.4.1
Variational Formulation
The well-reservoir equations are solved for p and T in the reservoir and for
pa, pt, ua, ut and Ta, Tt (annulus and tubing of the well). In order to assure the
mathematical consistence of the formulation, we need to state the regularity
properties of these variables. Let us say that the pressure and temperature fields
of the reservoir and the velocity, pressure and temperature ones of the well are
solution to the variational problem in the domain = r x w which we define
belonging to the following solution subspace:
:
{( p, T , ua , ut , pa , pt , Ta , Tt ) | p r ; T r ; ua , ut
; pa , pt
; Ta , Tt
Where:
: { p
: {T
(r )}
: {uw
(w ) | uw |w uw}
: { pw
(w )}
: {Tw
(r ) | p |r pr }
Where
k
(w )}
f
k f
() f | f L2 (), L2 ();...;
L2 ()
x
x
L2 () f : f d
81
Ta ,Tt
such that p r , T
a , ut )
r ,(u
w,
w ,( pa ,
p a , p t w , and
pt )
and
T ,T
a
subjected to:
In reservoir:
ct p
r
p
dv p t k p t gz dv
t
r
p k p gz nda 0
t
p dv c T p p C T p dv
C T Tt dv TT
t
t
p t
pf
C p TT t k p t gz dv
r
T k T dv
1 T p (t k (p t gz ))Tdv
T
T kT T nda 0
r
In well annulus:
Aa
( yu)a
dz 2 Rci u a ( yu)I dz 0
z
w
u 2 2 f u2
pa
dz p a a a Fa a a a g sin dz 0
z
Dha
z
p a
Aa
( yuC
T
) T a dz Ta (2 Rci (1 )U af (T f Ta ) 2 RtiU ta (Ta Tt ))dz
z
w
p a a
ag R
s
w
ao a
In well tubing:
p
a dz 0
z
Ta
are
yu t
ut
p t
u 2 2 f u2
pt
dz p t t t Fa t t t g sin dz 0
z
Dt
z
At
yuC T
At
p t
At
82
dz 0
Tt
dz 2 RtiU ta (Ta Tt ) At yu t g sin dz
z
w
tg R
s
yuC k ( yuC ) H
T
p JT t
p t
lat , go
t
to pt
w
Tt
p
t dz 0
z
83
N 1
,
N
hi h1
h
Where
(4.21)
h1 and hN are the first and the last element size respectively.
Rr p ct p
h
Rr T C p
c f
p
p p
p p
dvh t k r
k z t gsin dvh
h
t
r r
z z
p
p
kz p t gsin dah
kr p
r
z
T
p
dvh t TT
dvh
t
t
h
p
T p C T t dv
f
pf
T p
T p
C p T t kr
kz
t gsin dvh
h
z z
r r
84
p 2
p p
1 T T t kr kz
gsin
dvh
h
r
T T
T T
T
T
kTr
kTz
dvh kTr T
kTz T
dah
h
h
Rwua Aa ua
yu a
z
Rw pa p a
h
RwTa Aa
u 2 2 f u 2
pa
dzh p a a a Fa a a a gsin dzh
h
z
Dha
z
yuC
p a
T
Ta a dzh
z
ag R
pa
s
Aa Ta yuC p k JT
yuC p hlat , go
a
a
h
z
p
ao a
p
a
z
Ta
dzh
Rwut At ut
h
Rw pt p t
h
RwTt At
yu t
dzh
u 2 2 f u 2
pt
dzh p t t t Fa t t t gsin dzh
h
z
Dt
z
yuC T Tz dz
t
p t
tg R
p
s
At Tt yuC p k JT t yuC p hlat , go
t z
t
h
to pt
Tt
p
t
z
dzh
(4.22)
Limited to the domain h.
85
f
f k 1 f k
(4.23)
g k, f
g t , f k 1 f k f k 1 g t , f k 1 t
t
t
This equation shows that the system of equations must be solved in order to
get the values of the unknowns in the later time f k 1 . The implicit time fully
coupled system from the well-reservoir equations is solved using Newtons
method (Kelley 2003). It is a powerful iterative technique despite of its small
radius of convergence which requires a good initial guess. A simple linear
extrapolation was chosen to achieve this requirement:
c k 1 2c k c k 1
(4.24)
R c 0,
(4.25)
is solved from its associated Jacobian matrix and a residue vector through a
sequence of iterations of the linear system:
J ck ck R ck
c k 1 c k c k
(4.26)
Where
Ji, j c k
Ri
,
c k j
is the Jacobian matrix elements and tol is the required tolerance for the
chosen stopping method. Due to its flexibility a numerical Jacobian matrix was
used instead of an analytic one even at the expense of performance and the
possibility of a small numerical error. The numerical Jacobian is computed using
the following steps:
86
R c1
for j from 1 to # columns do
cp j c j
J # rows , j c
R cp j R c j
cp j c j
end for
The resulting sparse linear system is solved using a preconditioned GMRES
iterative algorithm. Approximate minimum degree reordering was chosen as the
preconditioner.
5
Validation and Case Studies
5.1
Segment Validation
The models for one reservoir layer and one well segment were validated
against a commercial multiphysics modeling and simulation tool (Comsol TM) and
a dynamic multiphase flow simulator (OlgaTM) independently and without valves.
Afterwards a fully coupled simulation for the well-reservoir layer/segment is
compared with Figure 4.5. The configuration used in the model for one
layer/segment is shown in Figure 5.1. It is used in the validation procedure
(without ICVs) and case studies (with ICVs).
88
5.1.1
Vertical Well
The well segment experiencing steady state flow of 500 bpd of oil
production in the tubing and in the annulus at the top of the segment is simulated
for pressure and temperature. The simulation results are presented as: pa, pt, Ta
and Tt profiles which are validated against OlgaTM using data reported in
(Muradov 2010) and reproduced as square dots in the Figure 5.2. As expected,
pressure profiles are dominated by hydraulic head and a uniform influx from the
annulus. Temperature profiles clearly show differences for tubing and annulus and
a good agreement with the reference data. The quality measures is reported as the
Euclidean Norm of the relative error which is 8.9 10-4 for Ta and 1.9 10-3 for Tt .
Figure 5.2 - Temperature and pressure in a vertical well with 500 bpd of oil production in
the tubing and in the annulus.
The same simulation is repeated for 500 bpd of oil/water in the tubing and
in the annulus at the top of the segment. The results are validated against Olga TM
giving the Euclidean Norm of the relative error 1.6 10 -4 for Ta and 6.2 10-3 for Tt
for this case.
89
Figure 5.3 - Temperature and pressure in a vertical well with 500 bpd of oil/water
production in the tubing and in the annulus.
5.1.2
Cylindrical shaped reservoir
The cylindrical shaped reservoir is simulated for oil and oil/water up to 48
hours of production time after a step rate change of 2000 [bpd] of the total flow
done at the surface. During the simulations 10% of leak is allowed in z direction
and gravitational effect and geothermal gradient are considered. The parameters
used for the simulation are presented in Table 5.1 except by reservoir thickness:
150 [ft], reference temperature: 154.8 [F], equivalent radius: 1000 [ft], and
saturation: So=1 and So=0.5 respectively to run the simulations. A commercial
multiphysics modeling and simulation tool (ComsolTM) is used as a reference tool
for: mesh size and time step tuning. The best match after the tuning procedure is
shown in Figures 5.3 and 5.4 for both cases. It was achieved using a minimum
time step of 6 minutes and mesh size 45/45 with density variation over r direction
calculated by equation 4.21 using h45/h1=1/15. The quality measure was again
based on the Euclidean norm of the relative error which were for oil production:
5.6 10-3 for p(r=Rwb,t), 5.3 10-5 for p(r,t=48h), 1.1 10-3 for T(r=Rwb,t), and 1.5 10-5
for T(r,t=48h).
90
In the case of oil/water production the Euclidean norm of the relative error
were: 3.4 10-4 for p(r=Rwb,t), 7.5 10-3 for p(r,t=24h), 4.2 10-4 for T(r=Rwb,t), and
5.9 10-4 for T(r,t=24h).
91
Figure 5.5 - Pressure and temperature transients for 2000 [bpd] and 1000 [bpd].
5.1.3
Case Studies
The model derived captures transient behavior in a multilayer vertical well
equipped with intelligent completion. Basically, it is a reasonable example of an
IW application. In an intelligent field scenario, real-time analysis tools are used
for reservoir management leading to production optimization, better reservoir
characterization and flow allocation. Through the use of appropriate tools it is
possible, in theory, to do pressure and temperature transient analysis without
intervention or well shut-in. Automatic well tests were applied in a valid set of
continuously measured pressure during unplanned shut-in periods proving that it
is possible (Olsen and Nordtvedt 2006).
Well-testing uses generally pressure transient responses due to production
rate changes to gather information about reservoir properties. Temperature
transients can also be used for this purposes as indicated by (Sui, EhligEconomides et al. 2010) for conventional wells and by (Muradov and Davies
2010) for horizontal IWs. Therefore, it is possible in principle to have the same
92
5.1.3.1
Base Case
The three zones are producing 1000 bpd of oil each with uniform flow
distribution and the well choked. A transient of 3000 bpd is then created by
opening the surface choke and all layers have to adapt to this new total rate of
6000 bpd. In addition, as all reservoir properties are equivalent at the different
layers, the same behavior is expected towards to the new constant total flow rate.
All ICVs are kept fully open during the whole numerical experiment. The result is
93
shown in Figures 5.7 to 5.11. The temperature shifts show the expansion and
Joule-Thomson effects in all zones. The valves and fluid mixture in the upper
layers are also present. It should be noted that the presence of annulus and tubing
in the model cannot be neglected for appropriate understanding of the temperature
behavior.
94
5.1.3.2
Two-Phase at Middle Zone
The second case study has the upper and lower zones producing 1000 bpd of
oil and the middle zone producing 50% of water and 50% of oil which is also
producing 1000 bpd. A transient of 3000 bpd is created by opening the surface
choke and the total flowrate is kept constant as in the base case. The new fluid in
the annulus of the middle layer has a stronger heat capacity with a weaker JouleThomson effect when compared with the fluid in the tubing. The heat cannot be
95
96
5.1.3.3
Different Permeability at Middle Zone
The third case study has a total flowrate of 3000 bpd but now the middle
zone has a higher permeability. All zones are producing oil when a 3000 bpd
transient is created by opening the surface choke. As in the base case the total
flowrate of 6000 bpd is kept constant. The temperature profiles and transient
presented in Figures 5.16 to 5.18 show the effect of a higher flowrate at the
middle layer resulting in a similar behavior of the previous case but weaker this
97
98
5.1.3.4
Zonal Transient
The last case study has a total flowrate of 5000 bpd which is kept constant
during the experiment. Upper and lower zones are producing 1000 bpd and the
middle one 3000 bpd of oil. A zonal transient of 2000 bpd is created at the middle
zone by changing the ICV flow area. The temperature profiles and transient
reproduce the behavior found in a real well (Valiullin, Ramazanov et al. 2009), as
presented in Figure 4.2. The temperature transient becomes weaker as the upper
99
layers are approached and this effect is even stronger for vertical wells (Figure
5.19 to 5.22).
100
101
6
Data Analysis
A decade after the first IW had been installed, an integrated modeling and
optimization framework, shown in Figure 6.1, was proposed by Brouwer
(Brouwer, Ndal et al. 2004). This paper describes how established concepts from
control engineering and meteorology can be used for production and reservoir
management. Since then, research and development were initiated in areas of
reduced order modeling, optimization, monitoring systems, data analysis and
assimilation.
Data analysis can be defined as the procedure used to transform data into
knowledge along the value chain path for making decisions as shown in the
pyramid chart of the Figure 6.2. Data analysis and assimilation aims to provide
the necessary information for model calibration and uncertainty minimization to
avoid unreliable decision making. Typically software tools commercially
available support all data analysis procedures: data acquisition, pre-processing,
post-processing and interpretation without integrated modeling and optimization
in mind.
103
Data acquisition and pre-processing deals with issues such as: sampling,
denoising, outlier removal, compression and time synchronization. As mentioned
earlier, pre-processing is done by application, e.g. gas lift optimization and IW
have its own application tools. The integrated modeling and optimization
framework aims to use appropriate pre-processing to improve the quality of the
data analysis and assimilation procedure. One way of doing so is to perform in
one place the pre-processing of the real-time data that is used by the different
applications. This implies that the data from the production system are viewed as
a single data set. Denoising, outlier removal and missing data can then be treated
in a traceable manner, assuring the same level of quality control to all
measurements. Further, time synchronization of the measured values will be more
robust. This can be used to improve predictions of mutual influences of processes
with different time scales rather than todays practice of only using it to adjust the
correct sample rate for each process (decimation) which is essential for field scale
analysis (Aggrey, Davies et al. 2007). As can be seen in Figure 6.3 the centralized
pre-processing can potentially improve the quality of the post-processing
predictions. Although this approach has not yet been fully explored, a successful
application of it can be found in offshore operations and maintenance
(Friedemann, Varma et al. 2008).
104
105
106
Compression/decimation;
Time synchronization;
Feature extraction.
They are used due to necessity of reliable data, free of outliers, missing data
and measurement noise to improve the accuracy and effectiveness of the postprocessing and interpretation procedures. The additional procedures included refer
to the mentioned issues, especially in subsea applications, as bandwidth
limitation, time synchronization, data overload, and spatially distributed sensing.
Feature extraction was also included to capture influences among processes.
Denoising/missing data interpolation/outlier removal procedure is studied
by using RPCA (Robust Principal Component Analysis) via PCP (Principal
Component Pursuit) to approximately restore and separate the signal from noise
and outliers which is assumed sparse (Lin, Chen et al. 2010). Missing data are
allowed and the procedure is done simultaneously through a surrogate convex
optimization approach.
107
including
different
ones,
surveying
the
same
phenomena.
7
Pre-processing Theory Background
Real monitoring systems suffer of various types of data corruption. They
might be caused by installation, telemetry, sensors and aging, being classified as:
random, systematic, and gross errors. Specifically for downhole monitoring we
emphasize connection, coupling, and placement induced errors as the ones that
nothing can be done after installation without a costly intervention.
In this chapter, first we describe random, systematic, and gross errors, with
some real examples, and afterwards how to model them. Second, we present
common techniques used in downhole monitoring system to treat noise and
outliers introducing RPCA via PCP (Cands, Li et al. 2011) which is, roughly
speaking, a signal separation problem, as one of the pieces of our framework.
Finally, we present a feature extraction technique based on DM (Coifman and
Lafon 2006) and our full pre-processing framework.
7.1
Corrupted Measurement Data
Random, systematic and gross errors are the major types of errors in
measurement systems. As the name states the random errors are characterized by
unknown and unpredictable changes. They are assumed to be additive and
modeled using the equation 7.1:
yt ,noise yt r
(7.1)
Where:
r
N (0, 2 )
0.8
0.6
0.4
0.2
100
200
300
400
500
600
700
800
900
1000
109
110
Pressure
300
295
290
285
p [kgf/cm2]
280
275
270
265
260
255
250
200
400
600
Time [hours]
800
1000
1200
Figure 7.3 - Noise and drift of a real PDG (the green signal is the clean one).
I
1 1
1
ln
T T0 DL I 0
(7.2)
Where:
DL Differential Loss
I anti-stokes and stokes intensity ratio
I0 anti-stokes and stokes intensity ratio at reference temperature (T0)
Lets assume that reference temperature and differential loss have a
negligible impact on the temperature uncertainty. From the reference paper, the
noise of stokes and anti-stokes signals are modeled by the equation 7.3:
(7.3)
111
2
1 2 I AS 2
AS 2 S
Is
IS
2
(7.4)
The equation 7.4 shows that the temperature uncertainty varies with signal
intensity which is function of the position. As interpretation methods used for
distributed temperature sensing depend on temperature gradient they are directly
impacted. Therefore this systematic error should be taken into account to
minimize misinterpretations. It is worth mentioning that signal processing as trace
averaging and moving averaging within trace normally used to minimize noise is
not able to remove this kind of effect and it should be removed before any signal
processing be done. An example of this effect in a real distributed temperature
measurement can be seen in Figure 7.4. The temperature signal was gathered from
a DTS installed in a producer horizontal well (onshore) where we are interested in
the 505 m of the horizontal length starting at 854 m (MD). The lines black and
blue corresponds to the temperature signals after signal processing with and
without systematic error correction due to noise dependence with position.
Distributed Temperature
80
75
T [C]
70
65
60
55
50
200
400
600
800
1000
1200
1400
Depth [m]
Figure 7.4 - Temperature gradient profile error due to reading unit noise model in a real
temperature measurement (blue line without correction).
The last major type of error is the gross errors. They occur when the
instrument goes on complete failure and they are understood as errors associated
112
(7.5)
It should be noted that the outlier is assumed to be an additive error over the
original signal and the switching distribution is typically Bernoulli or Markov.
7.2
Denoising and Outlier Removal
All major types of measurement errors described in the last section can be
minimized through the use of filters. There are only 5 of them heavily used in
downhole monitoring systems signals: mean filter, median filter, maximumminimum filter, moving average filter and wavelet filter. In addition to them we
introduced a non-conventional RPCA via PCP as a new type of filter for
simultaneous denoising and multiple outlier removal technique.
7.2.1
Conventional Filters
In this section we describe the most used filters in downhole monitoring
systems especially for transient analysis applications. As mentioned earlier they
consist of 5 filters:
The mean filter is the simplest one and consists in smoothing the signal by
taking the average value of the last w measurements. This filter can be modeled by
the equation 7.6:
yt ,mean
1
yt , w ... yt
w 1
(7.6)
113
result is then compared with the clean signal through L2 norm of the relative error
giving 6.48x10-2.
70
60
50
40
30
20
10
0
20
40
60
80
100
The median filter works in a similar way of the mean one but instead of
taking the mean of the last w values it takes the median of them. This filter can be
modeled by the equation 7.7:
yt ,median
t , w
(7.7)
As before using the synthetic signal and noise, and with the same w, we
evaluate the performance of the median filter making a comparison between the
filtered and the clean signals. The result gives 1.84x10-1 (L2 norm of the relative
error).
70
60
50
40
30
20
10
0
20
40
60
80
100
114
yt ,max min
yt
yt yt ,min , yt ,max
yt ,min
yt yt ,min
yt ,max
yt yt ,max
(7.8)
20
40
60
80
100
yt ,mva
1
yt w ... yt
w 1
(7.9)
115
70
60
50
40
30
20
10
0
20
40
60
80
100
The wavelet filter has proven to be valuable for denoising in particular the
wavelet shrinkage method proposed by (Donoho and Johnstone 1994) which
provides min-max optimal solution. Today many wavelets threshold techniques
are available but we focus on those studied by (Olsen 2011) which is aligned with
this thesis goals.
Waveshrink method is based on the principle that noise has smaller wavelet
coefficients than the underlying signal. Therefore if we apply a threshold function
to penalize those coefficients, while preserving the signal, we eliminate the noise.
Basically it has three steps: wavelet transform the noisy signal, apply a shrinkage
operator to the resulting coefficients, and apply the inverse of the wavelet
transform to restore the signal. Mathematically it can be expressed as:
yt W 1 D W yt ,noise ,
(7.10)
Where:
W wavelet transform
D shrinkage operator
- Threshold
W-1 inverse wavelet transform
116
correct choice of the wavelet family and scaling function, the shrinkage operator
which can be hard, soft, or a function (linear, Firm, Garrote, SCAD Smoothly
Clipped Absolute Deviation, etc.), the threshold estimator, and the noise estimator
can be completely different. An extensive work by (Olsen 2011) indicated
Spline39 wavelet, SCAD shrinkage rule, RED (Regression Estimated Decline)
threshold estimator, and RMAD (Robust Median Absolute Deviation) noise
estimator as recommended choices for pressure transient analysis.
The performance of the Waveshrink method is illustrated using the same
signal and noise as before for the hard and soft shrinkage operator (Figure 7.9)
which can be mathematically expressed as:
t ,noise
0.5
0.5
-0.5
-0.5
-1
20
40
60
80
100
-1
20
(7.10)
40
60
80
100
117
70
60
50
40
30
20
10
0
20
40
60
80
100
Figure 7.10 - Wavelet filter performance for the hard threshold (red is the clean
signal).
70
60
50
40
30
20
10
0
20
40
60
80
100
Figure 7.11 - Wavelet filter performance for the soft threshold (red is the clean
signal).
7.2.2
Robust PCA via Principal Component Pursuit
In order to understand this robust PCA (Principal Component Analysis)
algorithm, we firstly describe the PCA problem as a low rank minimization
problem: given a matrix D, where columns are the observation vectors corrupted
by Gaussian noise, find the optimal low-rank approximation in l2-sense.
Mathematically if we know r we can formulate the problem as:
min D L
L, N
2
F
s.t. rank L r
(7.10)
If we do not, as:
min rank ( A)
L, N
DL
2
F
(7.11)
118
Where
D=L+N;
L low rank signal;
N Gaussian noise.
The optimal solution for the former is given by: L UH V T where
min
L, N
L*
DL
(7.12)
1/ , 1/
S1/ 1/ , 1/
A recent work by (Cands, Li et al. 2011) showed that if we assume lowrank signal and sparse outliers, an exact recovery of the signal is expected with
high probability from arbitrary and completely unknown outliers corruption
patterns if a convex optimization problem called PCP (Principal Component
Pursuit) is solved. It is worth mentioning that this problem is different from the
119
mxn
(7.13)
L, N
Where
1
;
max(m, n)
The exact recovery is achieved if the following bounds are provided for L
and O (Cands, Li et al. 2011):
rank ( L)
r min m, n
log max m, n
, O 0 o mn
(7.14)
Where
- coherence of D 1 ;
o , r positive constants.
Notice that
1
is not a tuning parameter which means that
max(m, n)
under the assumptions works with high probability for recovering any low-rank,
incoherent matrix (Cands, Li et al. 2011).
Before we go further we present the relaxed version of the problem 7.13
(Zhou, Li et al. 2010) which includes noise and can also deal with missing data:
min L * O 1 s.t. D ( L O)
L ,O
(7.15)
120
A closed form solution for this problem is not known and numerous
approaches are available to find the minimizer. An extensive comparison of the
solvers: Iterative Threshold, Accelerated Proximal Gradient, Augmented
Lagrange Multiplier, and Inexact Augmented Lagrange Multiplier can be found in
(Lin, Chen et al. 2010), leading to Inexact Augmented Lagrange Multiplier as the
faster solver for the same accuracy to the RPCA problem. The latter was selected
and used throughout this thesis. It worth mentioning an alternative approach
presented by (Becker, Cands et al. 2011) for this problem using TFOCS, a
general purpose first-order conic solver, which produces a comparable solution
regarding relative error and can be used to prototype other algorithms.
The augmented Lagrange multiplier method when applied to the problem
7.14 gives the following optimization problem:
min L * O 1 Y , D ( L O)
L ,O
D ( L O)
2
F
(7.16)
Where the third term enforces the relaxed constraint via the matrix of
Lagrange multipliers Y and fourth term makes the objective function strictly
convex, improving the convergence.
The inexact augmented Lagrange multiplier algorithm (Lin, Chen et al.
2010) which solves 7.16 is presented below:
121
Input: D,
Y0
D
; O0 0; 0 0; 1; k 0
J D
(U , S ,V ) svd ( D Ok k1Yk )
Lk 1 US 1 S V T
k
Ok 1 S 1 D Lk 1 k1Yk
k
Yk 1 Yk k D Lk 1 Ok 1
k 1 k
k k 1
end
Output: L, O
The bounds provided for this example are rank(L) = 5 and card(O)=20%. As
can be seen in Figure 7.15 at less than 30 iterations it has been converged to the
specified tolerance.
122
8600
8500
6
8400
Card(O)
Rank(L)
8300
8200
3
8100
8000
7900
7800
10
15
20
25
30
10
15
20
25
30
Iteration
Iteration
10
-1
-2
10
-3
||L-L
hat F
|| /||L||
10
10
10
10
-4
-5
-6
10
15
20
25
Iteration
30
123
7.3
Feature Extraction
A general multi-sensor feature extraction framework based on Diffusion
Maps (Coifman and Lafon 2006) is presented in this section. Two important
aspects of this method which were extensively studied in (Lafon 2004) are used:
density-invariant embedding and extension of a given embedding. In the approach
we assume that all sensors are measuring different features of the same physical
phenomenon. Since different features might have different characteristic scales,
they are chosen empirically during the training stage (training signal embedding).
New signal embedding extension is calculated and classified accordingly using KNN (k-nearest neighbor classifier). This completes the general idea of the
framework which is explained in the next two sub-sections.
124
7.3.1
Diffusion Maps
DM (Coifman and Lafon 2006) is a non-linear dimensionality reduction and
clustering technique suitable for non-uniformly distributed datasets. The technique
unifies the analysis of spectral clustering and spectral embedding algorithms
through a probabilistic framework. The general idea is introduced by a notion of
diffusion distance which is the distance of a random walk between two nodes on a
graph. Notice that this definition can reveal different local structures at different
times. The graph nodes are then non-linearly embedded on the eigenvector
coordinates of a normalized graph Laplacian converting the diffusion distance into
Euclidean distance in the embedded space. Therefore it gives a probabilistic
meaning for this distance which depends on the dataset density and geometry. DM
when compared with other techniques is robust to noise and computationally
inexpensive for small and medium datasets.
7.3.1.1
Diffusion Maps Algorithm
Given a data set X xi i 1 , xi
n
D xi , x j
Where
D distance defined for the dataset, e.g., Euclidean distance;
characteristic scale (kernel bandwidth).
Afterwards a random walk (diffusion process) is induced by computing a
row-normalized Markov matrix (transition probability matrix) taking into account
the influence of the density which gives the new kernel:
W , ( xi , x j )
W ( xi , x j )
q ( x i ) q ( x j )
Where
q xi W xi , x j
j
125
Wsym ( xi , x j )
W , ( xi , x j )
v( x i )0.5 v( x j )0.5
Where
v , ( xi ) W , ( xi , x j )
j
Wsym l t x ltt l t x
1 0 1 ... l t
lt
1t
x x
l (t )
k 1
2t
k
t xi t x j
Where
t 1t 1 xi ,..., lt l xi for some t 0 is the DM of the dataset xi
T
7.3.1.2
Toy Example
A Trefoil Knot with Gaussian noise is used as a representative example of
the potential of the DM as a clustering algorithm. The data set X is threedimensional with 400 data points (Figure 7.18):
126
x sin(3t ) r
X y sin(t ) 2sin(2t ) r
z cos(t ) 2 cos(2t ) r
Where
0 t 2
N (0,0.05)
2
1
0
-1
-2
-3
2
1
0.5
-1
-2
-0.5
-1
127
1.5
0.5
-0.5
-1
-1.5
-1.5
-1
-0.5
0.5
1.5
7.3.2
Feature Extraction Framework
The aim of our feature extraction framework is to capture the underlying
and mutual influences of the production processes in the context of integrated
production framework. The multi-sensor is based on the spectral and density
invariant embedding of the DMs framework using Mahalanobis distance. It is
appropriate to multivariate analysis for the training stage and allows the DMs to
be computed simultaneously for the different signals. Each scale parameter i for
each required feature i of the signal is chosen based on the Singer Measure
(Coifman and Lafon 2006):
SM ( i ) e
D 2 xi x j
128
129
130
computing capacity. In addition, the data overload due to the increasing number of
sensors make this problem even worse as mentioned in Chapter 6. On the other
hand, data analysis might be impacted by these algorithms especially by the lossy
ones. The most frequent approach used is taking less number of measurements by
removing the ones with no significant information content. It is accomplished by
discarding measurements with assumed small variations from the time intervals.
This task is completed through a specified threshold and it is signal dependent.
Notice that data analysis algorithms which need equally spaced measurements as,
e.g., PCA and Waveshrink are impacted by this approach. We see this problem as
a change in the data acquisition by using compressive sensing (Cands, Romberg
et al. 2006) where the signal is acquired compressed and can be reconstructed
lossly, solving the problem. It is worth mentioning that in this case the denoising
and outlier removal proposed in this text should be different.
The following chapter describes the work by (Olsen 2011) regarding preprocessing techniques which is used as a reference work for validation purposes.
Pressure and temperature data generated by the non-isothermal dynamic IWreservoir model developed are corrupted by noise and outliers and used as the
input for both approaches. Only pressure is really compared for denoising/outlier
removal and transient identification. Transient temperature data is not available in
the literature or in Olsens work. Case studies complete the chapter showing the
results.
8
Pre-processing Framework Validation and Case Studies
The Chapter 8 is divided in: signal generation, validation and case studies.
Signal generation presents the synthetic pressure and temperature signals
generated from the noise/outlier model used and the one developed in Chapter 4.
Validation is accomplished through the comparison with the results from the work
by (Olsen 2011) which has an extensive study of automatic filtering and transient
identification on real-time transient pressure signals. Our feature extraction
approach was applied for transient identification but as described earlier it is
general and can be used for different applications. Finally, case studies evaluate
denoising/outlier removal and transient identification on transient temperature
signals, and transient identification on both. It is worth mentioning that
temperature transient analysis is limited and considered in early stage of
development (Muradov and Davies 2011). In addition, there are almost none real
data available in the literature for IW applications.
8.1
Noise, Outlier, and Signals
Transient signals in real wells might be generated due to shut-ins or zonal
shut-ins in case of IWs. They can be used in context real-time production
monitoring as an opportunity well-test. Automatic filtering and transient detection
for this application provide the appropriate pre-processing to be used for well-test
analysis enabling more frequent tests and consequently more information to
optimize reservoir management.
Real pressure and temperature signals in a dual zone injector IW and its
producer pair can be seen in the Figures 8.1 and 8.2 respectively. Notice that only
the lower zone was closed creating a transient. Even with the graph scale used it is
possible to see the noise and the outliers. It is a subsea well with a water depth of
1168 m and downhole sensors installed at 2644.1 m and 2656.5 m (TVD).
132
Pressure
340
320
300
p [kgf/cm2]
280
260
240
220
200
180
200
400
600
800
1000
Time [hours]
1200
1400
1600
1800
Figure 8.1 - Injector-producer pair pressure signal (dark blue and light blue are lower and
upper zones respectively; black producer well).
Temperature
80
70
60
T [C]
50
40
30
20
10
200
400
600
800
1000
Time [hours]
1200
1400
1600
1800
Figure 8.2 - Injector-producer pair temperature signal (red and green are lower
and upper zones respectively; pink producer well).
The pressure and temperature signals used in the validation and case studies
in this thesis were generated synthetically from the Chapter 4 model. It is assumed
that those signals are representative of the ones measured in an IW. We choose to
generate synthetic signal for two main reasons: the lack of transient pressure and
temperature signals available in the literature for IWs and the complete control
over the true noise level used in these signals. To be fair with the comparisons
made against Olsens work we do not generate rate changes with less than 10
measurements, the time interval is kept under 3-360 seconds, noise distribution is
133
normal with mean zero and variance within 1 to 10, and the outliers distribution is
uniform with level within 5 to 100 times the noise level.
Three different synthetic sets of pressure and temperature signals were
generated as representative signals from real IWs. They are corrupted by noise
and outliers and the same amount of both are added in each set of pressure and
temperature signals as they come from the same sensor. All three sets were
generated by 10 rate changes at random within 2000 to 500 bpd. Each signal has
sample rate of 6 minutes and 27000 measurements. Outliers were chosen from
uniform distribution with level varying from 10 to 100 times of the noise level and
number within 0.3% to 5% of the total number of measurements and noise is
assumed to be additive white Gaussian noise with variance within 1 to 6. Partial
pressure and temperature of signals 1, 2 and 3 can be seen in Figure 8.3.
The quality of the filtering is measured by the RMSE (Root Mean Squared
Error) divided by the noise level estimate (standard deviation). This metric was
introduced by (Olsen 2011) to compare different de-noised signals with different
noise levels. Outlier removal and transient identification performance are
measured by the percentage of success in the removal and in the identification
after the outlier removal/denoising procedures. It is considered an error when the
identification occurs after three measurements from where it actually happens.
RMSE
2
1
yt yt ,clean
N t
(8.1)
134
Where
noise standard deviation.
8.2
Denoising/Outlier Removal and Transient Identification
The wavelet analysis to denoise pressure signals and detect transients was
first introduced by (Kikani and He 1998). Since then, a lot of work have been
done such as, e.g., the work by (Athichanagorn, Horne et al. 1999) which
introduced the seven step procedure for processing data from downhole sensors,
the work by (Ribeiro, Pires et al. 2008) which showed that there is no universal
wavelet transform to be applied in any dataset due to strong dependence on
decomposition level and wavelet type, and the work by (Olsen 2011) who did an
extensive study on automatic outlier removal, denoising, data reduction, and
transient identification on pressure and rate signal for well test analysis. The latter
addressed the problem presented by (Ribeiro, Pires et al. 2008) and was used as
reference work to ours for validation purposes.
8.2.1
Denoising/Outlier Removal
The work by (Olsen 2011) developed a method to detect and remove
multiple outliers without requirements regarding the signal considered. The
outliers are detected and removed by a median filter with noise estimation from
the Waveshrink method. The latter is used afterwards for conventional denoising
purposes. In the opinion of the author, this method addresses the two main
drawbacks found in wavelet analysis in the petroleum industry literature: single
outlier removal and dependence upon the signal considered. The idea is
schematically illustrated in Figure 8.5.
The median filter considered is a moving window filter with a window size
of n measurements which is usually small compared to signal measurements. If
135
the value in the middle of the window is higher than the median plus a given
threshold, it will be considered an outlier and therefore removed. The threshold is
calculated using the noise estimator proposed in (Donoho and Johnstone 1994) as
being the MAD (Median Absolute Deviation) divided by 0.6745 which is
evaluated at the first decomposition level in the wavelet transform but using only
the absolute values of the non-zero detail coefficients. In Olsens work it is called
a RMAD divided by 0.6745 to make it different from the original.
Mathematically, the median filter can be defined as:
y
yt ,median t
median( w) yt 3
median( w) yt 3
Where:
= RMAD/0.6745;
w - window size.
Notice that the maximum window size recommended is 19 measurements
w 1
from which 9 outliers can be removed
and it is empirically defined in the
2
scope of the well-test analysis (Olsen 2011).
The Waveshrink method was selected as the denoising filter for pressure
and rate signals since wavelet analysis was introduced to well-test analysis by
(Kikani and He 1998). However, only with the work by (Pico, Aguiar et al. 2009)
and (Olsen 2011) an extensive study was done regarding wavelet selection,
shrinkage rule, threshold estimator, noise estimator, and primary resolution level
to choose. The best selection for the Waveshrink method in Olsens work were
based on average results of the smallest error in the root-mean-squared sense
between the clean signal and the noisy signal for more than 500 synthetic pressure
or rate signals. The characteristics of those signals were chosen as representative
ones from real-time pressure and rate signals gathered from North Sea wells.
The Waveshrink method (Donoho and Johnstone 1994) is based on the
principle of shrinking the wavelet coefficients towards zero in order to remove
noise. The main idea is to transform signal yt to the wavelet domain, shrink its
empirical wavelet coefficients towards zero, and transform the shrunk coefficients
back to the signal domain. Mathematically, it can be stated as:
yt W 1 D W yt ,noise ,
136
Where:
W wavelet transform;
D shrinkage operator;
threshold function;
W-1 inverse wavelet transform.
The shrinking operator might be hard, soft or a function and operates on the
wavelet coefficients through a threshold estimator which is function of the noise
and the resolution level considered. In Olsens work the best options for pressure
and rate signals in the scope of the well test analysis are:
Bi-orthogonal spline39 wavelet for pressure (Figure 8.6 and Table 8.1):
3
2
1
0
-1
-2
1.5
1
0.5
0
-0.5
-1
hN
hN
3
128
3
64
1
8
19
64
45
64
19
64
1
8
3
64
3
128
1
4
1
2
1
4
137
hN
1
80
3
8
3
8
1
80
1
2
1
2
7
32
11
16
11
16
7
32
1
80
3
8
3
8
1
80
hN
1
32
gN
gN
1
32
sgn d j max d j ,0
d j 2
2 d j 3.7
d j 3.7
dj
Where
Threshold function: 0.51 j 3.25 , j 1,...,6 (j is the resolution level);
Noise estimation:
RMAD
;
0.6745
138
8.2.2
Transient Detection
In the literature there are a variety of transient detection methods. The
majority of them are based on pattern recognition, regression, threshold or wavelet
methods. The transient detection presented in this section describes the approach
used by (Olsen 2011) for automatic pressure transient detection in the scope of
well-test analysis. It is based on a mask-matching pattern recognition approach
performed right after outlier removal and denoising procedures (Figure 8.7).
139
Figure 8.8 - Denoising/outlier removal and transient identification results using pressure
signal.
Notice that our approach only uses the noise estimate and the assumption of
sparse outliers and low rank signal to denoise and outlier removal. Even for 1D
signal it is possible adequately use the technique by using time slices of the signal
140
8.4
Case Studies
We divided the case studies in two types. The first one shows that the
temperature signal can be used in the same way as the pressure signal in transient
detection. The other concludes that denoising/outlier removal and feature
extraction methods are appropriate for multidimensional signals of multiple
sensors as can be found in IW applications. Note that the great advantages of the
proposed methods are the possibility of treating nonuniform data and the no need
of a previous knowledge of the signal but the noise level. The computational
performance of these methods is reasonable and very comparable to the reference
ones. However, a more detailed study is needed, mainly for medium and high
dimensions data.
8.4.1
Denoising and Outlier Removal of Transient Temperature Signals
The goal of this case study is showing that isolated temperature signal can
be used in the correct transient detection. The temperature signals used are the
same as the ones described in section 8.1. They are first filtered and afterwards
used in the transient detection. The filter uses only the noise level estimator, the
feature extraction method, which was trained with temperature transient signals,
and the K-NN classifier that required order 6. The collected results are shown in
Table 8.4.
141
157
156
Temperature [degF]
155
154
153
152
151
150
100
200
300
400
500
Time [h]
8.4.2
Transient Detection through Pressure and Temperature Signals
Transient detection was tested with the three synthetic temperature signals
described in section 8.1. They were filtered as in the previous study case and the
results can be seen in Table 8.5. If we compare the results with the isolated
pressure and temperature results we can see an improvement in the percentage of
correct detection due to the additional dimension, in other words, the use of the
pressure signal in the detection process. Note that we can include the spatial
dimensional when distributed and quasi-distributed sensors are used. In this case,
nonuniform spatial distribution of sensors and nonuniform sample rate are not a
problem for the proposed algorithms. The end results for this case can be seen in
Table 8.5.
142
9
Conclusions and Future Works
Integrated production optimization calls for revisiting the way we deal with
the multiple pieces of the petroleum production puzzle. The system view
perspective should lead to better production system behavior prediction and fast
response to unexpected events. IW and monitoring systems in this perspective
might improve both if an adequately use of the technology available is performed.
In this thesis we focused on modeling and data analysis pre-processing for well
test applications as one example IW monitoring systems in the context of
integrated modeling and optimization framework. We address the problem of IW
monitoring system design, a fully coupled non-isothermal dynamic IW-reservoir
model, and partially the data analysis pre-processing framework through
denoising/outlier removal and feature extraction techniques. In the following
sections we summarize our achievements and the work which still to be done.
9.1
IW Monitoring Systems Design
A method to design IW monitoring systems for the integrated modeling and
optimization framework has been developed. It was applied to an IW application
using reactive control strategy and artificial lift as a representative example. The
main objective was to introduce interpretation model uncertainty as constrain to
adequately specify a fit for purpose IW monitoring system. A detailed quantitative
study for different applications is certainly a work to be done.
9.2
Fully Coupled Non-Isothermal Dynamic IW-Reservoir Model
A fully coupled dynamic non-isothermal two-phase well-reservoir
numerical model for vertical IWs and multilayer reservoirs in the scope of well
test analysis has been developed. The model was successfully applied to multizone temperature transient analysis of well total and zonal rate change. The
numerical studies showed that transient temperature due to zonal rate change
without closing is achievable in a multilayered reservoir. Thermal conduction and
crossflow between layers were included in the model but their impact was not
independently presented in case studies. Even though the heat transfer coefficients
144
allow the use of gas-liquid mixtures in the drift-flux model it was not completely
tested. Wellbore storage effect was neglected in our model since our study is
limited to the reservoir drainage area but real data provided by (Valiullin,
Ramazanov et al. 2009) showed that it can hide the early time effect of the
transient temperature. Inclusion of time dependent equations for the IW will bring
better understanding of fast transients but it is also a future work.
9.3
Data Analysis Pre-Processing
A new data analysis approach to IW applications with centralized preprocessing has been developed. Denoising and feature extraction techniques were
applied to the transient pressure generated by the model developed in this thesis
and compared with a reference study by (Olsen 2011). Transient temperature was
introduced in this context showing similar results regarding denoising/outlier
removal and transient identification. The techniques applied showed satisfactory
results for multi-sensor applications. As general techniques they can be used with
different applications to confirm their applicability in the centralized preprocessing approach. Data reduction/compression and time synchronization preprocessing procedures were presented but not evaluated. In addition,
computational load should be evaluated for different time scale applications and
data sizes but they are all considered a future work.
145
Bibliography
Aggrey, G. (2007). A Study of Intelligent Oil and Gas Fields' Real Time
Optimisation and its Value Quantification. PhD, Heriot-Watt University.
Aggrey, G. H., D. R. Davies, et al. (2006). Data Richness and Reliability in
Smart-Field Management - Is There Value? SPE Annual Technical
Conference and Exhibition. San Antonio, Texas, USA, Society of
Petroleum Engineers.
Aggrey, G. H., D. R. Davies, et al. (2007). A Novel Approach of Detecting Water
Influx Time in Multi-Zone and MultilateralCompletions using Real-Time
Downhole Pressure Data. SPE Middle East Oil and Gas Show and
Conference. Kingdom of Bahrain, Society of Petroleum Engineers.
Ajayi, A. A., D. Mathieson, et al. (2005). An Innovative Way of Integrating
Reliability of Intelligent Well Completion System with Reservoir
Modelling. Offshore Europe. Aberdeen, United Kingdom, Society of
Petroleum Engineers.
Al-Hadhrami, A. K., L. Elliot, et al. (2003). "A New Model for Viscous
Dissipation in Porous Media Across a Range of Permeability Values."
Transport in Porous Media 53: 6.
Al-Khalifah, A.-J. A., K. Aziz, et al. (1987). A New Approach to Multiphase Well
Test Analysis. SPE Annual Technical Conference and Exhibition. Dallas,
Texas, 1987 Copyright 1987, Society of Petroleum Engineers.
Alberts, G., S. Belfroid, et al. (2007). An Investigation Into the Need of a
Dynamic Coupled Well-Reservoir Simulator. SPE Annual Technical
Conference and Exhibition. Anaheim, California, U.S.A., Society of
Petroleum Engineers.
Alves, I. N., F. J. S. Alhanati, et al. (1992). "A Unified Model for Predicting
Flowing Temperature Distribution. in Wellbores and Pipelines." SPE
Production Engineering 7(4): 363-367.
Athichanagorn, S., R. N. Horne, et al. (1999). Processing and Interpretation of
Long-term Data from Permanent Downhole Pressure Gauges. SPE Annual
Technical Conference and Exhibition. Houston, Texas, Society of
Petroleum Engineers, Inc.
Bear, J. (1988). Dynamics of fluids in porous media, Courier Dover Publications.
Becker, S. R., E. J. Cands, et al. (2011). "Templates for convex cone problems
with applications to sparse signal recovery." Mathematical Programming
Computation 3(3): 165-218.
Birchenko, V. M., V. Demyanov, et al. (2008). Impact of Reservoir Uncertainty
on Selection of Advanced Completion Type. SPE Annual Technical
Conference and Exhibition. Denver, Colorado, USA, Society of Petroleum
Engineers.
Bird, J. M. and N. Frost (1966). "Formation Productivity Evaluation From
Temperature Logs." SPE Journal of Petroleum Technology 18(3): 301305.
Brezzi, F. and M. Fortin (1991). Mixed and Hydrid Finite Element Methods,
Springer-Verlag Berlin and Heidelberg GmbH & Co. K.
Brouwer, D. R., G. Ndal, et al. (2004). Improved Reservoir Management
Through Optimal Control and Continuous Model Updating. SPE Annual
146
147
148
149
150
151
Appendix A
Fanning Friction Factor
Fanning friction factor is a dimensionless number that represents a
dimensionless shear stress at the pipe wall (Figure A.1) and it represents onefourth of the Darcy friction factor:
fF
wall
1 2
u
2
Figure A.1 - Shear stress at the pipe
wall.
fF
16
Rm
R m 2300
Transitional flow:
f F f FL
R m R mL f f
R mT R mL FT FL
2300 R m 4000
1.1098
1
5.0452
5.8506
4log
log
0.8981
Rm
fF
2.8257 R m
3.7065D
R m 4000
152
Where:
m u m D
m
Rm
m j y j
j
It should be noted that for the annulus the hydraulic diameter is used to
calculate Reynolds number:
Dh
fF
16
1 0, 04304R e 0.6142
Rm
Turbulent flow:
f F f F0 1 0.0153R e0.3978
Where:
Re
I u I D
I
153
Appendix B
Drift-flux model
The drift-flux multiphase model was introduced by Zuber and Findlay for
two-phase flow in 1965 (Zuber and Findley 1965). For a gas-liquid flow in a
vertical pipe the superficial gas velocity (cross sectional area averaged velocity) is
expressed as a sum of the distribution coefficient (Co) and the gas drift velocity
(ugm). It captures not only the mechanisms of higher concentration of gas near the
center of the pipe (Co um) but also the tendency of buoyancy (ugm). This explains
why gas inside a vertical pipe moves faster than liquid:
ug Coum ugm
ug
usg
ul
usl
1 xg
xg
Where:
xg
Ag
Ag Al
Drift-flux model is used in this thesis through correlations (Ghajar and Tang
2010):
xg
usg
Co usg usl u gm
usg usl
Co
1
usg usl usg
0.1
Where:
- Superficial tension;
Angle with vertical axis.
0.25
154
A general two-phase heat transfer correlation (Kim and Ghajar 2006) for
estimating the heat transfer coefficients regardless of flow pattern, gas-liquid
combination and pipe inclination angle is also used. It is valid for liquid and
permanent gas without taking phase changing into account.
The general heat transfer correlation is described by:
0.1
1 Fp
htp Fp hl 1 0.55
1 Fp
0.4
Prg
Prl
0.25
0.25
I
*
0.25
Where:
Gas quality ():
mg
mg ml
mg g usg A
ml l usl A
g
g
l
2
Fp 1 xg xg tan 1
gD l g
I 1
*
gD 2 l g
sin
l ul D
l 1 xg
0.5
In the heat transfer coefficient of the liquid phase the correction due to the
viscosity term was considered negligible. The 1D model used to describe the
behavior of the annulus and tubing cannot capture this effect because it is
averaged within T(r=rw) to T(r=0) and therefore it is meaningless.
The drift-flux multiphase model is used for application where there is
gas/liquid and it is used only in the well segment (annulus and tubing).
155
Appendix C
Heat Transfer Coefficients
A reservoir layer/well segment is shown in Figure C.1 with its radial layers:
tubing, annulus, cement and formation. The heat transfer mechanisms among
them can be described by conduction, convection and radiation (neglected here).
The overall heat transfer coefficients used in the temperature model is taken into
account the individual conductive and convective barriers to transfer heat within
the wellbore. Heat transfer coefficients were used in the development of this thesis
assuming that well transient is faster than reservoir transient therefore it is
possible to reach thermal equilibrium allowing heat transfer coefficient to be
defined.
U ta
1
1 Rti Rto Rti
ln
ht Kt Rti Rto ha
156
ht
For annulus:
ha
ht 1.84
2 Rti
1/3
Rat or ht 3.66 for 2Rti
Kt
For annulus:
In this case we use superposition (Heaton, Reynolds et al. 1964) and divide
the problem in two:
- A pipe experiencing inner convection with outer insulated:
hainner 1.84
2 Rci Rto
Ka
Raa
1/3
Where
- L is the length of the well segment.
- A pipe experiencing outer convection with inner insulated. In this case the
orientation is important:
Vertical outer wall (Churchill and Chu 1975):
0.387 Raa1/6
haouter1 0.825
8/27
0.492 9/16
Pra
157
2 R R
to
ci
Ka
haouter 2
0.387 Raa1/6
0.6
8/27
0.559 9/16
Pra
2 R R
to
ci
Ka
ha
It should be noted that strong buoyancy effect is expected in the flow which
Gr
is close to unity, otherwise it may be ignored. In pure natural convection, the
Re 2
strength of the buoyancy-induced flow is measured by the Ra where less than 108
is laminar flow and above 1010 is turbulent flow.
The heat transfer coefficients h t and h a have to take into account the
simultaneous flow inside the tubing and inside the annulus. Stephan correlation
(Stephan 1962) is used for this purpose:
0.84
0.6
0.86 Rti 1 0.14 Rti
Rci
Rci
( R Rto )
h' h ci
Rti
Rti
Rci
158
Pr
Cp
K
Grashof Number:
( Rci Rto )3 g a 2 a Ta
Gra
Grt
a 2
( Rti )3 g t 2 t Tt
t 2
Raleigh Number:
Ra GrPr
Reynolds Number:
Rea
Ret
t ut 2 Rti
a
U af
1
1
1
ha U awb
U awb
1
Rci
K ca sin g
R
R
Rci
ln co
ln wb
Rci Kcement Rco
159
Appendix D
Cylindrical-Shape Reservoir Layer Equations
Two phase formulation for an adiabatic cylindrical-shaped reservoir layer
(Figure D.1) experiencing transient flow is derived in this appendix.
The following group of assumptions was used for the mass balance and flow
equations for the control volume shown in Figure D.2:
- Rock and fluids are slightly compressible;
- Porous media is considered homogeneous;
- Capillarity pressure can be neglected pw po p ;
- Fluid saturations are constant during the simulation period;
- Total compressibility times pressure is much smaller than one ct p
1 ;
- Immiscible fluids;
Figure D.2 - Control volume of the reservoir mass balance and flow equations.
160
Mass Balance:
( p S p )
t
p up 0
( p S p )
p u p u p p
t
t
pSp
cf
1
1 p
and c p
p
p p
ct
p
u c j u j p
t
j
Where:
ct c f c j S j
j
u u j
j
Darcys Equation:
up
krp
k p p gz
krp
up p k p p gz
Combining with mass balance equation yields:
ct
p
j k p j gz c j p j gz p
t
j
j
p
t k p t gz
t
Where:
are small
ct
161
t j
j
t j j
t
j
It should be noted that the equation derived has the same mathematical form
of the classic diffusivity equation from the well test analysis.
Energy balance
According to Bear (Bear 1988), the temperature behavior of the fluids in
porous media is based on the basic mass and heat transfer principles. Therefore, it
can be derived by the general energy balance equation (Bird and Frost 1966):
U
U u p u qT : u
t
Enthalpy definition:
H U
Fourier law:
qT kT T
Total conductivity (formation and fluids):
kT ks (1 )k f
U sU s (1 ) f U f
According to Al-Hadhrami (Al-Hadhrami, Elliot et al. 2003) the term
: u can be replaced by u p , resulting:
H p (1 ) f U f Hu kT T
t
162
Assumptions:
The control volume (Figure D.3) is in local thermal equilibrium (quasisteady-state assumption);
Rock density is constant:
dU f dH f C pf dT f
dH C p dT
1 T dp
Defining:
C p C p (1 ) f C pf
Making the assumptions and final arrangements, yields:
C p
T
p
p
T
c f p f C pf T uC pT 1 T u p kT T
t
t
t
j S j C pj (1 ) f C pf
j
p
p
S j j T
c f p f C pf T
t
t
j
t
j S j C pj u T 1 j T u p S j k j (1 )k f
j
j
j
The meaning of the equation terms can be summarized as:
LHS:
Term 1 - transient temperature variation;
Term 2 - transient formation and fluid expansion or compression;
RHS:
Term 3 - heat convection;
Term 4 - viscous dissipation and fluid expansion or compression;
Term 5 - heat conduction.
This equation is presented in chapter 4 as equation 4.2.