Академический Документы
Профессиональный Документы
Культура Документы
i=1
i
(X
i
)]
2
_
(3.2.1)
Then the transformations
(Y ),
1
(X
1
), ...,
p
(X
p
) are said to be optimal for the
regression if they satisfy that:
e
2
(
1
, ...,
p
) = min
,
1
,...,p
e
2
(,
1
, ...,
p
) (3.2.2)
The correlation coecient between the transformed response variable and sum of
the transformed predictor variables (under the constraints, E[
2
(Y )] = 1 and E[
2
s
(X)])
is given by:
9
(,
s
) = E[(Y )
s
(X)] (3.2.3)
where
s
(X) =
p
i=1
i
(X
i
)
The optimal transformations
(Y ),
1
(X
1
), ...,
p
(X
p
) fulll the maximal correla-
tion condition because, as shown in Breiman and Friedman [3], they satisfy that:
s
) = max
,
1
,...,p
(,
s
) (3.2.4)
Breiman and Friedman [3] also showed that these optimal transformations for
correlation are also optimal for regression.
3.3 Mechanistic Modeling of Fluid Temperature
Transients
3.3.1 Distributed Reservoir Temperature Model
Theory of Thermometry in a Fluid-saturated Porous Media
The ow of energy-carrying uids through a porous media has been studied for many
years. Bejan[2] and Bear[1] presented a comprehensive thermodynamic approach to
obtaining a representative model for temperature distribution in a porous media. The
model accounted for spatial distribution and well as transient eects in the formation.
Sharafutdinov[19], Filippov and Devyatkin[5] and Ramazanov and Nagimov[15] also
followed similar approaches in developing their models for temperature distribution
in a uid saturated porous stratum.
In a owing well, the pressure and ow rate measurements by permanent mon-
itoring gauges are not constant. For gauges placed close to the sandface ow area,
10
these changes reect the dynamics of the ow in the reservoir. These ow dynamics
cause a temperature eld to evolve in the reservoir, driven by thermodynamic eects
such as the Joule-Thomson heating (or cooling), adiabatic expansion, and heat of
phase transitions. Other eects namely the viscous dissipation, equal to the mechan-
ical power needed to extrude the viscous uid through the pore, as well as frictional
heating between the uid and rock matrix during the uid ow are also factors that
contribute to the evolution of a nonuniform temperature eld in the medium.
Joule-Thomson eect is the change in the temperature of a uid due to expan-
sion or compression of the uid in a ow process involving no heat transfer or work
(constant enthalpy). This change is due to a combination of the eects of uid com-
pressibility and viscous dissipation. The Joule-Thomson eect due to the expansion
of oil in a reservoir or wellbore results in the heating of the uid because of the value
of the Joule-Thomson coecient of oil - it is negative for oil. The coecient has a pos-
itive value for real gases and the consequent cooling eect is more prominent in gases.
Theoretically, the Joule-Thomson coecient for ideal gases is zero implying that the
temperature of ideal gases would not change due to a pressure change if the system
is held at constant enthalpy. Combined with other factors, on expansion of the uid
and subsequently ow of liquid oil and/or water out of the reservoir, the wellbore and
near wellbore areas in the reservoir become heated above the normal static reservoir
temperature. By convection, diusion and further generation of heat enerygy due to
these eects, a nonuniform temperature is created, which spreads into the reservoir.
Conversely, during no-ow conditions (shut-ins), the regions already heated lose heat
to the surrounding formation through diusion and result in a temperature decline
at a rate determined by the thermal diusivity of the medium.
11
Reservoir Temperature Model in One-dimensional Cylindrical Coordinate
System
Single-phase Formulation
To derive the energy equation for a homogenous porous medium, the energy equations
for the solid and uid parts are derived separately from the rst law of thermody-
namics, and averaged over an elementary control volume to obtain the general form
of the model. The consideration is for a nonisothermal ow of a nonideal uid in a
porous media. The change in kinetic and potential energies of the ow will be taken
as negligible.
For the solid part, assuming no internal heat generation per unit volume of the
solid material, the energy conservation equation associated with ow in an elastic
porous medium becomes:
s
c
s
T
t
= k
s
r
r
T
r
(3.3.1)
where (, c, k)
s
are properties of the solid matrix.
The energy conservation equation at any point in space, occupied by the uid is
given by:
f
c
pf
_
T
t
+ u
p
T
r
_
= k
f
r
r
T
r
+ T
p
t
+ Tv
p
r
v
p
r
+ vg (3.3.2)
where (, c, k)
f
are the uid properties and T is the temperature for both the uid
and solid. An assumption of local thermal equilibrium has been made between the
uid and the porous matrix. Bejan[2] notes that this assumption is valid only for
small-pore media such as geothermal and oil reservoirs. Bejan[2] also showed that
Eqn. (3.3.1) and Eqn. (3.3.2) can be combined by volumetric averaging to give the
nal form of the model as:
12
((1 )(
s
c
s
) + (
f
c
f
))
T
t
+
f
c
f
v
T
r
=
((1 )
s
+
f
)
r
r
r
T
r
+ T
p
t
+ Tv
p
r
v
p
r
+ vg (3.3.3)
On rearrangement and assuming negligible gravity eects, this becomes:
T
t
+ u
T
r
=
r
r
r
T
r
+ C
p
t
+ u
p
r
(3.3.4)
The form of Eqn. (3.3.4) is the convection-diusion type partial dierential equa-
tion with source/sink terms. The second term on the right hand side of Eqn. (3.3.4)
is the compressibility term, while the last term is the viscous dissipation term.
The mass balance equation takes the form:
f
t
+
1
r
(rv
f
)
r
= 0 (3.3.5)
The ow is assumed to obey Darcys law, and the equation for Darcy ow given
by
v =
k
p
r
(3.3.6)
completes the formulation.
Where:
u = Cv
v = supercial velocity,
m
s
C = volumetric heat capacity ratio,
c
f
cm
C
f
= volumetric heat capacity of uid,
J
m
3
K
c
m
= (1 )(
s
c
s
) + (
f
c
f
) , volumetric heat capacity of uid saturated rock,
J
m
3
K
= porosity
= density
13
=
T
C
f
, adiabatic expansion coecient,
K
Pa
=
T1
C
f
, Joule-Thomson coecient,
K
Pa
=
1
T
)
p
, thermal expansion coecient,
1
K
=
m
cm
, thermal diusivity,
m
2
s
= thermal conductivity,
W
mK
m
=
f
+ (1 )
s
, thermal conductivity of uid saturated rock
Eqns (3.3.4), (3.3.5) and (3.3.6) form the governing equations for one-dimensional
thermal transport in a homogenous porous medium. The assumptions made in de-
riving the equations were:
The medium is homogenous, such that the solid and uid permeating the pores
are evenly distributed throughout the porous medium.
The medium is isotropic such that permeability, k and thermal conductivity
do not depend on the direction of the experiment.
At any point in the porous medium, the solid matrix is in thermal equilibrium
with the uid in the pores.
Darcys law applies.
Two-phase Formulation
The assumptions made for the two-phase formulation are similar to those made in the
single phase case, with the addition of negligible capillary eects. The thermal model
in a one-dimensional radial coordinate system for the two-phase system becomes:
14
((1 )(
s
c
s
) + (
w
c
fw
s
w
+
o
c
fo
s
o
))
T
t
+ (
w
c
fw
s
w
v
w
+
o
c
fo
s
o
v
o
)
T
r
=
((1 )
s
+ (
w
s
w
+
o
s
o
))
r
r
r
T
r
+ (
w
c
fw
s
w
w
+
o
c
fo
s
o
o
)
p
t
+(
w
c
fw
s
w
v
w
w
+
o
c
fo
s
o
v
o
o
)
p
r
(3.3.7)
On rearrangement and with negligible gravity and capillary eects, this becomes:
T
t
+ u
T
r
=
r
r
r
T
r
+
p
t
+ J
p
r
(3.3.8)
The mass balance equation takes the form:
(
w
s
w
)
t
+
1
r
(rv
w
w
s
w
)
r
= 0 (3.3.9)
(
o
s
o
)
t
+
1
r
(rv
o
o
s
o
)
r
= 0 (3.3.10)
The Darcy ow equation becomes:
v
w
=
k
w
w
.(
p
w
r
) (3.3.11)
v
o
=
k
o
o
.(
p
o
r
) (3.3.12)
Where:
C = volumetric heat capacity ratio,
(wc
fw
sw+oc
fo
so)
cm
C
f
= volumetric heat capacity of uid,
J
m
3
K
c
m
= (1 )(
s
c
s
) +((
w
c
fw
s
w
) +(
o
c
fo
s
o
)) , volumetric heat capacity of uid sat-
urated rock,
J
m
3
K
=
_
wc
fw
sww+oc
fo
soo
cm
_
15
J
=
_
wc
fw
swvww+oc
fo
sovoo
cm
_
=
T
C
f
, adiabatic expansion coecient,
K
Pa
=
T1
C
f
, Joule-Thomson coecient,
K
Pa
=
1
T
)
p
, thermal expansion coecient,
1
K
=
m
cm
, thermal diusivity,
m
2
s
= thermal conductivity,
W
mK
m
=
f
+ (1 )
s
, thermal conductivity of uid saturated rock
subscripts:
w = water
o = oil
f = uid
Eqns 3.3.8 through 3.3.12 are the equations dening the formulation for the tem-
perature distribution in a reservoir during two-phase ow. Eqn. 3.3.8 is a convection-
diusion equation. Analytical solutions for convection-diusion equations have been
the subject of research and numerical solutions have serious issues with stability due
in part to the nature of the model - a combination of a hyperbolic convective transport
and parabolic diusion transport models. In this work, a semianalytical technique
used in ground water transport was used to solve this problem. The technique is
known as Operator Splitting and Time Stepping (OSATS), and was developed for the
solution of contaminant transport in ground water hydrology.[10][17]
3.3.2 Operator Splitting and Adaptive Time Stepping (OS-
ATS)
Kacur[10] and Remesikova[17] described methods for solving the convection-diusion
type problem using the operator splitting approach. The operator splitting method
16
breaks the model into two dierent parts, the transport part and the diusion part.
Then, at each time step, the nonlinear transport part and the nonlinear diusion
part are solved separately. The semianalytical nature comes from the fact that the
solution is obtained in time sequences, with the solution at a time step depending on
the solution at all previous time steps. Holden et al.[6] showed the theoretical basis
for this technique.
Kacur[10] showed a precise way of solving the following type of problems:
F(u)
t
+ v(x)
u
x
x
(D(x, t))
u
x
= f(x, u) (3.3.13)
with the boundary and initial conditions:
u(0, t) = c
o
(t) (3.3.14)
u(L, t) = 0, for x (0, L), t > 0 (3.3.15)
u(x, 0) = u
o
(x) (3.3.16)
First, the transport part is solved, which presents a hyperbolic problem of the
form:
F(u)
t
+ v(x)
u
x
= 0, t (t
j1
, t
j
) (3.3.17)
with boundary conditions of the form of Eqns. 3.3.14 and 3.3.15, and initial
condition of the form:
U(x, t
j1
) = u
j1
(3.3.18)
17
The solution obtained is denoted as u
1/2
j
:= U(x, t
j
). Then the diusion transport
part is solved, which has the form:
F(u)
t
x
(D(x, t))
u
x
= f(x, u), t (t
j1
, t
j
) (3.3.19)
With the same boundary condition but with initial condition given by the solution
of the convective part -
U(x, t
j1
) = u
1/2
j
(3.3.20)
Finally, the solution at that time step is put as:
u
j
:= U(x, t
j
) (3.3.21)
Which is the solution of Eqn. 3.3.19. This is continued until the solution at the
last time step is obtained, which becomes the nal solution of the model.
3.3.3 Solution of the Thermal Convection-Diusion Model
by OSATS
The OSATS approach was used to solve the thermal model Eqns. 3.3.4 and 3.3.8.
The methodology adopted was:
Decouple the model into two parts: the convection transport part and the
diusion part.
At each time step, rst solve the hyperbolic convection transport part, account-
ing for variable ow rate, as well as heat generation due to viscous dissipation,
frictional and Joule-Thomson eects.
Then solve the diusion part at the same time step, adaptively modifying the
time step to ensure stability if solution is numerical.
Continue until the last time step.
18
Solution of the Single-phase Formulation by OSATS
The following assumptions were made in solving the single-phase reservoir thermal
model:
Constant uid Joule-Thomson, adiabatic expansion coecient, and thermal
conductivity i.e. these parameters are assumed to be weak functions of temper-
ature
Constant uid viscosity and formation porosity
Negligible gravity eects
A. Solution of the convective transport part:
The convection equation with its initial condition becomes:
T
t
+ u
T
r
= C
p
t
+ u
p
r
(3.3.22)
T(t = 0) = T
o
(r) (3.3.23)
Using the method of characteristics, the solution yields:
dr
dt
= u(r, t) = Cv(r, t) (3.3.24)
Such that for constant rate,
r
2
= r
2
1
+
qC
h
(t t
1
) (3.3.25)
Another form of Eqn. 3.3.25 starts by solving for the pressure terms independently
from the material balance equation, and using the Darcy equation to replace velocity
in Eqn. 3.3.24. If the total compressibility of the porous medium is considered
negligible, then the pressure equation reduces to:
19
1
r
r
r
p
r
= 0 (3.3.26)
With boundary conditions:
p(r = r
e
) = p
e
(3.3.27)
p(r = r
w
) = (t) (3.3.28)
where r
e
= external radius and r
w
= wellbore radius.
This yields:
p(r, t) = p
e
+
p
e
(t)
ln
_
re
rw
_ . ln
_
r
r
e
_
(3.3.29)
And:
p
r
= [
p
e
(t)
ln
_
re
rw
_ ].
1
r
(3.3.30)
Applying this to Eqn. 3.3.24 by replacing the velocity with the Darcy law equiv-
alent gives the form:
r
2
= r
2
1
2(p
e
t s(t)) (3.3.31)
where:
=
kC
ln R
,
R = ln
re
rw
,
s(t) =
t
_
0
()d
The parameter (t) which is the sandface pressure (bottomhole pressure) in the well
bore is readily obtainable from solutions of classical pressure transient problems for
dierent reservoir models.
20
Alternatively, where the total compressibility of the system is not negligible, the
solution of the pressure equation comes in the form:
p =
q
4kh
Ei
_
r
2
w
c
t
4kt
_
(3.3.32)
Continuing the method of characteristics solution for the temperature yields,
dT
dt
=
t
_
0
dp
dt
dt + (
)
t
_
0
p
d (3.3.33)
with
= C.
This becomes:
T(r, t) = T
0
(r) [p(r, 0) p(r, t)]
ln R
t
_
0
() ln
_
r
r
e
_
d (3.3.34)
Ramazanov and Nagimov [15] showed that by using the average time theorem,
the integral on the left hand side of Eqn. 3.3.35 can be closely approximated when
an optimal average time is used. Therefore, combining with Eqn. 3.3.31, and apply-
ing the average time theorem, the nal approximate solution for wellbore sandface
temperature becomes:
T(r
w
, t) = T
0
(r
1
) [p(r, 0) (t)]
ln R
[(t) (0)] ln
_
_
r
2
1
2(p
e
z s(z)
r
e
_
(3.3.35)
where:
r
1
=
_
r
2
w
2(p
e
t s(t) (3.3.36)
[0 z t]
z = average time.
21
An optimal choice for z must be used, and Ramazanov and Nagimov [15] suggested
t
2
< z < t.
B. Solution of the diusion part:
The form of the diusion problem is:
1
T
t
=
1
r
T
r
+
2
T
r
2
(3.3.37)
0 < r <
with initial and boundary conditions:
T(t = 0) = F(r) (3.3.38)
T(r = 0) = finite (3.3.39)
Ozisik [14] using the method of integral identity described by Masters [11], showed
that the solution to Eqn. 3.3.37 is of the form
T(r, t) =
1
2t
b
_
r
_
r
exp
_
r
2
+ r
2
4t
_
F(r
)I
o
_
rr
2t
__
dr
(3.3.40)
Where 0 < b < is equivalent to the thermal diusivity length, and takes the
form b =
t.
Therefore in operator splitting and time stepping approach, at each time step,
Eqn. 3.3.35 is evaluated for the solution of the convective part at that step. Then,
this solution forms the initial condition F(r) in the evaluation of the diusion solution,
Eqn. 3.3.40. The nal solution obtained from Eqn. 3.3.40 is taken as the temperature
of the system at that time step.
22
Solution of the Two-phase Formulation by OSATS
A. Solution of the convective transport part:
The convection equation with its initial condition is:
T
t
+ C
T
r
=
p
t
+ J
p
r
(3.3.41)
T(t = 0) = T
o
(r) (3.3.42)
The solution follows closely the approach used for single phase formulation.
By method of Characteristics,
dr
dt
= C (3.3.43)
For constant rate, let
q
T
= q
o
+ q
w
(3.3.44)
q
o
=
o
T
q
T
(3.3.45)
and
q
w
= q
T
q
o
= q
T
_
1
o
T
_
(3.3.46)
where:
o
= oil mobility
_
ko
o
_
w
= water mobility
_
kw
w
_
T
=
o
+
w
The solution to Eqn. 3.3.43 becomes
r
2
= r
2
1
+
q
T
h
_
w
c
fw
s
w
(1
o
T
) +
o
c
fo
s
o
T
_
(t t
1
) (3.3.47)
23
Again, we obtain another form of the solution Eqn. 3.3.47 by solving the pressure
equation, and using the Darcy equation to replace velocity in Eqn. 3.3.43. Assuming
total compressibility of the porous medium is negligible, then the pressure equation
reduces to:
1
r
r
r
p
r
= 0 (3.3.48)
with boundary conditions
p(r = r
e
) = p
e
(3.3.49)
p(r = r
w
) = (t) (3.3.50)
where r
e
= external radius and r
w
= wellbore radius.
Therefore,
p(r, t) = p
e
+
p
e
(t)
ln
_
re
rw
_ . ln
_
r
r
e
_
(3.3.51)
and
p
r
= [
p
e
(t)
ln
_
re
rw
_ ].
1
r
(3.3.52)
Applying this to Eqn. 3.3.43 by replacing the velocity with the Darcy law equiv-
alent and solving gives
r
2
= r
2
1
2(p
e
t s(t)) (3.3.53)
where
=
{
w
c
fw
s
w
w
+
o
c
fo
s
o
o
}
c
m
ln R
R = ln
r
e
r
w
24
s(t) =
t
_
0
()d
The parameter (t) which is the sandface pressure (bottomhole pressure) in the
well bore is readily obtainable from solutions of classical pressure transient problems
for multiphase ow, and for dierent reservoir models.
The temperature model therefore becomes,
dT
dt
=
t
_
0
dp
dt
dt + (
)
t
_
0
p
d (3.3.54)
where
=
wc
fw
swww+oc
fo
sooo
wc
fw
sww+oc
fo
soo
This yields the solution
T(r, t) = T
0
(r)
ln R
t
_
0
() ln
_
r
r
e
_
d (3.3.55)
Using the average time theorem, the integral on the left hand side of Eqn. 3.3.55
can be closely approximated when an optimal average time is used. Therefore, the
nal approximate solution for wellbore sand face temperature becomes:
T(r
w
, t) = T
0
(r
1
)
[p(r, 0)(t)]
ln R
[(t) (0)] ln
_
_
r
2
1
2(p
e
z s(z)
r
e
_
(3.3.56)
with
r
1
=
_
r
2
w
+ 2(p
e
t s(t))
0 < z < t, z = average time. An optimal choice for z must be used.
25
B. Solution of the diusion part:
The form of the diusion problem is
1
T
t
=
1
r
T
r
+
2
T
r
2
(3.3.57)
0 < r <
with initial and boundary conditions:
T(t = 0) = F(r) (3.3.58)
T(r = 0) = finite (3.3.59)
using the method of p function and integral identity again, we get the solution
as
T(r, t) =
1
2t
b
_
r
_
r
exp
_
r
2
+ r
2
4t
_
F(r
)I
o
_
rr
2t
__
dr
(3.3.60)
Where 0 < b < is equivalent to the thermal diusivity length, and takes the form
b =
t.
The nal form of the solution is obtained by following the operator splitting and time
stepping methodology used in the single-phase solution.
3.4 Numerical Solution
A fully numerical solution of the the model formulation would require a full dis-
cretization of the model. However, direct numerical solutions convection-diusion
type PDEs with source/sink terms have been known to be very unstable. A way out
26
was to use Operator Splitting and Time Stepping (OSATS), in which case convec-
tive transport part was solved analytically (since it is easier to solve this way) and
numerical discretization methods were used the diusion transport part.
The analytical solution of the convective transport problem has been shown in
preceding sections. In discretizing the diusion part, the nite dierence scheme was
adopted, and the coordinate system was taken to be cartesian. This gave:
1
T
t
=
1
r
T
r
+
2
T
r
2
(3.4.1)
Which became (in two-dimensional cartesian coordinate system) :
1
T
t
=
2
T
x
2
+
2
T
y
2
(3.4.2)
1
T
n+1
i,j
T
n+1
i,j
=
1
x
2
T
n
i+1,j
T
n
i1,j
+
1
y
2
T
n
i,j+1
T
n
i,j1
(3.4.3)
subject to the same boundary and initial conditions.
Therefore, in the OSATS routine, at each time step, the convective problem was
solved analytically, while the diusion problem was numerically approximated in the
central dierence formulation shown in Eqn. 3.4.3 and solved in an explicit scheme.
3.5 Wellbore Temperature Transient Model
Permanent downhole monitoring tools are usually located some hundreds of feet (200
- 300 ft) above the perforation/production zone. The tool placement constraint is one
that is imposed by the design of the completions, and the optimal location for pressure
and temperature data management would be a position as close to the perforation as
possible, to give measurements that are comparable with their sandface values. This
calls for a coupling of the reservoir temperature model to a wellbore model to account
27
for heat loss that may occur between the uid in the wellbore and the surrounding
formation when the uid ows from the sandface to the gauge location.
The wellbore model used in this work was obtained from the work of Izgec et al.
[8]. The solution to the model is a modication of Rameys model [9] to account for
heat transfer at shut-in times when ow rate is zero and heat transfer in the wellbore
is only by conduction into the formation.
Izgec et al. [8] showed that the distribution of temperature in a wellbore can be
obtained, as a fucntion of depth, by
T
f
(r, t) = T
ei
+
1 e
aL
R
t
L
R
_
1 e
(zL)L
R
_
g
G
sin +
(g sin )
c
p
Jg
c
_
+e
(zL)L
R
(T
fbh
T
ebh
)
(3.5.1)
where
L
R
is called the relaxation parameter, dened by:
L
R
=
2
wcp
_
rtoUtoke
ke+rtoUtoT
D
_
with
T
D
= ln [e
0.2t
D
+ (1.5 0.3719e
t
D
)]
t
D
t
D
=
t
r
2
w
= wellbore inclination angel
=
dp
dz
T
fbh
= uid temperature entering the sandface from formation (solution to the
reservoir temperature model)
T
ei
= geothermal temperature of formation at gauge location
28
T
ebh
= geothermal temperature of formation at bottomhole
3.6 Coupling the Reservoir Model to the Wellbore
Model
The issue of gauge placement, usually at some distance away from the perforation calls
for the coupling of the reservoir model to a wellbore model to account for heat loss
during ow between the perforation and the gauge location. In principle, the closer
the gauge to the perforation, the closer the overall model would be to the reservoir
model and the better the reservoir model can be used for further reservoir/formation
analysis.
The two models are coupled through the temperature of the uid at the bottomhole.
The bottomhole temperature is estimated using the distributed reservoir model, and
used as an input into the wellbore model to estimate the uid temperature at the
gauge location. The sensitivity of this overall model to the distance between the
gauge and the perforation will be revisited in a sensitivity analysis test to determine
the viability of using the model in reservoir studies, since the further away the gauge
is from the perforation, the less sensitive the solution will be to the reservoir model,
and more sensitive to the wellbore model.
Chapter 4
Qualitative Evaluation and
Sensitivity Analysis
4.1 Optimal Transformation for Estimation of Func-
tional Relationship
In order to show that a functional relationship may exist between temperature as a
response variable, and rate and pressure as the predictor variables, the nonparamet-
ric regression method known as Alternating Conditional Expectations (ACE) was
applied to a eld data set. ACE yields optimal transformations of the variables,
and the correlations between these transformations have been shown to be optimal
for regression between the variables. These transformations are also useful in estab-
lishing the existence of a functional relationship between the response and predictor
variables. Using a eld data set obtained from permanent downhole monitoring tool
in a well, the ACE method was applied to the pressure, rate and temperature data
to establish the existence or otherwise of a correlation and functional form for their
29
30
relationship. Figures 4.1 and 4.2 show the plots of rate, pressure and temperature
data, and the plot of the regression on the optimal transformations.
Figure 4.1: Flow rate data (left), pressure data (right) from DAT.1 data set.
Figure 4.2: Temperature data (left) from DAT.1 data set, optimal regression (right).
The optimal transformation functions showed a correlation coecient of 0.99.
This means that temperature is well correlated with ow rate and pressure, that a
functional relationship may exist between temperature, and rate and pressure and
this functional form can be extracted from any representative data set.
31
4.2 Qualitative Evaluation of Model - Single-phase
Case
Having seen, from ACE analysis, that a functional form may exist between temper-
ature as a dependent variable, and rate and pressure as the independent predictor
variables, the mechanistic models developed for their relationship were tested quali-
tatively for a check of reproducibility of trends seen in the data used. Using arbitrary
but typical and physically meaningful values of the model parameters, the following
results were generated for qualitative evaluation of the model and the solution strat-
egy adopted here. The model were also checked for reproducibility of transient trends
seen in the measured data.
The input data sets used were ow rate information from two real elds, and here
called DAT.1 and DAT.2 data sets. These were obtained from permanent downhole
gauges (PDGs). The data sets consist of PDG measurements of ow rate, pressure and
temperature with time for dierent wells in dierent elds. Using the representative
ow rate data as input, and thermal model developed, the temperature prole was
simulated for each representative data input set.
Figures 4.3, 4.4, and 4.5 show that the model and the solution qualitatively cap-
tured the changes, eects and trends in the data, in acceptable details. The overall
shapes are reproduced by the formulation/solution using arbitrary model parameters.
This forms a motivation for performing sensitivity analysis on the model parameters,
for using the model in parameter estimation in an inverse problem and for subsequent
uncertainty analysis.
32
Figure 4.3: Qualitative comparison using DAT.1 (rst representative transient).
Figure 4.4: Qualitative comparison using DAT.1 (second representative transient).
33
Figure 4.5: Qualitative comparison using DAT.2.
4.3 Sensitivity Analysis
Many variables/parameters are present in the model formulation and uncertainties
in their values present a challenge in further processing and utilization of the formu-
lation and solution methodology presented in this work, hence the need to test the
sensitivity of the formulation and solution to dierent values of these parameters.
The following parameters were tested for sensitivity of the solution to their values:
the porosity of the formation, Joule-Thomson coecient of the uids, reservoir thick-
ness, uid viscosity, thermal conductivity of rock and uid, permeability, thermal
diusivity length, distance of permanent downhole gauge from the perforation and
the geothermal gradient.
34
Figure 4.6: Sensitivity to porosity (top) and uid Joule-Thomson coecient (bottom).
35
Figure 4.7: Sensitivity to thermal diusivity length (top) and uid thermal conductivity
(bottom).
36
Figure 4.8: Sensitivity to reservoir permeability (top) and uid viscosity (bottom).
37
Figure 4.9: Sensitivity to gauge placement.
It is clear from Figures 4.6 to 4.9 that the temperature formulation and solution is
sensitive to most of the model parameters. The parameters with the most prominent
sensitivity (> 50% in temperature estimation for <= 25% change in the value of the
parameters) were the uid Joule-Thomson coecient which smoothed the responses
to small intermittent rate changes, the porosity of the formation, the height of the
formation, the thermal diusivity length, uid viscosity and reservoir permeability.
However, since some of the parameters such as the formation thickness and geothermal
gradient can be estimated with acceptable certainty from other means such as well
logs, viscosity from laboratory measurements, as well as the gauge distance from
the perforation face, the parameter space for the inverse problem was reduced to
the space of formation porosity, uid Joule-Thomson coecient, thermal diusivity
length and permeability. Porosity and permeability are highly correlated (as can be
seen also in the sensitivity plots). Therefore, in the inverse problem for parameter
38
estimation, permeability of the medium was specied, reducing the nal parameter
space to formation porosity, formation thermal diusivity length and uid Joule-
Thomson coecient. Sensitivity to the gauge distance presents a problem in using the
formulation presented in this work for possible reservoir parameter estimation. The
temperature of the uid recorded at the gauge location is a combined eect of thermal
transient processes in the reservoir which delivers uid of changing temperature (with
constant/changing rate) to the wellbore from the reservoir, and the heat loss to the
external formation during ow up the wellbore to the gauge location. This makes the
magnitude of heat loss in the wellbore very important, and calls for a careful use of
the formulation developed here if the gauge distance is very large. This is because,
at a very large distance away from the formation, the magnitude of the heat loss in
the wellbore will mask the eect of the reservoir thermal transients and hence lead to
poor reservoir parameter estimation. By repetitive trials, the optimal distance of the
gauge from the perforation to ensure obtaining representative reservoir parameters is
< 100m (300ft). This however is the average distance currently used in many eld
applications.
Diusivity length issues - single-phase
The challenge presented by the optimal selection of the diusivity length parameter,
b, became apparent. Since this parameter depends on the thermal diusivity and the
length of shut-in time (diusion-dominated heat transfer period), dierent shut-in
regimes required dierent optimal diusivity length because of diering shut-in time
durations.
As a test for this, the ow rate from DAT.1 data set (800 hrs, 466000 data points)
was used as input to the model to predict the temperature over the entire duration
of the measurement. Uniform thermal diusivity length was assumed over several
39
transients and revealed a complete loss of the diusion behavior later in time. Using
a uniform but dierent diusivity length, b = 11m, over the data region 0 - 350,000
(data point counter on x-axis)(Figure 4.10), b = 8m over the data region 200,000 -
400,000,(Figure 4.11) and b = 12m over 300,000 - 400,000 (Figure 4.12). These plots
are shown only for qualitative reasons since model parameters were specied arbi-
trarily and were generated to see the behavior of the dierent shut-in regions with
dierent diusivity length scale, as well as to reveal the eects later in time.
40
Figure 4.10: Temperature estimation with b = 11m, over 0-350,000 data points (left), actual
data (right).
Figure 4.11: Temperature estimation with b = 8m, over 200,000-350,000 data points (left),
actual data (right).
41
Figure 4.12: Temperature estimation with b = 12m, over 300,000-400,000 data points (left),
actual data (right).
Figure 4.13: Temperature estimation with b = 10m, over 100,000-200,000 data points (left),
actual data (right).
The results show that while the model has the ability to predict the temperature
prole in the reservoir, the accuracy of that prediction depends on the diusivity
42
length that characterizes the behavior of the prole at shut-in periods. No one dif-
fusivity length value will characterize the entire model over a long period of time
with recurring transients of dierent durations. Therefore, the optimal choice of this
length scale must be made and would not be one uniform value over several tran-
sient periods or over data taken for a relatively long time. Figure 4.13 shows that an
optimal selection should be found over each representative transient, separately and
independent of previous or subsequent shut-ins.
4.4 Sampling the Distribution of the Parameter
Space
The nature of the distribution of the parameter space is not known explicitly since the
model is nonlinear and the solution is semianalytic. Monte Carlo simulations was per-
formed to generate the one-dimensional and two-dimensional marginal distributions
of the parameter space. The distributions were then sampled for identication of the
optimal search space for parameter estimation, as well as to check for multimodalities
in the distribution of model parameters, using the method of volumetric probabilities
introduced by Tarantola [21]. The two-dimensional marginal distribution for the ra-
dial system, with porosity and oil Joule-Thomson coecient as parameters is shown
in Figure 4.14.
43
Figure 4.14: Two-dimensional marginal distribution of porosity and oil Joule-Thomson
coecient from DAT.1.
Figure 4.14 shows that the distribution of the parameter space, in two-dimensional
marginal distribution sense is unimodal. The plots also show that the optimal pa-
rameter space for both porosity and oil Joule-Thomson coecient as captured by the
model is within the feasible range and the values are physically realistic.
Chapter 5
Results and Discussion
The model solution, unique to the boundary condition chosen in the formulation, was
matched to the temperature data using the ow rate as input. In the case of eld
data collected over long periods in time, because of thermal diusivity length issues,
representative transient regions were selected, to ensure that a constant diusivity
length could be used. In the optimization routine, the parameters that were perturbed
to establish the match were porosity , oil Joule-Thomson coecient , uid thermal
conductivity
f
(in some instances as a check) and the optimal diusivity length
b. The values of these parameters obtained at the optimal match were taken to be
estimates of the optimal values of the parameters.
5.1 Synthetic Data
As a check on the procedure, synthetic data were generated using the model devel-
oped in this work. The synthetic data were generated in three forms:
normally distributed random noise (with mean 20% of the range of the ow
44
45
rate from DAT.1 data set) was added to the ow rate data, and this was used
as input to generate a temperature data set using the single-phase model
again, normally distributed random noise was added to the temperature data
obtained above to create a second noisy temperature data set
using a dierent transient region in DAT.1 data set, the step above was repeated
to generate a third temperature data set with appropriate noise added to it,
using the single-phase model
the two-phase model was also used to generate a temperature data set
Each of the four temperature data sets above was used as true measurement in
an inversion step to attempt to reestimate the model parameters used in generating
the data sets. This was done to test the robustness of the formulation and solution
strategy, and to examine the possibility of using them in an inverse model for param-
eter estimation.
46
Figure 5.1: Single-phase: temperature match using rate data with random noise.
Figure 5.2: Single-phase: temperature match using rate data with random noise, and true
temperature with random noise.
47
Figure 5.3: Single-phase: temperature match with random noise added to the true tem-
perature.
Figure 5.4: Single-phase: temperature match using rate data with random noise, with
dierent transient region.
48
Figure 5.5: Two-phase: temperature match using rate data with random noise, and true
temperature with random noise.
For Figure 5.1, the data were generated using = 0.3, = 4.5 10
8
(K/pa)
and b = 7.0m. The optimal parameter values after the match were = 0.227, =
3.5 10
8
(K/pa) and b = 7.3m.
In Figure 5.2, the data were generated using = 0.3, = 4.5 10
8
(K/pa) and
b = 7.0m and the match occurred at = 0.231, = 3.27 10
8
(K/pa) and b = 7.2m.
In Figure 5.4, the data were generated with = 0.3, = 4.5 10
8
(K/pa) and
b = 8.0m and the match occurred at = 0.259, = 1.51 10
8
(K/pa) and b = 8.32m
Finally, in the two-phase case shown in Figure 5.5, the data were generated with
= 0.3, Sw = 0.45 and b = 2.7m and the match occurred at = 0.145, S
w
= 0.493
and b = 3.0m.
The plots show good matches between the true data and simulated results for
the tests on both models (single- and two-phase models). Also, the values of the
49
model parameters obtained after the matching were close to the original values of
the parameters. This shows that the formulation with the solution method used was
able to closely reestimate the parameters used in generating a synthetic data set and
so lends support to the attempt to use the model formulation in an inverse step for
reservoir parameter estimation by matching the model to real eld temperature data.
5.2 Field Data
5.2.1 Single-phase System
Two dierent sets of eld data were used to test the model, as well as obtain model
parameters at optimal match of the model to the data. The data sets have been named
DAT.1 and DAT.2. DAT.1 set is an 800-hr long measurement, each measurement
taken every 6 secs. DAT.2 is 24-hr long with each measurement taken every second.
Again, representative transient regions were selected, to ensure a constant diusivity
length and the parameters for used in the inverse problem were porosity , oil Joule-
Thomson coecient , uid thermal conductivity
f
(in some instances as a check)
and the thermal diusivity length b.
The following plots were obtained using the semianalytical solution technique (Op-
erator Splitting and Adaptive Time Stepping [OSATS]) discussed in Chapter 3.0, and
the values of the model parameters at optimal match are shown in the caption of each
gure.
50
Figure 5.6: Matching DAT.1 using OSATS ( = 0.196, = 1.1610
7
(K/Pa), b = 9.2m).
Figure 5.7: Matching DAT.1 using OSATS ( = 0.21, = 9.0 10
9
(K/Pa), b = 6.12m).
51
Figure 5.8: Matching DAT.1 using OSATS ( = 0.235, = 4.48 10
8
(K/Pa), b =
7.817m).
Figure 5.9: Matching DAT.1 using OSATS ( = 0.2, = 7.7 10
8
(K/Pa), b = 5.2m).
52
Figure 5.10: Matching DAT.1 using OSATS ( = 0.228, = 1.0910
9
(K/Pa), b = 3.0m).
Figure 5.11: Matching DAT.1 using OSATS ( = 0.21, = 4.310
8
(K/Pa), b = 5.16m).
53
Figure 5.12: Matching DAT.2 using OSATS ( = 0.182, = 5.95 10
8
(K/Pa), b =
5.92m).
Figures 5.6 to 5.11 show good matches between the single-phase model and DAT.1
data set. Figure 5.12 also show the match using the single-phase model on DAT.2
data set. Both direct search (Genetic algorithms) and gradient based (Levenberg
Marquardt) optimization techniques were used in the inverse problem to run each
case, and the optimal parameters at the matches were always approximately equal
for each optimization algorithm used. These values of the model parameters (also
reservoir and uid properties) obtained at optimal match are physically meaningful,
and the porosity values are well within the range of porosity values for carbonate
reservoirs (for DAT.1 data set), the Joule-Thomson coecient values also satisfy the
range of values of the coecient that have been obtained for dierent types of crude
oil. The thermal diusivity length correlation to the duration of shut-in for each
transient is also very obvious. The longer the shut-in (when input ow rate is zero),
the longer the optimal thermal diusivity length estimated from the inverse problem.
54
A fully numerical solution of the the model formulation was also attempted
for comparison with results from the semianalytical solution. It must be noted
that numerical solution to convection-diusion problems, especially where there are
source/sink terms, are known to be very unstable. A way out was to use Opera-
tor Splitting and Time Stepping (OSATS), in which case numerical discretization
methods were used to solve each of the decoupled convection and diusion transport
parts.
Figure 5.13: Matching DAT.1 using fully numerical solution ( = 0.286, = 1.53
10
8
(K/Pa), b = 2.0m).
55
Figure 5.14: Matching DAT.1 using fully numerical solution ( = 0.287, = 9.0
10
9
(K/Pa), b = 2.6m,
f
= 0.11(W/m.K),
s
= 4.0(W/m.K)).
Figures 5.13 and 5.14 show that the numerical scheme could also match the data.
The model parameters at optimal match are comparable to the values obtained using
the more stable semianalytic solution scheme.
5.3 Two-phase System
The two-phase formulation has the uid saturations included among the model pa-
rameters. The saturation variables in the formulation are static (not modeled to
change with time). However, since permanent monitoring devices take measurements
over long periods of time, saturation changes with time can be modeled as a time-lapse
problem, in which eective values of saturations are estimated at each time.
Saturation data to fully test this formulation was not readily available at the time
of writing this report. Therefore, data from the single-phase system (DAT.1) was
used to test the formulation. The intent of the inversion was to check if the inversion
56
process would drive the water saturation to the specied critical value when data
acquired from single-phase oil ow is used. Therefore, as in the single-phase case, the
model developed here, unique to the boundary condition chosen in the formulation,
was matched to the temperature data using the ow rate as input. Representative
transient regions were selected, to ensure a constant diusivity length and the param-
eters for estimation were porosity , water saturation S
w
, and the optimal thermal
diusivity length b. The critical water saturation value used was S
wc
= 0.2.
Figure 5.15: Matching using two-phase model ( = 0.25, S
w
= 0.3, b = 5.36m).
57
Figure 5.15 shows the plots of the match of the two-phase model to data single-
phase system. Since the data set used was that from a single-phase oil ow, the
inversion optimization step was expected to drive the water saturation to its critical
value at optimal match. The initial value of water saturation was set at 0.6. Then a
gradient-based search was used to obtain the match of the model to the data. The
estimated value of water saturation at optimal match (according to tolerance set on
the optimization routine) was S
w
= 0.28. The algorithm drove the water saturation
towards the critical water saturation specied, at each iteration. Using direct search
(Genetic algorithm), similar results were also obtained.
5.4 Potential Applications
The models developed in this study have been shown to have the potential to help
characterize a reservoir using only rate and temperature data from any permanent
downhole monitoring source. Until recently, temperature data measured have been
largely ignored. Specically, this study provides a method for the estimation of poros-
ity in homogenous elds, as well as potential for estimation of saturation distribution
in a reservoir using temperature data only. Being rate-dependent, the temperature
data could also oer a way to estimate sandface owrate by inverting the temperature
model to estimate owrate.
Spatial distribution of porosity and saturation elds
(heterogeneous eld preliminary study)
The formulations and their solutions allow for local point-to-point estimation of tem-
perature in a discretized spatial grid. Each local estimation is dependent on the local
values of the model parameters, hence allowing for the estimation of local (grid) val-
ues of model parameters, each depending on the value of the temperature at the grid.
58
This forms the basis for the estimation of heterogeneous parameter elds. Results
can serve as secondary conditioning data for porosity and saturation elds in reservoir
modeling. Examples of such estimations are shown in Figures 5.16 and 5.17.
Figure 5.16: Three-dimensional spatial temperature distribution at a time instant using
DAT.1 data set.
Data from interference well testing or horizontal wells could be used in check-
ing/validating these characterization of temperature elds in space.
59
Figure 5.17: Two-dimensional spatial temperature distribution at a time instant using
DAT.1 data set.
Chapter 6
Scope for Further Study
6.1 Updating the Boundary Conditions
Only one boundary condition has been specied in the solution obtained in this
problem (bounded for convective, innite radial for diusion). Dierent reservoir
congurations will require dierent boundary condition (e.g. where the reservoir is
bounded by an aquifer), and will require the specication of the appropriate boundary
conditions in the solution.
6.2 Updating the Model to Treat Three-phase Sys-
tems (Gas-Oil-Water)
The formulations presented in this study considered single-phase oil, and two-phase
oil-water systems. In a single-phase gas system, some conditions and assumptions
made must be relaxed to accommodate specic physical conditions that come with
gas systems. For example, gas densities are not constant and are specically aected
60
61
by changes in temperature. Similarly, gas thermal conductivity is a strong function
of temperature as well as gas viscosities. Most of these conditions were relaxed in
building the single-phase oil model. Similar considerations must also be made in
developing the three-phase gas-oil-water model.
6.3 Reservoir Models with Fluid Injection or Phase
Transition
The formulation in this study assumed no uid injection and considered temperature
changes due strictly to uid expansion and viscous dissipation. However, the idea
behind the model can be extended to also account for cases where there is uid
injection into the reservoir. Similarly, phase transitions (e.g. liberation of free gas in
the reservoir) can also be considered for a more complete picture of thermal eects
in the reservoir.
6.4 Cointerpretation of Pressure, Rate and Tem-
perature
This model presents an additional source of information in transient analysis - tem-
perature. Using ACE, it was shown that a functional relation exists between temper-
ature, rate and pressure. This presents the possibility of cointerpreting rate, pressure
and temperature.
62
6.5 Heterogenous Fields
As shown earlier, if the appropriate data is available (e.g. temperature distribution
in space from a horizontal well, or temperature measured at dierent wells producing
from the same reservoir - interference testing), the spatial distribution of the model
parameters could be estimated and used to characterize the heterogenous elds of
porosity and saturation.
Nomenclature
b = thermal diusivity length, m C = volumetric heat capacity ratio,
(wc
fw
sw+oc
fo
so)
cm
C
f
= volumetric heat capacity of uid,
J
m
3
K
c
m
= (1 )(
s
c
s
) +((
w
c
fw
s
w
) +(
o
c
fo
s
o
)) , volumetric heat capacity of uid sat-
urated rock,
J
m
3
K
p = pressure, Pa q = owrate
m
3
s
r = radius
T = temperature, K
T
fbh
= uid temperature entering the sandface from formation (solution to the reser-
voir temperature model)
T
ei
= geothermal temperature of formation at gauge location
T
ebh
= geothermal temperature of formation at bottomhole
t
D
= dimensionless time
t
r
2
w
U
to
= overall heat transfer coecient,
J
(secKm
2
)
u = Cv
v = supercial velocity,
m
s
Greek
= porosity
= density
=
T
C
f
, adiabatic expansion coecient,
K
Pa
=
T1
C
f
, Joule-Thomson coecient,
K
Pa
=
1
T
)
p
, thermal expansion coecient,
1
K
=
m
cm
, thermal diusivity,
m
2
s
= thermal conductivity,
W
mK
m
=
f
+ (1 )
s
, thermal conductivity of uid saturated rock
63
64