I Society of petroleum Engineers I
SPE 50576
History Matching of Laboratory Coreflooding
A. E. Fincham, Elf Exploration UK plc., and B. Ferreol, Elf Exploration
Copyright 1998, Society of Petroleum Engineers Inc.
This paper was prepared for presentation at the 1998 SPE European Petroleum Conference
held in The Hague, The Netherlands, 2c-22 October 1998.
Ttis paper was selected for presentation by an SPE Program Committee following review of
information conteined in an abstract submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
correction by the author(s). The material, as presented, does not necessarily reflect any
poaitian of the Society of Petroleum Engineers, its officers, or members. Papers presented at
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of thie paper
for commercial purposes without the written consent of the Society of Petroleum Engineers is
prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300
words illustrations may not be copied. The abstract must contain conspicuous
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U. S.A., fax 01-972-952-9435.
Abstract
This paper describes a new program (Caropt) which analyses
data from dynamic core flooding experiments and
simultaneously estimates relative permeabilities and capillary
pressures. Relative permeability curves are parameterised
using a new modified form of standard spline interpolation
that allows more flexibility in their functional form. Caropt
has proven to be both robust and efficient in practice.
Examples are given of the application of Caropt to two
laboratory coreflooding experiments. The results produced
with Caropt show that capillary pressure has a significant
effect on the predicted relative permeability curves.
Introduction
Relative permeabilities are empirical factors that are used to
correct the single phase Darcy’s law for application to
multiphase flow. The relative permeability of any phase is an
increasing function of the phase saturation. Relative
permeabilities are important input data for reservoir simulation
studies and have a strong influence on reservoir behaviour.
Laboratory coreflooding experiments are used to find the
relative permeability of reservoir rock. Steady-state
experiments can be used but these are difficult and expensive
to perform. Dynamic coreflooding experiments are a more
cost-effective way to measure the relative permeability curves
and are in widespread use throughout the industry. When
capillary pressure effect are negligible the JBN1 method can be
used for these calculations. If capillary pressure effects are
significant then the calculations are considerably more
complex. A possible solution is to use flow rates that are
sufficiently high to make capillary pressure effects negligible,
95
Experiments
Production.
but these flow rates will be many times greater than those
occurring in the reservoir and may not give representative
results2-4.
If the coreflood experiment is run at low flow rates that
are representative of the flow rates found in the reservoir and
if capillwy pressure effects are important, it is necessary to use
a computer simulation model to solve the non-linear ptitial
differential equations that describe the flow behaviour. Such a
simulator can predict the results of an experiment when the
relative permeability and capillary pressure curves are given as
input data. We require to solve the inverse problem of finding
the relative permeability curves that correspond to the
experimental results. In theory repeated simulations using trial
and error methods to find the relative permeability and
capillary pressure curves could solve the inverse problem, but
in practice this is very time consuming and it is difficult to
achieve a good match. An alternative approach is to formulate
the problem as an optimisation problem with an objective
function that is the weighted sum of squares of the differences
between the measured and simulated data values. The
resulting nonlinear optimisation problem can be solved using a
standard numerical optimisation algorithm. The solution gives
the parameters that define the relative permeability and
capillary pressure curves. This method is similar to the
automatic history matching methods that are used in matching
reservoir production data with reservoir simulation.
Several authors4-7 have described automatic methods for
solving the inverse problem and there are a few commercially
available computer programs. We found that the existing
programs did not meet our requirements. The main limitations
were a lack of robustness and the restrictions on the functional
form of the relative permeability curves. We decided to
develop an in-house program that would specifically address
these perceived weaknesses.
In this paper we describe the techniques used in this
program (called Caropt). Caropt links together two well-
established programs: an Elf in-house simulator called Z2C
and a commercial optimisation package developed at Stanford
University called Npsol. The robustness of Caropt relies on the
robustness of these two programs. To increase the flexibility
of functional forms for the relative permeability curves we
have introduced a modification to the standard spline
 2 A. E. FINCHAM, B. FERREOL
interpolation that preserves the monotonicity of the
interpolated function at points where there is high order
curvature. This modified spline enables a greater range of
functional forms to be considered without generating
unrealistic interpolated values that can lead to a calculation
failure in the simulator. The performance of Caropt is
illustrated by two examples that are based on real
experimental data.
The Coreflood Problem
The experiment. In a dynamic coreflood experiment a core
of reservoir rock is saturated with oiI at residual water
saturation and then water or gas is injected until no more oil is
produced from the core. This can take more than two weeks.
In a waterflood experiment the water is injected at a steady
flow rate and in a gasflood experiment the injection pressure is
controlled to give a fixed pressure drop across the core. In the
waterflood experiment the measured data consists of the oil
production and the pressure drop across the core. In the
gasflood experiment the measured data consists of the oil
production and the gas flow at the outlet. In some experiments
the saturation profile along the core is also measured. This is
usually done at times when the flow has stabilised and the
saturation profile is steady because the time taken to measure
the complete profile can be significant and inaccuracies will
result when the profile is changing. It is possible to measure
saturation profiles under dynamic conditions but the apparatus
to do this is more expensive,
A more complex form of this basic experiment is performed in
Elf when capillary pressure effects need to be taken into
account. A series of step changes called ‘bumps’ are made in
the injection rate (for waterfIoods) or pressure drop across the
core (for gasfloods). A bump is made after the flow has
stabilised at the previous level and there may be up to four
bumps during the experiment. The initial flow rate
corresponds to expected reservoir flow rates and the bumps
may increase the flow rate substantially – the maximum factor
between final and initial flow rate is generally less than 10.
The aim is to have a flow rate high enough to eliminate
capillary pressure effects by the end of the experiment if
possible. This form of experiment helps to decouple the effects
of capillary pressure from the relative permeability in the
calculations. An example of the measured data from a
waterflood experiment is shown in Fig. 1. This experiment
lasts for 17 days and has two bumps in the water injection rate.
The large changes that can occur following a bump makes the
matching problem more difficult than an experiment that uses
a single rate.
Mathematical formulation. The objective function used for
matching the measured data has a contribution from each of
the measured variables. The oil production term is of the form:
96
SPE 50576
The summation is over all the times at which
measurements are taken. Typically there will be between 20
and 40 times at which the oil production is measured
throughout the experiment.
There is a similar term JP that arises from the pressure drop
measurements in a waterflood. In a gasflood there is a term J~
that arises from the measurement of gas production flow.
The contribution from saturation profile measurements is a
little more complicated because the water or gas saturation at a
number of positions along the core is measured. Typically
between 10 and 20 saturation values may be measured. The
calculated results from the simulator are at discrete points
along the core, which in general do not coincide with the
points at which saturation is measured. It is necessary to
interpolate calculated values to produce values at the
measurement points. In Caropt we use linear interpolation.
The contribution to the objective function is of the form:
N
‘S ‘~3f(s~(xj)– Ssi”
~ (Xj))’
i=l Nx j=,
The overall objective function is the sum of the
contributions from each measured variable:
for a wateflood
J= JO+ JP+J$
J= JO+ J~+J~ for a gas$ood
J is a function of the parameters that define the relative
permeability and capillary pressure curves. The method used
to parameterise these curves is very important. The form of
parameterisation will determine the range of functional forms
for the curves and the number of parameters required will
determine the time required to solve the optimisation problem.
Ideally we would like a parameterisation that combined
economy in numbers of parameters with flexibility in the
range of functional forms.
“Parameterisation. In the simulation program Z2C, which
forms the basis of the simulation model in Caropt, the input
for relative permeability and capillary pressure curves are
tabulated values and any value on a curve is calculated by
interpolation from the tabulated values. It is natural to use
these tabulated values as the parametric description of the
curve. The original version of Z2C used linear interpolation but
this has a disadvantage that a large number of parameters are
required to accurately represent a relative permeability curve.
An alternative that has been put forward by other authors5’cis
to use spline interpolation which can represent curves
accurately with many fewer parameters. We found, however,
that when we used standard splines to interpolate tabulated
data from relative permeability curves, there could be
problems in regions where the curves have a high order
curvature - an example is shown in Flg 2. The spline can not
interpolate the tabulated data without a significant overshoot
SPE 50576 HISTORY MATCHING OF LABORATORY
that produces a non-physical curve, and in particular there is a
region where the spline becomes negative. We found a
solution to this problem through modifying the standard spline
fit to prevent overshoot.
The standard spline fitting aIgorithm (see for example
Numerical Recipess) that fits a curve y(x) to tabulated values
(Xi,yi), calculates values for the second derivative y at the
points xi. It can be shown that it
(Yi+l- Yi)l
y,” , y,+,” S2.0. I .-——.
-
(Xi+l-Xiy
then the sign of the derivative of the spline y’ does not change
in the interval (X,X+l). By changing the values of M to satisfy
this inequality we can ensure that there will be no overshoot.
We have found that in practice we can relax the condition so
that we use the value 6.0 for the coefficient rather than 2.0.
This value has been found to give satisfactory results for all
the relative permeability curves that we have investigated and
the relaxed condition gives smoother curves than the strict
condition. The effect of the modification is that the curve no
longer satisfies the spline condition of continuity of the first
derivative at the nodes where the modification is applied. For
smooth relative permeability curves that can be fitted
adequately by standard splines, the modification condition is
usually satisfied and no change is necessary, so the modified
spline is identical to the standard spline. The modified spline
tit to the relative permeability data is compared with the
standard spline in Fig. 2.
Capillary pressure curves are generated using linear
interpolation. The form of these curves is not so well suited to
spline interpolation and the accuracy of representation is not
such an important issue as for relative permeability curves.
To ensure monotonicity of the relative permeability and
capillary pressure curves we must impose constraints on the
parameters that define the curves. For the oil relative
permeability curve these constraints are of the form :
‘ro,i < ‘ro,i+l for ‘w,i 2 ‘w,i+l
There are similar constraints on the parameters defining
the water/gas relative permeability curves and the capillary
pressure curve.
Solution by optimisation
The best match to the data is found by minimizing the
objective function J with respect to the parameters defining the
relative permeability and capillary pressure curves. The
optimisation is constrained by the linear constraints on the
parameters. In Caropt the constrained optimisation is
performed by an optimisation package Npsol that was
developed at Stanford University. The objective function and
the gradients of the objective function with respect to the
parameters are calculated by Z2C simulations. A special
97
COREFLOODING EXPERIMENTS 3
version of Z2C using modified spline interpolation has been
developed for this purpose. The gradients are calculated by
finite difference and each parameter requires a separate Z2C
simulation. Almost all the time required to perform an
optimisation is spent performing the Z2C simulations.
To improve the performance of the optimisation we have
used both transformation and scaling. The variables xi in the
optimisation that correspond to the relative permeability
parameters are transformed from kr,i to (l/kr,i ) and the
variables corresponding to both the relative permeabilities and
the capillary pressures are scaled using the starting values xo,i:
Xj’= X,/Xo,i
Effective scaling requires a starting guess that is a good
approximation. This can be achieved by an approximate fit
using a simple functional form for the relative permeability
curves – an exponential form for example. We have other
programs that can be used to do this with sufficient accuracy.
Parameter estimation errors. When an optimum
solution has been found it is possible to calculate confidence
intervals for the optimum parameters. The uncertainty in the
actual value of the parameters is partly due to the experimental
errors and partly due to the error arising from the
approximations made in simulation model. We can estimate
the combined effect of these errors from the accuracy of the fit
that has been achieved. Using a linear approximation we can
then calculate the covariance matrix C for the parameters. The
confidence interval for a parameter q can then be calculated
from the formula:
Details of the derivation of this formula can be found in
Numerical Recipesg. Kerig and Watson10 describe the
application of error estimation to relative permeability
calculations.
Experiences with Caropt
Caropt has been in regular use for more than 6 months at our
coreflood laboratories. The conclusions that have been made
based on this experience areas follows:
(i) Caropt is able to calculate relative permeabilities
from the results of complex coreflood experiments
with bumps. These calculations were previously too
difficult to achieve by trial and error methods.
(ii) There is an improvement in the quality of the results
produced because many different hypotheses can be
tested when matching the experimental resuits.
(iii) The quantification of errors by Caropt identifies the
saturation band where the resolution of the reIative
permeability curves is weak. This can highlight
where further data may be useful in improving the
results.
4 A. E. FINCHAM, B. FERREOL SPE 50576

Caropt has been used on a variety of water flood and gasflood Conclusions
experiments and has proven to be very robust and efficient in 1. Caropt has proven to be a robust and efficient computer
solving all these problems. program for the analysis of the results from dynamic
coreflooding experiments.
Example problems 2. A modification to the standard cubic spline interpolation
We discuss here the solutions made with Caropt for two real has extended the range of functional forms for relative
experiments a gasflood and a water flood. In both these permeability curves that can be considered.
experiments capillary pressure effects proved to be important 3. The application of Caropt to actual experimental data has
and Caropt was successful in resolving these effects in the shown that predicted relative permeability curves are
simultaneous calculation of the relative permeabilities and changed significantly when capillary pressure effects are
capillary pressures. Independent measurements of capillary accounted for in the calculation.
pressure were made to substantiate the solutions found by —
Caropt. TABLE 1 - CORE PROPERTIES AND
OPERATING CONDITIONS FOR
EXAMPLE 1
Example 1 -gasflood. The basic data for this experiment is Core properties
shown in table 1. The experiment lasted for 17 days and there L. (rnd) 4680
was a bump at 170 hours. A previous experiment on this core 4 0.2
L (cm) 11.3
had given results that were not consistent with other results 19.48
A(cm2)
using less permeable samples from the same facies. It was Sm 0.166
suspected that this was due to capillary pressure effects. This Operating conditions
new experiment was performed with the same initial pressure .0178

--_._J.
Pg(cP)

drop as before but now with the addition of a bump. The data VO(CP) 12
dp initiaI (kpa) 5
from this new experiment were analysed with Caropt and the dp final (kpa) 15 -.. —
results are shown in Figs. 3-5. A first set of optimum relative
permeabilities found with Caropt was smoothed manually and
reoptimised with the relative permeability parameters fixed for
water saturations below 0.6. This gives a slightly worse fit
than the first optimum but with smoother curves, which are TABLE 2 - CORE PROPERTIES AND
generally preferred. Caropt predicts larger relative OPERATING CONDITIONS FOR
permeabilities than a standard JBN analysis at water EXAMPLE 2
saturations greater than 0.6 and also a smaller value for SO~ Core properties
hti (red) 656
(Fig. 3). The overall effect is to increase the value of k,~at So, 0.1802
@
by a factor of about three. These differences can be attributed L (cm) 19.26
to tie neglect of capillary pressure in the JBN analysis. The A(cm2) 19.53
capillary pressure curve calculated by Caropt compares well Sti 0.17
Operating conditions
with the capillary pressure measured by mercury injection on a 1.0428
Pw(cp)
nearby plug (Fig. 4), and the Caropt results are in agreement PO(CP) 2.05
with the trend defined by the experiments on other samples Q“j initial (cc/see) 0.1667
from the same facies. Qinjtinal(cc/see) 0.8333

Example2 -waterflood. The data for this experiment is

shown in table 2. The waterflood was done with synthetic oil
after nettability restoration. The experiment lasted for 17 days
and there were two bumps - the first at 250 hours and the
second at 267 hours. The imbibition capillary pressure was
measured by centrifuge on a nearby plug. The results from
Caropt are shown in Figs. 6-8. Caropt relative permeabilities
are generally greater than the relative permeabilities calculated
by the JBN method and also the Caropt SO,is smaller (Fig. 6);
again this can be attributed to capillary pressure effects. No
smoothing of the Caropt relative permeabilities was done in
this example. The shapes of the capillary pressure curves from
Caropt and centrifuge measurements compare well (Fig. 7),
but the plateau is different.

98
 SPE 50576 HISTORY MATCHING OF LABORATORYCOREFLOODING EXPERIMENTS 5

Nomenclature 24870 presented at the 1992 Annuat Technical Conference and

optimum parameter estimate
a = Exhibition, Washington,DC, Oct. 4-7
& = estimation error for a 5. Ucan,S., Civan,F., and Evans,R.D.: “Uniqueness and
Simultaneous Predictability of Relative Permeability and
C = covariance matrix in parameter estimation
Capillary Pressure by Discrete and Continuous Means,” JCPT,
dp = pressure drop across core (kPa) Apr. 1997,36,52
J = overall objective function 6. Kerig,P.D., and Watson, A.T.: “A New Algorithm for Estimating
Jg = contribution to J from gas flow terms Relative Perrneabilities from Displacement Experiments,”
Jo= contribution to J horn oil production terms SPERE, Feb. 1987,103
Jp = contribution to J from pressure drop terms 7. Bourbiaux,B., Chardaire,C., Chavent,G., Jaffie,J., and LIu,J.:
Js = contribution to J from saturation profile terms “Simultaneous Estimation of Relative Permeabilities and
kO~Wi= reference absolute permeability – measured with oil Capillary Pressure,” SPE paper 19680 presented at the 1989
Annual Technical Conference and Exhibition, San Antonio, Oct.
at initial water saturation
8-11
k;= relative permeability 8. Press,W.H. et al: Numerical Recipes in C, Cambridge University
L = length of core (cm) Press, Cambridge (1992), 113
NOb,= number of observations of produced oil 9. Press,W.H. et ak Numerical Recipes in C, Cambridge University
NP= number of observations of saturation profiles Press, Cambridge (1992), 695
N. = number of measurement points on a saturation 10.Kerig,P.D., and Watson, A.T.: “Relative-PerrneabiIity Estimation
profile from Displacement Experiments: An Error Analysis,” SPERE,
Qi.j = water injection rate (cclsec) Mar. 1986, 175
S = saturation
SO,= residual oil saturaton
S~i = initial water saturation S1Metric Conversion Factors
V.= volume of produced oil (cc) Cpxl.o E-03 = Pas
w = weighting factor in optimisation in x 2.54 E+OO= cm
~ = viscosity (cp) md x 9.869233 E-04 = pmz
~z = porosity cu in. x 1.638706 E+O1 = cc
XI = Chi square statistic for 1 degree of freedom psi x 6.894757 E+OO= kPa

Subscripts and superscripts

g = gas
o = oiI
S = saturation
w = water
obs = measured value
sim = simulated value

Acknowledgements
The authors would like to thank Elf Exploration Production
and Elf Exploration UK for permission to present this paper.
We also thank G.Hamon and R.Bissell of Elf, CSTJF for
useful discussions during the development of Caropt.

References
1. Johnson,E.F., Bossler,D.P., and Naumann,V.O.:“Calculationof
Relative Permeabilityfrom DisplacementExperiments,”Trans.
AIME (1959) 216,370
2. Declaud,J.P., Guy,M., Iffly,R., and Labastie,A.: “Effect of Flow
Rate and Nettability on Water-Oil Relative Permeabilities and
Capill~ Pressure,” SPE Paper 9236 presented at the 1980
Annual Technical Conference and Exhibition, Dallas, Sept. 21-
24
3. Brown,C.E., Gamble,I,J.A., and Heaviside,J.: “Relative
Permeability for Intermediate Nettability Reservoirs,” SPE
paper 16968 presented at the 1987 Annual Technical Conference
and Exhibition, Dallas, Sept. 27-30
4. Fasanino,G., Lee,R.L., and Ouenes,A.: “Simulated Annealing for
Interpreting Gas/Water Laboratory Corefloods,” SPE paper

99
6 A.E.FINCHAM, B.FERREOL SPE 50576

0.15-
’80, 1- 50
> 0.13-

k 0.07-
a
g 0.05-
~ 0.03-
K 0.01.

4“0’ 0.4 0.5 0.6 0.7 0.8

Water saturation

Fig 1.- Experimental data measured in a waterflood experiment.
There are 25 pairs of measurements during the experiment. Flow
bumbs occur at 15000 and 16000 reins after the start of the
experiment.
Fig. 2- Comparison of the modified spline and standard spline
fit to parameter values representing an oil relative permeability
curve. The standard spline overshoots and becomes negative.
The modified spline decreases steadily with increasing water
saturation.

.-x
I.00E+OO

I.00E-01
1
●
●

I
.* 1.00E-02
.=

~ 1.00E-05
$

1 .00E-06
4
0 0.2 0.4 0.6 0.8 1
1.00E-07 -~---- —z.---——-T------- =
o 0.2 0.4 0.6 0.8 1
Gassaturation
,— .—-..-..
Gas saturation

L— —,

Fig. 3- Relative permeabilities calculated by Caropt ( solid line) Fig. 4 - Comparison of capillary pressures calculated by
compared to curves calculated using the JBN method (dotted Caropt with measurements using mercury injection and
line) for example 1 corrected for IFT

(a) oil production - cumulative (b) gas production

1 10 100 1000 10000100000 1 10 100 1000 i~ Im

time (reins) tim (m”ns)
—. ,.. .. . —..—.—.
l—s~mulated ■ Experimental ~ I—simulat~ = w~~
i. ..—.-. .—-— .-—

Fig. 5- Two graphs showing the final fit of the simulated oil and gas production data to the experimentally measured data.
100
SPE 5G576 HISTORY MATCHING OF LABORATORYCOREFLOODING EXPERIMENTS 7

l.ooEm -

1.00E-01 - -..

7.~E-02 -

1.WE-OS -

1.COE-C4-

1.OEa - ,0
,’
,‘ .- Water saturation
l.~E-00 i , .—
0 0.2 0.4 0,6 0.8 1 .. ——. —— —.—
Water Saturation

Fig. 6- Relative permeabilities calculated by Caropt ( solid Fig. 7- Comparison of capillary pressures calculated
line) compared to curves calculated using the JBN method by Caropt with measurements using centrifuge
(dotted line) for example 2 corrected for [FT.

(a) oil production - cumulative

w,
80
1
60 j

40-1

0
1 100 IOQOO

Time (reins)
.._ Time(mine)
—
/— Simlated x Hperimnta~
—. —sirruldd X @mtd

Fig. 8- Two graphs showing the final fit of the simulated oil production and preaaure drop results to the experimentally
measured data.

101