Physica A 314 (2002) 103108

www.elsevier.com/locate/physa
Using percolation theory to predict
oil eld performance
P.R. King
a;
, S.V. Buldyrev
b
, N.V. Dokholyan
b; c
, S. Havlin
d
,
E. Lopez
b
, G. Paul
b
, H.E. Stanley
b
a
Department of Earth Science and Engineering, Imperial College, Prince Consort Road,
SW7 2BP, London, UK
b
Center for Polymer Studies, Boston University, Boston, MA 02215, USA
c
Department of Chemistry and Chemical Biology, Harvard, Cambridge, MA 02138, USA
d
Minerva Center & Department of Physics, Bar-Ilan University, Ramat Gan, Israel
Abstract
In this paper, we apply scaling laws from percolation theory to the problem of estimating
the time for a uid injected into an oileld to breakthrough into a production well. The main
contribution is to show that when these previously published results are used on realistic data
they are in good agreement with results calculated in a more conventional way but they can
be obtained signicantly more quickly. As a result they may be used in practical engineering
circumstances and aid decision making for real eld problems.
c 2002 Elsevier Science B.V. All rights reserved.
PACS: 64.60.A
Keywords: Percolation; Scaling; Oil recovery; Flow in porous media
1. Introduction
Oil reservoirs are extremely complex containing geological structures on all length
scales. These heterogeneities have a signicant impact on hydrocarbon recovery. The
conventional approach to estimating recovery is to build a detailed geological model (of
around 10 million numerical grid cells), populate it with ow properties, coarse grain it
and then perform a ow simulation. In order to estimate the uncertainty in production a
number of possible geological realisations are constructed (with associated probabilities)

Corresponding author. Fax: +44-20-759-47444.

E-mail address: [email protected] (P.R. King).
0378-4371/02/$ - see front matter c 2002 Elsevier Science B.V. All rights reserved.
PII: S 0378- 4371( 02) 01088- 9
104 P.R. King et al. / Physica A 314 (2002) 103108
and this procedure repeated many times. A simple order of magnitude estimate of com-
puting times (given todays model sizes and computing speeds) indicates that this could
take many hundreds of days. Clearly this is completely impractical for many purposes.
Given this practical limitation a number of approaches have been taken, for example
improved coarse graining methods [1,2], fast simulation [3,4] and so on. In this paper
we adopt a dierent perspective. We simplify the geological model and ow physics
such that quasi-analytical predictions of uncertainty can be made extremely quickly.
The advantage is that the eects of the complex geometry which inuence the ow
can be readily estimated. Clearly the disadvantage is that much of the ow physics and
subtleties of the heterogeneity distribution are missed. Whilst it is the aim of future
research to address those issues we show, in this paper, that this simple model can
already give reasonable estimates of the production performance when applied to a real
data set.
We start by simplifying the rock heterogeneity by assuming that the permeability
can be split into good rock (i.e., nite, non-zero permeability) and poor rock (low
or zero permeability). For all practical purposes the ow takes place just in the good
rock. It is the interconnectivity of the permeable rock that controls the ow. The spatial
distribution of the sand is also governed by the geological process but can frequently
be considered as independent or of a short range correlation. Hence, the problem of the
connectivity of the sandbodies is precisely a continuum percolation problem. The place
of the occupancy probability p of percolation theory is taken by the volume fraction
of good sand (the net to gross ratio). This percolation view of sandbody connectivity
has been used before [5] but here we look not just at the static connectivity but also
at the dynamic displacement on this percolating system.
The second simplication is of the ow physics. Here, we shall assume that the
displacement is like passive tracer transport. In other words we have single phase ow
from injector to producer (we only consider a single well pair) and we assume that
the injected uid is passively convected along these streamlines. To be specic, we
shall consider the time to breakthrough (or the rst passage time for a passive tracer)
as the measure of performance. These are gross simplications which enable us to use
the scaling laws of percolation theory [6] to determine production performance and its
associated uncertainty.
2. Flow model
To simplify the model we shall assume that the permeability is either zero (shale)
or one (sand). The sandbodies are cuboidal. They are distributed independently and
randomly (i.e., as a Poisson process) in space to a volume fraction of p. Further, we
shall assume that the displacing uid has the same viscosity and density as the displaced
uid. This has the advantage that as the injected uid displaces the oil the pressure eld
is unchanged. This pressure eld is determined by the solution of the single phase ow
equations ((KP) =0). The injected ow then just follows the streamlines (normals
to the isobars, pressure is P) of this ow. In dimensionless units the permeability (K)
is either zero or one as described before. The boundary conditions are xed pressure
P.R. King et al. / Physica A 314 (2002) 103108 105
of +1 at the injection well and 0 at the production well. In this work, we shall only
consider a single well pair separated by a Euclidean distance r. The breakthrough time
then corresponds to the rst passage time for transport between the injector and the
producer.
For a given model of the reservoir, we can then sample for dierent realisations
of the locations of the wells (or equivalently for the same well locations for dierent
models of the reservoir with the same underlying statistics) and plot the distribution
of breakthrough times. This is the conditional probability that the breakthrough time is
t
br
given that the reservoir size (measured in dimensionless units of sandbody length)
is L and the net to gross is p, i.e., P(t
br
|r; L; p). In previous studies [7,8] we have
shown that this distribution obeys the following scaling:
P(t
br
|r; L; p)

1
r
d
t

t
br
r
d
t

g
t
f
1

t
br
r
d
t

f
2

t
br
L
d
t

f
3

t
br

d
t

; (1)
f
1
(X) = exp (ax

) ;
f
2
(X) = exp (bx

) ;
f
3
(X) = exp (cx

) :
Currently the best estimates of the various coecients and powers (as found from
detailed computer simulations on lattices and theory, see Andrade et al. 2000) in this
are:
d
t
= 1:33 0:05; g
t
= 1:90 0:03; a = 1:1; b = 5:0; c = 1:6(pp
c
)2:6(pp
c
) ;
= 3:0; = 3:0; = 1:0 and =|p p
c
|

= 4=3; p
c
= 0:668 0:003
(for continuum percolation):
In this paper, we will not discuss the background to this scaling relationship but con-
centrate on how well it succeeds in predicting the breakthrough time for a realistic
permeability eld. However, it is worth spending some time describing the motivation
behind the form of the various functions. The rst expression (f
1
) is an extension to
the expression developed by others (see [6] for a detailed discussion) for the short-
est path length in a percolating cluster between two points. The breakthrough time is
strongly correlated with the shortest path length (or chemical path).
To this there are some corrections for real systems. In a nite size system very large
excursions of the streamlines are not permitted because of the boundaries so there is a
maximum length permitted (and also a maximum to the minimum transit time). This
cut-o is given by the expression f
2
. Away from the percolation threshold the clusters
of connected bodies have a typical size (given by the percolation correlation length,
) which also truncates the excursion of the streamlines. This leads to the cut-o
given by the expression f
3
. The multiplication together of these three expressions is an
assumption that has been tested by Dokholyan et al. [7]. Also a more detailed derivation
of this form is given there and the references therein. Here, we shall concentrate on
106 P.R. King et al. / Physica A 314 (2002) 103108
using this scaling form to make predictions about the distribution of breakthrough times
for a realistic data set.
3. Application to a real eld
We took as an example a deep water turbidite reservoir. The eld is approximately
10 km long by 1:5 km wide by 150 m thick. The turbidite channels, which make up
most of the net pay (permeable sand) in the reservoir, are typically 8 km long by
200 m wide by 15 m thick. These channels have their long axes aligned with that of
the reservoir. The net to gross ratio (percolation occupancy probability, p) is 50%. The
typical well spacing was around 1:5 km either aligned or perpendicular to the long axis
of the eld. In order to account for the anisotropy in the shape of the sand bodies and
the eld we rst make all length units dimensionless by scaling with the dimension
of the sand body in the appropriate direction (so the eld dimensions are then L
x
; L
y
and L
z
in the appropriate directions). Then scaling law, Eq. (1), can be applied with
the minimum of these three values (L =min(L
x
; L
y
; L
z
)). The validity of using just the
minimum length has been previously tested [9].
The real eld is rather more complex than this, and a more realistic reservoir de-
scription was made and put into a conventional ow simulator. We could then enter
these dimensions into the scaling formula, Eq. (1). It should be noted that rst the di-
mensionless units were converted into real eld units to compare with the conventional
simulation results. Using these data we nd breakthrough times of around 1 year. The
full probability distribution of breakthrough times from the scaling law is given by the
solid curve in Fig. 1.
In addition conventional numerical simulations were carried out for the eld. We
could then collect the statistics for breakthrough times for the various well pairs to
compare with this theoretical prediction. Not all pairs exhibited breakthrough in the
3 1
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0 2 4
Breakthrough Time (years)
F
r
e
q
u
e
n
c
y
Fig. 1. Comparison of probability distribution of breakthrough times for example reservoir obtained from
percolation theory (smooth curve) and from full eld model (histogram).
P.R. King et al. / Physica A 314 (2002) 103108 107
timescale over which the simulations were run and there were only three injectors so
there were only 9 samples. The histogram of breakthrough times is also shown in Fig. 1.
Clearly with such a small sample these results cannot be taken as conclusive however,
certainly they are indicative that the percolation prediction from the simple model is
consistent with the results of the numerical simulation of the more complex reservoir
model. The agreement with the predictions is certainly good enough for engineering
purposes. We would hope that if the simulation had been run for longer and more well
pairs had broken through that better statistics could have been collected. The main
point being that the scaling predictions took a fraction of a second of cpu time (and
could be carried out on a simple spreadsheet) compared with the hours required for
the conventional simulation approach. This makes this a practical tool to be used for
making engineering and management decisions.
4. Post breakthrough behaviour
So far we have only discussed the time to breakthrough. However, it is also important
to know how the oil rate declines once breakthrough has occurred. We shall study
this for only a simple system. First we consider the homogeneous case (p = 1). If
we consider two wells in an innite system then we simple need to solve Laplaces
equation (
2
p=0) with the boundary conditions that the pressure is + at the injector
well (placed at (x; y) coordinates (r=2; 0)) and at the producer (at (r=2; 0)).
Strictly speaking we should account for the nite wellbore diameters and pressures,
but this is a minor correction. Also we assume that the wells operate at constant
pressure. We could also use constant rate boundary conditions but this does not alter
the essential results.
We can then calculate the entire pressure eld analytically, either by using conformal
maps or by making the simple change of variables x=r=2 sinh =(cosh cos ); y=
r=2sin =(cosh cos ), then pressure is associated with the coordinate and the
streamfunction with the coordinate . This enables us to calculate the transit time
along streamline which is
t
br
( ) =
r
2
4K
[1 + ( )cot ]
sin
2

(where K is the permeability of the eld). Asymptotically this implies that

t
br

2
r
2
K( )
3
as :
The oil production rate is proportional to which implies that asymptotically the
production rate declines like V(t) (r
2
=t)
1=3
. This is for an innite system. We would
expect this rate to decay exponentially when the streamlines see the boundaries of a
nite system.
This is the case for a homogeneous reservoir. For the percolating system we expect
to see rather more complex behaviour inuenced by the nite boundaries or by the
nite cluster sizes away from threshold. At the percolation threshold, we expect the
asymptotic decay of the oil production rate to be a power law, but with a dierent
108 P.R. King et al. / Physica A 314 (2002) 103108
power from the
1
3
found for the homogeneous case. So for p =p
c
we conjecture the
following asymptotic decay:
V(t )

r
d
B
t

:
Here, d
B
is the fractal dimension of the backbone of the percolating cluster (d
B
=
1:6432 0:0008 [10]). The new exponent is found to be 0:63 0:05 from numerical
simulations. Hence percolation theory is able to give us information, not just about
breakthrough times, but also post breakthrough behaviour.
5. Conclusions
We have applied results obtained earlier for the scaling law for breakthrough time
distributions for oileld recovery to realistic eld data. We have shown that by making
a number of simplifying assumptions we can readily use previous results from per-
colation theory to make extremely rapid estimates of the uncertainty in breakthrough
time. The agreement between the theory and the conventional simulation approach is
accurate enough for engineering purposes and therefore makes it a practical tool for
supporting decision making.
Acknowledgements
The authors thank L Gantmacher who did the calculations of postbreakthrough be-
haviour for the homogeneous case. They would also like to thank J. Andrade for helpful
discussions and BP for nancial support and permission to publish this paper.
References
[1] P.R. King, The use of renormalisation for calculating eective permeability, Trans. Porous Media 4
(1989) 3758.
[2] P.R. King, A.H. Muggeridge, W.G. Price, Renormalisation calculations of immiscible ow, Trans.
Porous Media 12 (1993) 237260.
[3] F. Bratvedt, K. Bratvedt, C.F. Buchholz, L. Holden, H. Holden, N.H. Risebro, A new front tracking
method for reservoir simulation, SPERE 19805, 1992.
[4] M.R. Thiele, M.J. Blunt, F.M. Orr, Modeling Flow in Heterogeneous Media using Streamtubes Part 1
(299339) & Part 2 (367391), In Situ 19 (1995).
[5] P.R. King, The connectivity and conductivity of overlapping sandbodies, in: A.T. Buller, E. Berg,
O. Hjelmeland, J. Kleppe, O. Torsaeter, J.O. Aasen (Eds.), North Sea Oil and Gas Reservoirs II,
Graham & Trotman, London, 1990, pp. 353358.
[6] S. Havlin, D. Ben-Avrahim, Diusion in disordered media, Adv. Phys. 36 (1987) 695798.
[7] N.V. Dokholyan, Y. Lee, S.V. Buldyrev, S. Havlin, P.R. King, H.E. Stanley, Scaling of the distribution
of shortest paths in percolation, J. Stat. Phys. 93 (1998) 603613.
[8] Y. Lee, J.S. Andrade, S.V. Buldyrev, N.V. Dokholyan, S. Havlin, P.R. King, G. Paul, H.E. Stanley,
Traveling time and traveling length in critical percolation clusters, Phys. Rev. E 60 (1999) 34253428.
[9] J.S. Andrade, S.V. Buldyrev, N.V. Dokholyan, S. Havlin, Y. Lee, P.R. King, G. Paul, H.E. Stanley,
Flow between two sites in percolation systems, Phys. Rev. E 62 (2000) 82708281.
[10] P. Grassberger, Physica A 262 (1999) 251.