Vous êtes sur la page 1sur 9

SPE 26978

Statistical Secondary Recovery Model


Roberto R. Barbieri, SPE, Amoco Argentina Oil Co.

Pwmlsslan to copy h rwttieW

* an kfmet of no mom than 200 w0rd8. Wstmtbn*

WY WI k COPW. Tho bstrael should eontaln eonsplamus cknwlc8$mont M wlmm

nd whom Uw papof was

P-M.

Writ.LlbrWlm,
SPE, P.0, Sat

S3363S,Rlchmdson,

TX 750U3S3S, USA, fax 01-2$4-9S2-S435.

Abstract
A secondaryrecovery model has been developed to predict

waterflood performance for different reservoir properties and


design conditions. A causal model based on simple and
multiple regression equations uses eight input variables to
estimate injeetion rate, ultimate secondary reserves, response
time and yields peak oil rate, peak year and the production
profile as a hmction of time. It is used for seeondary reserve
booking, to develop production profiles for project economics
and authorization for expenditures and to assess technical risk
by means of the simulation teehnique.
Introduction
Secondary reeoveryby water injection plays an important role

in the San Jorge Basin, where Amoco injects 32 Mm3wpd in


18 waterflood projects.
Performance prediction during design and the monitoring
of these floods have been based on different semiempirical
models. This model type has been employed because the
characteristics of the reservoirs and the available data
precluded the use of any theoretical approach. However, 20
additional projects were identified and a more reliable
prediction model was required. After the analysis of
geometrical data, fluid and rock properties and production
and injection rates, several models were evaluated, which
included analogical, theoretical and semiempirieal models,
with the latter being the most reliable.
This novel approach relies on historical data from 12
wateffloods Icwatedin the San Jorge Basin, but can be used in
other areas once the regression coetllcients for the particular
reservoirs are estimated. This causal statistical model predicts
waterflmxt performance for different reservoir properties and
design conditions, with a set of equations developed using
simple and multiple regression, They were developed by

44

omitting those variables with no significant effect, estimating


the regression eoeftlcients, finding the most effective
prediction equations and determining their strength by
correlation analysis.
The model employs six geometrical factors and two
reservoir quality parameters to generate five output variables.
Input variables are reservoir depth, total net sand thickness,
pore volume, number of sand layers, number of injectors and
producers, porosity and primary recovery factor. Output
variables are injectivity, secondary recovery factor, response
time, project life and recovered resemes after injecting 28
percent of the required number of pore volumes (R28).
By applying these output variables we estimate the injection
rate, the secondaryrecovery, the number of pore volumes to
inject to recover the ultimate secondary reserves, the peak oil
rate and year and the production profile as a function of time
(Figure 1).
The Procedure
A mathematical model can be defined as a symbolic

representation or abstraction of reality, and as such is a


function ~f the variables and the time period involved, the
solution methodology and the functional objective. It may be
short, medium or long term; deterministic or probabilistic and
for optimization or simulation. In our case it is a long term
causal model, that uses several simple and multiple regression
equations, developed for forecasting purposes.
The development process omits those variables with no
significant effect, evaluates the cause-effect relationship
between variables, estimates the regression coefficients, and
selects the most etkctive prediction equations Subsequently,
correlation analysis determines the relationships strength.
A regression equation relates a dependent variable with one
or more independent variables by means of regression
coefficients, determined by the minimum squares technique.
But the equation does not explain, in general, everything
about the variability in the dependent variable and therefore
an error exists. The cause for this may be an incomplete
knowledge of the variables, random effects or improper
equation selection. This error, known as Standard Error of
Estimate (Sy), is calculated as the positive square root of the
variance about the regression line (Sy2). This factor leads to
the preelection of a candidate equation with tit inversely

SPE Advanced Technology

Series, Vol. 4, No. I

proportional to Sye
Once the relationship between the dependent and
independent variables has been determined, the second step
involves measuring the degree to which the variables are
related. This is done by means of correlation analysis using
the dimensionless Determination Coetlicient (R2), which is
the proportion of the dependent variable variance explained
by the regression equation. Although a high R2 is used
extensively it should not be used alone because the variance
about the regression line may also be high.
Another ve~ important tool employed is Prediction Sum of
Squares (PRESS), which indicates the prediction error for one
data point when it is not used to generate the equation.
PRESS should be as small as possible to obtain equations
with good forecasting ability.
One problem in multiple regression is the effect called
multicolinearity - the correlation that exists among the
independent variables. This is undesirable in prediction
equations because if multicolinearity is high the regression
wfflcients are less reliable. The statistical parameter that
addresses this concern is DET, a global multicolinearity
indicator calculated as the correlation matrix determinant,
with 0.1 recommended as a minimum. To determine the
spxific independent variable that causes the problem the
statistical parameter Variable Inflation Factor (VIF) is
calculated. When VIF is greater than 10 the variable
inclusion is evaluated.
One important consideration during the equation
development is the number of independent variables to be
included, The Prudence Principle stated by the statistician
Maurice Kendall indicates that the smaller this number the
better the model results, but it is also important not to dkard
valuable information. The statistical tool that takes this effect
into account is called Cp, which is a function of the number
of variables in the equation and of the variance about the
regression line (Sy2). It is recommended that Cp be similar to
the number of variables and, if possible less than five times
that number.
Regression and correlation analyses seek prediction
equations with high R2, low Sy2, low PRESS, DET greater
than 0.1, VIF less than 10 and Cp no greater than five times
the number of independent variables.

taken into account by means of primary recovery factor


(PRF), which is comprised of several variables such as the
porosity, water and oil saturations, reservoir temperature and
pressure, absolute and relative permeabilities, fluid properties,
and other less significant variables. For this reason PRF is the
suitable quality indicator of our multizone heterogeneous
reservoirs.

The Equations

This equation indicates logical aspects of seconda~ recovery.


Recalling the heterogeneity of the reservoirs, SRF is greater
when the number of injectors to producers increases and
when the ratio of injectors to pore volume is greater (both
geometrical ratios). Also SRF increases when the reservoir
quality increases, as indicated by porosity and PRF.

The model employs eight input variables, six related to


project geometry and two to reservoir quality. These variables
lead to six equations and the suitable cumulative production
profile as a fimction of the cumulative water injected in the
reservoir. The selection is done from a sample of ten
different profiles correlated using simple regression.
The geometrical variables are the number of injectors,
number of producers, pore volume (rock volume times
porosity), average reservoir depth, total net sand thickness
and number of sand layers to flood. The reservoir quality is

SPE Advanced Technology

Series, Vol. 4, No. I

Injectivity. The injectivity (1) prediction equation depends


on reservoir depth and the ratio between total net sand
thickness and pore volume.
The equation is a second degree polynomial for depth and
first degree polynomial for total net sand thickness per pore
volume. The reason for the second degree polynomial is that
the injectivity trend, based on the results of step rate tests,
indicates increasing injectivities to a maximum, and then a
decreasing trend as a function of reservoir depth. This effect
is accurately correlated by a second degree polynomial, as
indicated by the following formula with the regression
coefficients listed in Table 1. With injectivity (I) expressed as
injection rate px pore volume (Q/PV) and I proportional to
total net sand thickness per pore volume (H/PV) it follows
that I is proportional to net sand thickness.
I =a +b(D/1000) +c(D/1000)2 +d (1OOOIVPV)
..............(1)
b>()

c<(-)

d>()

Secondary Recovery Factor. Secondary recovery factor


(SRF) is a function of reservoir quality and geometrical
factors. Independent variables are the ratio of injectors to
producers, the ratio of injectors to pore volume, porosity and
PRF. The multiple regression equation for SRF is given by:
Iw

Iw
SW

a+b .--. + c
P

1000 -..-+

d POROSITY+ e PRF

Pv
.,...,.........(2)

b~

C>()

d>()

e>o

ResponseTime. The time necessary for the flood to respond


(RT), after water injection initiation, has been correlated with
the average -net sand thickness (IUS) and with the primary

45

recovery factor (PRF) using multiple regression. The equation


for RT is inctkated below.
H
RT=a+b----+cpRF

correlation. This variable has been correlated using multiple


regression, as indicated by the following formula.
R28 = a+ b(H/S)+c(IW/P)

+ d(D/1000)

...............(5)

...............(3)

The equation indicates the greater the average sand thickness


and the higher the reservoir quality (PRF), the lower the
response time.
We will see later that when average net sand thickness
increases, the recovery process is less efficient, project life
increases and recovery for a given number of pore volumes
injected decreases. This eikct is not consistent with the
response time equation for homogeneous reservoirs, but in the
observed heterogeneous reservoirs viscous fingering and
stripping effkcts increase the time required to recover the
ultimate secondary reserves<
Project Life Project life (LIFE) has been correlated with
PRF and porosity, and the geometrical factors reservoir depth
and average net sand thickness. The equation, using multiple
regression is:
LIFE = a + b PRF + c (D/1000) + d (H/S)2 + e POROS17Y
...............(4)
b<()

c<()

d~

e<t)

This equation indicates reserves are recovered faster and


therefore the project life is shortened when reservoir quality is
high, i.e. high PRF and POROSITY. Obsemed reservoir
performance indicates that re.cove~ is more efficient for
deeper sands yielding a shorter LIFE and, as previously
shown a higher average sand thickness results in a less
efllcient recovery and increases LIFE.
Reserves for 28% of the Number of Pore Volumes to
Inject (RX).
Cumulative production profiles are not

dependent on the waterflood geographical location. As shown


in Figures 2 and 3, the eurrudative production profiles of
different areas may have similar behavior. The key parameter
that quantifies the differences is R28. It is the fraction of
ultimate secondruy reserves recovered when water injection
(after RT) amounts to 28XOof the total number of pore
volumes to inject, as shown in Figure 4. Twenty eight percent
was selected for two reasons, First, the 12 waterflood of this
study have already attained 28Aand are supported by reliable
data, Second, at 28%..suffkient differencesexist between the
cumulative production profiles to yield a more accurate

46

Formula 5 indicates that the greater the ratio H/S the smaller
the resewes recovered for 28Aof the number of pore volumes
to inject, which is consistent with the project life equation
(Eq.4). When average net sand thickness is greater, the
recovery is less eflicient, decreasing R28 and increasing
project life.
The effect of the ratio of injectors to producers on R28 is
evident as well as the proportionality of R28 to resemoir
depth
Cumulative Production Profile Models, Ten different
profiles have been developed for this Statistical Secondary
Recovery Model, each one characterized by the R28
parameter. The projects are VH H, VH IV, VH IX, AG Bl,
AG B17, CG I, CG IIW, CG V, CD IV and Z I. The
cumulative production profiles have been correlated with the
number of pore volumes of water injected (after RT) using
simple regression equations. This number has been
normalized with the total number of pore volumes to inject
(PVlt) necessary to recover the ultimate secondmy reserves
(SRt). These profiles area fimction of the unknown variables
related to rock and fluid properties, hence the need to identifjf
the most suitable profile. All the equations have the same
structure shown in Eq.6, but with different coefficients and
exponents (Table 1). The excdlent correlation of this
polynomial for the case shown in Figure 5, is found in all
cases.

5
SR
.-.. a+xbi
SRt
i=l

(l+ !!!.)ri
PvIt

.,............(6)

The importance of R28 and the use of the correct cumulative


production profile model is depicted in Figure 6, which shows
the expected end results for the same input data. All the
projects shown have the same values for input data,
independent of R28. Ultimate swondmy reserves for these
projects amount to 901 Mm30, but present values (at 15
percent discount rate) are 310, 414, 465 and 533 Mm30 for
profile models with R28s equal to 31, 55, 61 and 72
respectively. It is evident that an improper profile sdection
may result in inaccurate project economics.
Calculation Steps

The use of this model to simulate waterflood performance is

SPE Advanced Technology Series, Vol. 4, No, 1

straightforward, The steps are as follows:


1,

Estimate Injectivity and multiply by pore volume to


calculate the injection rate.

2.

Estimate the ResponseTime (RT).

3.

Estimate Project Life, deduct response time, multiply


by the injection rate and divide by the pore volume to
calculate the total number of pore volumes to inject
(PvIt), necessary to recover the ultimate secondary
reserves (SRt), up to the economic limit.

4.

Estimate R28.

5.

Estimate the Secondary Reeovery Factor and


multiply by the pore volume to determine the ultimate
secondmy resaves (SRt).

6.

For each time period multiply the injection rate by the


number of years (after RT) and divide by the pore
volume, to estimate the number of pore volumes of
water injected (WI).

7.

For each time period determine the ratio between pore


volumes of water injected and total pore volumes to
inject (PvI/PVIt) using PVI determined in Step 6 and
PVIt in Step 3.

8.

With R28 select the cumulative production profile


model to be used and with the calculated PWPVIt
estimate the ratio of recovered to ultimate secondaV
reserves.

9.

With the ratio from Step 8 and the ultimate secondary


reserves (SRt) from Step 5, calculate the cumulative
secondary production for each time period.

10.

Calculate yearly production by difference between


cumulative productions for year n and year n-1.

11.

Calculate average
production by 365.

daily

rate

dividing

yearly

project scheduling and to predict water requirements


It is useful to investigate the effect of input variables, some
related to reservoir quality (uncontrollable) and others to
project design (controllable). The controllable variables such
as number of injectors, number of producers, sands to be
flooded and injection rate can be selected to optimize
secondary recovery and production profiles.
Uncontrollable variables are generally stochastic and known
as ranges or as frequency distributions. With uncontrollable
variables, it is recommended to use probabilistic simulation,
using for example Monte Carlo or Latin Hypercube sampling
techniques, to develop secondmy recovery and production
profile estimates. The model has been loaded into a user
friendly spreadsheet suitable for probabilistic simulation,
enhancing its applicability scope. Technical risk can then be
assessed using different approaches from a simple sensitivity
analysis to the sophisticated simulation technique.
Figures 7 and 8 depict ultimate secondary reserves and the
production profile for a 20 year waterflood, estimated using
the Monte Carlo sampling technique and 1500 simulation
passes. Triangular and uniform distributions were used for
number of injectors and producers, both dependent on the
pore volume sampled on each pass, with a 0.8 correlation
coefficient. Truncated lognormal distributions were employed
for pore volume, depth and prima~ recovery factor and
truncated normal for porosity and reservoir thickness. The
number of sands follows a uniform distribution dependent on
the pore volume, with a correlation coefficient equal to 1.0.
Conclusions
A Statistical Secondary Recovery Model, developed using

regression and correlation analyses, proved to be the most


reliable approach to predict waterflood performance in the
heterogeneous reservoirs of the San Jorge Basin. This model
type was seleeted because the characteristics of the reservoirs
and the available data precluded the use of any theoretical
model. Also, the analogical approaches were not reliable. The
model can be used for secondwy reserve booking, to develop
production profiles for project eeonomics and capital
budgeting, to predict water requirements and to assess
technical risk using the simulation technique. It is also usetid
to evaluate the effects of reservoir quality variables
(uncontrollable) and to investigate the effect of geometrical
and operational variables (controllable).

Applications

This model yields a waterflood forecast as do theoretical or


empirieal models. However, the key difference is that it has
been tailored for heterogeneous resemoirs, with incomplete
input data availability, that precludes the use of theoretical
approaches.
Our company uses this model for booking secondary
reserves, to predict performance, to evaluate improvements in
wateffloods already installed, for capital budgeting, for

SPE Advanced Technology

Series, Vol. 4, No. 1

Nomenclature
= injectivity, m3wpd/Mm3 of pore volume
I

D=
H=
s

Pv

[w

average reservoir depth, meters below sea


level
total net sand, meters
number of sand layers
pore volume (rock volume times porosity),
Mm3
number of injection wells

47

number of production wells


POROSITY = porosity, percentage
PRF = primary recove~ factor, m30/m3 of pore
volume
LIFE = project life since start of water injection,
number of years
SR
= cumulative secondmy production, Mm30
SRt = ultimate secondary reserves, Mm30
SRF = secondary recovery factor, m30/m3 of pore
volume
PVI = number of pore volumes injected
PVIt = number of pore volumes required to recover
the ultimate secondmy reserves
RT = response time, number of years
PY = peak year, number of years
CE = contract end, number of years
PR = peak rate, m30pd
PCE = contract end production, m30pd
PEL = economic limit production, m30pd
= water injection flowrate, m3wpd
Q
ri
= exponents
bi
= polynomial coefficients
P

Technological Institute, He holds a BS degree in Chemical


Engineering and an MBA in Financial Management, both from
the Buenos Akes University.

Acknowledgments
1 thank Amoco Argentina for permission to publish this

paper, Special appreciation is expressed to professor Roberto


M. Garcia for the modeling assistance, to Jorge L. Mustoni
and the Comodoro Rivadavia staff for their valuable input and
to Dave A, Boatwright and Lynn C. Gregor for reviewing the
manuscript.
References
1. Amoco:Statistical Secon&uy Recovev Model (Volume I),
Comodoro Rivadavia,1992.
2. Croxton F.E., Crowden D.J. and Kleinss S.: Applied Genem/
Statistics, PrenticeHall Inc., Englewood Cliffs, USA, 1967.
3. Garcia R. M. and Mennoz O. L. : Amilisis de Regresir5n
Mtiltiple, Universidad de Buenos Aires, 1992.
4. Walpole R. E. and Myers R. H. : Probabilidad y Estadistica,
McGraw Hill, M4xico, 1992.
5. Wonnacott T.H. and Wonnacott R,J.: Regression ; A Second
Conrse in Sfafisflcs, John Wiley& Sons, New York, 1981.
S1 Metric Conversion

ft X3.048
bblx 1.589873

Factors

E-01 = m
E-01= m3

Roberto R. Barbieri is an engineer for Amoco Argentina Oil


Co. His career began in 1980 as facilities engineer and later
he moved into operations, reservoir and economics. He is a
professor in Project Economics at the Buenos Aires

48

SPE Advanced Technology Series, Vol. 4, No. I

Table 1-Regms&on Ccefficienta and Statistical lndicalorx


Cceffiiient
injectivity
,8
b

Max

1.09620463E+O0
1

1.332137 B9E-02

Coefficient
$kccmdw

-1.42380910E-01
-6,W083898E-O

hfh

Power

(I)

Max I

Coefficient

Power

Response
Time

[RT)

2.07234644E+O0

-4.74O5B7O4E-O2

7,370 11482E-02

0,143

0,889

-1.17151362E-01

2.50

6.86

0.424

1.301

5,420

-1.32193409E+01

Q,039

0.092

1.019

39,187

1
I

0.620

4.798O2453E-O2
2.5081 6518E-03

16.8

25.3

4.017501169E-OZ

0.039

0.108
l,~oE_ol

1.50E-03

1.000

0.189

4.0

5.0

3.0

IQ
Sy

0.938

0,997

0.923

0.03

000

0.04
R?8

Pmjeet Life (llFE)


a
b

6.77500593E+01
-1.35523700E+02

0,039

0.108

2,36269249E+01
-3.794 17926E+O0

2.50

6,86

-1. B4977658E+01

0.424

1.170

5.93377002E+01

0.100

0.889

2.83793233E+O0

2,00

6.86

1.14950301

0.424

1.301

-6.594 B2746E-01

16.8

25.3

0.290

Rz

0,878

M
h4
M

4.0
0.892

5.41
Coefficient

hl
b~

0.574

5.0

Sy

E+01

5.80E+02

1.1OE+O3

DJH
CP

Vft 8- R28 = 72.4


-4.75231905E+01

5,48
Power
R2 = 0999
-1
-~

E+03

1.60916413E+03
-1.16136654E+03
3.281 96592E+02
CG IIw - P28 = 59.0

8.49238555E-01

hl

1.028W1482E+0 1

Coefficient
VR IX-

WE =61.1

1.02909441

3,67299473E+02
-1.09576981

E+OO

-3
-4

-6.53716830E+01

fu = 1.000

2.45064290E+01
CD IV - R26 = 54.6

-4
-6
-B

E+02

-12

= 49.7

Rz = 1.000

= 1.000

-6
-0

h3
b4

9.41251840E+OI
-7.95972561 E+01

-8
-lo

133770823E+0
1
-I. B936831OI?+O1

-lo

h5

Z.59731947E+01

-12

R 35273830E+O0

-12
62 = 1.000

1.2679571 2E+O0

4.113B21541E+01

b2
h3

-5.13917116E+01
3.3113926 BBE+OI

b4

-1.0841

1957E+01

2,16860646E+O0

-lo

h5

1.41245252E+O0
rX 1- P28 = 24.2

L3171027OE+OO

-12

-6

-1.67135149E+01

-8

5.93337794E+O0
VH Iv - ma

= 30.7

-[0
= 1,000

-4
-6
-8

-6.92019850E+01
1.67277366E+02
-2.03563445E+02
1.1 9290532E+02
-Z.64121307E+OI

-1
-2
-3
-4
-5

FL?= i .000

-2,05023857E+02
5,93627222 F,+02

-1
-2

b3

-8.48191308E+02

-3

-114594834E+OI

-4

2.02819798E+01

2.926245 B8E+OI

h]
b2
b4

E+OO

13621 465BE+OI

_2

3.169743471!+O0
-1.34629152E+OI

1.261 39905E+OI

h]

-6.91563661

AG B17 - MI

-10

7.91378216E-01
- i

-1.19892065E+O0

=42.7

-6
-B

1.B9235469E+02
4E+01

-6

ZI-R28

-4

5,4205791

1.15714927E+O0

R2=looo

fi+OI

-1.64917031

-2751

-1.40485450E+01

1.93551406

-9.8610 Z851E+OI

-12

-5. 163BOB55E+01

(X V - R2f3 = 46.2

Power
F2 = 1,000

-10

-4

17356E+O0

Coefficient
AG El - R28 = 59,9
7.32273503E-01

3.29893843E+O0
-2.71906511 E+O1
6,37304090E+01

-5

Power
R2 = 1.000

h2

Max

1,301

CP

PREX

Min

0.424

1.30E-01
0,012

Min

t?
PRH

Power

Factor
(SW)

Rec

5.R9524632E+02
-1.59198094

SPE Advanced Technology

E+02

Series, Vol. 4, No. I

-4
-5

49

OUTPUT VARIABLES
Mm30

m30pd

PR,~_ _

I
I
I
PCE

RT

PY

Years
Production

CE

* Cumulative Production

SAME GEOGRAPHICAL

Figure 1

LOCATION

SR6Rt (%)
100

75

50-

25-

10

20

30

--will

50
60
Pvl/Pvit(%)

40

70

80

90

100

*VHIV-VHIX
Figure

DIFFERENT

GEOGRAPHICAL

LOCATION

SIVSRt(%)

100
80
60

A
40
.. e
20n-.

10

20

30

40

50

Pvl/Pvlt

I-AGBI

60

70

80

90

100

(/6)

*CGMA-VHIXI
Figure

50

SPEAdvanced

Technology

Series, Vol. 4, No. I

RESERVES

VS

PORE VOLUMES INJECTED

SRISRt (%)
100
R28
75

50
R28
1

25

PVI = 28?40
o
10

20

30

40

50

60

70

80

100

90

Pvl/Pvlt (%)
Figure 4

RESERVES

VS

PORE VOLUMES INJECTED

SR/SRt (%)
100%

75

50

R2=

1.000

25
0
0

20

60

80

100

Pvl/Pvlt (Ye)
&
Figure

PRODUCTION

PROFILE MODEL EFFECT

m30pd
600

R28 = 72
400

200

04812162024283236

40444852

Pvl/Pvlt (%)
SPE Advanced Technology

Series, Vol. 4, No. 1

Figure

6
51

SIMULATED SECONDARY RESERVES


%0

12

MEAN = 480
10

STD.DEV. = 125
8
6
4
2
0

650

450

250

850

Mm3cr
Figure 7

SIMULATED PRODUCTION PROFILE


m30pd
300

I
, s ,95th

II

PERCENTILE

!
1

100

1
/

---$th PERCENTILE -

Ov

EXPECTED - \

P ~

----

10
Years

12

14

16

--

18

20

Figure 8

SPEAdvanced

Technology

Series, Vol. 4, No, I

Vous aimerez peut-être aussi