You are on page 1of 6

SPE 157996

Optimizing Water Injection Rates for a Water-flooding field


Feilong Liu, Charlie Guthrie and David Shipley, Chevron Energy Technology Company

water injection). These results suggest that this systematic


Copyright 2012, Society of Petroleum Engineers
method may provide a way to optimize the water injection
This paper was prepared for presentation at the SPE Annual Technical Conference and Exhibition held target rates.
in San Antonio, Texas, USA, 8-10 October 2012.

This paper was selected for presentation by an SPE program committee following review of
information contained in an abstract submitted by the author(s). Contents of the paper have not been I. Introduction
reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The Waterflooding is by far the most widely used secondary
material does not necessarily reflect any position of the Society of Petroleum Engineers, its officers, or
members. Electronic reproduction, distribution, or storage of any part of this paper without the written recovery method in the oil industry. As the name implies,
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
waterflooding involves injecting water into an oil reservoir
contain conspicuous acknowledgment of SPE copyright. and driving the oil into the production well. Currently,
waterflooding is responsible for a big portion of world oil
Abstract production, and successful waterflood management is hugely
Setting proper water injection rates for the injection wells important to the world oil supply.
is a key factor to successfully operate an oil field under water Waterflood management is a broad topic involving a
flooding. The success of such activity could (a) reduce water range of activities, some of which are performed only once or
cycling at field, section and pattern levels; (b) improve twice during the life of the waterflood; others are performed
water/oil ratio (WOR) and areal sweep efficiency; (c) periodically based on analyses of recurring measurement
improve oil production and recovery by directing water data. Among which, setting water injection target rates for the
injection to specific zones and areas; and (d) reduce OPEX by injection wells is one of the key periodical activities. The
improving water utilization. success of such activity could (a) reduce water cycling at
Typically, the onsite engineers adjust injection rates using field, section and pattern levels; (b) improve water/oil ratio
heuristics. While this does improve performance we feel that (WOR) trend and areal sweep efficiency; (c) improve oil
a more systematic approach can be developed which will lead production and recovery by directing water injection to
to further gains. In this paper, we present a systematic specific zones and areas; and (d) reduce OPEX by targeted
method, using the linear programming, to optimize the water water utilization.
injection target rates. Typically, the oil production in a waterflood field is
In this method, the reservoir is considered to be a system constrained by a combination of the reservoir condition,
which can be modeled as a collection of continuous-time flowing pipeline network and surface facilities. Adjusting the
impulse responses that convert injection rates into a water injection rates can help control the oil production from
production rate. A very simple two parameter parametric the producer. Hence, how to determine the optimal water
model, like diffusivity-filter, has been used to quantify the injection target rates, subject to all kind of constraints at the
injector-producer continuous-time impulse responses channel field, to maximize the oil production is important to the
model and the Extended Kalman Filter has been used to onsite engineers. However, due to the complex dynamic
establish the allocation factors between injectors and interaction between reservoir, injection wells, production
producers in the water-flooded field. Subject to constraints, wells and surface facilities, this is not an easy task.
including the total available water amount, the maximum and So far, some optimization approaches [2] on adjusting
minimum injection rates, the maximum total production fluid water injection rates have been proposed, but few of them
for a producer and a gauge setting, a linear programming have been piloted or applied in an operating field. The onsite
optimizer has been applied to determine the optimized water engineers typically set rates using heuristics. What they
injection rate, based on the established allocation factors. regularly do is to look at the production history and bubble
This method was pilot tested on a Chevron oilfield for 3 map, and then decide where to inject more water and where
months. The decline curve for 6 months and for 2 months of to inject less. These approaches, while helpful, leave room for
historical oil production data have been calculated. The 3 further improvement.
month pilot test result indicated that the optimized oil To overcome these difficulties, we propose a systematic
production matches the historical 6-month decline curve very method, using the linear programming, to optimize the water
well with about 22% less total daily water injection. Also we injection target rates in which [3-4]:
saw about 2% incremental production above the historical 2-
month decline curve (again with about 22% less total daily
2 SPE 157996

1) The reservoir is considered to be a system which can A diffusivity filter [1] is normally used as a continuous-
be modeled as a collection of continuous-time time injector-producer channel model to quantify how the
impulse responses that convert injection rates into a reservoir converts the injection rates into total production
production rate. A very simple two parameters rate. Learning from expert knowledge (i.e., Chevron
parametric model like diffusivity-filter has been Engineers) and the shape of a diffusivity filter, this
used to quantify the injector-producer continuous- continuous-time impulse response has been assumed to be the
time impulse responses channel model. The continuous-time uni-modal function depicted in Fig. 3, which
Extended Kalman Filter has been used to establish can be empirically characterized as
the allocation factors between injectors and
producers in the waterflooding field. h(t ) = bte − at . (2)
2) Subject to constraints, including the total available
water amount, the maximum and minimum injection
Physically, a characterizes the delay time of the physical
rates, the maximum total fluid production for a
channel, which directly relates to many factors of the channel,
producer and a gauge setting, a linear programming
such as permeability, distance, etc. The larger that a is, the
optimizer has been applied to determine the
optimized water injection reallocation rate, based on shorter the delay time, the larger the allocation factor and the
the established allocation factors. larger the impact from the injectors.
Because only sampled data (i.e., periodic measurements
The rest of this paper is organized as follows. Section II of injection and production rates) are available for processing,
describes the reservoir model, injector-producer channel the continuous-time impulse response is sampled, and a
model, allocation factor and the optimization procedure; discrete impulse response has to be used in practice. As
Section III summarizes the pilot test at a Chevron field; and, depicted in Fig. 4, the discrete reservoir impulse response
Section IV draws conclusions. between the jth injector and producer can be formulated as

γ j z −1
Lj
II. Procedure
A. Reservoir Model =
H j ( z) =
(1 − α j z −1 ) 2 ∑ h (i)z
i =1
j
−i
(3)
The reservoir can be considered to be a system which can
be modeled as a collection of independent Injector-Producer
subsystem that convert injection rates into a production rate, In equation (3) the middle part is an auto-regressive (AR)
where each subsystem is modeled as a continuous-time model that has two equal poles, and the right part is a
impulse response. Hence, for a reservoir with N injectors and moving-average (MA) model. Note that α = e − aT and
a single producer, the reservoir is represented by N γ = bα T are the two parameters that characterize the AR
independent subsystems, and its reservoir model is depicted model, where T is the sample period; and h j (i ) ( j = 1,  , l )
in Fig. 1, where: i1 (t ), i2 (t ),  , iN (t ) , are the actual injection
and l are the weights and length of the MA model. Recall
rates that flow into the reservoir, n1 (t ), n2 (t ),  , nN (t ) are that the AR model can be interpreted as an infinite-length
the corresponding measurement noises for measuring moving average (MA) [divide the denominator of (3) into its
injection rates, and im,1 (t ), im,2 (t ),  , im, N (t ) are the measured numerator]. Hence, to better approximate the AR model, the
injection rates. Similarly p (t ) , n p (t ) and pm (t ) are the actual value of l is usually much greater than 2. Because of that, the
AR model will be used to estimate the model parameters,
production rate, the corresponding measurement noise for whereas the MA model will be used for optimization.
measuring production rate and the measured production rate,
respectively. The channel production rate p cj (t ) is the rate C.Allocation Factor [3-4]
produced by injector j, which represents the amount of From the standpoint of systematic modeling, when we do
production rate in p (t ) caused by injector j. the inverse z-transform on the reservoir model in (1), the
relation between the fluid production rate and water injection
Let P ( z ) , Pjc ( z ) , and I j ( z ) be the z-transforms of the rates at injectors can be formulated as:
production rates, the channel production rate produced by
injector j, and injection rates of the jth injector; and H j ( z ) N N Lj
th
be the reservoir impulse response between the j injector and =
p ( n) ∑=
=j 1
Pjc (n) ∑∑ h (i) I (n − i)
=j 1 =i 1
j j (4)
producer. As depicted in Fig. 2, the reservoir model can be
then formulated as,
As we may know, at fields, the onsite engineers change
N N injection rates I j (n − i ) infrequently, usually holding them
=
P( z ) c
j
=j 1 =j 1

= P ( z) ∑ H j ( z)I j ( z) (1)
constant at a target injection rates, I j . Consequently, (4)
could be simplified into
B. Injector-Producer Channel Model for Each Subsystem
SPE 157996 3

N N  Lj  where N is the total number of injection wells.


=
p ( n) ∑=
Pjc (n) ∑ ∑ I j  h j (i ) 
i 1 
(5) 2) Minimum and Maximum injection rate. The
minimum injection is to assure that the reservoir
=j 1 =j 1=  
pressure is higher than bubble point pressure; while
the maximum injection rate is to assure that the
The allocation factor, AF j , from the jth injector to the
injection will not break the wellbore or damage the
producer is then defined as well etc. This constraint can be formulated as:

 Lj 
( I j )min ≤ I j ≤ ( I j )max (11)
γ j z −1
=AF j  =
 i =1 ∑
h=
j (i ) 

=
H j ( z ) | z 1=
(1 − α j z −1 ) 2
|z 1 (6) where ( I j )min and ( I j ) are the minimum and
max
  maximum injection rate for the jth injection well.
3) Maximum production rate for a producer. This
which shows how much the injection rates from the jth constraint sets to assure that the total production rate
injector will inpact on the fluid production at a producer. The will not exceed the allowed facility capacity of the
total fluid production in (5) can be simplified into surface facility. This constraint can be formulated
N N as:
=
p ( n) ∑=
Pjc (n) ∑I j AF j (7) N
=j 1 =j 1
It can be seen from (3) and (6) that the area under each
=
pk (n) ∑I j =1
j AFk , j ≤ ( pk )max (12)

discrete impulse response has been used to evaluate the


allocation factor for each injector; hence, the greater the area,
where ( pk )max is the maximum capacity of the
the more an injector impacts a producer. surface facility, such as rod pump.
As explained in [3-4], the unknown parameters α j and 4) Maximum production rate for a gauge setting. This
constraint is similar to the maximum production rate
γ j are estimated from an EKF, and the allocation factors for a producer. We can’t allow the total production
from each injector to a producer is then calculated by (6) rate for the gauge setting to exceed its maximum
using the estimated parameters α j and γ j . Both the capacity. Let’s assume that a gauge setting is
connected by K producers, this constraint can then
unknown parameters and allocation factors are time-varied. be formulated as:
K K N
D. Optimization
According to (7), the oil production from the kth
=
=

pk (n)
k 1
∑∑ I
=
k 1 =j 1
j AFk , j ≤ ( GS )max (13)
producer, Oilk , and the total oil production from M wells,
Oil , can be formulated as
where ( GS )max is the maximum capacity of the
surface gauge setting facility.
N
=Oilk So )k pk (n) ( So )k ∑ I j AFk , j
(= (8) Combining all those constraints, the optimization
j =1 problem is then described as:
and a) Unknown variables: I j for all injectors;
M M N b) Objective function: maximize the total oil production,
=
Oil ∑=
Oilk ∑ ( So )k ∑ I j AFj (9)  M N 
=k 1 =k 1 =j 1 i.e., Max ( Oil ) = Max  (
 k 1 =j 1 ∑
So ) k I j AF j  ;
 ∑
=  
Respectively, where ( So )k is the oil cut for the kth producer. c) Constraints:
Recall from the introduction, the oil production in a N

waterflood field is constained by a combination of the ∑I


j =1
j ≤ ITotal (10)
reservoir condition, flowing pipeline network and surface
facilities. The common constraints in the field are:
1) Total available water for injection. The total
( I j )min ≤ I j ≤ ( I j )max (11)
available water for injection to a field is usually N

determined by field-level strategies, considering =


pk (n) ∑I
j =1
j AFk , j ≤ ( pk )max (12)
many important issues, such as the subsidence, the
OPEX on buying fresh water etc. This constraint can K K N
be formulated as:
N
=
=

pk (n)
k 1
∑∑ I
=
k 1 =j 1
j AFk , j ≤ ( GS )max (13)

∑I
j =1
j ≤ ITotal (10)
Typically the oil cut for a producing well will not change
4 SPE 157996

very quick; hence we assume the ( So )k are constant over a decline curve over 6 months and 2 months historical oil
production data have been calculated. The 3 month pilot test
short period. Because the objective function and constraints
result showed that the optimized oil production matches the
are all linear, this problem is then a problem of linear
decline curve over 6 month history very well with about 22%
optimization under constraints. Therefore, linear
less total daily water injection; and about 2% increment over
programming is adopted to solve this problem.
the decline curve of 2 month history, in addition to about
22% less total daily water injection. All results suggest that
III. Pilot Test this systematic method may provide a way to optimize the
This work has been piloted in a Chevron oilfield starting
water injection target rates.
on 09/03/2009. A group of 46 injectors, 23 shallow zone
injectors and 23 deep zone injectors, have been carefully
Nomenclature
selected by the onsite engineers. Considering the boundary
EKF = Extended Kalman Filter
issue, only 12 producers surrounded by these 46 injectors
AF = Allocation Factor
have been selected for observing the production performance.
AR = Auto-regressive Model
The allocation factors between injectors and producers
MA = Moving Average Model
were estimated by the EKF using 1 year historical production j = 1,  , N : Injector index
data. Based on the reservoir and surface facility condition, the
total available water for injection, the minimum and i1 (t ), i2 (t ), , iN (t ) : Actual Injection Rates
maximum injection rates, the maximum production rate for n1 (t ), n2 (t ), , nN (t ) : Measurement Noise of Injection Rates
all producers, and maximum production rates for all gauge
im ,1 (t ), im ,2 (t ), , im , N (t ) : Measured Injection Rates
settings were calculated. All these data will fit into the
optimizer to generate the optimal injection target rate for this p (t ) : Actual Production Rate
pilot area over the testing period. These optimal injection n p (t ) : Measurement Noise of Production Rate
target rates were then sent to the onsite engineer for their
daily injection control. pm (t ) : Measured Production Rate
The injection target control is depicted in Fig. 5; the total p cj (t ) : Channel Production Rate Caused by Injector j
fluid production rate and total oil production rate are depicted T: Sample period
in Fig. 6; and the 2 months’ decline curve analysis for total n: Discrete time step index
oil production rate is depicted in Fig. 7-8, respectively.
h(t ) : Channel model
We observe that:
1) from Fig. 5, the injection control is good over the H ( z ) : z-transform of channel model h(t )
testing period. Before the pilot, i.e., 09/03/2009, the P ( z ) : z-transform of total production rate p (t )
average total actual injection rate is around 9000 bbl I i ( z ) : z-transform of injection rate ii (t )
per day, while after pilot, this became around 7000
bbl per day. In about 7% of injectors, the actual Pjc ( z ) : z-transform of the jth channel production rate p cj (t )
injection rate is more than 10% different from the a and b: two parameters of the continuous channel model
target injection rates; α j and γ j : two parameters of the discrete channel model
2) from Fig. 6, the total oil production after 9/3/2009 fit
very well with the 6 months’ decline curve
(determined by historical data from 3/3/2009 to References
[1] Albertoni, A. and Lake, L.W. Inferring Interwell Connectivity
9/3/2009). This means, the total optimized oil Only From Well-rate Fluctuations in Waterfloods. SPE Res Eval &
production matches the decline curve of 6 month Eng 6 (1): 6–16. SPE-83381-PA, 2003
history very well with about 22% less total daily [2] X. Liang, D. B. Weber, T. F. Edgar, L. W. Lake, M. Sayarpour,
water injection; A. AI-Yousef, “Optimization of Oil Production Based on A
3) from Fig. 7-8, the decline curve is generated for the Capacitance Model of Production and Injection Rates”, 2007 SPE
most recent 2 months history data (i.e., 7/3/2009 to Hydrocarbon Economics and Evaluation Symposium, Dallas, Texas,
9/3/2009). Comparing with such decline curve, we USA, 2007 (SPE 107713)
can actually see about 2% total oil production [3] Feilong Liu, Jerry Mendel and Amir Mohammad Nejad,
“Forecasting Injector/Producer Relationships From Production and
increment with about 22% less total daily water
Injection Rates Using an Extended Kalman Filter”, SPE Journal
injection; (December 2009), 653-664 (SPE-110520).
4) both analysis indicated the benefit of this work, [4] Feilong Liu, “Intelligent Signal Processing for Oilfield
which is either cut the OPEX cost on injecting less Waterflood management”, Ph.D Thesis, University of Southern
water, or increase the production and cut the cost. California, 2008

IV. Conclusions
This paper has presented a systematic method, using the
linear programming, to optimize the water injection target
rates.
This work was tested on a Chevron oilfield, and the
SPE 157996 5

h j (i )
i1 (t ) h1 (t )

im ,1 (t )
p1c (t )
n1 (t ) ∑

n p (t )
i2 (t ) p2c (t )
h2 (t )
p(t ) pm (t )
n2 (t ) ∑
im ,2 (t )
Σ Σ

 pNc (t )
i
iN (t ) hN (t )
Figure 4: The sampled impulse response h j (i ) (the solid dots
im , N (t ) represent the non-zero samples).
nN (t ) ∑

Figure 1: The Reservior Model for reservoir with N injectors


and a single producer.

I1 ( z ) H1 ( z )
P1c ( z )

c
I2 ( z) H 2 ( z ) P2 ( z )

∑ P( z )

Figure 5: Step change of total injection rate, and the numbers
PNc ( z ) of injectors within ±10%, <-10% and >+10% of target.
I N ( z) H N ( z)

Figure 2: The Reservoir Model in z-Transform Domain.

Figure 6: The total oil and fluid production before and after
the pilot test on 09/03/2009, and the corresponding linear and
exponential decline curves of last 6 months’ historical total
Figure 3: One example of the impulse response between an oil production.
injector and a producer.
6 SPE 157996

The Decline Curves and their prediction on total oil production


600

550

500
Total Oil Production

450

400

Actual Oil Prd


350 Hyperbolic
Exponential
Linear
300
01/01/09 04/01/09 07/01/09 10/01/09 01/01/10
Date

Figure 7: The total oil production before and after the pilot
test on 09/03/2009, and the corresponding linear and
exponential decline curves of last 2 months’ historical total
oil production.

Figure 8: The total oil production before and after the pilot
test on 09/03/2009, and the corresponding linear and
exponential decline curves of last 2 months’ historical total
oil production.

You might also like