June 7-10, 2015.

Whistler,
9th International Symposium British Columbia,Control
on Advanced Canadaof Chemical Processes
9th International Symposium on Advanced Control of Chemical Processes
June
9th 7-10, 2015. Whistler,
International Symposium British
on Columbia,Control
Advanced Canadaof Chemical Processes
Available
June 7-10, 2015. Whistler, British Columbia, online at www.sciencedirect.com
Canada
June 7-10, 2015. Whistler, British Columbia, Canada

ScienceDirect
IFAC-PapersOnLinewith
Well Placement Optimization 48-8 (2015) 057–062
Geological Uncertainty Reduction
Well
Well Placement Optimization with Geological Uncertainty Reduction
Well Placement Optimization with Geological Uncertainty
Placement Optimization with Geological Uncertainty Reduction
Reduction
Shahed Rahim, Zukui Li*
Shahed Rahim,  Zukui Li*
Shahed Rahim, Zukui Li*
Shahed 
Rahim, Zukui Li* University of Alberta,
*Department of Chemical and Materials  Engineering,
*Department of Chemical and 
Materials Engineering, University of Alberta,
Edmonton,
*Department of Alberta,
ChemicalCanada T6G 2V4Engineering,
and Materials (E-mail: zukui@ualberta.ca)
University of Alberta,
*Department
Edmonton, of Alberta,
ChemicalCanada and Materials
T6G 2V4Engineering, University of Alberta,
(E-mail: zukui@ualberta.ca)
Edmonton, Alberta, Canada T6G 2V4 (E-mail: zukui@ualberta.ca)
Edmonton, Alberta, Canada T6G 2V4 (E-mail: zukui@ualberta.ca)
Abstract: Well placement optimization aims to determine optimal well locations so that the economic
Abstract:
benefit from Well placement
oil placement optimization
productionoptimization
can be maximized. aims to determine optimal well locations so that the economic
Abstract: Well aims to Geological uncertainty
determine optimal well has a significant
locations so that impact
the economic on the
benefit
Abstract:
optimal from Well
well oil placement
production
placement optimization
plan canandbetherefore aimshas
maximized. to Geological
determine
to be optimal
uncertainty
considered well
in the locations
has so that impact
a significant
well placement the economic
on the
optimization
benefit from oil production can be maximized. Geological uncertainty has a significant impact on the
benefit
optimal from
problem. well oil production
placement
A geological plancan andbetherefore
maximized. has Geological
to be considereduncertaintyin the haswell
a significant
placement impact on the
optimization
optimal well placementrealization
plan and reduction
therefore framework
has to be for well
considered placement
in the well under geological
placement uncertainty
optimization
optimal
problem.
is proposed well placement
A geological
in this work.planThe
realizationandobjective
thereforeisframework
reduction has
to to be for
considered
optimally well
select in small
placement
a the well
under placement
subsetgeological
of optimization
uncertainty
realizations and
problem. A geological realization reduction framework for well placement under geological uncertainty
problem.
is proposed
incorporate A them
geological
in thisinto realization
work.
the wellThe reduction
objective
placement isframework
to optimally
optimization for well
problem, select
soplacement
asatosmall
reduceunder
subset
the geological uncertainty
of realizations
computational and
efforts.
is proposed in this work. The objective is to optimally select a small subset of realizations and
is proposedthem
incorporate
A reservoir in this
case into
study work.
the wellThe objective
placement is toselected
optimization optimally
problem, select
so asatosmallreduce subset of realizations
the computational and
efforts.
incorporate them into thedemonstrates
well placement that the
optimization smaller
problem, sosubset of realizations
as to reduce is a veryefforts.
the computational good
incorporate
A reservoir
representation them
case into
study
of astudy the well placement
demonstrates
largerdemonstrates optimization
that
superset of realizations the selectedproblem,
smaller
and cansmaller so as
subset
significantly to reduce
of the computational
realizations is a very efforts.
good
A reservoir case that the selected subset decrease the computational
of realizations is a very good time
A reservoirwith
representation
associated case
oftheastudy demonstrates
largerplacement
well that the problem.
superset optimization
of realizations selected
and cansmaller subset decrease
significantly of realizations is a very good
the computational time
representation of a larger superset of realizations and can significantly decrease the computational time
representation
associated withofthea well largerplacement
superset optimization
of realizations and can significantly decrease the computational time
problem.
© 2015, IFAC
associated
Keywords:
associated
with (International
the well
geological
with the well Federation
placement
uncertainty,
placement well ofplacement,
Automatic
optimization
optimization Control) Hosting
problem.
optimization,
problem. by Elsevier
uncertainty Ltd. All rights reserved.
reduction
Keywords: geological uncertainty, well placement, optimization, uncertainty reduction
Keywords: geological uncertainty, well placement, optimization, uncertainty reduction
Keywords: geological uncertainty, well placement, optimization, uncertainty reduction
 computationally demanding task and impractical for larger
1. INTRODUCTION 
 computationally
realistic reservoirs demanding
with multiple task andwells,impractical
a smaller for larger
subset of
1. INTRODUCTION computationally demanding task and impractical for larger
For oil reservoir operations,1. INTRODUCTION the production amount of oil computationally
realistic
realizations reservoirs
are demanding
with
generally task
multiple
selected and
wells, impractical
and a smaller
used in for larger
subset
the of
well
1. INTRODUCTION realistic reservoirs with multiple wells, a smaller subset of
For oil reservoir operations, the production amount of oil realistic
realizations reservoirs
are
placement optimization with
generally multiple
selected wells,and a smaller
used in subset
the of
well
greatly
For oil depends
reservoir on the well the
operations, locations
productionand the amount geological
of oil realizations are generallymodel selected to account
and usedforingeological the well
For oil
greatly
property reservoir
depends operations,
on
of the reservoir. the well the production
locations and amount
the of
geological oil realizations
placement
uncertainty. are
Thus, generally
optimization reducing selected
model theto and
account
number used for
of in the
geologicalwell
greatly depends on the To wellachieve
locationsthe maximum economic placement optimization model to account for geological
and the geological
greatly depends
property of on the To
the reservoir. wellachieve
locations and
is the
the maximum geological
economic placement optimization
uncertainty.
for realizations Thus, model the
reducing to account
number for
of geological
benefit,
property of well placement
the reservoir. Tooptimization
achieve the maximum necessary
economic uncertainty. forThus, flow simulation
reducing becomes
the number an important
of geologicalstep in
property of
benefit,
determining the
wellthe reservoir.
placement
best Tooptimization
locations achieve
for the maximum
placing is necessary
wells in a economic
reservoir. for well uncertainty.
realizations
placementforThus,
flow reducing becomes
simulation
optimization. the
Yeten number an al.
et of(2003)
important geological
step in
used
benefit, well placement optimization is necessary for realizations for flow simulation becomes an important step in
benefit, well
determining
Reservoir the placement
flow best locations
simulation optimization
for
is placing
commonly is used
wells necessary
in a reservoir.
in realizations
well
for multiple
well placementfor flow
equiprobable simulation
optimization. becomes
geological etan al.
Yeten realizations important
(2003) step
in in
used
the
determining the best locations for placing wells in a reservoir. well placement optimization. Yeten et al. (2003) used
determining
Reservoir
placement the best
flow locations
simulation
optimization for
problems.is placing
commonly
The wells
well in
used a reservoir.
in well
positions are well placement
multiple
determination equiprobable
of optimization.
objective geologicalYeten realizations
function et
of al.
well (2003) in used
placement the
Reservoir flow simulation is commonly used in well multiple equiprobable geological realizations in the
Reservoir
placement flow simulation
optimization problems.is output
commonly well used
The variable of ininterest
positions well multiple equiprobable
are determination of objective geological
function realizations
of well uncertainty in the
placement
determined by maximizing
placement optimization the
problems. The well positions are optimization
determination toof account objectiveforfunctionthe geological
of well placement
placement
determined optimization
by maximizing problems.
the The
output well
variable positions
of are
interest determination
optimization
present associated toof objective
account for function
the of
geological well placement
uncertainty
such as the by
determined cumulative
maximizing oil production
the output(COP) variable or netof interest optimizationin to a reservoir.
account for Wang theet geological
al. (2012) uncertainty
selected a
determined
such
value the by
as(NPV) maximizing
cumulative
generated oil
by the
production
a output(COP)
reservoir variable
flow or netof interest
simulator. present
The optimization
associated
smaller in to
subset account
a reservoir.
of for
Wang
realization theet
to geological
(2012) uncertainty
al.quantify selected
geologicalaa
such as the cumulative oil production (COP) or net present associated in a reservoir. Wang et al. (2012) selected
such
value as the
(NPV) cumulative
generated oil
by production
a reservoir (COP)
flow or net
simulator. present
The associated
smaller
uncertainty in
subset a
in well reservoir.
of Wang
realization et
to al. (2012)
quantify selected
k-meansa
usinggeological
objective
value (NPV) function
generated for by thea wellreservoirplacement optimization
flow simulator. The smaller subset of placement
realizationoptimization
to quantify geological
value
objective
problem (NPV)
is generated
function
evaluated for
by by
thea well
running reservoir
the flow simulator.
placement
reservoir optimization
flow simulator smaller
The uncertainty
clustering. subset
in well
K-means of placement
realizationoptimization
clustering to cumulative
uses quantify usinggeological
k-means
field oil
objective function for the well placement optimization uncertainty in well placement optimization using k-means
objective
problem
with givenis function
evaluated
well for
positions. the
by running
As awellthe placement
result, reservoir
the optimization
flow
computational simulatortime uncertainty
clustering.
production in well
K-means
which placement
clustering
requires to be optimization
uses cumulative
calculated for using
every k-means
field
possibleoil
problem is evaluated by running the reservoir flow simulator clustering. K-means clustering uses cumulative field oil
problem
with
for given
the is well
flow evaluated by
positions.
simulator running theincreases
As a result,
significantly reservoir flowthe
the computational
with simulator
sizetime
of clustering.
productionofwhich
locations K-means
well and clustering
requires to be is
therefore uses
calculated cumulative
computationally field
for everyintensive.
possibleoil
with given well positions. As a result, the computational time production which requires to be calculated for every possible
with
for
the given
the flowwell
reservoir grid positions.
simulator
and the As a result,
significantly
number of the computational
increases
wells to with
be the size
placed. Intime production
locations
of Yasari
the et of which
al. well
(2013) requires
and used to be is
therefore
robust calculated
wellcomputationally
placement for every possible
intensive.
optimization
for the flow simulator significantly increases with the size of locations of well and therefore is computationally intensive.
for
the the flow
reservoir simulator
grid and significantly
the number ofincreases
wells
usedto with
be the
placed. size
In the locations
of Yasari et of
al. well
(2013) and therefore
used robust iswell
computationally
placement intensive.
objectiveoptimization
literature,
the reservoirvarious
grid andmethods have been
the number of wells toinbewell In the under
placement
placed. Yasariuncertainty
et al. (2013) using
useda risk
robustweighted
well placement function
optimization for
the reservoir
literature,
optimization grid
various
to and the
methods
determine number
have of
been
optimal wells
used
welltoinbe placed.
well
positions In
placementof thea Yasari
under
multiple et al.
uncertainty(2013)
realizations. used
using a
Theyrobust
risk well
weighted
selected placement
a objective
subset optimization
of function
realizationfor
literature, various methods have been used in well placement under uncertainty using a risk weighted objective function for
literature,
optimization various
to methods
determine have been
optimal used
well in well
positionsplacementof a
wella from under
multiple uncertainty
realizations.
a superset using a
They
by calculating risk weighted
selected
the NPVa for a objective
subset
all the of function
realizationfor
reservoir.
optimization In most cases, the optimal
to determine objective well function for the of
positions multiple realizations. They selected subset of realizations
realization
optimization
reservoir. In to
most determine
cases, the optimal
objective well
functionpositions
for the of
well a multiple
from a
usingabase realizations.
superset
case byby
well They
calculating selected
position andthe NPV a
then used subset
for all theof realization
realizations
placement
reservoir. In optimization
most cases,problem is to maximize
the objective function for the theNPV or from
well superset calculating the NPV for allranking to select
the realizations
reservoir.
placement In mostetcases,
optimization the objective
problem function
is to maximize for
the theNPV well
or
in the fromsmall
using abase
superset
case by ofcalculating
well position andthe NPV
then usedforYang
allranking
theetrealizations
al.to select
COP (Nasrabadi
placement optimization al., 2012).
problem Optimization
is to maximize methodsthe NPV used or using basesubset case well realization.
position and Similarly,
then used ranking to(2011)
select
placement
COP
well optimization
(Nasrabadi
placement et include:problem
al., 2012). mixedis to integer
maximize
Optimization methodsthe NPV used or
programming using
the
in combined basesubset
small case well
Steam positionGravity
ofAssisted
realization. and then
Similarly, used
Drainage Yangranking
et al.to(2011)
(SAGD) select
well
COP (Nasrabadi et al., 2012). Optimization methods used in the small subset of realization. Similarly, Yang et al. (2011)
COP
well (Nasrabadi
placement
(Rosenwald and et al., 2012).
include:
Green, 1974), Optimization
mixed integermethods
gradient-based the small subset
combined
used in production
programming
optimization Steam
and ofAssisted
realization.
placement Similarly,
Gravity
optimization Drainage Yanguncertainty
under et al. (2011)
(SAGD) well
by
well placement include: mixed integer programming combined Steam Assisted Gravity Drainage (SAGD) well
well finite
placement
(Rosenwald
using and Green,
difference include: 1974),
method mixed integer
gradient-based
(Bangerth et al., programming
optimization
2006), genetic combined aSteam
production
selecting and
subset Assisted
placement
of Gravityusing
optimization
realizations Drainage
under (SAGD)
uncertainty
traditional well
rankingby
(Rosenwald and Green, 1974), gradient-based optimization production and placement optimization under uncertainty by
(Rosenwald
using and
finite difference
algorithms Green,
(Bittencourt 1974),
method and gradient-based
(Bangerth
Horne, et1997),
al., 2006), genetic production
optimization
simulated selectingbased
method a and placement
subset
on theof NPV optimization
realizations
of all using
the under
realizations uncertainty
traditional for ranking
a by
base
using finite difference method (Bangerth et al., 2006), genetic selecting a subset of realizations using traditional ranking
using finite
algorithms
annealing difference
(Bittencourt method and (Bangerth
Horne, et al.,
1997), 2006), genetic
simulated
swarm case selecting
method a
based
scenario. subset
on theof realizations
NPV of all using
the traditional
realizations for ranking
a base
algorithms (Beckner
(Bittencourt and Song, and Horne,1995) and particle
1997), simulated method based on the NPV of all the realizations for a base
algorithms (Beckner
annealing
optimization (Bittencourt
(Onwunalu andand and
Song, Horne,
1995) 2010),
Durlofsky, and1997),
particle
etc. swarm case
simulated method based on the NPV of all the realizations for a base
scenario.
annealing (Beckner and Song, 1995) and particle swarm case In scenario.
this study, reservoir well placement optimization
annealing (Beckner
optimization (Onwunalu andand Song, 1995) 2010),
Durlofsky, and particle
etc. swarm case scenario.
optimization
The complexity (Onwunalu
of the and placement
well Durlofsky,optimization
2010), etc. problem considering In this study, reservoir
geological well placement
uncertainty is studied optimization
based on a
optimization (Onwunalu and Durlofsky, 2010), etc. In this study, reservoir well placement optimization
Thefurther
is complexity
increased of theby well placement uncertainty
incorporating optimizationassociated
problem novel In this
considering
methodstudy, for reservoir
geological
geological well placement
uncertainty
uncertainty is studied
reduction. optimization
basedThe on wellaa
The complexity of the well placement optimization problem considering geological uncertainty is studied based on
Thefurther
is
with complexity
increased
geological ofproperties
the
by well placement
incorporating
of the optimization
uncertainty
reservoir. problem novel
associated
Geological considering
placementmethod geological
for geological
optimization uncertainty
uncertainty
problem is is studied
solved reduction.
using basedThe on wella
derivative
is further increased by incorporating uncertainty associated novel method for geological uncertainty reduction. The well
is
withfurther increased
geological by
propertiesincorporating
of the uncertainty
reservoir. associated
Geological
generally free novel
placementmethod
optimizationfor geological
optimization method. uncertainty
problem is solved reduction.
using The well
derivative
uncertainty
with geologicalin well placement
properties of optimization
the reservoir. is Geological placement optimization problemGeological
is solved using uncertainty
derivative is
with geological
uncertainty
considered byin properties
well placement
incorporating of the
multiple reservoir.
optimization
geological is Geological
generally
realizations placement
free
considered optimization
optimization
by usingmethod. problem
method.
a reduced is solved
Geological using
subset uncertainty derivative
uncertainty
of geological is
uncertainty in well placement optimization is generally free optimization Geological is
uncertainty
considered
of the byin incorporating
reservoir wellin placement
the multiple
optimizationoptimization
geological
model. isHence,
generally
realizations the free optimization
considered
realization by using
from a method.
a reduced
superset Geological
of subset uncertainty
realization ofin geological
the is
well
considered by incorporating multiple geological realizations considered by using a reduced subset of geological
considered
of by
the reservoir
calculation of incorporating
COP inorthe NPV multiple
optimization
is based geological
on model.
the flow realizations
Hence,
simulation considered from
realization
the placement by using
optimization a model.
a superset reduced An subset
of realization optimal ofin realization
geological
the well
of the reservoir in the optimization model. Hence, the realization from a superset of realization in the well
of the
calculation
on reservoir
multiple COPinorthe
ofgeological NPV optimization
is based on
realizations. model.
the flowHence,
However, simulation realizationmethod
placement
the reduction from a superset
optimization
using model.
geologicalof realization
An
property optimalof in the
therealizationwell
reservoir
calculation of COP or NPV is based on the flow since simulationflow placement optimization model. An optimal realization
calculation
on multipleof
simulation for COP
geological
a or NPV
large is based
realizations.
number of on the flow since
However,
realizations simulation
is a flow
very placement
reduction
and static optimization
method
measures model.in An
usingis geological
used property
selectingoptimalofthetherealization
reservoir
subset of
on
on multiple
multipleforgeological realizations. However,
However, since flow reduction method using geological property of the reservoir
simulation geological
a large realizations.
number of realizations since
is a flow very reduction
and staticmethod measures usingis geological
used in property
selecting ofthethesubset reservoir of
simulation for a large number of realizations is a very and static measures is used in selecting the subset of
simulation© 2015
Copyright a large number of realizations is a very 57 and static measures is used in selecting the subset of
for IFAC
2405-8963 ©
Copyright © 2015, IFAC (International Federation of Automatic Control)
2015 IFAC 57 Hosting by Elsevier Ltd. All rights reserved.
Copyright
Peer review©under
2015 responsibility
IFAC 57 Control.
of International Federation of Automatic
Copyright © 2015 IFAC
10.1016/j.ifacol.2015.08.157 57
IFAC ADCHEM 2015
58
June 7-10, 2015. Whistler, BC, Canada Shahed Rahim et al. / IFAC-PapersOnLine 48-8 (2015) 057–062
realization. The well placement optimization was applied
using the subset of realizations obtained from the optimal
realization reduction method on a reservoir with a fixed
number of wells. Comparison studies with other geological
realization selection method are performed to demonstrate the
effectiveness of the proposed method.
The rest of the paper is organized as follows. Section 2
discusses the well placement optimization model. Section 3
provides the model used to select a smaller subset of
realizations from a larger superset of realizations. Results and
discussions of applying the well placement optimization
model under geological uncertainty are provided in Section 4.
The paper is concluded in section 5.
2. WELL PLACEMENT OPTIMIZATION UNDER
GEOLOGICAL UNCERTAINTY
Well placement optimization is a computational intensive
task. To evaluate the performance of a certain well placement
plan (i.e., the decision variables), a reservoir flow simulation
is performed for multiple geological realizations. So it is a
simulation based optimization problem. Since there is no
explicit objective function of the decision variables,
derivative free optimization method is desired. Specifically,
the derivative free optimization solver NOMAD is used in
this work. NOMAD implements the Mesh Adaptive Direct
Search (MADS) algorithm for constrained blackbox
functions. The MADS algorithm is an extension of the
pattern search method for nonlinear constrained optimization
problems and therefore is a derivative free method (Audet et
al., 2009). In this work, the objective function for the well
placement optimization problem is designed as maximizing
the risk averted expected cumulative oil production from a set
of realizations as given by (1).
NR
Max C O Prisk  C O PExpected    p i ( C O Pi  C O PE xp ected )
i 1
where the expected COP is given by
NR
C O PE xpected   p i C O Pi
i 1
In (1) and (2), NR is the number of realizations used to
determine the geological uncertainty, pi is the probability of a
geological realization i, COPi is the cumulative oil
production of realization i, γ is the risk averted factor (set as
0.1 in this work). The blackbox function is the reservoir
simulator which determines the COP value based on the
positions of the producer wells.
The robust well placement optimization used in this study is
summarized in a flow diagram as given in Fig. 1. The steps in
the well placement optimization under uncertainty are:
 Generate a large number of geological realizations
using geostatistical method.
 Select a smaller subset of those realizations using
realization reduction. The proposed realization
reduction model minimizes the probability distance
between the discrete distribution represented by the
superset of realizations and the reduced discrete
distribution represented by the selected realizations.
 Using the selected subset of realizations, perform
well placement optimization by maximizing the
objective function as given by (1). Each function
evaluation calls on the reservoir simulator to
calculate the COP.
 The optimal well locations using the subset of
realizations are obtained when the stopping criteria
for the optimizer are satisfied.
Robust well placement optimization is also performed using
all the realizations in the superset to obtain optimal well
locations for all the realizations. The well placement plan
using the reduced subset of realizations obtained from the
proposed method is compared to the well placement plan
using all the realizations in the superset.
Generate a superset of
geological realizations
Select subset of
realization using proposed
realization reduction
method
Perform robust well Perform robust well
placement optimization placement optimization
using all realizations using selected realizations
2
Obtain optimal well Obtain optimal well
locations for all locations for selected
(1) realizations subset of realizations
(2) Fig. 1. Workflow for well placement optimization under
uncertainty
3. GEOLOGICAL UNCERTAINTY REDUCTION
Geological uncertainty exists because it is not possible to
know the exact geological properties of every section of a
realistic reservoir. Techniques such as well exploration and
core holes can give an idea of the geology property of that
particular area of the reservoir. However, the geological
parameters of the area between the exploration wells or core
holes will still be unknown. As a result, geological
uncertainty will always exist for a reservoir. Reservoir
performance can be quantified by flow simulation, which
provides production parameters of interest such as the COP
and the NPV, etc. All the production parameters depend on
the geological properties of the reservoir. It is very important
to incorporate geological uncertainty in a reservoir model;
58
 IFAC ADCHEM 2015
June 7-10, 2015. Whistler, BC, Canada Shahed Rahim et al. / IFAC-PapersOnLine 48-8 (2015) 057–062 59
otherwise it may result in an incorrect prediction of
production parameters. To represent the geological
uncertainty, multiple geological realizations are usually
generated using geostatistical tool so as to obtain a broad
range of possible geological properties for a reservoir.
However, reservoir flow simulation cannot be run for all the
possible realizations due to significant computer processing
times. Therefore, in practice only a small number of
geological realizations are chosen to perform reservoir
simulation, so as to obtain a reservoir performance model
which incorporates geological uncertainty.
An optimal realization reduction model based on mixed
integer linear optimization (MILP) technique is used to select
a smaller subset of realizations from the superset of
realizations (Rahim et al., 2014). The proposed algorithm
uses reservoir geological properties and static measures to
quantify the dissimilarity between realizations, and uses
Kantorovich distance to quantify the probability distance
between the superset and the subset of realizations. The
objective is to find out the optimal subset which has a similar
statistical distribution characteristic to the superset of
realizations. The MILP model which selects the subset of
realizations from a superset of realization is given by the set
of equations below.
The objective function of the realization reduction algorithm
is to minimize the Kantorovich distance between the original
distribution and the reduced distribution
orig
min D Kan  p i
di
i I
where di represents the cost of removing a realization i (i.e.,
transporting and distributing its probability mass to preserved
realizations). This cost is quantified by a weighted
summation of the transported probability mass,
d i   ci , i ' vi , i '  i  I (4)
i ' I
where the weight is the dissimilarity c i , i ' between realizations
i and i’ given by
c i ,i '   m ik  m i ' k +    ict   i ' ct  i, i '
k c ,t
The dissimilarity between realizations is computed using the
geological properties and the static measures. m ik is the
value of the k type static measure for realization i,  ic t is the t
type geological property value of cell c in the reservoir grid
for realization i, λ is a weight parameter (set as 0.01) which
reflects the contribution of geological property data in the
dissimilarity calculation. The static measures used in the
calculation of the dissimilarity between realizations include
the average net permeability, the average net porosity, the
average net irreducible water saturation, the fraction of net
cells, the net pore volume, and the original oil in place and
net oil in place. For example, the average net permeability is
calculated as K net =  c k c I cnet  c I cnet , where binary
indicator parameter I cnet is used to denote whether a cell c in
the reservoir grid is net ( I cnet  1 ) or not ( I cnet  0 ), k c
denotes the permeability of cell c. The idea of net cell stems
from the fact that if a section of the reservoir rock has very
low porosity and permeability value, then that section of the
rock will be unable to carry any oil through it. For a complete
definition of the other static measures, the reader is referred
to (Rahim et al., 2014). All the static measures used in the
realization reduction method are properties of the reservoir
and independent of the location of wells within the reservoir.
The following constraints are also included in the proposed
MILP model for optimal geological realization
reduction/selection. First, if a realization i is removed (yi=1),
then all of its probability mass should be transported
(  i ' I vi , i ' =1 ). If a realization i is selected/preserved (yi =0),
then its probability mass should not be transported to any
realization (  i ' I vi , i ' =0 ).
v i ,i '
= yi i  I (6)
i ' I
Furthermore, if a realization i ' is removed ( y i ' =1), then no
probability mass can be transported to it ( vi , i ' =0 for any i). If
a realization i ' is selected ( y i ' =0), then probability mass can
be transported to it ( 0  v i , i '  1 ).
0  vi ,i '  1  y i '  i, i '  I (7)
If the total number of realizations to be removed is given by
R then the following equation ensures that R realizations are
removed
 yi  R  i  I (8)
(3) i I
The next set of equations ensure that at least 2 realizations are
selected from the subset I S B and subset I SW
 1  y   2 i
 i  I SB (9)
i I S B
 1  y   2 i
 i  I SW (10)
i I S W
where the subset I S B has 2 realizations which is identified
using the following steps. For each static measure, the
realizations corresponding to the top 3 highest static measure
values are identified. Those identified realizations ID are
combined into a superset, and from which the 2 most frequent
(5)
realizations are selected to from the set I S B . Same idea is
used to identify the subset I SW which has top 2 most frequent
realizations which represent the potential worst performance.
The idea behind incorporating these constraints is to ensure
the potential worst and best case realizations from the
superset of realizations is included in the selected subset of
realizations. With the selected realizations (i.e., y i ) and the
probability mass transportation plan (i.e., v i , i ' ), the new
probability of realizations in the reduced distribution p in ew
can be evaluated as follows:
new orig orig
p i '  (1  y i ' ) p i '   vi , i ' p i i '  I (11)
i
Finally, the complete optimization model is composed of (3)
to (11) and it is a mixed integer linear optimization problem.
This problem can be solved using MILP solver such as
CPLEX (IBM, 2010).
59
IFAC ADCHEM 2015
60
June 7-10, 2015. Whistler, BC, Canada Shahed Rahim et al. / IFAC-PapersOnLine 48-8 (2015) 057–062

4. CASE STUDY threshold permeability is set as ko = 3×10-13 m2 to determine

whether a cell is a net or non-net cell. Static measure based
To illustrate the proposed geological realization reduction
ranking method was applied next to obtain a subset of 10
method and demonstrate its application in well placement
selected realizations from the superset of 100 realizations. In
optimization, a two dimensional reservoir model with 50 × 50
the ranking based methods, all the 100 realizations in the
grid size (2500 cells) and 5m × 5m cell size is investigated in
superset are sorted in ascending order based on the static
this section. The reservoir has 5 fixed vertical injector well
measure values. 10 realizations are evenly selected from the
placed at grid positions: [8 45], [16 45], [24 45], [32 45] and
sorted list with ranks 1, 12, 23, 34, 45, 56, 67, 78, 89, 100. In
[40 45]. The number of vertical producer wells was fixed as
this study, static measures of Net Pore Volume (PVnet) and
5. The objective function for the well placement optimization
Original Oil in Place (OOIP) were used to perform realization
was evaluated using Matlab Reservoir Simulation Toolbox
reduction using the ranking based method. Equations for
(MRST) (Lie et al., 2012) on different geological realizations.
PVnet and OOIP are provided as follows:
MRST provided the COP for each Producer well location
plan. The simulation time horizon for the simulator was set as net
3000 days divided into 10 equal periods. The case study PV net  V  c c
Ic (13)
c
parameters used by MRST are provided in Table 1.
OOIP  V  c c
(1  S c ) (14)
c
Table 1. Case study parameters
where, Vc is the volume of reservoir cell c, ϕc is the porosity
Parameter Value of cell c, Sc is the irreducible water saturation of cell c and
Initial pressure po 5080 psi Icnet is an indicator to see if cell c is a net cell (Icnet=1) or non-
Oil viscosity μo at po 1.18 cp net cell (Icnet=0).
Water viscosity μw at po 0.325 cp
Oil density ρo 865 kg/m3
Water density ρw 929 kg/m3
Relative permeability exponent for oil no 2
Relative permeability exponent for water nw 2
Residual phase saturation for oil Sro 0
Residual phase saturation for water Srw 0
Relative permeability for oil kwmo at Sro 1
Relative permeability for water kwmw at Srw 1

In this study, a superset of 100 realizations were generated

for realization reduction. For each realization, porosity values
of the reservoir grid were generated in MRST using a built-in
function ‘GaussianField’ with a range parameter of [0.1,
0.5]. The function creates an approximate Gaussian random
field by convolving a normal distributed random field with a
Gaussian filter with a standard deviation of 2.5 (Lie et al.,
2012). Permeability values were further generated from the
porosity values using Karmen-Cozeny relationship (Lie et al.,
2012)
3
1 c
kc  2 2
(12)
2 A (1   c )
v

In (12), kc is the permeability of cell c, ϕc is the porosity of

cell c, Av is the surface area of spherical uniform grains with
a constant diameter of 10 and τ is tortuosity with value 0.81
(Lie et al., 2012). In the case study, the well placement
optimization results using a subset of realizations from the
proposed method are compared to subset of realizations Fig. 2. (top) Well placement plan using selected realizations
obtained using static measure based ranking method and from proposed method; (bottom) Well placement plan using
random selection.10 realizations were selected for the subset full set of realizations
of realizations. In random selection, 10 realizations are
arbitrarily selected from the superset of realizations. The decision variables for the case study were the X and Y
To evaluate the static measures for different geological locations of the producer wells to be placed. The objective
realizations, the threshold porosity is set as ϕo = 0.3 and was to maximize the risk averted expected cumulative oil
production after 3000 days of the simulation period. The well

60
 IFAC ADCHEM 2015
June 7-10, 2015. Whistler, BC, Canada Shahed Rahim et al. / IFAC-PapersOnLine 48-8 (2015) 057–062 61

placement optimization problem was simulated using a
system with 3.2GHz Intel Core i5 processor and 8 GB
memory. Well placement plans obtained from the
optimization on the case study using the subset of realizations
and the superset of realizations are provided in Fig. 2. In Fig.
2, the fixed injector wells are denoted by blue dot and the
producer well locations are denoted by red dot.
It is evident from Fig. 2 that the producer well placement
plan using a subset of realizations from the proposed
realization reduction method is very similar to the producer
locations used from the well placement optimization using all
the realizations. The computational time for the well
placement optimization and the mean and variance of the
COP from the final prodcuer well location of using the subset
of realizations using different methods and the original
becomes very close to the expected COP from the full set of
realizations.
Similarly, Fig. 4 provides the plot of standard deviation of
COP and error in the standard deviation of COP versus the
number of iterations used by the NOMAD optimizer for the
different realization reduction methods. Fig. 4 further
confirms that the amongst the realization reduction methods,
the proposed method has a standard deviation of the COP
closest to the standard deviation of the COP calculated with
all the realizations in the superset.
4
x 10
1.16
1.14

superset of all the realizations are given in Table 2.

1.12 ALL

Expected COP (m3/day)

Proposed
Table 2. Case study results OOIP
1.1 PVnet
COP
Mean Simulation Random
standard
COP time 1.08
deviation
(m3/day) (hours)
(m3/day) 1.06

All realizations 11529 1000.4 25

1.04
Selected
realizations from 11537 940.3 2.5 1.02
proposed method 0 100 200 300 400 500 600 700 800 900 1000
No. of iterations
Selected
realizations from 11554 1402.5 2.5 1000
PVnet ranking Proposed
900 OOIP
Selected
PVnet
realizations from 11472 1474.9 2.5 800 Random
OOIP ranking
Expected COP error (m3/day)

700
Selected
600
realizations from 11310 1235.4 2.5
random selection 500

400

300
Table 2 shows that the well placement optimization results
using the subset of realizations selected by the proposed 200
realization reduction method has the closest mean and
100
variance values of COP when compared to the mean and
variance COP obtained using all the realizations in the well 0
0 100 200 300 400 500 600 700 800 900 1000
placement optimization method. More importanlty, the well No. of iterations

placement optimization problem using a subset of realization
takes one-tenth of the computational time with the same 1000
iterations of optimization.
Fig. 3 provides the expected COP versus the number of
iterations used by the NOMAD optimizer for the different
realization reduction methods. It is clear that the expected
COP of the well placement plan using the subset of
realizations comes closest to the expected COP of the well
placement plan using all the realizations as the number of
iterations increases. The expected COP error is the absolute
difference between the expected COP of the well plans using
different realizations reduction methods with the expected
COP of the well plan using all the realizations in the superset.
It is clear from Fig. 3 that as the number of iterations
increases, the expected COP of the proposed method
Fig. 3. (top) Expected COP versus number of iterations using
different realization reduction methods; (bottom) Error in
expected COP versus number of iterations using different
realization reduction methods
5. CONCLUSION
In this study, a framework for well placement optimization
with geological uncertainty reduction was proposed. The well
placement optimization was formulated as a risk averted
optimization problem by considering geological uncertainty.
The optimization problem is solved using the derivative free
optimization method. Geological uncertainty was
incorporated into the robust optimization model which is
formulated based on a set of optimally selected realizations.
The subset of realizations was selected from a superset of

61
IFAC ADCHEM 2015
62
June 7-10, 2015. Whistler, BC, Canada Shahed Rahim et al. / IFAC-PapersOnLine 48-8 (2015) 057–062

realizations using a mixed integer linear optimization model REFERENCES

with an objective of finding out the optimal subset which has
Audet, C., Le Digabel, S.,Tribes C. (2009).
a similar statistical distribution characteristic to the superset
NOMAD user guide. Technical Report G-2009-37, Les
of realizations. The realization reduction model is
cahiers du GERAD.
independent of well positions and depends on the reservoir
Bangerth, W., Klie, H., Wheeler, M.F., Stoffa, P.L.,
geology. Results from case studies show that the well
Sen, M.K. (2006). On optimization algorithms for the
placement optimization problem using the proposed
reservoir oil well placement problem. Comput Geosci, 10,
realization reduction method is very efficient. The well
303-319.
placement plan obtained using the small subset of realizations
Beckner, B.L., Song, X. (1995). Field development
and the well placement plan for the superset of realizations
planning using simulated annealing – optimal economic well
are very similar and have similar mean COP values.
scheduling and placement, SPE Annual Technical
Significant reduction in the computational time was achieved
Conference and Exhibition Dallas, Texas.
by using a subset of realizations in the well placement
Bittencourt, A.C., Horne, R.N. (1997). Reservoir
optimization problem. Comparison studies show that the
development and design optimization, SPE Annual Technical
results of the proposed method are also superior to the
Conference and Exhibition San Antonio, Texas.
traditional ranking method and the random selection method.
IBM. (2009) User’s Manual for CPLEX V12.1.
ftp://public.dhe.ibm.com/.
1500 Lie, K.A., Krogstad, S., Ligaarden, I., Natvig, J.,
Nilsen, H., Skaflestad, B. (2012) Open-source MATLAB
1400 implementation of consistent discretization on complex grids.
Comput Geosci, 16, 297-322.
1300 Nasrabadi, H., Morales, A., Zhu, D. (2012). Well
Standard Deviation

placement optimization: A survey with special focus on

1200 ALL
Proposed
application for gas/gas-condensate reservoirs. Natural Gas
OOIP Science and Engineering, 5, 6-16.
1100
PVnet Onwunalu, J.E., Durlofsky, L.J. (2010). Application
Random
1000
of a particle swarm optimization algorithm for determining
optimum well location and type. Comput Geosci, 14, 183-
900 198.
Rahim, S., Li, Z., Trivedi, J. (2014). Reservoir
800 geological uncertainty reduction: an optimization based
0 100 200 300 400 500 600 700 800 900 1000
No. of iterations method using multiple static measures. Mathematical Geosci
(under review).
500 Rosenwald, G.W., Green, D.W. (1974). A method
450
for determining the optimum location of wells in a reservoir
using mixed-integer programming. SPE Journal, 14, 44-54.
400
Proposed
Wang, H., Echeverria-Ciaurri, D., Durlofsky, L.J.,
350 OOIP Cominelli, A. (2012). Optimal well placement under
Standard Deviation erorr

PVnet
300 Random
uncertainty using a retrospective optimization framework.
SPE Journal, 17 (1), 112-121.
250
Yang, C., Card, C., Nghiem, L., Fedutenko, E.
200 (2011). Robust optimization of SAGD operations under
150 geological uncertainties. SPE Reservoir Simulation
Symposium, Woodlands, Texas.
100
Yasari, E., Pishvaie, M.R., Khorasheh, F.,
50 Salahshoor, K. (2013). Application of multi criterion robust
0 optimization in water-flooding of oil reservoir. Petroleum
0 100 200 300 400 500 600 700 800 900 1000
No. of iterations Science and Engineering, 109, 1-11.
Yeten, B., Durlofsky, L.J., Aziz, K. (2003).
Fig. 4. (top) Standard deviation of COP versus number of
Optimization of nonconventional well type, location and
iterations using different realization reduction methods;
trajectory. SPE Journal, 8, 200-210.
(bottom) Error in standard deviation of COP versus number
of iterations using different realization reduction methods

ACKNOWLEDGEMENTS
The authors would like to acknowledge the financial support
from Natural Sciences and Engineering Resource Council
(NSERC) of Canada Discovery Grant.

62