You are on page 1of 11

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/248254264

Calculation OOIP in oil reservoir by pressure matching method using genetic


algorithm

Article  in  Journal of Petroleum Science and Engineering · February 2009


DOI: 10.1016/j.petrol.2008.12.008

CITATION READS

1 1,325

2 authors:

Feridun Esmaeilzadeh Ehsan Nourafkan


Shiraz University University of Lincoln
190 PUBLICATIONS   1,543 CITATIONS    33 PUBLICATIONS   200 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Hydrate study View project

Petroleum Reservoirs Geomechanics and Related Issues: Reservoir Rock Failure and Sand Production Phenomena (Mathematical Modeling and Numerical Simulation
Approach) View project

All content following this page was uploaded by Ehsan Nourafkan on 18 September 2018.

The user has requested enhancement of the downloaded file.


Journal of Petroleum Science and Engineering 64 (2009) 35–44

Contents lists available at ScienceDirect

Journal of Petroleum Science and Engineering


j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / p e t r o l

Calculation OOIP in oil reservoir by pressure matching method using


genetic algorithm
F. Esmaeilzadeh ⁎, E. Nourafkan
Chemical and Petroleum Engineering Department, School of Engineering, Shiraz University, Shiraz, Iran

a r t i c l e i n f o a b s t r a c t

Article history: This paper describes the optimization pressure matching method with a genetic algorithm for the estimation
Received 6 January 2007 of original oil in place (OOIP). In this method, the pressure of a reservoir is obtained based on the material
Accepted 1 December 2008 balance equation by minimizing the difference between calculated and reservoir pressure in order to
optimize reservoir parameters suchas OOIP, aquifer constant and water influx data calculated by the aquifer
Keywords:
function.
original oil in place
The high non-linearity of the pressure matching method makes deterministic classic optimization methods
pressure matching method
modified aquifer model inefficient and unlikely to be successful. Firstly, the solution is dependent on the initial guesses due to the
genetic algorithm non-convexities of the system equations, and secondly, the convergence of the solution is not always
guaranteed.
Therefore, an effective optimization strategy combining a two stage approach, genetic algorithm for the
initialization and identification of the search zone followed by a simplex search method used to refine the
solution, is proposed.
Three examples are provided to illustrate the effectiveness of this work by comparing the results of various
methods (graphical methods and volumetric estimation) with the results of the pressure matching method.
© 2008 Elsevier B.V. All rights reserved.

1. Introduction Graphical methods estimate the reservoir parameters directly


through the cumulative production of reservoir and PVT data. For
The material balance equation (MBE) is one of several methods water drive reservoirs in which cumulative water influx data are not
used for estimating reservoir parameters such as OOIP, OGIP and available, various aquifer models with different precisions were
aquifer constant (reservoir characterization). The material balance presented to estimate the cumulative water influx data. Some of the
equation allows critical decisions concerning depletion plans and more important aquifer models usually considered are the Small Pot
development strategies regarding the reservoir by estimating reser- aquifer for weak water drive reservoirs, the Hurst aquifer and
voir parameters such as OOIP and OGIP to be made. Everdingen–Hurst finite/infinite radial aquifers (Dake, 1978; Hurst,
Several attempts have been made to apply the material balance 1943; Van Everdingen and Hurst, 1949). The selection of the
equation to various types of reservoirs and to solve the equation appropriate aquifer should be based on the reservoir under study
analytically and graphically in order to obtain the initial oil in place and and the water influx amount.
the ratio of the initial free gas to oil in the reservoir. Schilthuis (1936) first Although, in this work we cannot find any data for testing the
developed a general material balance equation for all types of reservoir. model for a strong water drive reservoir, by selecting a suitable aquifer
Havlena-Odeh (1963) showed the material balance equation as a the proposed model is capable of handling the strong water drive.
straight line to obtain OOIP for a natural water drive reservoir. Campbell Tehrani (1985) introduced a pressure matching method for
and Campbell (1978) suggested a new method to estimate the depletion estimating OOIP and gascap size which is quite accurate and can be
mechanism of a reservoir. Several papers have been published for the used for all kinds of reservoirs. Pressure history matching is the key
estimation of the reservoir parameters using graphical methods (Wang stage of the reservoir characterization consisting of the adjustment of
et al., 1992; Hurst, 1973; Omole and Ojo, 1993). the measurements data such as oil, gas and water production, as well
as pressure recorded against time into MBE to achieve an optimal
match between calculated and measured history. Pressure matching is
⁎ Corresponding author. Department of Chemical and Petroleum Engineering, Faculty
of Engineering, P.O. Box 7134851154, Namazi Square, Shiraz, Iran. Tel.: +98 711 230
an optimization problem by defining an objective function that
3071; fax: +98 711 628 7294. minimizes the difference between historic data and those calculated
E-mail address: esmaeil@shirazu.ac.ir (F. Esmaeilzadeh). by the MBE.

0920-4105/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.petrol.2008.12.008
36 F. Esmaeilzadeh, E. Nourafkan / Journal of Petroleum Science and Engineering 64 (2009) 35–44

Table 1
The maximum and minimum of variables for case 2. The GAs parameters considered for
case 2

Variable Minimum Maximum


OOIP (N) 0 532.89
Aquifer constant (U) 0 5.265 × 10− 4
Matrix coefficients (ωi) 0 2
Hurst aquifer constant (a) 0 12

GAs parameters
Maximum number of generations 500
Crossover type Two point crossover
Mutation type Uniform mutation
Number of individual per variable 10
Elite children 2
Crossover fraction 0.8
Fig. 1. Principle of roulette wheel selection. Mutation probability 0.01
Number of children per pair of parents 2
Individual form Binary

Pressure matching is a nonlinear inverse problem, as the


relationship between the reservoir parameters and the data is highly porous rock (Schilthuis, 1936). It is written from the start of
nonlinear (Oliver et al., 2001). This aspect of the problem causes production to any time (t) as follows:
great difficulty to classic optimization methods and may result in the
   
method leading to a poor history matched model (as discussed in   Cw Swi  Cf
N ðBt −Bti Þ + G Bg −Bgi + NBti + GBgi Δpt + NBti + GBgi Δpt ð1Þ
Ballester and Carter, 2006). In addition, these methods are sensitive 1−Swi 1−Swi
to the choice of the initial guess and may often lead to a convergence

failure. + We + WI BIw + GI BIg = Np Bt + Np Rp −Rsoi Bg + Wp Bw
Consequently, in this work we used the hybrid optimization
method which can be applied for a nonlinear problem and is not
where N represents initial oil in place (STB), Np represents cumulative oil
sensitive to starting points. A genetic algorithm (GAs) is used to
produced (STB), G represents initial gas in place (SCF), GI represents
perform a preliminary search in the solution space and find the
cumulative gas injected into reservoir (SCF), We represents water influx
neighborhood of the global optimum. The best solution of the GAs was
into reservoir (bbl), WI represents cumulative water injected into
then used as initial guess for simplex search optimization to converge
reservoir (STB), Wp represents cumulative water produced (STB), Bti
towards the final solution.
represents two-phase formation volume factor (bbl/STB), Boi represents
GAs was first used in Petroleum Engineering in 1980 (Goldberg,
initial oil formation volume factor (bbl/STB), Bgi represents initial gas
1985). Velez-langs (2005) presented many papers in the literature
formation volume factor (bbl/SCF), Bt represents two-phase formation
that describe genetic algorithm applications in seismology, well log
volume factor (bbl/STB), Bg represents gas formation volume factor (bbl/
analysis, reservoir flow simulation, hydraulic fracturing design and
SCF), Bw represents water formation volume factor (bbl/STB), Big
reservoir permeability. In Bush and Carter (1996), a GAs approach
represents injected gas formation volume factor (bbl/STB), BIw repre-
to match the simulation model to the measured history of a
sents injected water formation volume factor (bbl/STB), Rsoi represents
reservoir adjusting some of the model parameters is discussed. In
initial solution gas–oil ratio (Mscf/STB), Rso represents solution gas–oil
other work, Romero et al. (2000) described the implementation
ratio (Mscf/STB), Rp represents cumulative produced gas–oil ratio (Mscf/
reservoir characterization by conditioning the reservoir simulation
STB), Cf represents formation compressibility (psia− 1), Cw represents
model to the production data (history matching) on a structural
water isothermal compressibility (psia− 1), Swi represents initial water
model.
saturation, p(t) represents current reservoir pressure (psia), and ΔPt =Pi −
The structure of this paper is as follows. The following section
P(t) represents reservoir pressure drop (psia).
describes the material balance equation, the pressure matching and
the objective function formulation. The details of the GAs framework
and a step by step description of the hybrid optimization are then
described. Finally, the results of model for three different fields are
presented.

2. General material balance equation

A general material balance equation that can be applied to all


reservoir types was first developed based on a tank model filled with

Fig. 3. The trend of alternation of optimum fitness function value in optimization case 2
Fig. 2. Two point crossover operator. using genetic algorithm.
F. Esmaeilzadeh, E. Nourafkan / Journal of Petroleum Science and Engineering 64 (2009) 35–44 37

In an oil reservoir, the ratio of the initial gascap volume to the through the [0,2] domain by optimization. For calculation of the
initial oil volume is defined as: aquifer constant, much information like viscosity, porosity, net
thickness and absolute permeability from the reservoir are needed
initial gas cap volume GBgi (Hurst, 1973; Omole and Ojo, 1993). But in the present method, the
m= = ð2Þ
initial oil volume NBti values of the aquifer constant are obtained through the optimization
process without using the mentioned data. In this way, the formal
Therefore, the general material balance equations for the oil mathematical description of the objective function is given by the
reservoir can be expressed as follows: equation below.

F = NEt + We ð3Þ
minΦðp; pcal ; θÞ

F = Np ⌊Bo + Rp −Rsoi Bg ⌋ + Wp Bw ð4Þ
in which
Et = Eo + mEg + Efw ð5Þ
θ = ½N; m; U and ω
Eo = Bt −Bti ð6Þ


n pðiÞ−pcal ðiÞ
Bti ΦðθÞ = ∑ j j ð12Þ
Eg = Bg −Bgi ð7Þ i=1 pðiÞ
Bgi
 
Cf + Cw Swi where, Ф(θ) is the objective function in terms of difference between
Efw = ð1 + mÞBo Δpt ð8Þ
1−Swi the field pressure and those calculated from the following equation by
  trial and error, and the output response of the objective function
Bt = Bo + Bg ðRsoi −Rso Þ ð9Þ recognizes the fitness function value.
We = USðp; t Þ ð10Þ

 ⁎ 
pcal ðiÞEfw + USðpcal ðiÞ; t Þ = F ðiÞ + N pi Efw −Eo ðiÞ−mEg ðiÞ
where F represents cumulative reservoir void age (bbl), Et represents  
⁎ C + Cw Swi ð13Þ
overall oil expansion (RB/STB), Eo represents cumulative oil expansion Efw = ð1 + mÞBo f
1−Swi
(RB/STB), Eg represents cumulative gas expansion (RB/STB), Efw is
cumulative water and formation expansion (RB/STB), U represents
aquifer constant, S(p,t) represents aquifer function and is different for 4. System solution
various aquifers, Bo represents oil formation volume factor (bbl/STB),
and Boi represents initial oil formation volume factor (bbl/STB). Bo is 4.1. General principle
measured by the differential expansion test, while Bt is measured by
constant composition expansion (CCE). In absence of water influx A large scale of design, control, scheduling or other engineering
data, an aquifer model must be selected to calculate a series of influx problems results in the solution of optimization problems. Optimiza-
values. tion methods are grouped as the stochastic method, which uses the
objective function for finding the optimum, and the mathematic
3. Pressure matching method method, which uses the first or the second derivative of the objective
function to reach the optimum. The first methods are slow and the
General evaluation steps using the pressure matching calculation second methods have failed results for the nonlinear function. In
approach are expressed as follows: addition, these methods are sensitive to the choice of the initial guess
and may often lead to a convergence failure.
• Assume values for N, m, U and a of a particular aquifer model S(p,t).
Over the last 20 years genetic algorithms have been shown to be
• The first pressure point p(1) at the first scheduled data point can be
effective over a wide range of problems, although they were initially
calculated by balancing the material balance equation while
specifying the unknown parameters.
• Second pressure point p(2) can be found with the second scheduled
data and p(1).
• The procedure continues until all the pressure points are found.
• Calculated pressure must match the reservoir pressure. The
correction reservoir parameters that minimize the difference
between the calculated and the reservoir pressure data by the
optimization process can be found.
The influx values are of insufficient quality to determine the OOIP
with confidence. These accrue primarily because the aquifer models
cannot adequately represent the actual aquifer behavior, and also
because of the uncertainty of the input data. In this work, in order to
decrease the error of water influx estimation, the modified aquifer
model is used as follows:

S Vðp; t Þ = ωi Sðp; t Þ ð11Þ

where, ωi is the matrix of coefficients to correct the calculated values


of the aquifer function. The exact value of coefficients is specified Fig. 4. The trend of OOIP alternation in optimization of case 2 using simplex method.
38 F. Esmaeilzadeh, E. Nourafkan / Journal of Petroleum Science and Engineering 64 (2009) 35–44

Fig. 5. The General working framework of hybrid optimization.

proposed as an academic investigation. Genetic algorithm is a The individuals of the offspring population are produced by four
stochastic technique that has been utilized many times in history operator elitism, selection, crossover and mutation which will be
matching optimization (Ballester and Carter 2006; Schulze et al., discussed below.
2002). As previously mentioned, in this work a hybrid optimization
method is applied. 4.3. Elitism operator

4.2. Genetic algorithm The better individual in the parent population can be cloned
directly to the offspring population by the elitism operator. These
The concept of GAs was first introduced by Holland (1975), and since individuals are known as elite children.
then a number of advanced forms of simple GAs have been reported by
several authors (Konak et al., 2006; Bunnag and Sun, 2005).
The GAs optimization starts with a population of individual Table 2
potential solutions of a problem. As the optimization progresses, the PVT properties of oil reservoir, case 1

current population of the solution is known as the parent population Days Pressure (psia) Bo (RB/STB) Rs (Mscf/STB) Bg (RB/Mscf) Bw (RB/STB)
and is progressively replaced by a new population known as offspring. 0 2855 1.2665 0.5010 0.9201 1.0222
The fitter individuals who have a lower fitness function value in the 305 2779 1.2677 0.5010 0.9637 1.0224
parent population are favored over weaker individuals to survive 700 2627 1.2681 1.0502 1.0502 1.0228
1285 2457 1.2554 1.0977 1.0977 1.0232
longer and produce stronger offspring for the next generation. The
1465 2402 1.2512 1.1146 1.1146 1.0233
genetic algorithm covers almost all parts of the solution space in each 2005 2223 1.2383 1.2010 1.2010 1.0237
generation of search, and by contuses generation (evolution of 2365 2080 1.2278 1.2825 1.2825 1.0240
algorithm), the response point is directed to the optimum value. 2905 1833 1.2074 1.4584 1.4584 1.0246
Anyone wishing to learn more about the fundamentals should consult 3325 1665 1.1949 1.6112 1.6112 1.0250
3595 1460 1.1802 1.8526 1.8526 1.0254
one of the introductory texts, such as Michalewicz, (1994).
F. Esmaeilzadeh, E. Nourafkan / Journal of Petroleum Science and Engineering 64 (2009) 35–44 39

Table 3 Table 4
Cumulative production data, case 1 Results of optimization, case 1

Days Pressure Cumulative oil Cumulative gas Cumulative water Parameters


(psia) produced (mmSTB) produced (mmscf) influx(STB) Optimization Method OOIP (N) Aquifer constant (U) ADa
0 2855 0.000 0.000 0 GA 20.16 0.00227 3.87 × 10− 6
305 2779 0.192821 94.513 0 GA + Direct Optimization 20.05 0.00229 3.78 × 10− 6
700 2627 0.633942 312.064 0 n
1285 2457 1.314880 710.670 4
a
AD = 1
n ∑ j ðpðiÞ−p cal ðiÞÞ
pðiÞ j×100.
i=1
1465 2405 1.524400 850.934 7
2005 2223 2.152960 1355.720 26
2365 2080 2.572000 1823.250 60 Individual ← Individual + 1;
2905 1833 3.200560 2732.860 822 sum ← sum + PIndividual;
3235 1665 3.584680 3397.740 11135
3595 1460 4.003720 4216.120 97443 end
select Individual;

4.4. Selection operator 4.5. Crossover and mutation operator

A stochastic sampling method called the proportional selection or The crossover operator comprises the basic mechanism for
qroulette wheelq is used here as the selection operator (Shopova and redistributing genetic characteristics between the individuals. It is
Vaklieva Bancheva, 2006). Its operation is illustrated in Fig. 1. applied to two parent individuals and creates two new offspring
According to this scheme, the probability for the selection of each individuals. The new offspring will have a structure in which both
individual is assumed to be as slots of a Roulette wheel that is parents, genetic coded, are used.
proportional to the fitness value of the individual. Each slot is as wide The crossover adopted here is the two point crossover technique.
as the probability for the selection of the corresponding individual. According to it, two random integers R and R′ between 1 and the
Therefore, the individuals with fitter fitness values should have a number of genes in an individual are produced. The function selects
greater probability of being selected to undergo crossover and genes numbered less than or equal to R from the first parent, genes
mutation. The following function is used to calculate respective numbered from R + 1 to R′, inclusive from the second parent, and
selection probabilities: genes numbered greater than R′ + 1 from the first parent. An
illustration of two point crossover is shown in Fig. 2.
i 
∑ Φ Nj ; mj ; Uj ; ωj The fraction of each population, other than elite children, that is
PIndividual ðiÞ =
j=1
ð14Þ made up of crossover children is specified by the crossover fraction, Pc.
Npop  A crossover fraction of 1 means that all children other than elite
∑ Φ Nj ; mj ; Uj ; ωj
j=1 individuals are crossover children, while a crossover fraction of 0
means that all children are mutation children.
where, Φ represents objective function value, Npop represents the For example, if the population size is 20, the elite children are 2,
individuals number in the population, and PIndividual represents the and the crossover fraction is 0.8. There are 18 individuals other than
probabilities selection of each individual. elite children, so the algorithm rounds off 0.8 ⁎ 18 = 14.4 to 14 to get the
Roulette wheel selection emphasizes the selection of the better number of crossover children. The remaining 4 individuals, other than
individual in the population and exerts extreme pressure on the elite children, are mutation children.
search process. The roulette wheel selection algorithm is shown Like the crossover procedure, the mutation operation is performed
below: on the entire population. The mutation operation creates a new
offspring from only one individual by changing one or more genes in it.
Individual ← 1; The mutation technique chosen here is uniform mutation type,
sum ← PIndividual; which has a two-step process. First, the algorithm selects a fraction of
R ← random number (0,1); the genes of an individual for mutation, where each gene has a
while sum b R do mutation rate, Pm, of being mutated. In the second step, the algorithm
generates a random number and replaces each selected gene if the
random number is smaller than the mutation rate, Pm.
The GAs program is used, coupled with the pressure matching
method (Eqs. (1)–(10)) in order to optimize the OOIP. GAs need some
input data to optimize, including:
• GAs parameters: the parameters of a number of an individual for
each variable, number of iteration, type of crossover and mutation,

Table 5
Comparison results, case 1

Sw 20.8%
Cf 2.28 × 10− 6 psi− 1
Cw 26 × 10− 6 psi− 1

Eclipse result
OOIP ≅20 million STB
Method ADa
Small pot aquifer (this work) 20.05 million STB 0.25
FE (Pletcher, 2002) 30 million STB 50
OOIP (pot aquifer, Pletcher, 2002) 20.3 million STB 1.5
Fig. 6. Comparison results of the calculated pressure by modified pressure matching
a
method with the field data, case 1. AD = (|Calculated Result − Eclipse Result|) / Eclipse Result × 100.
40 F. Esmaeilzadeh, E. Nourafkan / Journal of Petroleum Science and Engineering 64 (2009) 35–44

Table 6 Table 10
PVT properties of oil reservoir, case 2 Comparison results, case 2

Days Pressure (psia) Bo (RB/STB) Rs (Mscf/STB) Bg (RB/Mscf) Rp (Mscf/STB) Sw 5%


0 2740 1.404 0.650 0.930 0 3 × 10− 6 psi− 1
365 2500 1.374 0.592 0.980 0.7600 Cw 4 × 10− 6 psi− 1
730 2290 1.349 0.545 1.070 0.8450 Volumetric data OOIP (mmstb)
1095 2109 1.329 0.507 1.170 0.9205 OOIP by volumetrics 312
1460 1949 1.316 0.471 1.280 0.9751 Method ADa
1825 1818 1.303 0.442 1.390 1.0250 Optimum Hurst aquifer (this work) 320.14 2.60
2190 1702 1.294 0.418 1.500 1.0650 Schilthuis (graphical method) 464.32 48.82
2555 1608 1.287 0.398 1.600 1.0950 Hurst (graphical method) 459.21 47.18
2950 1535 1.280 0.383 1.700 1.1200 a
AD = (|Calculated Result − Volumetric Result|) / Volumetric Result × 100.
3285 1480 1.276 0.371 1.760 1.1450
3650 1440 1.273 0.364 1.820 1.1600

• Problem variables maximum and minimum range allowed vari-


Table 7 ables: these minimum and maximum allowed variables are
Cumulative production data, case 2 dependent on the specific problem being considered. In the present
Days Pressure (psia) Cumulative oil produced (STB) work, optimizing parameters contain the N, m, U and ω. Selection of
0 2740 0 this range for variables is based on the initial estimation and the
365 2500 7.88 properties of the reservoir. For example, if the reservoir has a gas
730 2290 18.42 cap, the range for variable m is between 1 and 5. And if the gas cap is
1095 2109 29.15
small, then the range of variable m is between 0 and 1. The range of
1460 1949 40.69
1825 1818 50.14 ω factor for correcting the aquifer model has been chosen between 0
2190 1702 58.42 and 2. The OOIP is allowed to vary between 0 and the highest
2555 1608 65.39 maximum possible of OOIP. This maximum amount can be
2950 1535 70.74 calculated by MBE based on the hypothesis of the lack of gascap
3285 1480 74.54
3650 1440 77.43
and water influx in the reservoir.
A step-by-step description of the optimization in this work is
outlined as follows:
the probability crossover and mutation must be supplied to the GAs
program. GAs parameters must be varied in many GAs trials in order 1. At first, define the mathematic formulate of problem, objective
to achieve the specific process global optimum. Some parameters function and aquifer function.
such as the number of iterations, crossover, and mutation type are 2. Specify the maximum and minimum of variables. The range of
changed by the complexity of the problem. The final parameters of variables for case 2 is presented in Table 1.
GAs are defined based on convergence and the fitness function 3. Define the parameters of GAs. The GAs parameters considered for
value, which need numerous calculations and take much time. case 2 are presented in Table 1.

Table 8
Results of optimization, case 2

Optimized parameters with using modified aquifer model


Optimization method OOIP (N) Aquifer constant (U) Hurst aquifer constant (a) ADa
GA 320.17 3.30 × 10− 4 0.261 5.3 × 10− 5
GA + Direct Optimization 320.14 3.30 × 10− 4 0.262 4.9 × 10− 5

Optimized parameters without using modified aquifer model (ωi = 1)


GA 535.05 3.72 × 10− 6 5.35 1.338 × 10− 2
GA + Direct Optimization 549.66 −2.35 × 10− 5 165.29 1.277 × 10− 2
n
a
AD = 1
n ∑ j ðpðiÞ−p cal ðiÞÞ
pðiÞ j×100.
i=1

Table 9
Calculation results of the optimum coefficients of modified aquifer model and water influx into the reservoir, case 2

Optimized matrix coefficients, ωi, with using modified aquifer model


Optimization method ω0 ω1 ω2 ω3 ω4 ω5 ω6 ω7 ω8 ω9 ω10
GA 0 1.202 0.977 0.842 0.747 0.669 0.603 0.528 0.506 0.454 0.522
GA + Direct Optimization 0 1.201 0.978 0.843 0.747 0.669 0.604 0.529 0.506 0.455 0.523

Calculated water influx we (mmrb), with using modified aquifer model


GA 0 3.806 10.083 19.028 27.642 35.952 43.617 48.804 58.366 61.527 82.522
GA + Direct Optimization 0 3.803 10.083 19.037 27.631 35.928 43.661 48.865 58.330 61.625 82.584

Calculated water influx, we (mmrb), without using modified aquifer model (ωi = 1)
– 0 −0.0935 −0.346 − 0.729 − 1.221 − 1.8037 −2.460 − 3.178 −4.007 − 4.741 − 5.566

Calculated Water influx, we (mmrb), using Eq. (3) with volumetric OOIP value (N = 312)
Eq. (3) 0 3.4 12.2 23.09 34.6 45.91 56.53 66.12 74.47 81.53 87.29
F. Esmaeilzadeh, E. Nourafkan / Journal of Petroleum Science and Engineering 64 (2009) 35–44 41

Table 11
PVT properties of oil reservoir, case 3

Days Pressure (psia) Bo (RB/STB) Rs (Mscf/STB) Bg (RB/Mscf) Rp (Mscf/STB)


0 5869 1.3773 0.657 0.6311 0
349 5479 1.3609 0.6095 0.6586 0.7561
417 5335 1.3548 0.592 0.6701 0.850
526 5223 1.3501 0.5784 0.6794 0.8595
830 4923 1.3376 0.5422 0.7070 0.9341
936 4870 1.3353 0.5357 0.7125 0.958
1299 4650 1.3262 0.5093 0.7362 1.003
1660 4375 1.3148 0.4765 0.7697 1.059
2020 4080 1.3027 0.4414 0.8122 1.093
2378 3750 1.2893 0.4025 0.8694 1.1179

“fminsearch” is a simplex optimization method which uses the


Nelder–Mead method to minimize the objective function by tuning
the parameters values (Lagarias et al., 1998). This method finds the
Fig. 7. Comparison of the results of calculated water influx data with and without using
minimum of a scalar function of several variables, starting at an
modified aquifer model with those calculated by using Eq. (3) with the volumetric OOIP
value (N = 312 mmstb), case 2. initial guess. The simplex search method is thus used to refine the
GA solution. The final answer of GAs and the fminsearch routine
was shown in Tables 8 and 9. Fig. 4 shows the trend of OOIP
4. Perform the optimization through GAs in order to obtain the alternation in the optimization of case 2 using the simplex method.
problem response of these GAs parameter values. The initial
population for each pressure match variable (OOIP, gascap and Fig. 5 presents the general working framework of hybrid optimization.
matrix of coefficients, ωi) was synthesized by 10 random numbers
from a range of variables which cover all solution space. The 5. Results and discussion
calculation for each set of individual is done by MBE equation
through Eq. (13), and the fitness function value can result from The calculation results of the present method for three types of oil
Eq. (12). The random numbers of pressure match variables convert reservoir are compared with the graphical results and volumetric
to binary form individuals (0 and 1) which were used in the GAs estimation. The results of the optimization process in comparison with
optimization process. The set of individuals with the lower fitness other methods are presented as separate tables for each case.
function value have a greater chance to create next generation
individuals. The new population (offspring) was produced by elite 5.1. Case 1
offspring, and performance selection, crossover and mutation
operator on the fitter binary individual. Therefore, the general This is an under-saturated reservoir with a weak water drive
population thereby approaches optimum value from generation to mechanism. The cumulative productions and PVT data are listed in
generation. The process is repeated until a specified maximum Tables 2 and 3. The OOIP of the reservoir is calculated by Eclipse
number of generations can be reached. software (Pletcher, 2002). A Small Pot aquifer model is used to obtain
Since the trend of calculation in GAs is stochastic, nonconvergence the water influx data. The calculated results of the present work in
in the results or deviation from optimal values are probable. Thus, comparison with the field data are shown in Fig. 6. The calculated
in order to reach the desired result, the GAs must be repeated many results of this work with respect to the graphical methods in
times. Fig. 3 displays the trend of alternation of the optimum fitness comparison with the Eclipse software result show that the presented
function value in optimization case 2. method has fewer errors as compared to the other results. Table 4
5. GA cannot guarantee finding the optimum solution, but it is able to represents the optimum value of OOIP, and the aquifer constant.
determine a near-optimal solution. The best individuals obtained Table 5 shows the comparison results of the present work with the
from the GA will now serve as an initial point for the unconstrained other methods.
nonlinear optimization in order to converge towards the global
optimum (fminsearch routine in Matlab software). The routine 5.2. Case 2

This is a water drive reservoir without gascap (Dake, 1978). The


cumulative productions and PVT data are listed in Tables 6 and 7. In
this case, the effect of optimization on calculated water influx is

Table 12
Cumulative production data, case 3

Days Pressure Cumulative gas produced Cumulative oil produced


(psia) (mmscf) (STB)
0 5869 0.000 0
349 5479 5.9888 0.635
417 5335 15.5649 1.000
526 5223 26.8326 1.338
830 4923 39.6768 2.429
936 4870 51.3935 2.759
1299 4650 62.2173 3.979
1660 4375 71.6021 5.201
2020 4080 79.2288 6.491
2378 3750 89.8188 7.922
Fig. 8. Calculated water influx for different cases, case 2.
42 F. Esmaeilzadeh, E. Nourafkan / Journal of Petroleum Science and Engineering 64 (2009) 35–44

Table 13
Results of optimization, case 3

Optimized parameters with using modified aquifer model, emphasis m = 0.169


Optimization method OOIP (N) Initial gascap volume fraction (m) Aquifer constant (U) Hurst aquifer constant (a) ADa
−5
GA 22.61 – 3.92 × 10 11.82 8.75 × 10− 4
GA +Direct Optimization 22.40 – 3.86 × 10− 5 12.00 2.20 × 10− 5

Optimized parameters with using modified aquifer model


GA 22.27 0.094 4.185 × 10− 5 11.917 1.26 × 10− 5
GA + Direct Optimization 23.46 0.094 4.184 × 10− 5 11.902 5.47 × 10− 6

Optimized parameters without using modified aquifer model (ωi = 1), emphasis m = 0.169
GA 30.19 – 2.95 × 10− 5 7.150 5.11 × 10− 3
GA + Direct Optimization 33.62 – 1.786 × 10− 5 0.349 3.42 × 10− 3
n
a
AD = 1
n ∑ j ðpðiÞ−p cal ðiÞÞ
pðiÞ j×100.
i=1

Table 14
Calculation results of the optimum coefficients of modified aquifer model and water influx into the reservoir, case 3

Optimized matrix coefficients, ωi, with using modified aquifer model, emphasis m = 0.169
Optimization method ω0 ω1 ω2 ω3 ω4 ω5 ω6 ω7 ω8 ω9
GA 0 1.562 1.485 1.603 1.272 1.253 1.093 0.978 0.957 0.889
GA + Direct Optimization 0 1.589 1.478 1.602 1.275 1.255 1.091 0.977 0.955 0.880

Calculated water influx, we (mmrb), with using modified aquifer model, emphasis m = 0.169
GA 0 0.492 0.690 1.208 2.271 2.781 4.215 5.671 7.748 9.578
GA + Direct Optimization 0 0.500 0.675 1.187 2.238 2.738 4.136 5.570 7.602 9.321

Optimized ωi matrix coefficients, with using modified aquifer model


Optimization method ω0 ω1 ω2 ω3 ω4 ω5 ω6 ω7 ω8 ω9
GA 0 1.594 1.488 1.612 1.280 1.261 1.095 0.983 0.960 0.845
GA + Direct Optimization 0 1.765 1.601 1.648 1.283 1.231 1.067 0.956 0.936 0.876

Calculated water influx, we (mmrb), using modified aquifer model


GA 0 0.597 0.798 1.334 2.447 2.933 4.406 5.925 8.092 9.998
GA + Direct Optimization 0 0.603 0.794 1.325 2.444 2.915 4.390 5.914 8.084 10.068

Calculated water influx, we (mmrb), without using modified aquifer model (ωi = 1), emphasis m = 0.169
– 0 0.253 0.365 0.586 1.348 1.666 2.842 4.216 5.826 7.685

Calculated water influx, we (mmrb), using Eq. (3) with volumetric data (N = 22.36 and m = 0.169)
Eq. (3) 0 0.603 0.794 1.325 2.444 2.915 4.390 5.914 8.084 10.068

investigated with and without using a modified aquifer model. Table 8 listed in Tables 11 and 12. The optimization pressure matching method
represents the optimum value of OOIP, aquifer constant and Hurst is performed for the following cases:
aquifer constant with and without optimizing the calculated water
• Using the modified aquifer model with a constant value of m
influx data. The optimum coefficient of the matrix, ωi, the calculated
(m = 0.169).
water influx and the calculated water influx by volumetric OOIP
• Using the modified aquifer model and optimization value of m.
(N = 312 mmstb) using Eq. (4) are shown in Table 9. Table 10 shows the
• Optimization process without using the modified aquifer option.
comparison results of the present work with the other methods. Fig. 7
shows the calculated results of the optimized pressure matching
method with and without using the modified aquifer model in Table 15
Comparison results, case 3
comparison with the field data. As can be seen, the fluctuations of
calculated pressure have been observed for the unmodified aquifer Sw 22%
model. This trend causes more errors in the calculation of OOIP. In Cf 3 × 10− 6 psi− 1
Cw 3 × 10− 6 psi− 1
addition, the results show that the optimized pressure matching
method using the modified aquifer model is superior to those not Volumetric data OOIP (mmstb)
using it. The results of water influx calculation with and without using OOIP by volumetrics 22.36
the modified aquifer model are compared with those of the water Initial gascap volume fraction 0.169
influx data calculated by using Eq. (3) with the volumetric OOIP value
(N = 312 mmstb) shown in Fig. 8. As can be seen, the calculated water Method ADa
influx using the modified aquifer model is more in agreement with Optimum Hurst aquifer, m = 0.169 (this work) 22.40 0.17
Optimum Hurst aquifer with optimizing m (this work) 23.46 0.44
those of the volumetric OOIP with respect to the aquifer model.
Optimum initial gascap fraction m (this work) 0.094 4.4
FE, m = 0.169 36.9 65
5.3. Case 3 Campbell, m = 0.169 29.19 30
Schilthuis aquifer, m = 0.169 32.09 43
This is a small gascap reservoir with a water drive mechanism Hurst aquifer, m = 0.169 22.69 1.4

(Wang et al., 1992). The cumulative productions and PVT data are a
AD = (|Calculated Result − Volumetric Result|) / Volumetric Result × 100.
F. Esmaeilzadeh, E. Nourafkan / Journal of Petroleum Science and Engineering 64 (2009) 35–44 43

• The presented method is able to obtain the aquifer constant without


using the reservoir parameters such as viscosity, porosity, net
thickness and absolute permeability, however this information is
necessary for the calculation of the aquifer constant in other methods.
• Comparison results show that the accuracy of the water influx data
calculation is improved by modifying a simple aquifer function such
as the Hurst aquifer model, while in other models, for the exact
estimation of water influx data, a complex aquifer model such as the
EH finite/infinite model was needed.

Acknowledgement

The authors are grateful to the Shiraz University for supporting this
research.

References
Fig. 9. Comparison results of the calculated pressure with the field data for different
cases, case 3.
Ballester, P.J., Carter, J.N., 2006. Characterising the parameter space of a highly nonlinear
inverse problem. Inverse Probl Sci Eng 14 (2), 171–191.
Bunnag, D., Sun, M., 2005. Genetic algorithm for constrained global optimization in
continuous variables. Appl Math Comput 171, 604–636.
Bush, M.D., Carter, J.N., 1996. Application of a modified genetic algorithm to parameter
estimation in the petroleum industry. In: Dagli, C., Akay, E., Philip-Chen, C.L.,
Fernandez, B., Ghosh, J. (Eds.), Intelligent Engineering Systems through Artificial
Neural Networks, vol. 6. ASME Press, New York.
Campbell, R.A., Campbell Sr., J.M., 1978. Mineral property economics. Petroleum
Property Evaluation, Campbell Petroleum Series, vol. 3.
Dake, L.P., 1978. Fundamental of Reservoir Engineering. Elsevier Scientific Publishing
Company.
Goldberg, D.E., 1985. Computer-aided gas pipeline operation using genetic algorithms
and rule learning part I: genetic algorithms in pipeline optimization. Society of
Petroleum Engineers e-library (SPE 14590) vol. 14590.
Holland, J., 1975. Adaptation in Natural and Artificial Systems. MIT Press, Cambridge,
MA.
Havlena, D., Odeh, A.S., 1963. The material balance as an equation of a straight line. JPT,
Trans., AIME, vol. 228.
Hurst, W., 1943. Water influx into a reservoir and its application to the equation of
volumetric balance. Trans., AIME, vol. 151, p. 57.
Hurst, W., 1973. Technical Note the Material Balance Equation. Petroleum Reservoir
Engineer, Texas.
Fig. 10. Calculated water influx data for different cases, case 3.
Konak, A., David, Coit, W., Alice, Smith, E., 2006. Multi-objective optimization using
genetic algorithms: a tutorial. Reliab Eng Syst Saf 91 (No. 9), 992–1007.
Lagarias, J.C., Reeds, J.A., Wright, M.H., Wright, P.E., 1998. Convergence properties of the
Nelder–Mead simplex method in low dimensions. SIAM J Optim 9 (1), 112–147.
The results of the calculation show that the absolute errors of
Michalewicz, Z., 1994. Genetic Algorithm + Data Structure = Evolution Programs, 2nd Ed.
calculation decreased by using the modified aquifer model. The Springer-Verlag, New York.
optimized value of the parameters is shown in Table 13. The optimum Omole, O., Ojo, K.P., 1993. A New Method for Estimating Oil in Place and Gascap Size
values of ωi matrix, the calculated water influx data for three cases and Using the Material-Balance Equation. SPE. 026266.
Oliver, D.S., Reynolds, A.C., Bi, Z., Abacioglu, Y., 2001. Integration of production data into
the calculated water influx data using Eq. (3) with the volumetric data reservoir models. Pet Geosci Sci 7, 65–S73.
(N = 22.36 mmstb and m = 0.169) are shown in Table 14. The comparison Pletcher, J.L., 2002. Improvements to Reservoir Material-Balance Methods. SPE, pp. 49–59.
results of the present model with the other methods are shown in Romero, C.E., Carter, J.N., Zimmerman, R.W., Gringarten, A.C., 2000. A modified genetic
algorithm for reservoir characterization. Proc. of the 2000 SPE Seventh Interna-
Table 15. The calculated and field pressure data for the mentioned tional Oil and Gas Conference and Exhibition, Beijing, China.
cases are shown in Fig. 9. The results of the water influx calculation Schilthuis, R.J., 1936. Active oil and reservoir energy. Trans AIME vol. 118, 33.
with and without using the modified aquifer model are compared with Schulze, R.W., Riegret, Axmann, J.K., Hease, O., Rian, D.T., You, Y.L., 2002. Evaluationary
algorithms applied to history matching of complex reservoirs. SPE Reser Evalu Eng
those of the water influx data calculated by using Eq. (3) with the 5 (2), 163–175.
volumetric values (N = 22.36 mmstb, m = 0.169) shown in Fig. 10. Shopova, E.G., Vaklieva Bancheva, N.G., 2006. BASIC—a genetic algorithm for
engineering problems solution. Comput Chem Eng 30, 1293–1309.
Tehrani, D.H., 1985. An analysis of volumetric balance equation for calculation of oil in
6. Conclusions place and water influx. J Petrol Technol 1664–1670.
Van Everdingen, A.F., Hurst, W., 1949. Application of the Laplace transforms to flow
The optimization of the pressure matching method by using problem in reservoirs. Trans AIME 186, 304–324.
Velez-Langs, O., 2005. Genetic algorithms in oil industry: an overview. J Petrol Sci Eng
genetic algorithm (stochastic optimization) with the simplex Nelder–
47, 15–22.
Mead method (direct optimization) is investigated. The calculation Wang, B., Litvak, B.L., Bowman, G.W., 1992. OILWAT: Microcomputer Program for Oil
results for three fields using the pressure matching method are Material Balance with Gascap and Water Influx. SPE. 24437.
compared with the results of volumetric data and graphical methods.
The results show that good agreement will be achieved.
Meanwhile, the presented method can be applied for an oil Glossary
reservoir which cannot be solved by a conventional method. Some of
Bg: gas formation volume factor, bbl/SCF
the important features in the presented method are:
Bgi: initial gas formation volume factor, bbl/SCF
• This method is not sensitive to the number of unknown parameters Bo: oil formation volume factor, bbl/STB
Boi: initial oil formation volume factor, bbl/STB
and all parameters can be estimated by optimization, while the
Bt: two-phase formation volume factor, bbl/STB
graphical methods cannot be used for the reservoir containing three Bti: initial two-phase formation, bbl/STB
or more unknown parameters. Bw: water formation volume factor, bbl/STB
44 F. Esmaeilzadeh, E. Nourafkan / Journal of Petroleum Science and Engineering 64 (2009) 35–44

BIw: initial water formation volume factor, bbl/STB pchrom: probabilities selection of each chromosome
Co: oil compressibility, psia− 1 pi: initial reservoir pressure, psia
Cf: formation compressibility, psia− 1 Pm: mutation rate, dimensionless
Cw: water isothermal compressibility, psia− 1 Rsoi: initial solution gas/oil ratio, Mscf/STB
Efw: cumulative water and formation expansion, RB/STB p: current reservoir pressure, psia
Eg: cumulative gas expansion, RB/STB Rp: cumulative produced gas/oil ratio, Mscf/STB
E o: cumulative oil expansion, RB/STB Rs: solution gas/oil ratio, Mscf/STB
Et: overall oil expansion, RB/STB S(p,t): aquifer function, psia
F: cumulative reservoir void age, res bbl S′(p,t): modified aquifer function, psia
G: initial gas in place, SCF Swi: initial water saturation, dimensionless
GI: cumulative gas injected into reservoir, SCF U: aquifer constant, mmrb/psia
Gp: cumulative gas produced, SCF we: water influx into reservoir, bbl
m: ratio of OGIP to OOIP, dimensionless wI: cumulative water injected into reservoir, STB
N: initial oil in place, STB wp: cumulative water produced, STB
Npop: individual number in the population, dimensionless
Np: cumulative oil produced, STB Greek letters
OGIP: original gas inplace, SCF ω: matrix of correction coefficient, dimensionless
OOIP: original oil inplace, STB ΔPt: reservoir pressure drop, psia, =Pi − P(t)
p: reservoir pressure, psia Φ: objective function value, psia
Pc: crossover fraction, dimensionless
Pm: mutation rate, dimensionless
pcal: calculation pressure by MBE in pressure matching method, psia

View publication stats

You might also like