Вы находитесь на странице: 1из 15

Petroleum 4 (2018) 75e89

Contents lists available at ScienceDirect

Petroleum
journal homepage: www.keaipublishing.com/en/journals/petlm

Application of artificial intelligence to forecast hydrocarbon


production from shales
Palash Panja a, *, Raul Velasco a, Manas Pathak b, Milind Deo b
a
Energy & Geoscience Institute, 432Wakara Way, Suite 300, Salt Lake City, UT 84108, United States
b
Department of Chemical Engineering, University of Utah, 50Central Campus Dr., Salt Lake City, UT 84112, United States

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

Article history: Artificial intelligence (AI) methods and applications have recently gained a great deal of attention in
Received 15 June 2017 many areas, including fields of mathematics, neuroscience, economics, engineering, linguistics, gaming,
Received in revised form and many others. This is due to the surge of innovative and sophisticated AI techniques applications to
22 September 2017
highly complex problems as well as the powerful new developments in high speed computing. Various
Accepted 22 November 2017
applications of AI in everyday life include machine learning, pattern recognition, robotics, data pro-
cessing and analysis, etc. The oil and gas industry is not behind either, in fact, AI techniques have recently
Keywords:
been applied to estimate PVT properties, optimize production, predict recoverable hydrocarbons, opti-
Surrogate models
LSSVM
mize well placement using pattern recognition, optimize hydraulic fracture design, and to aid in reservoir
ANN characterization efforts. In this study, three different AI models are trained and used to forecast hy-
Oil recovery drocarbon production from hydraulically fractured wells. Two vastly used artificial intelligence methods,
Artificial intelligence namely the Least Square Support Vector Machine (LSSVM) and the Artificial Neural Networks (ANN), are
Unconventional reservoirs compared to a traditional curve fitting method known as Response Surface Model (RSM) using second
order polynomial equations to determine production from shales. The objective of this work is to further
explore the potential of AI in the oil and gas industry. Eight parameters are considered as input factors to
build the model: reservoir permeability, initial dissolved gas-oil ratio, rock compressibility, gas relative
permeability, slope of gas oil ratio, initial reservoir pressure, flowing bottom hole pressure, and hydraulic
fracture spacing. The range of values used for these parameters resemble real field scenarios from prolific
shale plays such as the Eagle Ford, Bakken, and the Niobrara in the United States. Production data
consists of oil recovery factor and produced gas-oil ratio (GOR) generated from a generic hydraulically
fractured reservoir model using a commercial simulator. The Box-Behnken experiment design was used
to minimize the number of simulations for this study. Five time-based models (for production periods of
90 days, 1 year, 5 years, 10 years, and 15 years) and one rate-based model (when oil rate drops to 5 bbl/
day/fracture) were considered. Particle Swarm Optimization (PSO) routine is used in all three surrogate
models to obtain the associated model parameters. Models were trained using 80% of all data generated
through simulation while 20% was used for testing of the models. All models were evaluated by
measuring the goodness of fit through the coefficient of determination (R2) and the Normalized Root
Mean Square Error (NRMSE). Results show that RSM and LSSVM have very accurate oil recovery fore-
casting capabilities while LSSVM shows the best performance for complex GOR behavior. Furthermore,
all surrogate models are shown to serve as reliable proxy reservoir models useful for fast fluid recovery
forecasts and sensitivity analyses.
© 2018 Southwest Petroleum University. Production and hosting by Elsevier B.V. on behalf of KeAi
Communications Co., Ltd. This is an open access article under the CC BY-NC-ND license (http://
creativecommons.org/licenses/by-nc-nd/4.0/).

* Corresponding author.
E-mail address: ppanja@egi.utah.edu (P. Panja).
Peer review under responsibility of Southwest Petroleum University.

Production and Hosting by Elsevier on behalf of KeAi

https://doi.org/10.1016/j.petlm.2017.11.003
2405-6561/© 2018 Southwest Petroleum University. Production and hosting by Elsevier B.V. on behalf of KeAi Communications Co., Ltd. This is an open access article under
the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
76 P. Panja et al. / Petroleum 4 (2018) 75e89

1. Introduction recovery were recently described by researchers [46e49]. Other


applications include the description of CO2 solubility [50] and
Surrogate models are particularly useful for quick predictions calcium carbonate [51] in brine sequestration processes.
given a range of input parameters. These models are used to Eight important parameters are considered as input data
forecast oil production and perform sensitivity and uncertainty that include geological parameters (initial reservoir pressure,
analyses. Polynomial equations and other non-linear equations rock compressibility, and permeability), operational parameter
known as response surface models (RSM) have been popularized (bottom hole pressure), completion parameters (fracture spacing),
for their simple mathematical structure and for easier imple- rock-fluid properties (Corey gas relative permeability exponent),
mentation. Recently, artificial intelligence applications have and fluid properties (initial solution gas-oil ratio and the
gained the interest of engineers and scientists due to their un- linear slope of solution gas-oil ratio versus pressure) which are
conventional ways of connecting input data to output. RSM selected from a previous study [52] where a mechanistic
coupled with a proper design of experiments [1] was proven to be study revealed these parameters to be highly significant. Six
an efficient and fast proxy model for forecasting production models (5 time-based models for production after 90 days, 1 year,
performance and analyzing uncertainties [2]. Oil rate and water 5 years, 10 years, and 15 years and one rate-based model when
cut results were also predicted using RSM [3]. Response surface oil rate drops to 5 bbl/day/fracture) of oil recovery and produced
models are widely applied to various aspects of reservoir GOR are developed for each surrogate model (RSM, ANN and
engineering including estimating initial hydrocarbon uncertainty LSSVM). Data is generated from a generic reservoir model
[4], production uncertainty [5e10], finding an optimal scheme for with one vertical hydraulic fracture placed in the middle of
well placement [7,11e14], history matching [13,15,16], and the reservoir using a commercial reservoir simulator. The math-
determining the dew point of water in natural gas processing unit ematical formulations and workflow to create these surrogate
[17]. Field cases have been studied using pattern recognition models are discussed in this article. The results obtained for all
techniques [18] to determine pressure and production variation models are compared using error analyses in terms of coefficient
according to well locations. of determination (R2) and normalized root mean square error
Even though researchers have developed numerical, analytical, (NRMSE).
and semi-analytical techniques to understand the physics
underlying the production from hydraulically fractured tight
2. Reservoir model
formations [19e22], many of these systems grow in complexity
rendering most of these methods inapplicable. The AI approach
Unconventional reservoirs such as shales and other tight
on the other hand is very useful when dealing with highly
formations are very complex in terms of possible natural
complex systems. At the cost of understanding the physical
fracture presence and heterogeneity. However, it is possible to
mechanisms taking place in tight formations, AI helps us analyze
build a homogeneous reservoir model using average properties
and forecast hydrocarbon production and assess performance.
if the variation is not very large. Typically, wells are drilled
In this study, two of the most common AI techniques namely,
vertically and then directed horizontally for 1 to 2 miles,
ANN and LSSVM as well a second order polynomial RSM are used
where as many as 100 vertical hydraulic fractures are induced
to predict oil and gas-oil ratio production from hydraulically
to generate high conductive flow paths to the wellbore.
fractured low permeability reservoir. The comparison of these
Simulating an entire reservoir model that consists of 100
three methods in terms of performance and accuracy is also
hydraulic fractures is very computationally expensive. Hence, only
discussed. The application of ANN started before LSSVM in the
a small representative portion of the reservoir is simulated where
early 90's, data from well tests were already being interpreted
production from a single vertical fracture is considered. The
using ANN [23,24]. Rock characteristics such as lithology were
reservoir properties are assumed to be homogeneous as listed in
determined from well logs using fuzzy neural networks [25].
Table 1.
Reservoir heterogeneity with respect to porosity, permeability,
The number of unique input parameter combinations could
and oil saturations were characterized from geophysical well logs
lead to an enormous number of experiments or simulations. The
such as gamma ray, bulk density, deep induction, etc. using ANN
Box-Behnken method [53] is chosen in this study to keep the
[26]. Thermodynamic properties from reservoir fluids such as
required number of simulations to a minimum. This simulation
bubble point pressures and formation volume factors at the
design is also suitable for second order response surface models.
bubble point have been predicted from four inputs: solution GOR,
Using the Box-Behnken design, 114 simulations are modeled for
reservoir temperature, oil gravity, and gas density using ANN,
eight input parameters in three levels (minimum, medium and
SVM and non-linear regression [27]. Similarly, crude oil viscosity
maximum) as shown in Table 2. The values of all input parameters
and solution GOR as functions of pressure have been determined
are converted to a 1, 0, and 1 scale using a linear relationship,
from 12 variables including compositions of oil, bubble point
except for matrix permeability and rock compressibility (where
pressure, bubble point viscosity [28], etc. using ANN. Calculations
logarithmic values are used instead).
of gas condensate dew point pressures were also made using gas
composition, temperature, and heavy fraction properties [29e31]
and condensate to gas ratio [32]. Results predicted by ANN for
asphaltene precipitation [33] showed promising results Table 1
Simulation parameters.
compared to experimental studies [34]. Oil rates have also been
measured in the pipe line using ANN for varying pressures and Reservoir top depth (ft) 12000
temperatures [35]. Various applications of LSSVM include Reservoir thickness (ft) 200
Reservoir width (ft) 750
porosity and permeability determination [36e39], water conning
Fracture width (ft) 0.05
in horizontal wells [40,41], well placement [40], gas-oil relative Fracture height (ft) 200
permeability curves [42], phase equilibrium calculations of hy- Fracture half-length (ft) 375
drates [43], oil flow rate predictions [44], and temperature- Fracture orientation Parallel to YZ plane
pressure relationship in natural gas production and processing Reservoir porosity (%) 5
Initial water saturation (%) 16
[45]. Wide applications of artificial intelligence in improved oil
P. Panja et al. / Petroleum 4 (2018) 75e89 77

Table 2
Input parameters and their values.

Variable Symbol Minimum (1) Medium (0) Maximum (þ1)

1 Matrix Permeability (nD) 1 10 225 5000


2 Gas Rel. Permeability Exponent, ng 2 1 2 3
3 Rock Compressibility (1/psi) 3 4  106 4  105 4  104
4 dRs/dp (SCF/STB/psi) 4 0.50 0.65 0.80
5 Initial Gas Oil Ratio, 5 800 1900 3000
Rsi (scf/stb)
6 Initial Pressure, Pi (psi) 6 4000 5250 6500
7 BHP (psi) 7 500 1000 1500
8 Fracture Spacing (ft) 8 60 180 300

Apart from the 114 simulation results that were used to train the Table 3
models, 30 additional simulations were ran to test the models. Number of parameters determined in each surrogate model.
Therefore, the training data is comprised of approximately 80% of
Metho Number of parameters Optimized parameter
the total data set (114 out of 144) while the testing portion
RSM 1 intercept all
is comprised of approximately 20% (30 out of 144). The list of
44 coefficients
simulations used to train and test the models can be found in LSSVM 1 Bias term (b) Regularization
Appendix (Tables A.3 and A.4). IMEX™ from the Computer 1 Regularization parameter (g) parameter (g)
Modeling Group, Calgary, Canada was used to conduct all black-oil 1 Kernel parameter (s) Kernel parameter (s)
simulations. The minimum number of simulation grid blocks 92 Support values (ai)
ANN 15 Biases ( 14 hidden layer þ1 output layer) All
necessary to obtain accurate results and avoid convergence issues
126 Weights (8X14 for hidden layerþ14 for
was used as prescribed by Panja et al. [54]. output layer)

3. Surrogate model

As mentioned earlier, three types of surrogate models e a bbl/day/fracture. In other words, five time-based models and one
Response Surface Model (RSM), a Least Square Support Vector rate-based model were developed as summarized below:
Machine (LSSVM) model, and an Artificial Neural Networks (ANN)
model - were developed and compared in this study. Simulation  Time based model: Models for oil recovery and GOR at 90 days, 1
results in terms of oil recovery and gas oil ratio (GOR) were year, 5 years, 10 years and 15 years.
obtained in two ways: by recording oil recovery and GOR  Rate based model: Models for oil recovery and GOR when oil
after certain production times and when the oil rate dropped to 5 rate drops to 5 bbl/day/fracture.

All unknown parameters in the surrogate models (RSM,


LSSVM and ANN) are obtained using an optimization routine
known as Particle Swarm Optimization (PSO) using Matlab
(MathWorks® Inc.). The same optimization routine was used
for all surrogate models to eliminate any performance bias.
Sometimes, unacceptable physical values such as negative oil
recovery factors or gas oil ratios are obtained using surrogate
models. To avoid this pitfall, logarithms of the outcomes (recovery
factors and gas-oil ratio) are used to build the models. A simplified
schematic of methodology used to develop the surrogate model is
shown in Fig. 1.
All unknown parameters are listed in Table 3. These parameters
are discussed in more detail in the upcoming sections.
The first two models (i.e. RSM and LSSVM) are discussed in
detail in a previous article [55]. Therefore, these two models are
intentionally discussed in brief here and the reader is referred to
the referenced article for more details.

3.1. Response surface model (RSM)

The response surface model is the most common method used in


many branches of engineering. Basically, an algebraic equation is
fitted to develop a relationship between input and output data.
During equation fitting with training data, the parameters (co-
efficients, intercepts etc.) are determined through an optimization
routine to minimize error. A second order polynomial equation is
Fig. 1. Surrogate model development schematic. chosen in this study. The equation is defined as:
78 P. Panja et al. / Petroleum 4 (2018) 75e89

Fig. 2. Workflow used to develop RSM and LSSVM. Modified from Panja et al. [55].

been widely applied in various fields because it is easier to


implement, speedy solution convergence, etc. On the other hand,
LSSVM has the inherent nature of overfitting to minimize error.
Various combinations of data training and testing sets such
as 90e10 (%), 85e15(%), 80e20(%), and 70e30 (%) were tried.
Eventually, a data set with 80% used for training and 20% used for
Fig. 3. Division of total data set into LSSVM training, optimization, and test data. testing yielded the best prediction capabilities in this study. The
same combination was used for the other two surrogate models
(RSM and ANN).
The input and output relationship in LSSVM is given by
Equation (2):
X
8 X
8 X
8
f ðX1 ; …; X8 Þ ¼ a0 þ ak Xk þ aij Xi Xj (1)
k¼1 i¼1 j¼i yi ¼ wT 4ðxi Þ þ bwhere xi 2Rp and yi 2R (2)

For 8 input variables, there are 8 interaction coefficients, ak, 36 s The final form of LSSVM is given by Equation (3):
order interaction coefficients, aij, and one intercept, a0, as shown in 2 3
Equation (1). A workflow to develop surrogate models (RSM and /
60 1 1 7
LSSVM) is shown in Fig. 2. 6 7
6 1 7
As part of the development of a model, validation is performed 61 Kðx1 ; x1 Þ þ g / Kðx1 ; xN Þ 7
6 7
using test data to assess robustness. An accepted error margin is set 6 7
6« « 1 « 7
for the surrogate model. In this fashion, surrogate models are 6 7
6 7
continuously improved unless the error reaches its acceptance limit. 4 1 5
1 KðxN ; x1 Þ / KðxN ; xN Þ þ g (3)
ðNþ1Þ X ðNþ1Þ
2 3 2 3
3.2. Least square support vector machine (LSSVM) b 0
6a 7 6y 7
6 1 7 6 1 7
6 7 ¼6 7
The Support Vector Machine (SVM) is usually used for classi- 4 « 5 4 « 5
fication and regression analysis. A modified form of SVM, namely aN yN
ðNþ1Þ X 1 ðNþ1Þ X 1
the least square support vector machine (LSSVM) is used in this
study. LSSVM is close to SVM formulation but solves a linear K (x,xi) is known as the kernel function which is chosen a priori.
system instead of a quadratic programming (QP) problem. It has The Radial Basis Function (RBF) kernel is used in this study as
P. Panja et al. / Petroleum 4 (2018) 75e89 79

shown in Equation 4 are linked through neurons. The most common feed forward
! architecture consists of one input layer, one or more hidden layers,
xi  xj2
  and one output layer as shown in Fig. 4.
Radial Basis Function : K xi ; xj ¼ exp  (4)
s2 Links between input and output are established through the
Where. internal computations in the hidden layers. The complexity and
non-linearity of the model are increased by increasing the number
xi: Input vector of ith data of hidden layers where the individual components of a layer are
b: Bias term known as nodes. In this study, there are nine input nodes (eight
g: Regularization parameter input parameters as listed in Table 2 and one bias) contained in
s: Kernel parameter one hidden layer. A weight was given to each connection for every
ai: Support values node and a bias term was added to each hidden and output node.
Bias and weight values used in this study are summarized for one
It is evident from Equations (2)e(4) that if the regularization output in Table 3. In the process of training the ANN model,
parameter, g, and the kernel parameter, s2, are provided, the all weights and biases are determined by minimizing the
bias term and all support values can be determined from a linear error between the predicted output and the training output via
relationship. This is accomplished by using an optimization tech- activation function at each the node. All output data is normalized
nique where initial g and s2 values are guessed and iteratively as shown in Equation (6):
improved as described in Fig. 2. For the optimization part, the
 
training data is further divided into LSSVM training data (80%) and Yj  Minimum Yj
optimization data (20%) as shown in Fig. 3. Yj; normalized ¼     (6)
Maximum Yj  Minimum Yj
LSSVM training is over once all parameters in Table 3 are found.
At this point, the model can be applied for any unknown input Computations in hidden nodes and output nodes are shown
vector using the RBF kernel as shown in Equation 5 in Fig. 5.
! As shown in Fig. 5 computation consists of two calculations:
X
N
x  x2i summation and transformation through activation functions
yðxÞ ¼ ai exp  þb (5)
i¼1
s2 where activation functions may be linear or non-linear. In
this study, sigmoid transfer function was used as shown in
Equation 7

1
3.3. Artificial Neural Networks (ANNs) Y¼ (7)
1 þ expð  xÞ
The Artificial Neural Networks (ANNs) algorithm was developed The sensitivity of the model to the number of hidden nodes
based on human learning processes through brain and nerve (neurons) was also investigated. As described earlier, the non-
networks. This is a connectionist technique where input and output linearity relationship between input and output data increases

Fig. 4. Basic structure of ANN with input, hidden, and output layers.
80 P. Panja et al. / Petroleum 4 (2018) 75e89

Swarm Optimization (PSO) which is discussed in the upcoming


sections.

3.4. Goodness of fit

There are various error measuring tools used in every branch of


science and engineering. Their uses are mostly dependent on the
model and purpose of the system. During the fitting portion of
the model (training the model) the Mean Square Error (MSE) is
set as the objective function to determine the optimized model
parameters using PSO. As the minimum value of MSE is the indi-
cation of a good match between experimental or simulated values
and modeled values, MSE is minimized during optimization in PSO.
The MSE is calculated between experimental or simulated values
and modeled values as shown in Equation (8).
Pn  2
i¼1 Yobs;i  Ymodel;i
MSE ¼ (8)
n
Yobs and Ymodel are the simulated and modeled values respec-
tively, n is the number of data sets.

Fig. 5. Input-to-output structure and calculations inside (a) hidden and (b) output nodes.

with the number of hidden nodes. However, increasing non-linearity


doesn't always guarantee higher prediction accuracy. To find out the
optimum number of hidden nodes (neurons), a sensitivity study was
conducted on the training and testing data for oil recovery and gas oil
ratio after 5 years of production as shown in Fig. 6.
It is evident from Fig. 6 that the R2 is close to unity for training
data. On the other hand, the R2 value for the test data increases
initially with the number of neurons for oil recovery and gas oil
ratio. A maximum R2 value can be clearly identified at 14 neurons
for the case of oil recovery. Therefore, 14 hidden neurons are used in
this study. The ANN parameters used in study are summarized
below:

 Number of neurons in the first layer (hidden layer) ¼ 14


 Number of neurons in the second layer (output layer) ¼ 1
 Number of weights ¼ Number of neurons in the first layer *
Number of input þ Number of neurons in the first layer *
Number of neurons in the second layer ¼ 126
 Number of biases ¼ Number of neurons in the first layer þ
Number of neurons in the second layer ¼ 15

The unknown parameters in the ANN structure are summarized


in Table 3. During training of the ANN, these 126 wt and 15 biases
are determined using an optimization routine, namely the Particle Fig. 6. Coefficients of determination using different number of hidden nodes for (a) Oil
recovery and (b) gas oil ratio at 5 years.
P. Panja et al. / Petroleum 4 (2018) 75e89 81

Table 4
Particle swarm optimization parameters in various surrogate models.

Surrogate Model Particle Swarm Optimization

Name Parameters Optimized Number of Particles for C1 C2 w Maximum Iteration


a single parameter

RSM 1 intercept 100 2 2 0.6 1000


44 coefficients
LSSVM 1 Regularization parameter (g) 100 2 2 0.6 1000
1 Kernel parameter (s)
ANN 15 Biases 100 2 2 0.6 1000
126 wt

In this study, the Normalized Root Mean Square Error (NRMSE) The NRMSE is defined in Equation (10),
and the coefficient of determination (R2) are adopted to measure pffiffiffiffiffiffiffiffiffiffi
the discrepancy between simulated data and model data. The MSE
NRMSE ¼ (10)
NRMSE is used over MSE to compare various models (time based Yobs;max  Yobs;min
and rate based models) in the same scale. The coefficient of
determination, R2, is defined as shown in Equation 9 Where Yobs,max is the maximum value and Yobs,min is the minimum
value of the observed data.
The value of R2 varies from 0 to 1. R2 values close to unity and
SSres
R2 ¼ 1  (9) small NRMSE values are indication of a good fit.
SStot
Where.
4. Optimization routine: Particle Swarm Optimization
P
SSres ¼ ni¼1 ð Yobs;i  Ymodel;i Þ2 , the residual sum of squares
Pn
SStot ¼ i¼1 ðY obs  Ymodel;i Þ2 , the total sum of squares Inspired by the motion of bird swarms, the Particle Swarm
P
Y obs ¼ 1n ni¼1 Yobs;i , the mean of observed values Optimization (PSO) routine was developed by Eberhart and

Fig. 7. Particle Swarm Optimization flow chart. Modified from Ahmadi et al. [58].
82 P. Panja et al. / Petroleum 4 (2018) 75e89

Kennedy [56]. In this method, each potential solution is treated as for food in a vast area.
particle. Each particle is characterized by its position and velocity. Two solutions, pbest and gbest, at any iteration during execution
The position of a particle is defined in a hyperspace whose of the algorithm are tracked. The local best or pbest is defined as
dimension is equal to the number of unknown parameters being the best position of a particle in the hyperspace as determined by
optimized as shown in Table 3. For example, in the case of ANN, the fitness value. The global best or gbest is the overall best value
particles fly in a 141-dimensional hyperspace. Several particles are by any particle so far in the population. At each iteration step, the
initially defined in hyperspace where they iteratively change their velocity is updated first and then position. Accelerating the
position to determine the optimum position. Fitness of a particle particle towards its pbest and gbest by updating velocity is done by
is determined by a fitness function such as the MSE. This algo- two separate random numbers (random 1 and random 2) as
rithm is similar to the method followed by a bird groups searching shown in Equation 11

(11)

Fig. 8. Comparison of RSM, ANN and LSSVM models using training data for (a) Oil recovery after 90 days (b), Oil recovery after 10 years (c), Oil recovery after oil rate drops to 5 bbl/
day/fracture (d), Gas Oil Ratio after 90 days (e), Gas Oil Ratio after 10 years (f), and Gas Oil Ratio after oil rate drops to 5 bbl/day/fracture.
P. Panja et al. / Petroleum 4 (2018) 75e89 83

The cognitive components guide the local search from its local
best (pbest) and the social component is responsible for global
search depending on the population best (gbest) [57]. In Equation
(11), vikþ1 is the velocity in the next iteration step which is partially
preserved from the current velocity by an inertia weight, wi (range
0.4e0.9). The acceleration coefficients (C1 and C2) for cognitive
and social components are chosen by trial and error. The position
of a particle for the next iteration step is updated by the following
equation:

pikþ1 ¼ pik þ vikþ1 $Dt (12)

The values of wi, C1, C2, and other parameters used in this study
are given in Table 4.
Initial position and velocity of each particle is randomly
distributed. After the initialization of positions and velocities of all
particles, fitness is calculated. In subsequent steps, positions and
velocities are updated iteratively by the local best and the global
best parameters as summarized in the flowchart shown in Fig. 7
[58].
The entire flowchart can be divided into four parts, namely,
initialization, fitness evaluation, condition check, and updates
of velocity and position. Acceptance of any particle as potential
solution is determined by its fitness value which is calculated in
each iteration step. As described earlier, one local best, pbest, and
one global best, gbest, are recorded during each iteration. The
number of iterations is only limited by time and computational
constraints; hence the maximum iteration number is defined by
the user.

Fig. 9. Fitness of RSM, ANN and LSSVM models for of oil recovery for training data (a)
5. Results and discussion Co-efficient of determination, R2 and (b) NRMSE.

Five time-based models (90 days, 1 year, 5 years, 10 years


and 15 years) and one rate-based model (5 bbl/day/fracture)
were trained using RSM, LSSVM and ANN. The objective of due to its complex behavior. Mainly at higher value of GOR
this study is to compare performance of three surrogate (obtained from 10 years model), flow becomes boundary
models. Production performance in terms of oil recovery and gas- dominated. On the other hand, lower GOR (obtained from
oil ratio are compared with simulation data. Since all time-based 90 days to 1 year models) occurs when flow is at the transient
models behave similarly, only two time-based models (one early linear regime. Overall, both oil recovery and GOR from surrogate
production model after 90 days and a long production model after models are in good agreement with simulation results. The errors
10 years) along with one rate-based model are discussed here. are calculated in terms of R2 and NRMSE as listed in Tables A.1 and
The fitness of a model with training data and test data are both A.2. For visual comparison, R2 and NRMSE for RSM, ANN and
discussed in this section. Once the model is trained, it is tested LSSVM models of oil recovery are shown in Fig. 9a and b
against an unknown data set (i.e., test data) to check for robust- respectively.
ness of forecasting capabilities. As discussed earlier, the fitness is R2 and NRMSE for oil recovery using RSM, LSSVM and ANN for
determined by two measures, the coefficient of determination all time- and rate-based models are greater than 0.95 and less
(R2) and normalized root mean square error (NRMSE) which are than 6% respectively. These values are evidence of well-trained
used here to compare different surrogate models. models.

5.1. Training 5.2. Testing

In this section, model fitness as compared with training data is Some models may have the tendency to overfit with training
evaluated and discussed. It is important to assess individual data and consequently fail to predict unseen test data with high
models to check for overfitting. Three surrogate models (RSM, accuracy. In this study, 20% of all data was used to check the fore-
LSSVM, and ANN) for oil recovery and gas oil ratio after 90 days, 10 casting capabilities of all developed models. Results for oil recovery
years of production and at a terminal rate (5 bbl/day/fracture) are and GOR after 90 days, 10 years production, and at terminal oil rate
shown in Fig. 8. (5 bbl/day/fracture) are shown in Fig. 10.
Typical high model fitness as compared with training data Although all models had high fitness compared with training
can be observed for all cases in Fig. 8. However, the capture of data, they showed relatively lower fitness when compared
the production behavior using surrogate models without with test data. Considering the fact that test data was not
apprehending the underlying physics is a great challenge. It is accounted for during training, the models show promising
difficult to model the GOR from low permeability reservoirs [59] forecasting capabilities without significant aberrations. R2 and
84 P. Panja et al. / Petroleum 4 (2018) 75e89

Fig. 10. Comparison of RSM, ANN and LSSVM models using test data for (a) Oil recovery after 90 days, (b) Oil recovery after 10 years, (c) Oil recovery after oil rate drops to 5 bbl/day/
fracture, (d), Gas Oil Ratio after 90 days (e), Gas Oil Ratio after 10 years (f), and Gas Oil Ratio after the oil rate drops to 5 bbl/day/fracture.

NRMSE are calculated as listed in Tables A.1 and A.2. The R2 and 6. Conclusions
NRMSE values for RSM, ANN, and LSSVM oil recovery models are
shown in Fig. 11 a and b respectively. Artificial intelligence tools aimed to predict oil recovery and
Except for a few cases, the forecast accuracy for all models are gas-oil ratio from hydraulically fractured tight formations can be
within decent ranges. As shown in the figures above, RSM shows successfully developed using simulation information as a training
higher accuracy predicting oil recovery followed by LSSVM. framework. In this study, three models were developed based on
As shown in Figs. 10 and 11, AI tools have the potential to RSM, ANN, and LSSVM to predict recovery from wells producing
predict oil recoveries and fluid ratios given a small training data under time-based (90 days, 1 year, 5 years, 10 years, and 15 years)
set. Large amounts of completion, geological, and production data and rate-based constraints (5 bbl/day/fracture). Eight key factors,
can indeed be used to train more robust models and complement namely, matrix permeability, gas relative permeability exponent,
current conventional tools to evaluate the potential of tight oil rock compressibility, initial gas-oil ratio, slope of solution gas-oil
reservoirs. The cost, however, is that AI skips the physical ratio versus pressure, initial pressure, flowing bottom-hole
description and understanding of the multiphase production pressure, and fracture spacing were considered as input
mechanisms in tight formations. This cost may not be too high to parameters for all cases. After all models were trained with the
pay since the current conventional understanding of these sys- same database, they were used to predict production for different
tems may not be sufficiently developed yet. In fact, researchers scenarios. Using simulation as a comparison basis, all models were
have recently reported on the discrepancies between conven- evaluated in terms of their oil recovery and producing gas-oil ratio
tional thinking and fluids under nanoconfinement in tight for- predictive capabilities. It was found that RSM and LSSVM have
mations [60,61]. better predictive capabilities for oil recovery than ANN. In addi-
tion, LSSVM exhibits the highest accuracy with respect to gas-oil
ratio prediction.
P. Panja et al. / Petroleum 4 (2018) 75e89 85

ng Exponent Of Relative Permeability Curve For Gas, -


NRMSE Normalized Root Mean Square Error, -
pbest Particle's Best Position
Pi Initial Reservoir Pressure, psi
PSO Particle Swarm Optimization, -
PVT Pressure-Volume-Temperature, -
R2 Coefficient Of Determination, -
Rsi Initial Gas/Oil Ratio, SCF/STB
RSM Response Surface Model, -
SSres Residual Sum Of Squares, Unit of Output
SStot Total Sum Of Squares, Unit of Output
SVM Support Vector Machine, -
Vk Velocity Of Particle, -
wi Inertia Weight, -
Ymodel,i Modeled Value, Unit of Output
Yobs,i Observed Data, Unit of Output
Yobs,max The Maximum Value Of Observed Data, Unit of Output
Yobs,min The Minimum Value Of Observed Data, Unit of Output

Appendix A. Supplementary Information

Table A1
Coefficient of determination (R2) of RSM, LSSVM and ANN for all models.

Output Model Training Data Test Data

RSM LSSVM ANN RSM LSSVM ANN

Fig. 11. Fitness of RSM, ANN and LSSVM oil recovery models for testing data: (a) Co- Oil Recovery 90 days 0.99 0.99 0.96 0.69 0.52 0.51
efficient of determination, R2 and (b) NRMSE. 1 year 0.98 0.99 0.98 0.78 0.69 0.53
5 years 0.99 0.99 0.99 0.63 0.81 0.60
10 years 0.99 0.99 0.98 0.91 0.9 0.72
15 years 0.99 0.99 0.99 0.97 0.93 0.84
Rate 0.98 0.98 0.99 0.57 0.54 0.48
Field-scale modeling and simulation of hydraulically Based
fractured ultra-low permeability reservoirs lead to very expensive Gas Oil Ratio 90 days 0.98 0.99 0.95 0.92 0.84 0.80
computational overhead in commercial simulators. Surrogate 1 year 0.98 0.98 0.96 0.93 0.91 0.90
reservoir models, on the other hand, are useful for quick oil 5 years 0.98 0.98 0.97 0.41 0.73 0.30
10 years 0.88 0.92 0.93 0.76 0.73 0.46
production forecast and assessment. Additionally, these models
15 years 0.83 0.77 0.84 0.79 0.75 0.32
can be used for risk and uncertainty analysis. Overall, Rate 0.84 0.88 0.92 0.68 0.45 0.43
artificial intelligence applications such as LSSVM have promising Based
applications in various aspects of production and reservoir
engineering.

Nomenclature

g Regularization Parameter, - Table A2


Normalized Root Mean Square Error (NRMSE) of RSM, LSSVM and ANN for all
s Kernel Parameter, - models.
ai Support Values, Unit of Output
Output Model Training Data Test Data
Y obs Mean Of Observed Values, Unit of Output
a0 The Intercept Of The Surrogate Model, Unit of Output RSM LSSVM ANN RSM LSSVM ANN
AI Artificial Intelligence, - Oil Recovery 90 days 1.9 1.9 3.5 16.5 20.3 20.7
aij Coefficient Of 2nd Order Interaction Of Inputs, - 1 year 2.4 2.3 2.5 12.4 14.7 18.1
ak Coefficient Of Independent Input, - 5 years 2.0 1.9 2.1 16.1 11.5 16.7
10 years 1.9 1.7 2.6 7.9 8.5 14
ANN Artificial Neural Networks, - 15 years 2.7 2.4 2.1 4.9 7.3 10.8
b Bias Term, Unit of Output Rate 3.5 3.3 2.4 20.7 21.2 22.6
BHP Bottom Hole Pressure, psi Based
C1 Acceleration Coefficient For Cognitive Components Gas Oil Ratio 90 days 2.6 2.0 4.6 8.7 11.8 13.3
1 year 3.3 3.3 4.2 7.9 9.3 9.7
C2 Acceleration Coefficient For Social Components
5 years 3.0 3.1 3.7 24.0 16.1 26.2
DOE Design Of Experiments 10 years 5.7 4.6 4.3 16.1 17.2 24.3
dRs/dp Slope Of Gas/Oil Ratio In PVT, (SCF/STB)/psi 15 years 5.8 6.8 5.6 14.4 15.5 25.7
gbest Population's Best Particle's Position Rate 5.2 4.5 3.8 14.1 18.4 18.8
GOR Gas/Oil Ratio, SCF/STB Based

LSSVM Least Square Support Vector Machine, -


MSE Mean Square Error, Unit of Output
86 P. Panja et al. / Petroleum 4 (2018) 75e89

Fig. A1. Data training comparison of RSM, ANN and LSSVM models for (a) Oil recovery after 1 year (b) Oil recovery after 5 years (c) Oil recovery after 15 years (d) Gas Oil Ratio after 1
year (e) Gas Oil Ratio after 5 years (f) Gas Oil Ratio after 15 years.
P. Panja et al. / Petroleum 4 (2018) 75e89 87

Table A3 Table A3 (continued )


List of simulations using Box-Behnken DOE used to train surrogate models.
Sr. No. Km ng Cf dRs/dp Rsi Pi Pwf Xf
Sr. No. Km ng Cf dRs/dp Rsi Pi Pwf Xf
(nD) e (1/psi) ((SCF/STB)/psi) (SCF/STB) (psi) (psi) (ft.)
(nD) e (1/psi) ((SCF/STB)/psi) (SCF/STB) (psi) (psi) (ft.)
74 225 2 4.00E-05 0.5 3000 5250 1000 180
1 10 1 4.00E-05 0.65 1900 5250 1000 180 75 225 2 4.00E-05 0.8 800 5250 1000 180
2 10 3 4.00E-05 0.65 1900 5250 1000 180 76 225 2 4.00E-05 0.8 3000 5250 1000 180
3 5000 1 4.00E-05 0.65 1900 5250 1000 180 77 225 2 4.00E-05 0.5 1900 4000 1000 180
4 5000 3 4.00E-05 0.65 1900 5250 1000 180 78 225 2 4.00E-05 0.5 1900 6500 1000 180
5 10 2 4.00E-06 0.65 1900 5250 1000 180 79 225 2 4.00E-05 0.8 1900 4000 1000 180
6 10 2 4.00E-04 0.65 1900 5250 1000 180 80 225 2 4.00E-05 0.8 1900 6500 1000 180
7 5000 2 4.00E-06 0.65 1900 5250 1000 180 81 225 2 4.00E-05 0.5 1900 5250 500 180
8 5000 2 4.00E-04 0.65 1900 5250 1000 180 82 225 2 4.00E-05 0.5 1900 5250 1500 180
9 10 2 4.00E-05 0.5 1900 5250 1000 180 83 225 2 4.00E-05 0.8 1900 5250 500 180
10 10 2 4.00E-05 0.8 1900 5250 1000 180 84 225 2 4.00E-05 0.8 1900 5250 1500 180
11 5000 2 4.00E-05 0.5 1900 5250 1000 180 85 225 2 4.00E-05 0.5 1900 5250 1000 60
12 5000 2 4.00E-05 0.8 1900 5250 1000 180 86 225 2 4.00E-05 0.5 1900 5250 1000 300
13 10 2 4.00E-05 0.65 800 5250 1000 180 87 225 2 4.00E-05 0.8 1900 5250 1000 60
14 10 2 4.00E-05 0.65 3000 5250 1000 180 88 225 2 4.00E-05 0.8 1900 5250 1000 300
15 5000 2 4.00E-05 0.65 800 5250 1000 180 89 225 2 4.00E-05 0.65 800 4000 1000 180
16 5000 2 4.00E-05 0.65 3000 5250 1000 180 90 225 2 4.00E-05 0.65 800 6500 1000 180
17 10 2 4.00E-05 0.65 1900 4000 1000 180 91 225 2 4.00E-05 0.65 3000 4000 1000 180
18 10 2 4.00E-05 0.65 1900 6500 1000 180 92 225 2 4.00E-05 0.65 3000 6500 1000 180
19 5000 2 4.00E-05 0.65 1900 4000 1000 180 93 225 2 4.00E-05 0.65 800 5250 500 180
20 5000 2 4.00E-05 0.65 1900 6500 1000 180 94 225 2 4.00E-05 0.65 800 5250 1500 180
21 10 2 4.00E-05 0.65 1900 5250 500 180 95 225 2 4.00E-05 0.65 3000 5250 500 180
22 10 2 4.00E-05 0.65 1900 5250 1500 180 96 225 2 4.00E-05 0.65 3000 5250 1500 180
23 5000 2 4.00E-05 0.65 1900 5250 500 180 97 225 2 4.00E-05 0.65 800 5250 1000 60
24 5000 2 4.00E-05 0.65 1900 5250 1500 180 98 225 2 4.00E-05 0.65 800 5250 1000 300
25 10 2 4.00E-05 0.65 1900 5250 1000 60 99 225 2 4.00E-05 0.65 3000 5250 1000 60
26 10 2 4.00E-05 0.65 1900 5250 1000 300 100 225 2 4.00E-05 0.65 3000 5250 1000 300
27 5000 2 4.00E-05 0.65 1900 5250 1000 60 101 225 2 4.00E-05 0.65 1900 4000 500 180
28 5000 2 4.00E-05 0.65 1900 5250 1000 300 102 225 2 4.00E-05 0.65 1900 4000 1500 180
29 225 1 4.00E-06 0.65 1900 5250 1000 180 103 225 2 4.00E-05 0.65 1900 6500 500 180
30 225 1 4.00E-04 0.65 1900 5250 1000 180 104 225 2 4.00E-05 0.65 1900 6500 1500 180
31 225 3 4.00E-06 0.65 1900 5250 1000 180 105 225 2 4.00E-05 0.65 1900 4000 1000 60
32 225 3 4.00E-04 0.65 1900 5250 1000 180 106 225 2 4.00E-05 0.65 1900 4000 1000 300
33 225 1 4.00E-05 0.5 1900 5250 1000 180 107 225 2 4.00E-05 0.65 1900 6500 1000 60
34 225 1 4.00E-05 0.8 1900 5250 1000 180 108 225 2 4.00E-05 0.65 1900 6500 1000 300
35 225 3 4.00E-05 0.5 1900 5250 1000 180 109 225 2 4.00E-05 0.65 1900 5250 500 60
36 225 3 4.00E-05 0.8 1900 5250 1000 180 110 225 2 4.00E-05 0.65 1900 5250 500 300
37 225 1 4.00E-05 0.65 800 5250 1000 180 111 225 2 4.00E-05 0.65 1900 5250 1500 60
38 225 1 4.00E-05 0.65 3000 5250 1000 180 112 225 2 4.00E-05 0.65 1900 5250 1500 300
39 225 3 4.00E-05 0.65 800 5250 1000 180 113 225 2 4.00E-05 0.65 1900 5250 1000 180
40 225 3 4.00E-05 0.65 3000 5250 1000 180 114 225 2 4.00E-05 0.65 1900 5250 1000 180
41 225 1 4.00E-05 0.65 1900 4000 1000 180
42 225 1 4.00E-05 0.65 1900 6500 1000 180
43 225 3 4.00E-05 0.65 1900 4000 1000 180
44 225 3 4.00E-05 0.65 1900 6500 1000 180
45 225 1 4.00E-05 0.65 1900 5250 500 180
46 225 1 4.00E-05 0.65 1900 5250 1500 180 Table A4
47 225 3 4.00E-05 0.65 1900 5250 500 180 List of simulations performed using random input parameter values used to test the
48 225 3 4.00E-05 0.65 1900 5250 1500 180 surrogate models.
49 225 1 4.00E-05 0.65 1900 5250 1000 60 Sr. No. Km ng Cf dRs/dp Rsi Pi Pwf Xf
50 225 1 4.00E-05 0.65 1900 5250 1000 300
51 225 3 4.00E-05 0.65 1900 5250 1000 60 (nD) e (1/psi) ((SCF/STB)/psi) (SCF/STB) (psi) (psi) (ft.)
52 225 3 4.00E-05 0.65 1900 5250 1000 300
1 1496 1.17 1.0E-05 0.62 2069 5738 743 300
53 225 2 4.00E-06 0.5 1900 5250 1000 180
2 361 1.27 3.8E-05 0.67 1456 4170 1417 180
54 225 2 4.00E-06 0.8 1900 5250 1000 180
3 31 1.35 8.0E-06 0.58 1625 4637 769 60
55 225 2 4.00E-04 0.5 1900 5250 1000 180
4 44 1.78 2.0E-05 0.59 1724 4560 1266 60
56 225 2 4.00E-04 0.8 1900 5250 1000 180
5 2475 2.66 1.9E-05 0.69 2408 5670 689 300
57 225 2 4.00E-06 0.65 800 5250 1000 180
6 12 2.61 1.4E-05 0.58 2820 6111 787 60
58 225 2 4.00E-06 0.65 3000 5250 1000 180
7 210 1.12 2.0E-05 0.75 1775 4861 591 180
59 225 2 4.00E-04 0.65 800 5250 1000 180
8 28 1.80 1.9E-05 0.79 2969 5951 1076 60
60 225 2 4.00E-04 0.65 3000 5250 1000 180
9 4390 2.05 6.0E-06 0.72 2531 5688 1183 300
61 225 2 4.00E-06 0.65 1900 4000 1000 180
10 840 1.83 5.4E-06 0.60 1460 4017 1047 300
62 225 2 4.00E-06 0.65 1900 6500 1000 180
11 225 2.31 4.0E-05 0.68 2106 5505 926 180
63 225 2 4.00E-04 0.65 1900 4000 1000 180
12 187 2.26 5.9E-06 0.53 1689 4967 1144 180
64 225 2 4.00E-04 0.65 1900 6500 1000 180
13 14 1.58 4.3E-06 0.77 2779 6290 1148 60
65 225 2 4.00E-06 0.65 1900 5250 500 180
14 694 1.86 1.5E-05 0.76 1539 4003 1179 300
66 225 2 4.00E-06 0.65 1900 5250 1500 180
15 13 1.03 3.0E-05 0.75 1273 5156 1136 60
67 225 2 4.00E-04 0.65 1900 5250 500 180
16 16 2.97 1.9E-05 0.58 1413 5061 1445 60
68 225 2 4.00E-04 0.65 1900 5250 1500 180
17 256 1.33 6.2E-06 0.68 1323 5152 709 180
69 225 2 4.00E-06 0.65 1900 5250 1000 60
18 18 1.21 9.4E-06 0.51 2281 5925 1209 60
70 225 2 4.00E-06 0.65 1900 5250 1000 300
19 1618 1.74 1.2E-05 0.63 1552 4806 736 300
71 225 2 4.00E-04 0.65 1900 5250 1000 60
20 1612 1.40 3.8E-05 0.59 2527 5962 619 300
72 225 2 4.00E-04 0.65 1900 5250 1000 300
21 892 1.98 5.7E-06 0.55 1566 5178 1107 300
73 225 2 4.00E-05 0.5 800 5250 1000 180
(continued on next page)
88 P. Panja et al. / Petroleum 4 (2018) 75e89

Table A4 (continued ) [18] S.D. Mohaghegh, J.S. Liu, R. Gaskari, M. Maysami, O.A. Olukoko, Application of
well-base surrogate reservoir models (SRMs) to two offshore fields in Saudi
Sr. No. Km ng Cf dRs/dp Rsi Pi Pwf Xf Arabia, case study, in: SPE Western Regional Meeting, Society of Petroleum
Engineers, Bakersfield, California, USA, 2012.
(nD) e (1/psi) ((SCF/STB)/psi) (SCF/STB) (psi) (psi) (ft.)
[19] R. Velasco, P. Panja, M. Deo, New Production Performance and Prediction Tool
22 25 1.68 2.9E-05 0.55 1900 4089 950 60 for Unconventional Reservoirs, in: Unconventional Resources Technology
23 604 2.90 1.8E-05 0.63 1614 4440 959 300 Conference, 1e3 August. San Antonio, Texas, USA, URTEC-2461718-MS, So-
24 251 2.84 9.5E-06 0.53 2944 5804 1162 180 ciety of Petroleum Engineers, 2016.
25 4237 1.11 6.2E-06 0.68 1710 5184 1270 300 [20] T.W. Patzek, F. Male, M. Marder, Gas production in the Barnett Shale obeys a
26 565 2.48 1.1E-05 0.64 1966 4382 850 300 simple scaling theory, Proc. Natl. Acad. Sci. 110 (49) (2013) 19731e19736.
[21] R.A. Wattenbarger, A.H. El-Banbi, M.E. Villegas, J.B. Maggard, Production
27 1448 1.54 1.2E-05 0.71 1393 4853 1162 300
analysis of linear flow into fractured tight gas wells, SPE-39931-MS, in: SPE
28 168 1.85 5.3E-06 0.71 2676 5518 916 180
Rocky Mountain Regional/Low-permeability Reservoirs Symposium, 5-8 April,
29 147 2.10 1.6E-05 0.69 1431 4479 1342 180
Society of Petroleum Engineers, Denver, Colorado, USA, 1998.
30 1692 2.89 6.7E-06 0.51 2672 5846 1333 300 [22] M. Nobakht, L. Mattar, S. Moghadam, D.M. Anderson, Simplified forecasting of
tight/shale-gas production in linear flow, J. Can. Petrol. Technol. 51 (06)
(2012) 11.
[23] A.U. Al-Kaabi, W.J. Lee, Using Artificial Neural Networks to Identify the Well
Test Interpretation Model (Includes Associated Papers 28151 and 28165 ),
1993, 8(03).
[24] I.R. Juniardi, I. Ershaghi, Complexities of Using Neural Network in Well Test
References Analysis of Faulted Reservoirs, Society of Petroleum Engineers, 1993.
[25] C.D. Zhou, X.-L. Wu, J.-A. Cheng, Determining reservoir properties in reservoir
[1] B. Yeten, A. Castellini, B. Guyaguler, W.H. Chen, A comparison study on studies using a fuzzy neural network, in: SPE Annual Technical Conference
experimental design and response surface methodologies, in: SPE Reservoir and Exhibition, Society of Petroleum Engineers, Houston, Texas, 1993.
Simulation Symposium, Society of Petroleum Engineers Inc, The Woodlands, [26] S. Mohaghegh, R. Arefi, S. Ameri, M.H. Hefner, A methodological approach for
Texas, 2005. reservoir heterogeneity characterization using artificial neural networks, in:
[2] T.C.A.D. Amorim, D.J. Schiozer, Risk analysis speed-up with surrogate models, SPE Annual Technical Conference and Exhibition, Society of Petroleum Engi-
in: SPE Latin America and Caribbean Petroleum Engineering Conference, So- neers, New Orleans, Louisiana, 1994.
ciety of Petroleum Engineers, Mexico City, Mexico, 2012. [27] E. El-Sebakhy, T.S.,S. Al-Bokhitan, Y. Shaaban, I. Raharja, Y. Khaeruzzaman,
[3] B. Li, F. Firedmann, A novel response surface methodology based on “ampli- support vector machines framework for predicting the PVT properties of
tude factor” analysis for modeling nonlinear responses caused by both crude-oil systems, in: Kingdom of Baharin: 15th SPE Middle East Oil & Gas
reservoir and controllable factors, in: SPE Annual Technical Conference and Show and Conference, 2007.
Exhibition, Society of Petroleum Engineers, Dallas, Texas, 2005. [28] M. Oloso, A. Khoukhi, A. Abdulraheem, M. Elshafei, Prediction of crude oil
[4] C.Y. Peng, R. Gupta, Experimental design in deterministic modelling: assessing viscosity and gas/oil ratio curves using recent advances to neural networks,
significant uncertainties, in: SPE Asia Pacific Oil and Gas Conference and in: SPE/EAGE Reservoir Characterization and Simulation Conference, Society
Exhibition, Society of Petroleum Engineers, Jakarta, Indonesia, 2003. of Petroleum Engineers, Abu Dhabi, UAE, 2009.
[5] J.P. Dejean, G. Blanc, Managing uncertainties on production predictions using [29] A. Rabiei, H. Sayyad, M. Riazi, A. Hashemi, Determination of dew point pres-
integrated statistical methods, in: SPE Annual Technical Conference and sure in gas condensate reservoirs based on a hybrid neural genetic algorithm,
Exhibition, Society of Petroleum Engineers, Houston, Texas, 1999. Fluid Phase Equilibria 387 (2015) 38e49.
[6] B. Corre, P. Thore, V.d. Feraudy, G. Vincent, Integrated uncertainty assessment [30] M.A. Ahmadi, M. Ebadi, Evolving smart approach for determination dew point
for project evaluation and risk analysis, in: SPE European Petroleum Confer- pressure through condensate gas reservoirs, Fuel 117 (2014) 1074e1084.
ence, Society of Petroleum Engineers Inc, Paris, France, 2000. [31] M.A. Ahmadi, M. Ebadi, A. Yazdanpanah, Robust intelligent tool for estimating
[7] E. Manceau, M. Mezghani, I. Zabalza-Mezghani, F. Roggero, Combination of dew point pressure in retrograded condensate gas reservoirs: application of
experimental design and joint modeling methods for quantifying the risk particle swarm optimization, J. Petrol. Sci. Eng. 123 (2014) 7e19.
associated with deterministic and stochastic uncertainties - an integrated test [32] M.A. Ahmadi, M. Ebadi, P.S. Marghmaleki, M.M. Fouladi, Evolving predictive
study, in: SPE Annual Technical Conference and Exhibition, Society of Petro- model to determine condensate-to-gas ratio in retrograded condensate gas
leum Engineers Inc, New Orleans, Louisiana, 2001. reservoirs, Fuel 124 (2014) 241e257.
[8] R. Venkataraman, Application of the method of experimental design to [33] M.A. Ahmadi, Neural network based unified particle swarm optimization for
quantify uncertainty in production profiles, in: SPE Asia Pacific Conference on prediction of asphaltene precipitation, Fluid Phase Equilibria 314 (2012)
Integrated Modelling for Asset Management, Copyright 2000, Society of Pe- 46e51.
troleum Engineers Inc, Yokohama, Japan, 2000. [34] M.A. Ahmadi, S.R. Shadizadeh, New approach for prediction of asphaltene
[9] J. Chewaroungroaj, O.J. Varela, L.W. Lake, An evaluation of procedures to es- precipitation due to natural depletion by using evolutionary algorithm
timate uncertainty in hydrocarbon recovery predictions, in: SPE Asia Pacific concept, Fuel 102 (2012) 716e723.
Conference on Integrated Modelling for Asset Management, Copyright 2000, [35] M.A. Ahmadi, M. Ebadi, A. Shokrollahi, S.M.J. Majidi, Evolving artificial neural
Society of Petroleum Engineers Inc, Yokohama, Japan, 2000. network and imperialist competitive algorithm for prediction oil flow rate of
[10] S.D. Mohaghegh, Quantifying uncertainties associated with reservoir simula- the reservoir, Appl. Soft Comput. 13 (2) (2013) 1085e1098.
tion studies using a surrogate reservoir model, in: SPE Annual Technical [36] A.A. Fatai Adesina Anifowose, Prediction of porosity and permeability of oil
Conference and Exhibition, Society of Petroleum Engineers, San Antonio, and gas reservoirs using hybrid computational intelligence models, in: Cairo,
Texas, USA, 2006. Egypt: North Africa Technical Conference and Exhibition, SPE, 2010.
[11] B. Guyaguler, R.N. Horne, Uncertainty assessment of well placement optimi- [37] F.A. Anifowose, A.O. Ewenla, S.I. Eludiora, Prediction of Oil and Gas Reservoir
zation, in: SPE Annual Technical Conference and Exhibition, Copyright 2001, Properties Using Support Vector Machines, International Petroleum Tech-
Society of Petroleum Engineers Inc, New Orleans, Louisiana, 2001. nology Conference, Bangkok, Thailand, 2011.
[12] E. Manceau, F. Roggero, I. Zabalza-Mezghani, Use of experimental design [38] F. Ammal, G. Al-anazi, D. Ian, Support-vector regression for permeability
methodology to make decisions in an uncertain reservoir environment from prediction in a heterogeneous reservoir: a comparative study, SPE Reserv.
reservoir uncertainties to economic risk analysis, World Pet. Congr. (2002). Eval. Eng. 13 (03) (2010).
[13] J.L. Landa, B. Güyagüler, A methodology for history matching and the [39] M.A. Ahmadi, R. Ahmadi, S.M. Hosseini, M. Ebadi, Connectionist model pre-
assessment of uncertainties associated with flow prediction, in: SPE Annual dicts the porosity and permeability of petroleum reservoirs by means of
Technical Conference and Exhibition, Society of Petroleum Engineers, Denver, petro-physical logs: application of artificial intelligence, J. Petrol. Sci. Eng.s
Colorado, 2003. 123 (2014) 183e200.
[14] P.E. Carreras, S.E. Turner, G.T. Wilkinson, Tahiti: development strategy [40] M.A. Ahmadi, A. Bahadori, A LSSVM approach for determining well placement
assessment using design of experiments and response surface methods, in: and conning phenomena in horizontal wells, Fuel 153 (2015) 276e283.
SPE Western Regional/AAPG Pacific Section/GSA Cordilleran Section Joint [41] M.A. Ahmadi, M. Ebadi, S.M. Hosseini, Prediction breakthrough time of water
Meeting, Society of Petroleum Engineers, Anchorage, Alaska, USA, 2006. coning in the fractured reservoirs by implementing low parameter support
[15] C. Yang, L.X. Nghiem, C. Card, M. Bremeier, Reservoir model uncertainty vector machine approach, Fuel 117 (2014) 579e589.
quantification through computer-assisted history matching, in: SPE Annual [42] M.A. Ahmadi, Connectionist approach estimates gaseoil relative simulation in
Technical Conference and Exhibition, Society of Petroleum Engineers, Ana- petroleum reservoirs: application to reservoir simulation, Fuel 140 (2015)
heim, California, U.S.A., 2007. 429e439.
[16] P.A. Slotte, E. Smorgrav, Response surface methodology approach for history [43] A. Eslamimanesh, F. Gharagheizi, M. Illbeigi, A.H. Mohammadi, A. Fazlali,
matching and uncertainty assessment of reservoir simulation models, in: D. Richon, Phase equilibrium modeling of clathrate hydrates of methane,
Europec/EAGE Conference and Exhibition, Society of Petroleum Engineers, carbon dioxide, nitrogen, and hydrogen þ water soluble organic promoters
Rome, Italy, 2008. using Support Vector Machine algorithm, Fluid Phase Equilibria 316 (2012)
[17] M.A. Ahmadi, R. Soleimani, A. Bahadori, A computational intelligence scheme 34e45.
for prediction equilibrium water dew point of natural gas in TEG dehydration [44] Reza Gholgheysari Gorjaei, R.S. Mohammad Torkaman, Mohsen Safari,
systems, Fuel 137 (2014) 145e154. Ghassem Zargar, A novel PSO-LSSVM model for predicting liquid rate of two
P. Panja et al. / Petroleum 4 (2018) 75e89 89

phase flow through wellhead chokes, J. Nat. Gas Sci. Eng. 24 (2015) 228e237. quantitative variables, Technometrics 2 (4) (1960) 455e475.
[45] M.-A. Ahmadi, M.Z. Hasanvand, A. Bahadori, A least-squares support vector [54] P. Panja, T. Conner, M. Deo, Grid sensitivity studies in hydraulically fractured
machine approach to predict temperature drop accompanying a given pres- low permeability reservoirs, J. Petrol. Sci. Eng. 112 (0) (2013) 78e87.
sure drop for the natural gas production and processing systems, Int. J. [55] P. Panja, M. Pathak, R. Velasco, M. Deo, Least square support vector machine:
Ambient Energy 38 (2) (2015) 122e129. an emerging tool for data analysis, in: SPE Low Perm Symposium, 5-6 May,
[46] M.A. Ahmadi, M. Masoumi, R. Askarinezhad, Evolving connectionist model to Society of Petroleum Engineers, Colorado, Denver, 2016.
monitor the efficiency of an in situ combustion process: application to heavy [56] R. Eberhart, J. Kennedy, A new optimizer using particle swarm theory, in:
oil recovery, Energy Technol. 2 (9e10) (2014) 811e818. Micro Machine and Human Science, 1995. MHS '95., Proceedings of the Sixth
[47] M.-A. Ahmadi, M. Masumi, R. Kharrat, A.H. Mohammadi, Gas analysis by in International Symposium on, 1995.
situ combustion in heavy-oil recovery process: experimental and modeling [57] C. Banerjee, R. Sawal, PSO with dynamic acceleration coefficient based on
studies, Chem. Eng. Technol. 37 (3) (2014) 409e418. multiple constraint satisfaction, in: International Conference on Advances in
[48] M.A. Ahmadi, M. Masoumi, R. Askarinezhad, Evolving smart model to predict Electronics Computers and Communications. Bangalore, India, 2014.
the combustion front velocity for in situ combustion, Energy Technol. 3 (2) [58] M.A. Ahmadi, R. Soleimani, M. Lee, T. Kashiwao, A. Bahadori, Determination of
(2015) 128e135. oil well production performance using artificial neural network (ANN) linked
[49] M.A. Ahmadi, M. Zahedzadeh, S.R. Shadizadeh, R. Abbassi, Connectionist to the particle swarm optimization (PSO) tool, Petroleum 1 (2) (2015)
model for predicting minimum gas miscibility pressure: application to gas 118e132.
injection process, Fuel 148 (2015) 202e211. [59] P. Panja, M. Deo, Unusual behavior of produced gas oil ratio in low perme-
[50] M. Ali Ahmadi, A. Ahmadi, Applying a sophisticated approach to predict ability fractured reservoirs, J. Petrol. Sci. Eng. 144 (2016) 76e83.
CO2solubility in brines: application to CO2sequestration, Int. J. Low-Carbon [60] M. Pathak, H. Kweon, P. Panja, R. Velasco, M.D. Deo, Suppression in the bubble
Technol. 11 (3) (2016) 325e332. points of oils in shales combined effect of presence of organic matter and
[51] M.-A. Ahmadi, A. Bahadori, S.R. Shadizadeh, A rigorous model to predict the confinement, in: SPE Unconventional Resources Conference, 15-16 February,
amount of Dissolved Calcium Carbonate Concentration throughout oil field Society of Petroleum Engineers, Calgary, Alberta, Canada, 2017.
brines: side effect of pressure and temperature, Fuel 139 (2015) 154e159. [61] R. Velasco, M. Pathak, P. Panja, M. Deo, What happens to permeability at the
[52] P. Panja, T. Conner, M. Deo, Factors controlling production in hydraulically Nanoscale? A molecular dynamics simulation study, in: SPE/AAPG/SEG Un-
fractured low permeability oil reservoirs, Int. J. Oil, Gas Coal Technol. 3 (1) conventional Resources Technology Conference, 24-26 July. Austin, Texas,
(2015) 18. USA: Unconventional Resources Technology Conference, 2017.
[53] G.E.P. Box, D.W. Behnken, Some new three level designs for the study of

Вам также может понравиться