You are on page 1of 8

Application of Artificial

Intelligence Techniques to
Estimate the Static Poisson’s
Ratio Based on Wireline
Log Data
Salaheldin Elkatatny Static Poisson’s ratio ( static) is a key factor in determine the in-situ stresses in the reser-
Department of Petroleum Engineering, voir section.  static is used to calculate the minimum horizontal stress which will affect
King Fahd University of the design of the optimum mud widow and the density of cement slurry while drilling. In
Petroleum and Minerals, addition, it also affects the design of the casing setting depth.  static is very important for
Post Box No. 5049, field development and the incorrect estimation of it may lead to heavy investment deci-
Dhahran 31261, Saudi Arabia; sions.  static can be measured in the lab using a real reservoir cores. The laboratory
Petroleum Department, measurements of  static will take long time and also will increase the overall cost. The
Cairo University, goal of this study is to develop accurate models for predicting  static for carbonate reser-
Cairo 12613, Egypt voirs based on wireline log data using artificial intelligence (AI) techniques. More than
e-mail: elkatatny@kfupm.edu.sa 610 core and log data points from carbonate reservoirs were used to train and validate
the AI models. The more accurate AI model will be used to generate a new correlation
for calculating the  static. The developed artificial neural network (ANN) model yielded
more accurate results for estimating  static based on log data; sonic travel times and bulk
density compared to adaptive neuro fuzzy inference system (ANFIS) and support vector
machine (SVM) methods. The developed empirical equation for  static gave a coefficient
of determination (R2) of 0.97 and an average absolute percentage error (AAPE) of
1.13%. The developed technique will help geomechanical engineers to estimate a com-
plete trend of  static without the need for coring and laboratory work and hence will
reduce the overall cost of the well. [DOI: 10.1115/1.4039613]

Keywords: static Poisson’s ratio, artificial intelligence, well logs, minimum horizontal
stress, neural network, support vector machine, adaptive neuro fuzzy inference system

1 Introduction (0.1–0.3), medium formations such as sandstone have  static range


of (0.2–0.3) and hard formations such as limestone have  static
Simulation and modeling are the key factors for reservoir
range of (0.3–0.4) [17]. The change in the geomechanical parame-
management and productivity enhancement [1,2], in addition
ters due to any operation is very important to be investigated. For
to the control of the hydrocarbon transportation [3]. The one-
example, injecting of carbon dioxide in Khuff Formation had a sig-
dimensional and three-dimensional geomechanical earth model
nificant change in the mechanical properties [18].
can be generated based on the static geomechanical parameters;
There are two methods to determine Poisson’s ratio values: (i)
static Young’s modulus and  static [4], in addition these parame-
the dynamic method used in field, in which the ultrasonic veloc-
ters affect the fracture complexity [5] and the rock properties after
ities are measured; and (ii) the static method used in laboratory, in
fluid-rock interaction [6,7].  static affects the deformation of the
which the deformation as a result of the application of a known
elastic materials, where it is defined as the ratio of lateral expan-
force is measured. The dynamic Poisson’s ratio (dyn ) of the rock
sion to longitudinal contraction [8].
can be calculated by the following equation:
The value of  static is used to calculate the stresses under the res-
ervoir conditions. The in-situ stresses are used to build the geome-
V 2  VS2
chanical earth models, which in turn are needed for the design of dyn ¼  P2  (1)
hydraulic fracture orientation and optimization [9] and well drilling 2 VP  VS2
design [10]. Chan et al. [11] stated that the dominant factor affect-
ing predictions or rock stresses is the in-situ modulus. The value of where VP is the compressional wave velocity in km/s and VS is the
 static varies with the lithology and rock properties such as bulk den- shear wave velocity in km/s
sity, porosity, pore structure, temperature, pore pressure, fluid satu- Laboratory measured  static value of the earth more truly repre-
ration, and the rock consolidation [12–14]. Fahrenthold and Gray sents the actual stresses behavior in the reservoir [19]; however,
[15] concluded that shales rock exhibited lower elastic parameters measuring the  static in the lab requires long time and also high
as compared to neighboring sandstone from a Gulf Coast cost in additional to the time and the cost required to obtain the
geopressured-geothermal well. For any elastic material, the value core from the reservoir section [20,21]. The data obtained from
of  static is 0–0.5 and for most of the rocks is 0.15–0.4 [16]. Soft the laboratory measurements are used to generate a correlation
formations such as unsaturated clays have  static values of with the log data and the developed equation then used to generate
a complete profile of  static through the reservoir section [9,21].
However, the cores are retrieved from the selected intervals so the
Contributed by the Petroleum Division of ASME for publication in the JOURNAL
OF ENERGY RESOURCES TECHNOLOGY. Manuscript received September 19, 2017; final
accuracy of these correlations are very limited.
manuscript received March 4, 2018; published online March 29, 2018. Assoc. Kumar et al. [22] used nonlinear regression technique to
Editor: Ray (Zhenhua) Rui. develop a correlation for static Poisson’s ratio with ultrasonic

Journal of Energy Resources Technology Copyright V


C 2018 by ASME JULY 2018, Vol. 140 / 072905-1

Downloaded From: https://energyresources.asmedigitalcollection.asme.org on 01/23/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use


longitudinal and shear wave velocities, which is valid only for iso- instead of decreasing the error of training. SVM works in way of
tropic material. They observed that Poisson’s ratio increases with solving linearly quadratic programming problems with a distinc-
the decrease of ultrasonic velocities. tive, optimum solution that improbable to be local minima [40].
Support vector regression was used to determine the  static SVM’s performance depends on many parameters that should be
based 600 core data with logs; pore pressure, minimum horizontal selected carefully to achieve a perfect prediction model.
stress, bulk density, overburden stress, sonic times, bulk density, Elkatatny et al. [41,42] proved that the ANN can be success-
and porosity [14]. The main disadvantage of their approach is its fully used to predict the real the drilling fluid rheological proper-
dependence upon minimum horizontal stress, pore pressure and ties in a real time. They developed different equations to
overburden stress. Whereas, minimum horizontal stress, pore determine the plastic viscosity, yield point, flow index, and con-
pressure, and overburden stress are estimated from the elastic sistency index. They concluded that ANN can be used to obtain
parameters. the rheological properties with a high accuracy (correlation coeffi-
Asoodeh [23] used committee machine intelligence system cient (R) >95% and the average absolute percentage error
approach to predict Poisson’s ratio. He used 600 data points from (AAPE) less than 8%). Elkatatny et al. [43–45] applied different
carbonate reservoirs and implemented multistage technique. In AI techniques to determine the reservoir permeability, the bubble
the first stage, he predicted shear wave travel time from well logs point pressure and the oil formation volume factors. They con-
and in next stage, he used predicted shear wave travel time log cluded that ANN yielded more accurate results as compared with
with other well logs to predict Poisson’s ratio. ANFIS and SVM for prediction the afetermentioned properties.
In the light of this literature survey, till now there is no general They changed the black box of the ANN to a white box by devel-
empirical equation to calculate the static Poisson’s ratio from the oping different equation for these properties based on the opti-
well log data directly. The goal of this study is to build an accurate mized ANN model.
model for  static using 610 data points of the wireline logs based Artificial neural network was used to predict the pump pressure
on artificial intelligence (AI) techniques. in a real time operation during the drilling operation [46]. Salehi
et al. [47] confirmed that the advanced ANN model can be used to
predict the collapse pressure of the casing in a field application.
1.1 Artificial Intelligence Techniques. Artificial neural net- Tariq et al. [48–51] and Parapuram et al. [52] developed differ-
work (ANN) is a computational method, which is derived from ent AI techniques to determine the rock mechanical properties,
the construction features of biological neural networks [24]. ANN rock strength, uniaxial compressive strength, and the failure
is made up of components that called neurons. Every neuron has a parameters for carbonate formation. They concluded that AI can
specific input/output and they are linked together to make a net- be used in with a high accuracy to determine the geomechanical
work of nodes, which form the biological neural networks properties of carbonate formation.
[25,26]. Weights and biases are used to control the input variables Artificial neural network was applied to determine the average
and determine the relationship between the neurons and the source oil flow rate using reservoir parameters and the wellhead flowing
[23]. Trial and error can be used to choose the number of hidden data with R of 0.97 and AAPE of 7.85% [53]. Solomon and Ade-
layers and neuron in each layer. Optimization is necessary to wale [54] used ANN to determine the fracture width in depleted
choose the suitable number of layers and neuron, because selec- reservoirs and they confirmed that ANN yielded comparable results
tion too much neuron results in over-fitting and selecting small with the existing two-dimensional fracture simulators. Respati and
number of neurons results in under-fitting [27,28]. The system Ardan [55] used ANFIS to predict the loss zones based on well his-
under study can be characterizes using ANN without physical tory and well logging. Back propagation artificial neural network
phoneme, where the nonlinearity between the input and output (BPANN) and least squares support vector regression (LS-SVR)
variables can be approximated by applying ANN [29] ANNs can was used to identify the relationship between the well log data and
be classified into several types based on the interconnection core mineralogy based brittleness. The obtained results of both LS-
between layers and neurons. The most popular types are cascade- SVR and BPANN model are competent for predicating brittleness,
forward back-propagation and feed-forward back-propagation while LS-SVR approach is more accurate than the BPANN method
[30]. Vineis and Rainoldi [31] illustrated that ANN required nom- at same conditions [56]. Rate of penetration (ROP) can be predicted
inal statistical training. ANN has different training algorithms and using ANN with a high accuracy based on the drilling parameters
it has the capability to detect all possible integration between pre- and the drilling fluid properties in vertical wells [57]. In horizontal
dictor parameters [32] wells, Mantha and Samuel [58] applied different AI technique to
Adaptive neuro fuzzy inference system (ANFIS) is a combina- predict the ROP with a high accuracy. They concluded that random
tion of the NN and fuzzy logic (FL) in a one platform and it forests technique and boosting helped increase accuracy and reduce
required evolutionary algorithm to optimize its variables [33]. FL errors of ROP prediction.
is a logical system that is a generalization of multivalued logic.
FL is nearly identical to the theory of fuzzy sets that belongs to
group of objects with boundaries in which membership is a prob- 2 Data Description
lem of degree [34]. It utilizes the back propagation algorithm and Actual laboratory measurement for more than 610 core samples
the least square to train the data to adjust the membership function from different carbonate reservoirs were collected. In addition,
which assist the fuzzy to learn the data to be modeled [35]. Walia the logging data (bulk density, compressional time, shear time)
et al. [36] stated that ANFIS has smaller conversion factor and was obtained at the same depth of the core samples. Table 1 lists
better learning capability. Uçar et al. [37] concluded that the main the statistical parameters of the available logs data and static Pois-
drawback of the ANFIS is the existing of the exponential term in son’s ratio. The bulk density range is 2.22–2.98 g/cm3; the com-
the number of consequent variables, which can be modified to pressional time range is 44.34–91.42 ls/ft; the shear time range is
enhance the system efficiency 73.19–151.51 ls/ft; and the static Poisson’s ratio range is
Support vector machine (SVM) is a group of learning machine 0.21–0.45.
that can be used for classification and regression problems. SVM The relative importance was determined by calculating the cor-
is considered as a generalized linear classifier and it also belongs relation coefficient between the input parameters and the static
to a regularization. SVM can be used to solve nonlinear complex Poisson’s ratio. Static Poisson’s ratio has a weak function of shear
problem [38]. SVM has a special features of optimum hyperplane time and bulk density and R was 0.1 and 0.28 for shear time and
which increase the space to training examples in a great dimen- bulk density, respectively.  static has a good function of compres-
sional feature space [39]. SVM is a new learning method that sional time and R was 0.53; while  static is a strong function of
depends on statistical learning theory in which the risk is mini- the ratio between shear time over compressional time (Dtc/Dts)
mized by decreasing the generalization error of the upper bound and R was 0.9.

072905-2 / Vol. 140, JULY 2018 Transactions of the ASME

Downloaded From: https://energyresources.asmedigitalcollection.asme.org on 01/23/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Table 1 Statistical analysis of the collected data (610 data Table 3 Statistical analysis of the testing data (183 data
points) points)

Statistical RHOB, DTC, DTS, Poisson’s Statistical RHOB, DTC, DTS, Poisson’s
parameter g/cm3 ls/ft ls/ft ratio parameter g/cm3 ls/ft ls/ft ratio

Minimum 2.22 44.34 73.19 0.21 Minimum 2.28 44.34 73.66 0.24
Maximum 2.98 91.42 151.51 0.45 Maximum 2.98 82.37 150.42 0.44
Range 0.76 47.08 78.33 0.24 Range 0.70 38.03 76.76 0.20
Mean 2.67 52.15 96.50 0.37 Mean 2.68 51.65 96.17 0.37
Standard deviation 0.13 6.71 10.97 0.04 Standard deviation 0.12 6.09 10.10 0.04
Kurtosis 1.26 7.01 5.48 2.17 Kurtosis 0.90 6.91 6.06 1.33
Skewness 0.43 2.27 1.71 1.00 Skewness 0.28 2.20 1.54 0.94

that SVM yielded a higher AAPE (3%) and R (0.87) as compared


Table 2 Statistical analysis of the training data (427 data with ANFIS and ANN for the training data.
points)
To assess the developed AI models, another set of data (183
Statistical RHOB, DTC, DTS, Poisson’s unseen data points) was used. The unseen data are 30% of the col-
parameter g/cm3 ls/ft ls/ft ratio lected data. Table 3 lists the statistical parameters of the testing
data.
Minimum 2.22 44.34 73.19 0.21 Figure 2 shows that for unseen data (183 data points), ANN
Maximum 2.98 91.42 151.51 0.45 model predicts  static with a R of 0.98 and AAPE of 1.11% for
Range 0.76 47.08 78.33 0.24 unseen data. ANFIS and SVM models yielded an AAPE of 1.68
Mean 2.67 52.36 96.65 0.37 and 1.73, respectively, when they were used for predicting  static
Standard deviation 0.13 6.96 11.33 0.04 from the log data.
Kurtosis 1.40 6.87 5.25 2.49 The obtained results confirmed that ANN is the more accurate
Skewness 0.49 2.26 1.75 1.03
AI model that can be used to determine  static with high accuracy
(R of 0.98 and AAPE of 1.3%) based on three log data.

3 Building Artificial Neural Network Model 3.1 Neural Network Architecture. The  static ANN model
To build the artificial inteillegce model, the input and output consists of three input parameters; bulk density, sonic times, one
data should be normalized, [41]. Seventy percent of the data (427 hidden layer, and  static as the only output. The ANN model was
data points) was used to train the ANN models. Table 2 lists the generated using the back propagation neural network algorithm.
statistical parameters of the training data. It is clear that the range The number of neuron was changed from 5 to 50 neurons and the
of the training data covered the range of the available data, which optimization process yielded that the lowest APPE and the heights
should be insured to build a strong ANN model. R between the actual and estimated values of  static was obtained
Figure 1 shows the high accuracy of predicting  static using when using 20 neurons.
ANN model and the R was 0.94 and AAPE was 2%. ANFIS pre- The best transfer function between the input and the hidden
dicts  static with R of 0.88 and AAPE of 2.92%. Figure 1 shows layer was Tan-sigmoidal, while the linear function was used

Fig. 1 Estimation of Poisson’s ratio using AI models for training data (427 data points)

Journal of Energy Resources Technology JULY 2018, Vol. 140 / 072905-3

Downloaded From: https://energyresources.asmedigitalcollection.asme.org on 01/23/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Fig. 2 Estimation of static Poisson’s ratio using ANN model for testing data (183 data points)

Table 4 Poisson’s ratio—ANN model complete description

Neural network parameters Ranges

Number of inputs 3
Number of outputs 1
Number of neurons 20
Number of hidden layer(s) 1
Training algorithm Levenberg Marquadt
Learning rate 0.12
Hidden layer transfer function Tan-sigmoidal
Outer layer transfer function Pure linear
Training ratio 0.70
Testing ratio 0.30

weights between input layer and hidden layer and w2 is the opti-
mized weights between the hidden and the outer layer. Table 5
lists the values for w1 and w2.
Fig. 3 Architecture of static Poisson’s ratio model The denormalized values of static Poisson’s ratio (staticn Þ can
be determined using Eq. (2) and normal values can be determined
using Eq. (3)
between the hidden layer and the output layer. Figure 3 illustrates 2 !3
X N
2
the architecture of the  static ANN model. Table 4 lists the variable staticn ¼ 4 w2i 5 þ b2
1 þ e ð 1i;1 n 1i;2 Cn 1i;3 Sn 1i Þ
parameters of the  static ANN model. 2 w q þw Dt þw Dt þb
i¼1
Levenberg–Marquardt back propagation algorithm was used as
a training algorithm in order to obtain the weights and the biases. (2)
When the model achieved the lowest APPE and the highest R
between the actual and estimated values, the weights and the static ¼ 0:11  staticn þ 0:34 (3)
biases are extracted to generate the best model.
The ANN black box was converted to a white box by develop-
ing Eq. (3). No need for the ANN model anymore. Engineers can
4 Development of the First Empirical Correlation for use Eq. (3) and the data available in Table 5 to estimate  static
Poisson’s Ratio Based on Artificial Neural Network from the log data. This is the first empirical equation to determine
 static form the well log data.
Model Applying Eqs. (2) and (3) on the unseen data (183 data points)
A new  static empirical correlation was developed based on the yielded R2 of 0.97 and AAPE of 1.13% when plotting the actual
optimized values of weights and biases. w1 is the optimized  static data versus the predicted  static data, Fig. 4.

072905-4 / Vol. 140, JULY 2018 Transactions of the ASME

Downloaded From: https://energyresources.asmedigitalcollection.asme.org on 01/23/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Table 5 Weights and biases for Eq. (3)

Hidden layer Weights between input and Weights between hidden and Hidden layer Output layer
neurons (N) hidden layer (W1) output layer (W2) bias (b1) bias (b2)

1 1.2314 2.9794 2.0121 0.3913 3.8005 0.0967


2 1.3902 2.5859 2.4137 0.2448 3.3980
3 3.4145 0.1898 1.4337 0.0680 3.0697
4 2.5968 0.9273 2.7916 0.1152 2.2065
5 3.3914 0.4693 1.6750 0.0649 2.1876
6 2.8713 2.1071 1.1225 0.4812 2.1325
7 0.4226 2.0808 3.2887 0.8226 0.8938
8 2.0472 1.6680 2.0529 0.5568 0.6840
9 2.1682 3.1313 0.3281 0.7791 1.0094
10 2.5987 0.8754 2.6584 0.1450 0.3550
11 2.4961 0.3705 2.7577 0.2773 0.4325
12 3.6297 0.7305 0.2771 0.1844 0.7534
13 2.4701 2.4706 1.6484 0.2709 0.7058
14 1.3511 2.2715 2.5952 0.3211 1.8720
15 2.4343 2.7492 0.1814 0.2044 2.2261
16 1.2946 2.9277 2.1830 0.0862 2.0350
17 2.2039 1.0213 3.0066 0.3858 2.4757
18 2.8993 1.4603 1.6259 0.1591 3.1477
19 0.9094 2.9271 2.2723 0.4253 3.3760
20 0.5098 3.0616 1.5846 0.7362 4.1947

4.1 Validation Using Field Data. The developed ANN


empirical correlation was used to provide a complete profile of
the static Poisson’s ratio for a carbonate reservoir based on well
log data; sonic travel times and bulk density. Figure 5 shows the
log data for well-1. The reservoir interval is around 350 ft. Zero
value is the top of the reservoir and 350 ft is the bottom of the res-
ervoir. The reservoir depth is shallow (between 6000 and 7000 ft
for well-1).
Figure 6 shows that, Eq. (3) can be used to obtain a complete
profile for  static with high accuracy. Fairly good match between
the estimated values and the five core data available for this well
was observed. R2 is 0.98 and AAPE is 2.0% when comparing the
actual  static (five data points) with the estimated ones.
To validate the developed model for  static for a deep carbonate
reservoir, the data for well-2 were used. Figure 7 shows the log
Fig. 4 Actual Poisson’s ratio versus predicted one using Eqs. data for well-2. The reservoir section was around 1000 ft.
(2) and (3) for the unseen data (183 data points)

Fig. 5 Input log data for well-1

Journal of Energy Resources Technology JULY 2018, Vol. 140 / 072905-5

Downloaded From: https://energyresources.asmedigitalcollection.asme.org on 01/23/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Fig. 6 Complete profile of the static Poisson’s ratio using Eq.
(3) for well-1 Fig. 8 Static Poisson’s ratio prediction using Eq. (3) for well-2

Fig. 7 Input log data for well-2

072905-6 / Vol. 140, JULY 2018 Transactions of the ASME

Downloaded From: https://energyresources.asmedigitalcollection.asme.org on 01/23/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Figure 8 shows that the predicted values of  static matched the [4] Chang, C., Zoback, M. D., and Khaksar, A., 2006, “Empirical Relations
actual core values (eight points). R2 is 0.93 and AAPE is 2.5% Between Rock Strength and Physical Properties in Sedimentary Rocks,” J. Pet.
Sci. Eng., 51(3–4), pp. 223–237.
when comparing the actual  static (eight data points) with the esti- [5] Hofmann, H., Babadagli, T., and Zimmermann, G., 2014, “Numerical Simula-
mated ones. Based on these results, it can be concluded that the tion of Complex Fracture Network Development by Hydraulic Fracturing in
developed  static equation can provide a complete profile of the Naturally Fractured Ultratight Formations,” ASME J. Energy Resour. Technol.,
 static based on three log data; bulk density, sonic travel times 136(4), p. 042905.
[6] Cui, G., Ren, S., Rui, Z., Ezekiel, J., Zhang, L., and Wang, H., 2018, “The Influ-
with high accuracy. ence of Complicated Fluid-Rock Interactions on the Geothermal Exploitation in
The main limitations of the developed model are (1) it is devel- the CO2 Plume Geothermal System,” Appl. Energy, in press.
oped for carbonate formation, for other lithology the model should [7] Cui, K., Qain, Y., Jeon, I., Anisimov, A., Matsuo, Y., Kauppinen, E., and Mar-
be rebuild to update the weights and biases. (2) It is working per- uyama, S., 2017, “Scalable and Solid-State Redox Functionalization of Trans-
parent Single-Walled Carbon Nanotube Films for Highly Efficient and Stable
fect within the range of the training data. The developed model Solar Cells,” Adv. Energy Mater., 7(18), p. 1700449.
cannot be used to estimate the static Poisson’s ratio out of the data [8] Jaegar, J., Cook, N. G., and Zimmerman, R., 2007, Fundamentals of Rock
range of the training data (below or above the data training data Mechanics, 4th ed., Wiley-Blackwell, Malden, MA.
range). [9] Gatens, J. M., Harrison, C. W., Lancaster, D. E., and Guidry, F. K., 1990, “In-
Situ Stress Tests and Acoustic Logs Determine Mechanical Propertries and
Stress Profiles in the Devonian Shales,” SPE Form. Eval., 5(3), pp. 248–254.
5 Conclusions [10] Nes, O., Fjær, E., Tronvoll, J., Kristiansen, T. G., and Horsrud, P., 2012,
“Drilling Time Reduction Through an Integrated Rock Mechanics Analysis,”
Actual laboratory measurements for static Poisson’s ratio with ASME J. Energy Resour. Technol., 134(3), p. 032802.
the basic log data were used to generate artificial intelligence [11] Chan, T., Hood, M., and Board, M., 1982, “Rock Properties and Their Effect on
Thermally Induced Displacements and Stresses,” ASME J. Energy Resour.
models to estimate a complete profile of the static Poisson’s ratio. Technol., 104(4), pp. 384–388.
Based on the obtained results, the following conclusions can be [12] Phani, K. K., 2008, “Correlation Between Ultrasonic Shear Wave Velocity and
drawn: Poisson’s Ratio for Isotropic Porous Materials,” J. Mater. Sci., 43(1), pp.
316–323.
(1) ANN, ANFIS, and SVM techniques can be used to predict [13] Ameen, M. S., Smart, B. G. D., Somerville, J. M., Hammilton, S., and Naji, N.
 static from well logs, where ANN model yielded more A., 2009, “Predicting Rock Mechanical Properties of Carbonates From Wireline
Logs (a Case Study: Arab-D Reservoir, Ghawar Field, Saudi Arabia),” Mar.
accurate results compared with ANFIS and SVM. Pet. Geol., 26(4), pp. 430–444.
(2) The developed empirical equation for  static gave a coeffi- [14] Al-Anazi, A., and Gates, I. D., 2010, “A Support Vector Machine Algorithm to
cient of determination R2 of 0.97 and AAPE of 1.13%. Classify Lithofacies and Model Permeability in Heterogeneous Reservoirs,”
(3) The black box of the ANN model was changed to a white Eng. Geol., 114(3–4), pp. 267–277.
[15] Fahrenthold, E. P., and Gray, K. E., 1988, “Compaction Performance of
box by extracting the weights and biases of the optimized Geopressured-Geothermal Reservoir Rock,” ASME J. Energy Resour. Tech-
ANN model. Any user can apply the developed correlation nol., 110(3), pp. 189–195.
without the need to have the ANN model. The developed [16] Gercek, H., 2007, “Poisson’s Ratio Values for Rocks,” Int. J. Rock Mech. Min.
correlation can be built and run using any available Sci., 44(1), pp. 1–13.
[17] Howard, G. C., and Fast, C. R., 1970, “Hydraulic Fracturing,” Monograph, Vol.
software. 2, 1st ed., Society of Petroleum Engineers of AIME, Richardson, TX.
(4) The developed mechanism will help geomechanical engi- [18] AL-Ameri, W., Abdulraheem, A., and Mahmoud, M., 2015, “Long-Term
neers to obtain a complete profile of  static without the need Effects of CO2 Sequestration on Rock Mechanical Properties,” ASME J.
for coring and lab work and hence will reduce the overall Energy Resour. Technol., 138(1), p. 012201.
[19] Canady, W. J., 2011, “A Method for Full-Range Young’s Modulus Correction,”
cost of the well. North American Unconventional Gas Conference and Exhibition, The Wood-
lands, TX, June14–16, SPE Paper No. SPE 143604.
Acknowledgment [20] Khaksar, A., Taylor, P. G., Fang, Z., Kayes, T. J., Salazar, A., and Rahman, K.,
2009, “Rock Strength From Core and Logs, Where We Stand and Ways to Go,”
The author would like to thank Dr. Mohamed Mahmoud for EUROPEC/EAGE Conference and Exhibition, Amsterdam, The Netherlands,
proof reading the paper. June 8–11, SPE Paper No. SPE 121972.
[21] Abdulraheem, A., Ahmed, M., Vantala, A., and Parvez, T., 2009, “Prediction of
Rock Mechanical Parameters for Hydrocarbon Reservoirs Using Different Arti-
Nomenclature ficial Intelligence Techniques,” SPE Saudi Arabia Section Technical Sympo-
sium, Al-Khobar, Saudi Arabia, May 9–11, SPE Paper No. SPE-126094.
AAPE ¼ average absolute percentage error [22] Kumar, A., Jayakumar, T., Raj, B., and Ray, K. K., 2003, “Correlation Between
AI ¼ artificial intelligence Ultrasonic Shear Wave Velocity and Poisson’s Ratio for Isotropic Solid Materi-
ANFIS ¼ adaptive neuro fuzzy inference system als,” Acta Mater., 51(8), pp. 2417–2426.
ANN ¼ artificial neural network [23] Asoodeh, M., 2013, “Prediction of Poisson’s Ratio From Conventional Well
Log Data: A Committee Machine With Intelligent Systems Approach,” Energy
R¼ correlation coefficient Sources, Part A: Recovery Util. Environ. Eff., 35(10), pp. 962–975.
R2 ¼ coefficient of determination [24] Arabjamaloei, R., and Shadizadeh, S., 2011, “Modeling and Optimizing Rate of
SVM ¼ support vector machine Penetration Using Intelligent Systems in an Iranian Southern Oil Field (Ahwaz
 dyn ¼ dynamic Poisson’s ratio Oil Field),” Pet. Sci. Technol., 29(16), pp. 1637–1648.
[25] Schalkoff, R., 1997, Artificial Neural Networks, McGraw-Hill, New York.
 static ¼ static Poisson’s ratio [26] Ali, J. K., 1994, “Neural Networks: A New Tool for the Petroleum Industry?,”
staticn ¼ denormalized value of static Poisson’s ratio European Petroleum Computer Conference, Aberdeen, UK, Mar. 5–17, SPE
VP ¼ compressional wave velocity in km/s Paper No. SPE 27561.
Vs ¼ shear wave velocity in km/s [27] Haykin, S. S., 1998, Neural Networks, a Comprehensive Foundation, Prentice
Hall, Upper Saddle River, NJ.
Dtc ¼ compressional time, ls/ft [28] Aalst, W. M. P., Rubin, V., Verbeek, H. M. W., Van Dongen, B. F., Kindler, E.,
Dts ¼ shear time, ls/ft and G€ unther, C. W., 2010, “Process Mining: A Two-Step Approach to Balance
q¼ bulk density, g/cm3 Between Underfitting and Overfitting,” Software Syst. Model., 9(1), pp.
87–111.
[29] Lippman, R. K., 1987, “An Introduction to Computing With Neural Nets,”
References IEEE ASSP Mag., 4(2), pp. 4–22.
[1] Rui, Z., Wang, X., Zhang, Z., Lu, J., Chen, G., Zhou, X., and Patil, S., 2018, “A [30] Goyal, S., and Goyal, G. K., 2011, “Cascade and Feedforward Backpropagation
Realistic and Integrated Model for Evaluating Oil Sands Development With Artificial Neural Network Models for Prediction of Sensory Quality of Instant
Steam Assisted Gravity Drainage Technology in Canada,” Appl. Energy, 213, Coffee Flavoured Sterilized Drink,” Can. J. Artif. Intell., Mach. Learn. Pattern
pp. 76–91. Recognit., 2(6), pp. 78–82.
[2] Guo, J., Luo, B., Lu, C., Lai, J., and Ren, J., 2017, “Numerical Investigation of [31] Vineis, P., and Rainoldi, A., 1997, “Neural Networks and Logistic Regression:
Hydraulic Fracture Propagation in a Layered Reservoir Using the Cohesive Analysis of a Case-Control Study on Myocardial Infarction,” J. Clin. Epide-
Zone Method,” Eng. Fract. Mech., 186, pp. 195–207. miol., 50(11), pp. 1309–1310.
[3] Rui, Z., Han, G., Zhang, H., Wang, S., Pu, H., and Ling, K., 2017, “A New [32] Burbidge, R., Trotter, M., Buxton, B., and Holden, S., 2001, “Drug Design by
Model to Evaluate Two Leak Points in a Gas Pipeline,” J. Natural Gas Sci. Machine Learning: Support Vector Machines for Pharmaceutical Data Analy-
Eng., 46, pp. 491–497. sis,” Comput. Chem., 26(1), pp. 5–14.

Journal of Energy Resources Technology JULY 2018, Vol. 140 / 072905-7

Downloaded From: https://energyresources.asmedigitalcollection.asme.org on 01/23/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use


[33] Tahmasebi, P., 2012, “A Hybrid Neural Networks-Fuzzy Logic-Genetic Algo- [48] Tariq, Z., Elkatatny, S. M., Mahmoud, M. A., Abdulraheem, A., Abdelwahab,
rithm for Grade Estimation,” Comput. Geosci., 42, pp. 18–27. A. Z., and Woldeamanuel, M., 2017, “Estimation of Rock Mechanical Parame-
[34] Alarifi, S., AlNuaim, S., and Abdulraheem, A., 2015, “Productivity Index Pre- ters Using Artificial Intelligence Tools,” 51st U.S. Rock Mechanics/Geome-
diction for Oil Horizontal Wells Using Different Artificial Intelligence chanics Symposium Held, San Francisco, CA, June 25–28, Paper No. ARMA
Techniques,” SPE Middle East Oil & Gas Show and Conference, Manama, 17-301.
Bahrain, Mar. 8–11, SPE Paper No. SPE 172729-MS. [49] Tariq, Z., Elkatatny, S. M., Mahmoud, M. A., Abdulraheem, A., Abdelwahab,
[35] Attia, M., Abdulraheem, A., and Mahmoud, M. A., 2015, “Pressure Drop Due A. Z., Woldeamanuel, M., and Mohamed, I. M., 2017, “Development of New
to Multiphase Flow Using Four Artificial Intelligence Methods,” SPE North Correlation for Unconfined Compressive Strength for Carbonate Reservoir
Africa Technical Conference and Exhibition, Cairo, Egypt, Sept. 14–16, SPE Using Artificial Intelligence Techniques,” 51st U.S. Rock Mechanics/
Paper No. SPE 175724. Geomechanics Symposium Held, San Francisco, CA, June 25–28, Paper No.
[36] Walia, N., Singh, H., and Sharma, A., 2015, “ANFIS: Adaptive Neuro-Fuzzy ARMA 17-428.
Inference System- a Survey,” Int. J. Comput. Appl., 123(13), pp. 32–38. [50] Tariq, Z., Elkatatny, S. M., Mahmoud, M., Abdulwahab, Z. A., and Abdulra-
[37] Uçar, T., Karahoca, A., and Karahoca, D., 2013, “Tuberculosis Disease Diagno- heem, A., 2017, “A New Technique to Develop Rock Strength Correlation
sis by Using Adaptive Neuro Fuzzy Inference System and Rough Sets,” Neural Using Artificial Intelligence Tools,” SPE Reservoir Characterization and Simu-
Comput. Appl., 23(2), pp. 471–483. lation Conference and Exhibition, Abu Dhabi, United Arab Emirates, May
[38] Guo, G., 2014, Support Vector Machines Applications, Y. Ma, and G. Guo, 8–10, SPE Paper No. SPE 186062.
eds., Springer International Publishing, Cham, Switzerland. [51] Tariq, A., Elkatatny, S. A., Mahmoud, M. A., Zaki, A., and Abdulraheem, A.,
[39] Anifowose, F. A., Ewenla, A. O., and Eludiora, S. I., 2011, “Prediction of Oil 2017, “A New Approach to Predict Failure Parameters of Carbonate
and Gas Reservoir Properties Using Support Vector Machines,” International Rocks Using Artificial Intelligence Tools,” SPE Kingdom of Saudi Arabia
Petroleum Technology Conference, Bangkok, Thailand, Nov. 15–17, SPE Paper Annual Technical Symposium and Exhibition, Dammam, Saudi Arabia, Apr.
No. IPTC 14514. 24–27, SPE Paper No. SPE-187974-MS.
[40] El-Sebakhy, E. A., Sheltami, T., Al-Bokhitan, S. Y., Shaaban, Y., Raharja, P. D., [52] Parapuram, G. K., Mokhtari, M., and Hmida, J. B., 2017, “Prediction and Anal-
and Khaeruzzaman, Y., 2007, “Support Vector Machines Framework for Predicting ysis of Geomechanical Properties of the Upper Bakken Shale Utilizing Artifi-
the PVT Properties of Crude Oil Systems,” SPE Middle East Oil and Gas Show and cial Intelligence and Data Mining,” SPE/AAPG/SEG Unconventional
Conference, Manama, Bahrain, Mar. 11–14, SPE Paper No. SPE 105698-MS. Resources Technology Conference, Austin, TX, July 24–26, Paper No.
[41] Elkatatny, S., Mahmoud, M., Tariq, Z., and Abdulraheem, A., 2017, “New URTEC-2692746-MS.
Insights Into the Prediction of Heterogeneous Carbonate Reservoir Permeability [53] Buhulaigah, A., Al-Mashhad, A. S., Al-Arifi, S. A., Al-Kadem, M. S., and Al-
From Well Logs Using Artificial Intelligent Network,” Neural Comput. Appl., Dabbous, M. S., 2017, “Multilateral Wells Evaluation Utilizing Artificial
epub. Intelligence,” Middle East Oil & Gas Show and Conference, Manama, King-
[42] Elkatatny, S. M., Tariq, Z., and Mahmoud, M. A., 2016, “Real Time Prediction dom of Bahrain, Mar. 6–9, SPE Paper No. SPE 183688.
of Drilling Fluid Rheological Properties Using Artificial Neural Networks [54] Solomon, O., Adewale, D., and Anyanwu, C., 2017, “Fracture Width Prediction
Visible Mathematical Model (White Box),” J. Pet. Sci. Eng., 146, pp. and Loss Prevention Material Sizing in Depleted Formations Using Artificial
1202–1210. Intelligence,” SPE Nigeria Annual International Conference and Exhibition,
[43] Elkatatny, S. M., 2017, “Real Time Prediction of Rheological Parameters of Lagos, Nigeria, July 31–Aug. 2, SPE Paper No. SPE 189068.
KCl Water-Based Drilling Fluid Using Artificial Neural Networks,” Arabian J. [55] Respati, P. S., Ardan, C., and Alfaqih, M. R., 2016, “Case Study: Forecast Per-
Sci. Eng., 42(4), pp. 1655–1665. formance of Potential Zone Using Artificial Intelligence AI in Deltaic Mature
[44] Elkatatny, S. M., and Mahmoud, M., 2017, “Development of New Correlations Field,” International Petroleum Technology Conference, Bangkok, Thailand,
for the Oil Formation Volume Factor in Oil Reservoirs Using Artificial Intelli- Nov. 14–16, Paper No. IPTC-18821.
gent White Box Technique,” Petroleum, in press. [56] Shi, X., Liu, G., Jiang, S., Chen, L., and Yang, L., 2016, “Brittleness Index Pre-
[45] Elkatatny, S. M., and Mahmoud, M., 2017, “Development of a New Correlation diction From Conventional Well Logs in Unconventional Reservoirs Using
for Bubble Point Pressure in Oil Reservoirs Using Artificial Intelligent Artificial Intelligence,” International Petroleum Technology Conference Bang-
Technique,” Arabian J. Sci. Eng., epub. kok, Thailand, Nov. 14–16, Paper No. IPTC-18776.
[46] Wang, Y., and Salehi, S., 2015, “Application of Real-Time Field Data to Opti- [57] Elkatatny, S., 2017, “New Approach to Optimize the Rate of Penetration Using
mize Drilling Hydraulics Using Neural Network Approach,” ASME J. Energy Artificial Neural Network,” Arabian J. Sci. Eng., epub.
Resour. Technol., 137(6), p. 062903. [58] Mantha, B., and Samuel, R., 2016, “ROP Optimization Using Artificial Intelli-
[47] Salehi, S., Hareland, G., Dehkordi, K. K., Ganji, M., and Abdollahi, M., 2009, gence Techniques With Statistical Regression Coupling,” SPE Annual Techni-
“Casing Collapse Risk Assessment and Depth Prediction With a Neural Net- cal Conference and Exhibition, Dubai, United Arab Emirates, Sept. 26–28, SPE
work System Approach,” J. Pet. Sci. Eng., 69(1–2), pp. 156–162. Paper No. SPE 181382.

072905-8 / Vol. 140, JULY 2018 Transactions of the ASME

Downloaded From: https://energyresources.asmedigitalcollection.asme.org on 01/23/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use

You might also like