Professional Documents
Culture Documents
This paper was prepared for presentation at the International Petroleum Technology Conference held in Doha, Qatar, 79 December 2009.
This paper was selected for presentation by an IPTC Programme Committee following review of information contained in an abstract submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the International Petroleum Technology Conference and are subject to correction by the author(s). The material, as presented, does not necessarily
reflect any position of the International Petroleum Technology Conference, its officers, or members. Papers presented at IPTC are subject to publication review by Sponsor Society
Committees of IPTC. Electronic reproduction, distribution, or storage of any part of this paper for commercial purposes without the written consent of the International Petroleum Technology
Conference is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not be copied. The abstract must contain conspicuous
acknowledgment of where and by whom the paper was presented. Write Librarian, IPTC, P.O. Box 833836, Richardson, TX 75083-3836, U.S.A., fax +1-972-952-9435.
Abstract
Permeability is one of the crucial parameters in dynamic reservoir modeling and simulation. Direct measurement of
permeability through coring and wireline formation testing is expensive and sometimes hard to achieve. In addition, the
interval of coring is always limited.
In this study available core and wireline log data of a heterogeneous carbonate reservoir located in sarvak formation of Iran
are used to predict permeability not only for the cored but un-cored wells too. To reach this purpose, the concept of rock
typing has been taken into consideration to overcome the problem of heterogeneity. Flow zone index (FZI) approach is
selected to determine the rock types of the reservoir understudy. Mathematical manipulation is then used to transform the
continuous FZI values to discrete ones known as discrete rock types (DRT). Wireline log data corresponde to each DRT are
individualized and subjected to statistical analysis to find their influence on the process of permeability prediction. Gamma
ray, sonic, density and neutron porosity logs have been chosen as input parameters for building artificial neural network
(ANN) models for permeability prediction. An individual ANN model is constructed for the process of estimating the
permeability for each DRT. The result of permeability prediction using this technique is highly satisfactory but dependent on
the successful prediction of FZI in uncored intervals/wells. Fuzzy logic is the approach that was used for estimating the FZI
by wireline logs data. Applying fuzzy logic provided accurate predicted FZI logs for uncored wells. By deriving DRTs from
the FZI log, relevant built ANN models for each DRT might be used for predicting permeability. Validation of the predictive
capability of the method in two cored wells (Blind-test wells) proved the estimation technique to be robust.
For the sake of comparison, permeability-effective porosity transform and multilinear regression are applied for permeability
prediction of the reservoir understudy. Results of applying these methods are considerably less than the results achived in this
work.
Introduction
Permeability is one of the crucial parameters in dynamic reservoir modeling and simulation. Direct measurement of
permeability through coring and wireline formation testing is expensive and sometimes fail to achieve [1]. Also these
measuremens do not provide a continuous profile of permeability along the well and continuous permeability values need to
be predcited indirectly from independent data. In recent years, different methodologies have been introduced to the petroleum
geosciences/engineering discipline to predict permeability from openhole logs. Most of these methods are based on using
artificial intelligence techniques such as artificial neural networks (ANN) and fuzzy logic. These approaches could enhance
the permeability prediction, but in the presence of heterogeneity the degree of accuracy and certainty of using these methods
reduces too. Heterogeneity is a problem that occurs in most of the carbonate reservoirs. Due to the chemical nature of
carbonates, some secondary diagenism processes such as acidic waters may cause dissolution in these systems and forms
fractures and vuggs. Existence of fractures and vuggs increases the permeability to large amounts while does not have
considerable effect on porosity values and therefore causes irregularties in the relationship between porosity and
permeability. This fact can be obsereved on conventional cross plots of porosity-permeability of heterogeneous reservoirs
where high values of permeability are corresponded to low or medium values of porosity.
2 IPTC 13732
Dividing the reservoir rock to discrete rock types can be a solution to the problem of heterogeneity. Rock typing is a process
for the classification of reservoir rocks into distinct units. These units are deposited under similar geological conditions and
undergoes through similar diagenetic alterations. If the rocks are properly classified and defined, a given rock type is
imprinted by a unique porosity/permeability correlation, capillary pressure profile, and set of relative permeability curves [2].
Different quantitative rock typing techniques exist in the literature that each of them has its own definition and applicability.
Among all of them, flow zone index (FZI) approach provides the most applicable results.
In this study, artificial neural networks are used to predict the permeability, but due to the heterogeneity of the reservoir
understudy, FZI approach is applied to enhance the prediction capability of the ANN models. For this purpose, an individual
ANN model is constructed for each discrete rock type (DRT). This idea reduces the scattering in the data set and therefore
increases the degree of accuracy of predicted permeability. But the power and applicability of this model is dependent on
accurate prediction of FZI in uncored intervals/wells. Fuzzy logic is implied for estimating FZI in uncored wells. This
approach provides a good synthetic FZI log for uncored wells and therefore DRTs can be obtained highly correlated with
those defined by cores. When DRTs are known, the constructed ANN models relevant to each rock type can be used for
predicting the permeability of that unit.
[3]
Figure 1: stratigraphic column of the cretaceous succession in the south west of Iran
Data provided for this study include conventional core and wireline logs data of five wells namely 1, 2, 3, 4, and 5 drilled in
the studied oil field. Available wireline logs are SGR (standard gamma ray), CGR (corrected gamma ray), RHOB (bulk
density), TNPH (thermal neutron porosity), Rs (medium resistivity), Rt (deep resistivity), Rxo (shallow resistivity), DT
(sonic), and VCLAY (volume of clay). These logs data are calibrated with cores at exact cored depths. For investigating the
reservoir in terms of uniformity/heterogeneity, conventional crossplots of core derived porosity and permeability and also the
porosity histograms relevant to each of the five wells are shown in figures 2 through 11. Porosity-permeability crossplots of
wells #3 and 4 indicates large irregularities in the porosity and permeability relationship while this non uniformity is less in
wells numbers 1, 2 and 5. On the basis of porosity histograms, one observation indicates that porosity distribution of Well#1
is quite different than the other four wells. On the other hand, wells numbers 3, 4 and 5 have almost similar distribution.
However, the histogram of well#2 shows a wide range of distribution. Based on this simple observation, it might be
concluded that the reservoir is a heterogeneous system.
IPTC 13732 3
10000.000
1000
1000.000
100
Permeability(mD)
Permeability(mD)
100.000
10 10.000
1.000
1
0.100
0.1
0.010
0 5 10 15 20 25 30 35
0.0 5.0 10.0 15.0 20.0 25.0 30.0
Porosity(%) porosity(%)
1000 100
100
10
Permeability(mD)
Permeability(mD)
10
0.1
0.1
0.01 0.01
0 5 10 15 20 25 30 35 40 0 2 4 6 8 10 12 14 16 18
Porosity(%) Porosity(%)
25% 100%
Frequency Cumulative
90%
1000.000
20% 80%
100.000 70%
Permeability(mD)
15% 60%
Frequency
Cumulative
10.000
50%
10% 40%
1.000
30%
0.100 5% 20%
10%
0.010
0% 0%
0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00
0.000
0.150
0.170
0.180
0.190
0.200
0.210
0.220
0.230
0.240
0.290
0.300
0.310
Cumulative
Frequency
Cumulative
10%
15% 50% 50%
8%
40% 40%
10% 6%
30% 30%
20% 4%
20%
5%
10% 2% 10%
0% 0% 0% 0%
0.000
0.040
0.060
0.080
0.100
0.120
0.140
0.160
0.220
0.240
0.260
0.300
0.000
0.030
0.050
0.070
0.090
0.110
0.130
0.150
0.170
0.190
0.210
0.230
0.250
0.270
0.290
0.310
0.330
0.350
0.370
10% 10%
80% 80%
70% 70%
8% 8%
60% 60%
Frequency
Cumulative
Frequency
Cumulative
6% 50% 6% 50%
40% 40%
4% 4%
30% 30%
20% 20%
2% 2%
10% 10%
0% 0% 0% 0%
0.000
0.020
0.030
0.040
0.050
0.060
0.070
0.080
0.090
0.100
0.110
0.120
0.130
0.140
0.150
0.160
0.170
0.180
0.190
0.200
0.210
0.220
0.230
0.240
0.250
0.260
0.300
0.000
0.020
0.030
0.040
0.050
0.060
0.070
0.080
0.090
0.100
0.110
0.120
0.130
0.140
0.150
0.160
0.170
0.180
0.190
0.200
0.210
0.220
0.230
0.240
0.250
0.260
0.270
0.280
0.290
0.300
Porosity (Fraction) Porosity (Fraction)
Figure 10: Porosity jistogram, Well#4 Figure 11: Porosity jistogram, Well#5
K
RQI ( m) = 0.0314 ............................................................................................................................................... (Eq. 1)
e
Then core derived porosity must be converted to normalized ones using [4],[5]:
e
z = ...................................................................................................................................................................... (Eq. 2)
1 e
RQI
FZI = .................................................................................................................................................................... (Eq. 3)
z
By taking the logarithm of both sides of (Eq.3), the following will be obtained.
log RQI = log z + log FZI ......................................................................................................................................... (Eq. 4)
This Equation yields a straight line on a log-log plot of RQI versus z with a unit slope. The intercept of this straight line at
z= 1 is the flow zone indicator. Samples with different FZI values will lie on other parallel lines. Samples that lie on the
same straight line have similar pore throat characteristics and, therefore, constitute a flow unit [5],[6]. Complete mathematical
derivation of FZI is represented in Appendix A.
For determining the rock type's distribution throughout the 3D geological model, the continuous FZI values calculated by Eq.
3 are subjected to mathematical manipulation obtained by geostatistical techniques to be converted to discrete ones. The
relevant equation is [7]:
Where C is a constant and depends on the normal distribution of the FZI data. It is usually about 10.6.
Figures 12 thorugh 14 show the RQI-z crossplots classified by DRT for wells numbers 3, 4 and 5. Wells numbers 1 and 2
are not considered for rock type determination because their numbers of data points are not satisfactory. As can be seen on
RQI- z plots, 10 discrete rock types are recognizable in this reservoir, starting from DRT 8 and ends at DRT 17. It should be
noted that some DRTs are integrated such as 6 and 7 to a one DRT (DRT 7) and DRTs 16, 17 and 18 to a one DRT (DRT
IPTC 13732 5
16). Ofcourse the frequency of each rock type is different. According to simple statistical analysis, the frequency of DRTs 9,
10, 11, 12 and 13 are higher than other rock types and these units constitute the main parts of the reservoir. Table 1 indicates
the rock type distribution in each well. Table 2 also indicates statistical analysis of data associated to each discrete rock types
for wells numbers 3, 4 and 5.
Well#2 3 8 3 5 7 2 0 0 1 0
Well#3 10 20 30 36 39 25 16 10 14 6
Well#4 2 7 4 4 4 6 2 0 2 1
Well#5 10 22 61 39 28 20 9 4 4 4
Table 2(a): Average petrophysical values for different rock types within Well#3
DRT No. Average porosity Average Permeability Average FZI
DRT 8 0.15382 1.04 0.282298
Table 2(b): Average petrophysical values for different rock types within Well#4
DRT No. Average porosity Average Permeability Average FZI
DRT 8 0.10315 0.121 0.296328
DRT 9 0.085871 0.281429 0.495416
DRT 10 0.0747 0.36575 0.726374
DRT 11 0.064793 0.913 1.30116
DRT 12 0.09015 5.1945 1.92748
DRT 13 0.071367 17.2545 3.31765
DRT 14 0.0775 33.654 5.84469
Table 2(c): Average petrophysical values for different rock types within Well#5
DRT No. Average porosity Average Permeability Average FZI
DRT 8 0.146114 0.5696 0.29523
DRT 9 0.110469 0.611043 0.475127
DRT8 1
DRT9
DRT10
10 DRT8
DRT11
DRT9
DRT12
DRT10
DRT13
DRT11
DRT14
DRT12
1 DRT15
DRT13
DRT16
RQI
0.1 DRT14
DRT17
RQI
Linear (DRT8)
Linear (DRT8)
Linear (DRT9)
Linear (DRT9)
Linear (DRT10)
0.1 Linear (DRT10) Linear (DRT11)
Linear (DRT11) Linear (DRT12)
Linear (DRT12) Linear (DRT14)
Linear (DRT13)
Linear (DRT14)
0.01 Linear (DRT15)
0.01
0.01 0.1 1 0.01 0.1 1
Linear (DRT16)
Linear (DRT17)
Normalized Porosity Normalized Porosity
Figure 12: Rock typing by FZI method, Well#3 Figure 13: Rock typing by FZI method, Well#4
10 DRT 8
DRT 9
DRT 10
DRT 11
DRT 12
DRT 13
1 DRT 14
DRT 15
DRT 16
RQI
DRT 17
Linear (DRT8)
To predict permeability in this study, at first a multiple regression model is constructed for wells numbers 3 and 5. Wireline
log data selected as input for the regression model are SGR (standard gamma ray), CGR (corrected gamma ray), RHOB (bulk
density), TNPH (thermal neutron porosity), Rs (medium resistivity), Rt (deep resistivity), Rxo (shallow resistivity), DT
(sonic), and VCLAY (volume of clay) and the output or response of the model is absolute permeability and logarithm of that.
Regression coefficients for predicting both of the permeability and logarithm of permeability of wells 3 and 5 are given in
table 3. Adjusted correlation coefficient (R2-adjusted) is the criterion that is considered to investigate the quality of the
prediction. According to R2s, the actual permeability data and the predicted ones do not have satisfactory correlation. Also it
should be noted that R2 of permeability prediction is slightly higher than that of logarithm of permeability prediction.
Figures 15 through 20 show the core derived permeability against predicted ones by multiple regression models for wells 3
and 5 and also the associated residual plots. Each graph of residual plots contains four diagrams. The first diagram that is a
normal probability plot investigates the normality assumption of the residuals distribution. Points on the plot should generally
form a straight line, if the residuals are normally distributed [10]. In figure 17 the form of the data are closer to a straight line
with respect to figure 20, indicating that the regression built for well #3 is better than that of well #5. Second diagram is the
histogram of the residuals. This graph is an exploratory tool to show general characteristics of the residuals, including typical
values, spread, and shape. A long tail on one side may indicate a skewed distribution. If one or two bars are far from the
others, those points may be outliers [10]. Although both figures 17 and 20 have outliers points, those of figure 20 are more
with respect to figure 17. The third graph is the graph of residuals versus fitted values. This plot should show a random
pattern of residuals on both sides of 0. If a point lies far from the majority of points, it may be an outlier. There should be no
recognizable patterns in the residual plot [10]. As it is obvious, figure 20 has a clear pattern that is like a straight line dipping
from north-west to south-east. The final graph of residual plots is a plot of all residuals in the order that the data was
collected. This plot helps to check the assumption that the residuals are uncorrelated with each other [10]. A concept of all
these plots is that the regression models can not provide satisfactory results for the problem of permeability estimation of the
understudy heterogeneous reservoir especially in the case of well#5.
1000 400
350
y=0.5854x+12.116
R2=0.5857
300
100
250
Permeability(md)
PredictedK(md)
200
ActualK
10
150 PredictedK
100
1
50
0.1 50
0.01 0.1 1 10 100 1000 2780 2800 2820 2840 2860 2880
ActualK(md) Depth(ft)
Figure 15: Permeability prediction by multiple regressions, Figure 16: Permeability prediction by multiple regressions,
Well#3 Well#3
100
50 100
0
10
PredictedK(md)
1 -100
0.1
-100 0 100 200 300 0 50 100 150 200
Residual Fitted Value 10
75
Residual
100
50
0
25
-100 0.1
0 0.01 0.1 1 10 100 1000
-100 -50 0 50 100 150 200 250 1 20 40 60 80 100 120 140 160 180 200
Residual Observation Order
ActualK(md)
Figure 17: Residual Plots for regression model, Well#3 Figure 18: Permeability prediction by multiple regressions, Well#5
8 IPTC 13732
400 90
200
Residual
Percent
50
100
300 10
0
Permeability(md)
1
0.1 -100
ActualK -100 0 100 200 300 0 40 80 120
200
PredictedK Residual Fitted Value
60
200
Frequency
Residual
0
40 100
100 20 0
Figure 19: Permeability prediction by multiple regressions, Figure 20: Residual Plots for regression model, Well#5
Well#5
The next approach that is used to predict permeability is artificial neural networks (ANN). For this purpose, a multilayer
feed-forward neural network consisting of two layers is applied. In this model, among nine available log data, four having the
highest correlation with logarithm of permeability are chosen as input for the network. Selected logs are RHOB, TNPH, DT,
SGR logs respectively. The first layer of the network is a hidden layer. This layer consists of 8 neurons. In this study, number
of the neurons in the hidden layer is determined by constructive method. In constructive method, the optimal number of
hidden layer nods is obtained by starting with a relatively small (may be 1) number of neurons and increasing the number of
neurons one by one untill there is no further significant improvement in model performance. Second layer of the network is
the output layer consisting of one neuron that is the absolute permeability. In this model, tangant of sigmoid function and
linear function are used as activation functions in hidden neurons and output neuron respectively. Algorithm used for training
the network and adjusting the weights and biases is back propagation. To predict permeability by this ANN model, available
data are divided into three subsets. The first subset that is used for training the network contains 60% of the whole data.
Second and third subsets are used for testing and validating the network performance and the data allocated to each of which
is 20% of the total. To avoid overtraining in this model, cross validation approach is implemented. In this method, the
training stops when the testing subset error starts to increase. Figures 21 through 24 show predicted permeability by ANN
model against measured ones for wells 3 and 5.
400 400
350 y=0.8565x+5.5096 350
2
300 R =0.7859 300
Perm eability,m d
PredictedK
MeasuredK Depth,ft
Figure 21: Permeability prediction by ANN, Well#3 Figure 22: Permeability prediction by ANN, Well#3
500 500
450 450
y=0.628x+5.349
400 2 400
R =0.7261
350
Perm eability,m d
350
PredictedK
300 300
250 250
MeasuredK
200 200
150 PredictedK
150
100 100
50 50
0 0
0 100 200 300 400 500
2820 2840 2860 2880 2900 2920 2940 2960
MeasuredK
Depth,ft
Figure 23: Permeability prediction by ANN, Well#5 Figure 24: Permeability prediction by ANN, Well#5
IPTC 13732 9
As can be seen, the correlation coefficient has improved considerably with respect to applying multiple regressions. Ofcourse
the estimated permeability suffers from inaccuracy yet.
PredictedK(log),md
1.5 2
R =0.7416
1.5
1
1
0.5
0.5
0
0
0.5
1 0.5
1.5 1
2 1.5
2 1 0 1 2 3 2 1 0 1 2 3
MeasuredK(log),md MeasuredK(log),md
Figure 25: Permeability prediction by regression, DRT10 Figure 26: Permeability prediction by regression, DRT11
Normal Probabilit y Plot Versus Fit s
3 99
2.5 y=0.816x+0.2586 90
1
2
Residual
2
Percent
R =0.8171 50
0
PredictedK(log),md
1.5 -1
10
1
1 -2
0.5 -2 -1 0 1 -2 -1 0 1 2
Residual Fitted Value
0
Hist ogram Versus Order
0.5
12
1
1
9
Frequency
Residual
1.5 0
6
2 -1
3
2 1 0 1 2 3
0 -2
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Figure 27: Permeability prediction by regression, DRT12 Figure 28: Residual plots for regression model, DRT10
No rmal Pro b abilit y Plo t Versu s Fit s No rmal Pro b ab ilit y Plo t Versus Fit s
99 99
50 1.0
90 90
25 0.5
Residual
Residual
Percent
Percent
50 0 50 0.0
-0.5
-25
10 10
-1.0
-50
1 1
-50 -25 0 25 50 0 25 50 75 100 -1.0 -0.5 0.0 0.5 1.0 -2 -1 0 1 2
Residual Fitted V alue Residual Fitted Value
Frequency
Residual
Residual
8 0 0.0
5.0
-0.5
4 -25
2.5
-1.0
-50
0 0.0
-40 -20 0 20 40 1 5 10 15 20 25 30 35 -1.2 -0.6 0.0 0.6 1.2 1 5 10 15 20 25 30 35
Residual O bser vation O r der Residual O bser vation O r der
Figure 29: Residual plots for regression model, DRT11 Figure 30: Residual plots for regression model, DRT12
10 IPTC 13732
3 2.5
2.5 2
2
logofPermeability,md
1.5
PredictedK(log),md
y=0.8631x+0.0955
1.5 2
R =0.9122 1
1
0.5
0.5 PredictedK
0
0
0.5 MeasuredK
0.5
1
1
1.5
1.5
2 2
2 1 0 1 2 3 2780 2800 2820 2840 2860 2880
MeasuredK(log),md Depth,ft
Figure 31: Permeability prediction by ANN, DRT10 Figure 32: Permeability prediction by ANN, DRT10
3
2.5
2.5 2
y=0.8472x+0.1615
logofPerm eability,m d
2
2 R =0.8647 1.5
PredictedK(log),md
1.5 1
0.5
1
0 PredictedK
0.5
0.5 MeasuredK
0 1
0.5 1.5
1 2
2780 2800 2820 2840 2860 2880
1.5
2 1 0MeasuredK(log),md
1 2 3 Depth,ft
Figure 33: Permeability prediction by ANN, DRT11 Figure 34: Permeability prediction by ANN, DRT11
3
y=0.9602x+0.1325 3
2.5 2
R =0.9537 2.5
2
2
logofPermeability,md
PredictedK(log),md
1.5
1.5
1 1
0.5 0.5
PredictedK
0
0
0.5 MeasuredK
0.5
1 1
1.5 1.5
2 2
2 1 0 1 2 3 2780 2800 2820 2840 2860
MeasuredK(log),md Depth,ft
Figure 35: Permeability prediction by ANN, DRT12 Figure 36: Permeability prediction by ANN, DRT12
distinctly different for most of the rock types. Although for some DRTs density functions look similar, all the rock types have
different mean interval transit time. The density functions of different rock types might be grouped into three clusters. In the
first cluster DRT 14 through 17 can be grouped as good rock. In the second cluster they are DRT 13, 10, 9, and 8, while DRT
11 and 12 can be classified one cluster.
The density functions of the density log data is shown in Figure 39 for all rock types. The obtained density functions are quite
different for most of the rock types. Although for some DRTs density functions look similar, all the rock types have different
mean and variance interval. The density functions of different rock types might be grouped into three clusters, namely
homogenous, slight heterogonous, and strong heterogeneous formations. In the first cluster DRT 14 through 17 can be
grouped as high density rocks. In the second cluster they are DRT 13,10,9, and 8, while DRT 11 and 12 as one cluster with
almost the same mean and a wide variance interval for both rock types. The third cluster has a wide range of bulk density
which is an indication of a very heterogeneous formation as mentioned above.
The density functions of the neutron porosity log data is shown in Figure 40 for all DRTs. The obtained density functions are
different for all the rock types and are similar to the bulk density log behavior classification.
The Rs density functions for the different DRTs are given in Figure 41. As can be seen, DRTs 13 up to 17 have almost the
same mean, while the other are quite different and have a wide variances. So based on this observation, two distinct clusters
might be defined. The same observation is obtained for the Rt logs data except for DRT 13 as shown in Figure 42.
The plotted data of the density function of Rxo log is shown in Figure 43. As can be seen there is no distinct difference
between the functions and an overlapping is almost exists. The same results were observed for the CGR and VCLAY plots as
shown in Figures 44, and 45 respectively. It should be mentioned that the CGR log is the formation shale indicator which is
obtained form the Gamma Ray contribution from Thorium and Potassium. VCALY data are also obtained from the CGR log
and it represents the volume of the shale.
Based on mentioned statistical analysis, it was concluded that SGR, CGR, Rxo, TNPH, and VCLAY logs are probably the
main logs that should be used for prediction. Ofcourse in another attempt, RHOB, TNPH, DT and CGR are also used as input
to the model and no significant difference in the results was observed.
DRT 16
0.05
Density
DRT 16
DRT 17
0.10 DRT 17
0.04 Mean StDev N
Mean StDev N
16.28 4.523 10
68.35 7.789 10
18.74 5.616 20
0.03 16.05 8.454 30
68.96 7.059 20
67.67 8.860 30
14.21 8.833 36
0.02 9.479 7.542 39
0.05 68.02 10.36 36
69.65 10.07 39
13.96 8.385 25
62.43 8.361 25
0.01 12.38 5.197 16
57.64 5.534 16
14.71 7.065 10
55.61 2.276 10
13.45 5.324 14
0.00 0.00 56.35 2.751 14
13.95 7.414 5
0 10 20 30 40 50 60 70 80 90 62.76 12.90 5
SGR DT
Figure 37: Density function for standard gamma ray log Figure 38: Density function for sonic log
Variable Variable
14 DRT 8
DRT 8
DRT 9 20 DRT 9
DRT 10 DRT 10
12 DRT 11
DRT 11
DRT 12 DRT 12
DRT 13 DRT 13
10 15 DRT 14
DRT 14
DRT 15 DRT 15
Density
DRT 16 DRT 16
Density
8 DRT 17
DRT 17
Mean StDev N
10 Mean StDev N
6 2.492 0.09990 10 0.1254 0.06649 10
2.490 0.09922 20 0.1355 0.05677 20
2.494 0.1202 30 0.1211 0.07272 30
4 2.473 0.1409 36 0.1217 0.07308 36
5 0.1345 0.07938 39
2.458 0.1458 39
2.556 0.1308 25 0.08179 0.05812 25
2 0.04602 0.03448 16
2.610 0.05778 16
0.03748 0.01873 10
2.617 0.02867 10
0.03565 0.01871 14
0 2.632 0.04749 14 0
0.07272 0.08747 5
2.25 2.40 2.55 2.70 2.85 2.555 0.1691 5 -0.075 0.000 0.075 0.150 0.225 0.300
RHOB TNPH
Figure 39: Density function for density log Figure 40: Density function for neutron porosity log
12 IPTC 13732
DRT 16
Density
DRT 16
0.04 DRT 17 DRT 17
Figure 41: Density function for medium resistivity log Figure 42: Density function for deep resistivity log
Density
DRT 16
DRT 17 DRT 17
Figure 43: Density function for shallow resistivity log Figure 44: Density function for corrected gamma ray log
40 Variab le
D RT 8
D RT 9
D RT 10
D RT 11
30 D RT 12
D RT 13
D RT 14
D RT 15
Density
D RT 16
D RT 17
20
M ean S tD ev N
0.3444 0.01079 10
0.3510 0.01968 20
0.3460 0.01998 30
10 0.3500 0.02424 36
0.3455 0.02718 39
0.3497 0.01815 25
0.3502 0.01720 16
0.3494 0.02072 10
0.3609 0.02078 14
0
0.3477 0.01917 5
0.28 0.32 0.36 0.40 0.44 0.48
V CLA Y
To perform fuzzy logic approach for predicting the FZI, well#3 is selected for training the model and wells numbers 4 and 5
are used as blind-test wells to test the prediction model. In this study, subtractice clustering technique (SC) is used for
clustering the data. Based on this method, potentiality of all the data points to be a cluster center will be determined as a
function of its distance to all other data points. After computing the potential of every data point, the data point with the
highest potential will be selected as the first cluster center. Then, an amount of potential will be subtracted from each data
point as a function of its distance from the first cluster center. When the potential of all data points has been revised, data
point with the highest remaining potential will be selected as the second cluster center and this process will be continued to
find the optimum number of clusters [14]. A critical parameter that should be determined in subtractive clustering technique is
radius of clusters. Very small radius causes the number of clusters to be high .Although large number of clusters may cause
the network to cover all the heterogeneities of the training data it would not have the good performance in test samples. The
problem is due to the overtraining of the model. So the best way to find the best radius is finding it experimentally and by the
trial and error.
After clustering the data by subtractice clustering technique, Takagi-Sugeno fuzzy inference susytem is applied for
constructing the model and then the process of prediction. Takagi-Sugeno fuzzy-modeling method is a framework for
generating fuzzy if/then rules from numerical data. This model consists of a set of fuzzy rules, each describing a local linear
input/output relationship [14].
IPTC 13732 13
Results of FZI prediction by fuzzy logic approach are illustrated in figures 46 through 49. As it can be seen, good correlation
exists between predicted FZIs and core derived ones validating the generated model.
25
30
20 y=0.8443x+0.1028
2
25
R =0.9591
FlowZoneInde x
20
15
PredictedFZI
15
PredictedFZI
10
10 MeasuredFZI
5
5
0 0
0 5 10 15 20 25 2680 2685 2690 2695
MeasuredFZI Depth,ft
Figure 46: FZI prediction by fuzzy logic, Well#4 Figure 47: FZI prediction by fuzzy logic, Well#4
30
30
25
25
y=0.8067x+0.4571
20 2
PredictedFZI
R =0.9136 20
FlowZoneIndex
15 15
PredictedFZI
10 10 MeasuredFZI
5 5
0 0
0 5 10 15 20 25 30 2800 2850 2900 2950 3000
MeasuredFZI Depth,ft
Figure 48: FZI prediction by fuzzy logic, Well#4 Figure 49: FZI prediction by fuzzy logic, Well#4
Model Validation
To verify the capability of the approach used in this study, two cored wells (wells numbers 4 and 5) are taken into
consideration. At first, FZI model generated for the well#3 is used to predict the log of FZI and 3D modeling of discrete rock
types for the wells numbers 4 and 5. Neural networks models that were constructed based on the data of well#3 for
permeability prediction are applied in this stage to predict the logarithm of permeability associated with DRTs 10, 11 and 12
that constitute the main parts of the reservoir rock. Figures 50 through 53 illustrate the results. It is obvious that applying this
approach provides better results than using multiple regressions or an individual neural network models. Ofcourse, validity of
this approach is highly dependent on the quality of wireline logs that should be used as inputs for the models.
To compare the results of applying this approach with another method, after modeling the rock types, permeability-effective
porosity transform is used instead of the neural network models to predict the permeability based on the effective porosity
derived from porosity logs. Although this approach provides acceptable results, its accuracy is lower than the approach used
in this study.
0 0.25
y=0.8784x0.0139
0.05 2
0.2
R =0.8796 y=0.8625x+0.0029
PredictedK(log),m d
2
0.15 R =0.8377
0.1
PredictedK(log),m d
0.1
0.15
0.05
0.2
0
0.25 0.05
0.3 0.1
0.25 0.2 0.15 0.1 0.05 0 0.05 0 0.05 0.1 0.15 0.2 0.25
MeasuredK(log),md MeasuredK(log),md
Figure 50: Permeability prediction by hydraulic unit approch, Figure 51: Permeability prediction by hydraulic unit approch,
DRT10, Well#5 DRT11, Well#5
14 IPTC 13732
0.5 0.4
0.45
0.3 y=0.9851x0.0095
0.4 y=0.9287x+0.0151 2
R =0.9389
PredictedK(log),m d
PredictedK(lo g),md
0.35 R =0.9153 0.2
0.3 0.1
0.25
0.2 0
0.15 0.1
0.1
0.05 0.2
0 0.3
0 0.1 0.2 0.3 0.4 0.5 0.3 0.2 0.1 0 0.1 0.2 0.3 0.4
MeasuredK(log),md MeasuredK(log),md
Figure 52: Permeability prediction by hydraulic unit approch, Figure 53: Permeability prediction by hydraulic unit approch,
DRT12, Well#5 DRT10, 11 and 12, Well#4
Conclusions
Based on the results of this study the following conclusions might be concluded:
1. In the case of heterogeneous carbonate reservoirs, conventional methods such as multiple regressions are not useful for
prediction processes.
2. Artificial neural networks provide better estimations of permeability with respect to multiple regressions in carbonate
reservoirs.
3. Flow zone index approach found to be useful for reservoir rock typing and defining the hydraulic units.
4. Differentiating ANN models for permeability estimation based on hydraulic units improves the results of prediction
process.
5. Accurate prediction of FZI for uncored wells plays an important role in applying hydraulic unit-based models for
permeability prediction.
6. Fuzzy logic found to be a powerful tool for predicting FZI by wireline logs.
7. Better estimated models could be generated if high quality wireline logs were available.
Nomenclature
AI = Acoustic impedance
b = regression coefficients
CGR = corrected gamma ray
DRT = district rock type
DT = sonic transit time
Fs = shape factor
FZI = flow zone indicator
HFU = hydraulic flow unit
K = permeability (md)
RHOB = bulk density
RQI = reservoir quality index
Rs = medium resistivity
Rt = deep resistivity
Rxo = shallow resistivity
SGR = standard gamma ray response
Sgv = surface area per unit grain volume
VCLAY = volume of clay
TNPH = thermal neutron porosity
= porosity, fraction
e = effective porosity
= tortuosity factor
References
1. Ghafoori, M. R., Roostaeian, M. and Sajjadian, V. A.,: "A State-of-the-Art Permeability Modeling Using Fuzzy Logic in a
Heterogeneous Carbonate (An Iranian Carbonate Reservoir Case Study), paper IPTC 12019, presented at the International Petroleum
Technology Conference, Kuala Lumpur, Malaysia, 3-5 December 2008.
IPTC 13732 15
2. Riyaz Kharrat, Ramin Mahdavi, Mohammad Hashem Bagherpour, Shahab Hejri,: "Rock Type and Permeability Prediction of a
Heterogeneous Carbonate Reservoir Using Artificial Neural Networks Based on Flow Zone Index Approach", paper SPE 120166 presented
at the 2009 SPE Middle East Oil & Gas Show and Conference, Kingdom of Bahrain, 15-18 March 2009.
3. Taghavi, A. A., Mork, A. and Kazemzadeh, E.: "Flow Unit Classification for Geological Modeling of a Heterogeneous Carbonate
Reservoir: Cretaceous Sarvak Formation, Dehluran Field, SW Iran", Journal of Petroleum Geology, Vol. 30(2), April 2007, pp129-146.
4. Shamsuddin H. Shenawi, Jerry P. White, Emad A. Elrafie and Khaled A. Kilany, "Permeability and Water Saturation Distribution by
Lithologic Facies and Hydraulic Units: A Reservoir Simulation Case Study", paper SPE 105273, presented at the 15 SPE Middle East Oil
& Gas Show and Conference, Bahrain International Exhibition Centre, Kingdom of Bahrain, 1114 March 2007.
5. Yaser Abdi, Mehdi Ghane and Ahmad Reza Haghighi,: "Integrated Reservoir Characterization and Modeling of One Iranian Naturally
Fractured Reservoir Using Laboratory and Field Data", paper SPE 111120, presented at the 2007 SPE/EAGE Reservoir Characterization
and Simulation Conference, Abu Dhabi, U.A.E. , 28-31 October 2007.
6. Amaefule, J.O., Altunba., M., Tiab, D., Kersey, D. G., and Keelan, D.K.,: "Enhanced Reservoir Description: Using Core and Log Data to
Identify Hydraulic (Flow) Units and Predict Permeability in Uncored Intervals/Wells", paper SPE 26436, presented at the 1993 SPE
Annual Technical Conference and Exhibition, Houston, Texas, 3-6 October 1993.
7. Guo, G., Diaz, M.A., Paz, F., Smalley, J. and Waninger, E.A., : "Rock Typing as an Effective Tool for Permeability and Water saturation
Modeling: A Case Study in a Clastic Reservoir in the Oriente Basin", paper SPE 97033, presented at the 2005 SPE Annual Technical
Conference and Exhibition, Dallas, Texas, U.S.A. 9-12 October 2005.
8. Babadagli, T. and Al-Salmi, S.,: "Improvement of Permeability Prediction for Carbonate Reservoir using Well Log Data", paper SPE
77889, presented at the SPE Asia Pacific Oil and Gas Conference and Exhibition, Melbourne, Australia, 8-10 October 2002.
9. Ceylan Yildirim Akbas, Determination of Flow Units for Carbonate Reservoirs by Petrophysical- Based Methods, a thesis submitted to
the graduate school of natural and applied sciences of Middle East Technical University, 146 pages, August 2005.
10. MINITAB 15 User's Guide 2006, multiple regressions and Residual Plot.
11. MATLAB User's Guide 2006, Neural Networks Tool Box.
12. Pablo E. Lecentre and Pablo M. Carrica,: "A Method to Estimate Permeability on Uncored Wells Based on Well Logs and Core data" ,
paper SPE 81058, presented at the SPE Latin American and Caribbean Petroleum Engineering Conference, Trinidad, west Indies, 27-30
April 2003.
13. Taghavi, A. A., : "Improved Permeability Estimation Through Use of Fuzzy Logic in a Carbonate Reservoir From Southwest Iran",
paper SPE 93269, presented at the 14th SPE Middle East Oil & Gas Show and Conference, Bahrain, 12-15 March 2005.
14. PetroFuzzy User's Guide 2008.
15. Dmitry Svirsky, Andrey Ryazanov, Michael Pankov, Yukos EP, Patrick WM Corbett, and Andrey Posysoev,: "Hydraulic Flow Units
Resolve Reservoir Description Challenges in a Siberian Oil Field", paper SPE 87056, presented at the SPE Asia Pacific Conference on
Integrated Modeling for Asset Management, Kuala Lumpur, Malaysia, 29-30 March 2004.
16. Djebbar Tiab, Erle C. Donaldson,: "Petrophysics, Theory and Practice of Measuring Reservoir Rock and Fluid Transport Properties",
Gulf Professional Publishing, second edition, U.S.A., 2004, pp.120-122 and 320-322.
Appendix A [6],[15],[16]
The derivation of the FZI equations are based on the assumption that porous medium can be represented by a bundle of
capillary tubes. The combination of Darcy's law and Poiseuille's law of straight cylindrical tubes produces a simple equation
which relates the porosity and permeability as given below:
r2
K= e ............................................................................................................................................................................(A-1)
8
The above equation correlates the geometrical characteristics of the pore size (radius) and pore shape. The value 8 in the
equation (A-1) is for cylindrical tubes. A tortuousity factor and the mean hydraulic radius were used by Kozeny and
Carmen to account for realistic porous media in the above equation, which resulted in the generalized form as given in
equation (A-2). The mean hydraulic radius was expressed through the surface area per unit grain volume, Sgv.
e3 1
K= .................................................................................................................................................(A-2)
(1 e ) Fs S gv
2 2 2
Where Fs is the shape factor, K is in m2 and e is in a fraction. The group Fs2 is known as the Kozeny constant. This
constant is not known for particular rocks.
This approach addresses variability of the Kozeny constant and the S2gv term by classifying the flow zone indicator, FZI,
which includes all major geological and geometrical characteristics of a porous medium.
1
FZI = .............................................................................................................................................................(A-3)
Fs 2 S gv2
16 IPTC 13732
So
e3 1
K= ..........................................................................................................................................................(A-4)
(1 e ) FZI 2
2
Dividing both sides by porosity (e) and taking the square root of the both sides, results in:
K 1
= e ........................................................................................................................................................(A-5)
e 1 e FZI
If the permeability is expressed in terms of millidarcies, then Reservoir Quality Index, RQI, can be defined as follows:
K
RQI ( m) = 0.0314 ..................................................................................................................................................(A-6)
e
e
z = ..........................................................................................................................................................................(A-7)
1 e
RQI
FZI = ........................................................................................................................................................................(A-8)
z
Where, FZI is Flow Zone Indicator (m). By taking the logarithm of both sides of equation (A-8), the following equation is
obtained: