Académique Documents
Professionnel Documents
Culture Documents
number of HFUs (or any user-defined are internally consistent and predictably different
number of HFUs); and from properties of other rocks 2.
regression analysis is performed using Amaefule et al.3 considered the role of the
the different HFUs as dummy variables mean hydraulic radius in defining hydraulic flow
to predict values permeability. units and correlating permeability from core data.
Their approach was essentially based on a modified
Development of Hydraulic Flow Unit Concept Kozeny-Carmen25 equation coupled with the
Although permeability values are best concept of mean hydraulic radius.
determined from core data, most wells are not
cored. Consequently, permeability values in
uncored wells are usually estimated from porosity Cross Sectional Area r
and permeability relationships developed from core r mh = = ,(2)
WettedPerimeter 2
data or from well test data. The general expression
for the conventional porosity-permeability
where r is the pore throat radius in m and rmh is
transform can be written as
mean hydraulic radius in m.
Kozeny4 and Carmen5 simulated a porous
log( k ) = a + b (1) medium as a bundle of capillary tubes. They
combined Darcys law for flow in a porous medium
and Poiseuilles law for flow in tubes. A tortuosity
There is no rigorous theoretical basis to factor was also included, because for a realistic
support the traditional crossplot of the logarithm of model of porous media the connected pore structure
permeability versus porosity, but an analogy can be is not straight capillary tubes, to give the following
made with the Kozeny-Carmen1 equation. When relationship between porosity and permeability
graphing permeability versus porosity data it is
assumed that permeability is log normally
distributed. However, correlation of two normally
e r e r 2 e r 2mh
k= = = ,.(3)
distributed variables does not necessarily establish 8 2 2 2 2 22
causality. Unlike permeability, porosity is generally
independent of grain size. where k is permeability, e is effective porosity, and
Fig. 1 shows a scatterplot of permeability is tortuosity.
versus porosity for core data from all the wells used The mean hydraulic radius can be related to
in this study. The plot shows a lot of scatter, which the surface area per unit grain volume, Sgv, and the
clearly indicates that porosity alone is not enough to effective porosity, e, by the following equation:
explain the permeability variation. Even if the
porosity-permeability data that were used came 1 e
from the same well, as in Fig. 2, the plot still shows S gv = .......(4)
a lot of scatter. The scatter of these plots could be r mh 1 e
attributed to the existence of more than one rock
type, with different fluid flow properties. Hence, the Combining Eqs. 3 and 4, gives the generalized
porosity permeability relationship is best achieved if Kozeny-Carmen1 equation.
rocks with similar fluid-flow properties are
identified and grouped together. Each group is
referred to as a hydraulic flow unit. 3e 1
k= (5)
1
( ) F s 2 S 2gv
2
A hydraulic flow unit is defined as the
e
representative volume of total reservoir rock within
which geological properties that control fluid flow
SPE 63254 PERMEABILITY ESTIMATION USING HYDRAULIC FLOW UNITS IN A CENTRAL ARABIA RESERVOIR 3
where k is in m2 and e is a fraction. The term Fs2 with significantly different FZI values will lie on
is known as the Kozeny constant, which is usually other, parallel, unit-slope lines. Samples that lie on
between 5 and 100 in most reservoir rock. The term the same straight line have similar pore throat
Fs2S2gv is a function of geological characteristics of attributes, and thereby, constitute a unique hydraulic
porous media and varies with changes in pore flow unit. Each line is an HFU and the intercept of
geometry.6 The determination and discrimination of this line with z = 1 is the mean FZI value for that
the Fs2S2gv group is the focal point of the HFU HFU.
classification technique. The basis of HFU classification is to identify
Amaefule et al.3 addressed the variability of groups of data that form a unit-slope straight lines
Kozenys constant by dividing Eq. 5 by effective on a log-log plot of RQI versus z. The permeability
porosity, e. of a sample point is then calculated from a pertinent
HFU using the mean FZI value and the
k e 1 corresponding sample porosity using the following
0.0314 = ,(6)
e (1 e) F s S gv equation:
1
FZI = ,.(7) HFU Classification Using Core Data
F s S gv
Identification of hydraulic flow units within
a reservoir is accomplished with core data.
reservoir quality index RQI as
Measurements of porosity and permeability under
k net confining stress from conventional cores are
RQI = 0.0314 , .(8) used to compute RQI, z, and FZI. Although there
e
should be one single FZI value for each HFU, a
distribution for each FZI around its true mean value
and normalized porosity z as results because of random measurement errors in
core analysis. When multiple HFU groups exist, the
overall FZI distribution function is a superposition
z = e , .(9)
1 e of the individual distribution functions around their
mean FZI. Identification of each mean FZI, or the
Eq. 6 becomes corresponding HFU, would require decomposition
of the overall FZI distribution into its constituting
RQI = z FZI .....(10) elements. This is a desuperposition problem, and
cluster analysis technique allows for such a
Taking the logarithm of both sides of Eq. 10 yields decomposition process.2
Three approaches, histogram analysis,
probability plot, and analytical cluster algorithm
log RQI = log z + log FZI .(11) have been used by previous investigators2,3 for
clustering core data into the appropriate HFU
groups. The first two methods can be called
Ideally, on a log-log plot of RQI versus z, graphical clustering methods.
all samples with similar FZI values will lie on a Graphical clustering methods of histogram
straight line with a slope of one, and data samples analysis and probability plots provide a general
4 FAHAD A. AL-AJMI, STEPHEN A. HOLDITCH SPE 63254
visual image of FZI distribution to determine the The third method is hierarchical cluster
number of HFUs and their mean FZI values. The analysis (e.g. Wards algorithm). In this method, the
clustering of FZI is performed on biases of the distances between data points (FZI values) are
logarithm of FZI because FZI values usually exhibit calculated, initially treating each sample data as a
log-normal distribution. This is attributed to the cluster. Next, the two clusters that are closest in
strong dependency of FZI on permeability. distance are merged and the distances of new
Since FZI distribution is a superposition of clusters from other clusters are calculated. The
multiple log-normal distributions, a histogram of process of distance calculation and merging of
log FZI should show n number of normal clusters is continued until the required number of
distributions for n number of HFUs. When clusters clusters is attained. The number of clusters is an
are distinctly separate, the histogram clearly input into the hierarchical cluster analysis. Usually,
delineates each HFU and provides its corresponding graphical methods such as the probability plot, can
mean FZI value. However, it is often difficult to be used to determine an appropriate number of
separate the overlapped individual distributions HFUs for a data set.
from a histogram plot. Therefore, the histogram Analytical hierarchical cluster analysis
method is not suitable for most field applications introduces bias to the process of HFU classification
because the transition zones between the different because the number of clusters should be known
HFUs often cloud the judgment on their identity.2 prior to clustering.
Fig. 3 shows a histogram of the log of FZI
for CTR-17. It is clearly difficult to identify or
determine the number of HFUs that exist in this Iterative Multi-Linear Regression Clustering
well from the histogram plot alone because of the Both the graphical and the analytical
superposition nature of the histogram plot. clustering methods will introduce bias to the
The probability plot (the cumulative distribution determination of HFUs. In this resaerch, we have
function) is the integral of the probability density developed a new clustering technique that is
function (histogram). The probability plot is unbiased and easy to apply. Moreover, this new
smoother than the histogram and hence the scatter clustering technique can be used to determine the
in the data is reduced in this plot and the optimal number of HFUs that exist in the reservoir.
identification of clusters becomes easier. A normal The iterative, multi-linear, regression
distribution forms a distinct straight line on a (IMLR) clustering technique is based on deciding
probability plot. Therefore, the number of straight the optimal center for each data cloud and isolating
lines in the probability plot may be used to indicate that set of data with a cluster identification tag. The
the number of hydraulic flow units in the reservoir. basic concept of this method is that a log-log plot of
Because mean FZI values cannot be calculated from RQI versus z will produce a straight line with a
probability plot, the representative FZI value of unit slope and an intercept that represent the mean
each HFU is obtained by averaging all the FZI FZI value for each HFU. Samples of different pore
values within the corresponding HFU limit. The throat attributes will lie on a parallel straight line.
superposition effect may shift or distort the straight This sounds like a very simple problem to solve.
lines in the probability plot. Fig. 4 shows a However, in this case there are two unknowns and
probability plot of the logarithm of FZI for Well single equation (Eq. 11). The two unknowns are the
CTR-17. intercept of the straight line, HFU mean FZI value,
Graphical clustering methods may carry and the second unknown is how the data samples
some biased errors because they are based on visual are distributed among the different HFUs. The only
interpretation. Moreover, the overlapped individual way to solve this problem is to guess the first
distributions and the transition zones between unknown and solve for the second, then recalculate
HFUs may cloud the judgment on their identity. the first unknown and repeat the process till we get
both unknowns within acceptable tolerance. This
SPE 63254 PERMEABILITY ESTIMATION USING HYDRAULIC FLOW UNITS IN A CENTRAL ARABIA RESERVOIR 5
problem is solved iterativly to optimize the Adding more HFUs should cause a reduction in the
placement of each straight line on the log-log plot SSE. However, as we continue to add HFUs, we
of RQI versus z that gives a minimum error sum of will reach to a point where error reduction is getting
squares (SSE). smaller and smaller. The SSE in this case can be
used as a criterion for determining when we have
The HFU classification process using core data enough HFUs to describe the data. The SSE will
requires the following steps: eventually get smaller if we fit enough straight
1- Compute the reservoir quality index (RQI) and lines. However, as we add more straight lines
normalized porosity (z) from core information (more HFUs), we soon reach the law of
using Eqs. 8 and 9. diminishing returns. In other words, it isn't worth
2- Plot RQI versus z in logarithmic space. adding any further straight lines because the change
3- Use a reasonable initial guess of the intercept of in SSE is negligible. A plot of error sum squares
each straight-line equation, the mean FZI value versus the number of HFUs is shown in Fig. 7. The
of each HFU. SSE is plotted on the vertical axis and the number
4- Assign core sample data to the nearest straight of HFUs is recorded on the horizontal axis. Fig. 7
line. clearly shows a declining SSE curve with adding
5- Recalculate the intercept of each HFU using more HFUs until we get to about the number 5
least-squares regression equations. where the SSE curve will be flat. This indicates that
6- Compare the new and old values of the intercept five HFUs is a fair representation of the number of
for every straight line. If the difference is not HFUs in the formation.
within the acceptable tolerance, update the
intercept values and go to Step 4. IMLR Unbiased Clustering
After determining the optimal HFUs that
The procedure above should be repeated should used to provide the best mapping of the
until the optimal location of each straight line is different formation HFUs, the IMLR method will
found in which the error sum of squares is a provide unbiased clustering of the sample data. The
minimum for the desired number of hydraulic flow IMLR method produces unbiased clustering
units. because the calculated parameters are based on the
least-square equations, which provide unbiased
Determination of Optimal Number of HFUs estimation of regression parameters.
The IMLR method can be used to provide On the basis of the optimal HFUs from the
the optimal number of hydraulic flow units that IMLR clustering technique, a combined RQI versus
exist in the formation. The conventional porosity z graph is made for all the wells in the study area,
permeability transform is based on the assumption as shown in Fig. 8. The unit slope lines are drawn
that there exists only one rock type and all the through segments of data according to mean FZI
sample data exhibit the same flow characteristics. values calculated for each group of data that belong
This is the same if we use only one hydraulic flow to the same HFU. The permeability porosity
unit in our analysis. Fig. 5 is a plot of reservoir relationship constructed from the HFUs is
quality index versus normalized porosity where one illustrated in Fig. 9.
HFU is used to fit the whole data set. Fig 6 shows a
plot of permeability versus porosity assuming single Porosity Permeability Relationship
HFU. The error sum of squares (SSE) in this case is With the classification of existing HFUs
1963. This plot clearly indicates that using the above, we need to represent the formation porosity-
traditional permeability porosity transform will permeability relationship in a single formula instead
produce inaccurate permeability estimation in the of having n number of equation for n number of
CTR field. The resulting coefficient of HFUs. This is done by using the different HFUs as
determination, R2, is 0.26. indictor variables in the following formula:
6 FAHAD A. AL-AJMI, STEPHEN A. HOLDITCH SPE 63254
permeability porosity relationship constructed for dependent and independent variable transformations
the different HFUs is illustrated in Fig 15. The that improve correlation.
selected five HFUs are deemed sufficient to The optimal transformation for FZI is shown
minimize the scatter around the HFU lines for the in Fig. 17. The irregular shapes of the estimated
whole range of permeability. A comparison transformation further reveal the complicated
between CTR-6 HFUs and grain size profiles relationship between core FZI and well logs.
plotted against depth is shown in Fig. 16. The
dashed line represents the grain size profile derived A linear regression in the transformed space results
from visual inspection of cores. The area curve in
represents the HFUs profile for Well CTR-6
derived using the IMLR clustering technique.
Considering the differences in methods used to * ( FZI) = 09969
. [
1 (ln_ GR) + 2 (ln_ Rd / s) + 3 (e) + 4 (Zone)
* * * *
]
derive these two classification systems, the , .(13)
correlation is quite satisfactory.
with R2 = 0.72, which indicates a good level of
HFUs Prediction Using Log Data correlation.
The next major task is to extend the concept Finally, FZI is predicted from the well log
of hydraulic flow units in wells where only well log data using the following equation derived by ACE:
measurements are available. Amaefule et. al.3
derived a fundamental relationship between
porosity and permeability based on FZI as follows:
FZI pre
j [ ( ) ( ) ( )
= * 1 1 ln _ GR j + 2 ln _ R d / s j + 3 ej + 4 ( Zone j )
* * * *
]
3
,......(14)
k = 1014 FZI 2
(1 ) 2
correlation coefficient of 0.64. Data points that 5- The statistically derived hydraulic flow units
belong to HFUs 1 and 2 show a lot of scatter. The were compared with the core description made
flow units 1 and 2 are considered as having poor at the well site by the well-site geologist. The
flow characteristics. grain size classes from core description match
The value of correlation coefficient may be affected very well with the statistically derived clusters
by a few aberrant pairs. A good alignment of a few from the HFU method. This match indicates that
extreme pairs can dramatically improve an these hydraulic flow units reflect the different
otherwise poor correlation coefficient. Conversely, rock types in the analyzed formation. Moreover,
an otherwise good correlation could be ruined by it shows the practical efficiency of the
the poor alignment of a few extreme pairs. procedure used to determine the number of
Moreover, it is important to note that correlation hydraulic flow units in the formation.
coefficient provides a measure of the linear 6- A computer program was developed to perform
relationship between two variables. If the hydraulic flow unit analysis. Three main
relationship is not linear, the correlation coefficient processing options were integrated: sensitivity
may be very poor summary statistics. run to determine the optimal number of HFUs;
HFU analysis based on the optimal number of
Conclusions HFUs or any user-defined number of HFUs;
regression analysis using the different HFUs as
1- Reservoir porosity permeability relationship is dummy variables to calculate the HFU methods
best achieved if rocks with similar fluid-flow predicted permeability.
conductivity are identified and grouped
together. Each group is referred to as a Acknowledgments
hydraulic flow unit. The first author gratefully acknowledges the support
2- Permeability estimation using the HFU of Saudi Aramco and thanks Saudi Aramco for
technique was improved by developing a new permission to publish this paper.
unbiased clustering algorithm to cluster the
different rock types on a basis of their flow Nomenclature
characteristics. Unlike the existing analytical = regression coefficient
and graphical cluster analysis, which introduces
bias to the HFUs determination, our IMLR = tortuosity
clustering algorithm introduces no bias. = viscosity
3- A procedure for determining the optimal
number of clusters that should be used in the = error tolerance
HFU technique was introduced. In this
core = core porosity
procedure, the error sum of squares was used as
criterion for determining the required number of t = sonic travel time
HFUs to describe the reservoir.
4- Permeability estimation using the HFU method ACE = alternating conditional expectation
was extended to uncored wells by implementing CNL = compensated neutron log
the alternating conditional expectation (ACE) 2
algorithm. ACE provides a data-driven approach e = ACE regression error
for identifying the functional forms for the well f = function
log variables involved in the correlation.
Permeability profiles generated by HFUs using Fs2 = Kozeny constant
well log data agree with core data, which FZI = flow zone indicator
illustrate the potential and applicability of the
HFU method. GR = gamma ray
SPE 63254 PERMEABILITY ESTIMATION USING HYDRAULIC FLOW UNITS IN A CENTRAL ARABIA RESERVOIR 9
20
10000
10
1000
Permeability, md
0
100 -.88 -.38 .13 .63 1.13 1.63 2.13
-.63 -.13 .38 .88 1.38 1.88
10
LFZI
1
Fig. 3- Histogram of the logarithm of FZI for Well CTR-17.
0.1
0.01
0 5 10 15 20 25 30 35
Porosity, %
10000
1000
Permeability, md
100
0.1
10
0.01
0 5 10 15 20 25 30 35
Porosity, %
1
Fig. 2- Scatterplot of core permeability vs. core porosity for
Well CTR-2.
RQI
0.1
0.01
0.01 0.1 1
Z
100 00 1 00 00
100 0 1 00 0
C ore Pe rm e ability , m d
1 00
P e rme ability , md
100
10
10
1
1 K = 0 .0 1 0 8
2 .6 1 4 9
2
R = 0 .2 5 6 1 0 .1
0.1
0 .0 1
0.01 0 10 20 30 40
0 5 10 15 20 25 30 35 Porosity , %
P orosity, % Fig. 9- Porosity permeability relationship derived by HFUs
Fig. 6- Fitting core permeability and porosity using a single method.
HFU.
100000
2400
10000
T o ta l E rro r S um o f S qua re s (S S
Kcal = Kcore
2
Calculated permeability
2000 R = 0.9275
1000
1600
100
1200
10
800
1
400
0.1
0
0.01
0 2 4 6 8 10
0.01 0.1 1 10 100 1000 10000 100000
N umbe r o f H FU 's Measured core permeability
Fig. 7- Total error sum of squares versus number of HFUs. Fig. 10- Crossplot of estimated permeability using HFUs
method versus core permeability.
10 10000
0 .3 7 5 1
K c a l = 3 .9 9 0 4 K c o r e
2
1000 R = 0 .3 7 5 1
C a lc ula te d p e rme a b ility
1 100
RQI
10
0 .1
1
0 .1
0 .0 1
0 .0 1
0 .0 1 0 .1 1
0 .0 1 0 .1 1 10 100 1000 10000
z
M e a s ure d c o re p e rme a b ility
Fig. 8- Clustering core data into the optimal number of HFUs
using the IMLR technique. Fig. 11- Estimated permeability using traditional transform
versus core permeability.
12 FAHAD A. AL-AJMI, STEPHEN A. HOLDITCH SPE 63254
y = 1 .0 6 3 1 x
2
N u m b er of H FU s SS E
R = 0 .9 7 6 2
100 1 1 9 6 3 .0
2 6 0 1 .6
10
3 3 1 8 .1
1 4 2 2 8 .6
5 1 5 0 .3
0 .1 6 1 2 5 .3
7 1 0 4 .3
0 .0 1
0 .0 1 0 .1 1 10 100 1000 10000
8 9 2 .3
M e a s ur e d P e r m e a b ility, md .
9 8 6 .7
10000
Table 2- Sample Output for the HFU Analysis
1000 y = 1 .1 1 8 8 x
0.9563 DEPTH POROSITY PERMEABILITY CLUSTER z RQI
2
P r e d i c te d P e r m e a b i l ity
R = 0 .9 5 6 3 ft. % md #
100 735.00 12.70 1.600 2 0.145 0.111
735.50 9.40 0.300 1 0.104 0.056
10
736.00 10.30 0.100 1 0.115 0.031
736.50 9.30 0.200 1 0.103 0.046
737.00 8.10 0.300 2 0.088 0.060
1 737.50 8.90 0.200 1 0.098 0.047
. . . . . .
0 .1 . . . . . .
. . . . . .
0 .0 1 . . . . . .
0 .0 1 0 .1 1 10 100 1000 10000 900.00 9.60 0.300 1 0.106 0.056
900.50 9.70 0.500 2 0.107 0.071
M e a s ur e d P e r m e a b il i ty, m d .
901.00 11.40 0.300 1 0.129 0.051
Fig. 13- Calculated permeability using the Wards clustering
algorithm for Well CTR-2.
100
Execute
Execute
10
Clear
1
R eg res s ion An alys is
N um ber of H FU 's
Execute 5 0.1
Clear 0.01
Data File:
c:\f32\hfuxls\core6.dat 0 5 10 15 20 25 30 35
Porosity, %
6 6 1 00 00 0
1 00 0
H F U P ermeab ility
4 4
1 00
3 3 10
HU1
1 HU2
2 2 HU3
0 .1 HU4
HU5
1 1
0 .0 1
0 .0 1 0 .1 1 10 1 00 1 00 0 1 00 00 1 00 00 0
Fig. 16- Relationship between statistically derived HFUs and
grain size classes from core description for Well CTR-6 [(1) C o re P ermeab ility
medium-coarse, (2) fine-coarse, (3) fine-medium, (4) fine- Fig. 21- Scatterplot of permeability estimated using HFUs
very fine, and (5) siltstone]. method vs. core permeability.
G R & C a lip er P H IE H F U P e rm e a b i l i t y , m d
3 .5 650
3 .0
2 .5
700
2 .0
* ( ln_FZI )
1 .5
1 .0 750
0 .5
0 .0
800
-0 .5
-1 .0
-1 .5 850
-3 .0 -2 .0 -1 .0 0 .0 1 .0 2 .0 3 .0 4 .0
_ln_FZI
900
1000
4 .0
1050
C a lc ula te d ln_ F Z
2 .0
HF U P erm .
P H IE
Core P erm .
1100
0 .0
-4 .0
-4 .0 -2 .0 0 .0 2 .0 4 .0
M e a s ure d ln_ F ZI
Fig. 18- FZI values predicted from well logs using ACE
versus core FZI.
14 FAHAD A. AL-AJMI, STEPHEN A. HOLDITCH SPE 63254
750
800
850
900
950
1000
HF U P erm .
P H IE
Core P erm .
1050