Professional Documents
Culture Documents
SPE 30978 State-Of-The-Art in Permeability Determination From Well Log Data: Part 1-A Comparative Study, Model Development
SPE 30978 State-Of-The-Art in Permeability Determination From Well Log Data: Part 1-A Comparative Study, Model Development
S -surface area per unit bulk volume if a=1. Equation 9 describe a line with negative slope -1/m and the
Sp-surface area per unit volume of pore space intercept (1/m)LogRw. The line estimates of cementation factor and
So-surface area per unit volume of solid material. formation-water resistivity wil then be used in the equation 5 for
calculation of formation factor and then water saturation in
The porosity function φ3/(1−φ)2 is that measure of rock texture producive zones.
which relates permeability to average grain diameter. These
formulations are valid for packs of uniformly sized spheres. T ixier 2 , 1 9 4 9 . Using empirical relationships between resistivity and
Another major drawback is that surface area can be determined water saturation, water saturation and capillary pressure, and
only by core analysis, and only with special equipment. capillary pressure and permeability, Tixier established a method for
determining permeability from resistivity gradients.
A r c h i e , 1 9 4 1 . In his classical paper1, though he did not provide a 2.3 2
K C(a ) (10)
permeability formula, Archie set the basis for quantitative log w o
interpretation. Analyzing the laboratory-determined resistivities of
a large number of brine-saturated cores from various sand forma- R 1
a (11)
tions, Archie introduced the concept of “formation resistivity D Ro
factor” in the form:
where:
Ro
F (4) C is a constant, normally about 20,
Rw ∆R is the change in resistivity (ohm-m),∆D is the change in depth
(ft), corresponding to ∆R
where Ro is the 100% water-saturated formation resistivity (ohm- ρw is formation water density (g/cm3),
m), and Rw is the resistivity of the brine (ohm-m). He established ρo is hydrocarbon density (g/cm3).
that the formation factor, F, is a function of the type and character Equation 10 and 11 can be rewritten as:
of the formation, and varies, among other properties, with the
porosity and permeability of the reservoir rock: 1
K 2 2.3 R
a ( ) (12)
F (5) 20 Ro(dw do) D
m
where a is the formation factor coefficient, and m, the “cementation The resistivity gradient is determined from a deep investigation
factor,” is the slope of the line representing the correlation under tool, lateral or focused logs, and corrected for borehole effects.
discussion (1.8 < m < 2.0 for consolidated sands). For water This method assumes that saturation exponent, n, is equal to 2.0,
saturations down to 0.15 or 0.20, the following approximate and that at any water saturation, capillary pressure is related to
equation has been found: permeability in the manner: Pc = f/(K)1/2. The model is physically
FRw limited in scope by the relative paucity of logs exhibiting valid oil-
Sw n (6) water contacts and the necessity for estimating the hydrocarbon
Rt
density as it exists in the reservoir. Also, the calculated permeabil-
where Rt is the resistivity of the formation, and n is the saturation ity is an average for the zone corresponding to the resistivity
exponent (n ≅ 2). The cementation factor, m, and the saturation gradient.
exponent, n, are the biggest sources of uncertainty in permeability
determination3. They can be obtained by laboratory measurements, Following the work of Wyllie and Rose3, Tixier developed a
which is seldom the case, or approximated according to some simpler model that is used more often than equation 12:
general guidelines and/or experience. Methods for deducting the
1 3
cementation factor have had a long history. We wil present here 2
K 250 (13)
only one method based on the establishment of a “water line” in a Swi
zone that is 100% water saturated15. In this case, equation 6
becomes: In this study we have used this equation as Tixier model.
Rt a
Rw m (7) W yllie & R o s e 3 , 1 9 5 0 . In their thorough analysis of the theoretical
basis of quantitative log interpretation, Wyllie & Rose expanded
which can be liniarized to: the empirical relationship proposed by Tixier, based on the
LogRt Log(aRw) mLog (8) following assumptions:
-irreducible water saturation Swi is a straight line function of grain that can be evaluated in terms of the statistically determined
surface area. parameters A, B, and C. He applied a reduced major axis (RMA)
-minimum water saturation computed in a reservoir is equal to the method of analysis to data obtained by laboratory measurements
irreducible water saturation Swi. conducted on 155 sandstone samples from three different oil fields
from North America. Based both on the highest correlation
Their model is: coefficient and on the lowest standard deviation, Timur has chosen
1 from five alternative relationships the following formula for
K Constant[ ]
(2
1
) (14) permeability.
Pc2F m
Sw 4.4
K 0.136 (19)
where the constant is equal to ( 21.2)d2/ts, for K in millidarcies, Swi2
capillary pressure, Pc, in psi, and interfacial tension, d, in dynes/cm.
The pore shape factor, ts, lies between 2.0 and 2.5 and may be and for residual water saturation:
taken 2.25. When Pc is unobtainable (i.e., in the absence of an oil-
1.26
water contact in the reservoir), they suggested the following Swi 3.5 1 (20)
correlation: K 0.35
1
Swi C[ ] C (15) with a standard error of 13% pore volume.
K 1/2F 0.67
where: This model is applicable where condition of residual water
C-textural constant with dimension of a length (which is of the saturation exists. Timur also assumed that a value of 1.5 for the
form φ3/(1−φ)2 ). cementation factor, m, holds in all cases.
C’-dimensionless constant which is related to percent of bound
water due to clay content of the reservoir rock. For clean sand C o a tes & D u m a n o i r 6 , 1 9 7 4 . An improved empirical permeability
C’=0. technique has been proposed by Coates & Dumanoir:
2w
C
K 1/2 (21)
Wyllie & Rose pointed out that “the value of K deduced from these w 4 Rw/Rti
equations cannot, however, be expected to have more than an order
of magnitude significance.” where:
850,000 Rw2 With the support of core and log studies, they adopted a common
K ( 3.5depth,ft) (17)
APIGravity FRoRti exponent, w, for both the saturation exponent, n, and cementation
exponent, m.
where K is obtained directly in darcies. This empirical relationship
was determined by multiple correlation from relatively few data. m n w (24)
The formula should not be used for high gravity crudes (API > 40o)
and for depths greater than 6500 ft. Equations 21, 22, and 23 are valid for clean, oil-bearing forma-
tions, with oil density equal to 0.8. When the hydrocarbon has a
T im u r 5 , 1 9 6 8 . Based on the work of Kozeny and Wyllie & Rose, density appreciably different from 0.8, the log readings of Rti are
Timur proposed a generalized equation in the form: multiplied, before entering equation 23, by the correction factor
given by:
B
K A (18) Rtcorr 2
Swi C 0.077 1.55 h
0.627 h (25)
Rtlog
4 BALAN, B., MOHAGHEGH, S., AMERI, S. SPE 30978
Coates & Dumanoir also presented a methodology for testing if the where Y is the dependent variable, X1,X2,...,Xp are the independent
formation is at irreducible water saturation. However, they note random variables and e is a random error (or residual) which is the
that if the reservoir is heterogeneous, it may fail that test and still amount of variation in Y not accounted for by the linear relation-
be at irreducible water saturation. If a formation is not at irreduc- ship. The parameters B1,B2,...,Bp. called regression coefficients, are
ible water saturation the assumed value of Rt from the log is less unknown and are to be estimated.
than Rti , and the resulting w from equation 23 is in error. To
overcome this, they classified the formation rocks in three litho- Taking the expectation of both sides of equation 29, we have:
logical classes (table 1 of their paper). An empirical relation was E(Y|X1,...,Xp) B0 B1X1 B2X2 ... BpXp (30)
then established between the clean matrix density (ρgcn), rock
classification, and Ri: where the expected value of the errors is zero. In this representa-
tion E(Y|X1,...,Xp) is the conditional mean, or expected value of Y,
given X1,X2,...,Xp. We can write the assumed relationship between
6
w 10 Y and X1,X2,...,Xp as:
(Rw/Rt)cn ( Swirr) (26)
cn G( gcn
2.6)3 Y E(Y|X1,...,Xp) e (31)
where by Y on the left-hand side of equation 31 is meant Y given
where G is a coefficient involving the matrix classification. The X 1,X2,...,Xp.
corrected value of (Rw/Rt)cn derived from equation 26 for a clean
formation is used in place of Rw/Rti in equation 23 to determine w The aim here is to estimate regression coefficients and thus E
when the formation is not at irreducible water saturation. (Y|X1,...,XP) or Y in terms of the N observations. There are many
accessible software packages that can solve this problem. In this
Another correction that Coates & Dumanoir provided is for shaly study, the dependent variable Y is the logarithm of permeability -
formations: since permeability seems to be lognormal- and the independent
variables X 1,...,Xp are well log variables. In their paper4, Wendt
and Sakurai established a general procedure for permeability
Rw RwVsh ( Swirr) w
( Swirr) cn prediction by multiple variable regression. They also pointed out
(27)
Rti Rsh cn (1 Vsh)w 1 the shortcomings of using this technique. When the regression
method is used for prediction, the distribution of predicted values
is more narrow than that of the original data set. Kendall and
This value of Rw/Rti is entered in equation 23 to obtain w. Stuart15 explained this, stating that the regression model “does not
purport to represent a functional relation between mathematical
The method of Coates & Dumanoir is the first that satisfies the variables or a structural relation between random variables; it
condition of zero permeability at zero porosity and when Swirr = either exhibits a property of a bivariate distribution or, when the
100%. Because of the corrections provided, this method can be regressor variables are not subject to error, gives the relation
applied to formations that are not at irreducible water saturation, between the mean of the dependent variable and the value of the
and to shaly formations. Values for the exponents m and n are not regressor variables.” That is, the regression provides the best
needed because they are found as a result of the computation. estimate on the average. The ability of a regression model to
predict the permeability extremes is enhanced through a weighting
C o a tes 7 , 1 9 8 1 . Coates and Denoo proposed the following formula scheme of the high and low values. But because of this, the
for permeability determination: predictor can become unstable and also statistically biased. The
2
(1 Swirr) assumption that the error is related only to the dependent variable
K 1/2 100 (28) (permeability measurements) and not to the independent variables
Swirr
(log variables), can be verified by comparing repeat runs of
where K is in milidarcies. properly calibrated instruments with the main runs of the logs,
provided that there is no bias in the measurement. Logs of accept-
This formula also satisfies the condition of zero permeability at able quality have errors with a relatively small unbiased scatter that
zero porosity and when Swirr = 100%. The formation must be at is a function of the physics of the tool, its response characteristics,
irreducible water saturation. and the borehole environment. If the deviations are indeed random,
then they would be expected to be normally distributed with a mean
value of zero.
2.Multiple Variable Regression
Multiple regression is an extension of the regression analysis that In the analysis of real logging data several initial remedial steps
incorporates additional independent variables in the predictive should be taken in order to maximize the validity and value of the
equation. Here, the model to be fitted is: analysis results15:
-the data should be environmentally corrected for systematic
Y B0 B1X1 B2X2 ... BpXp e (29) borehole effects;
SPE 30978 STATE-OF-THE-ART IN PERMEABILITY DETERMINATION FROM WELL LOG DATA: PART 1- A 5
COMPARATIVE STUDY, MODEL DEVELOPMENT
-the logs should be shifted wherever necessary to ensure common Virginia. Figure 1 shows the Granny Creek Field and the location
depth registration; of the eight wells (1107, 1108, 1109, 1110, 1126, 1128, 1130 and
-the logs should have a common vertical resolution, a compatibility 1134) that have been used in this study. This field is a well
requirement that may involve smoothing of finer resolution documented21-23, highly heterogeneous reservoir that has produced
measurement; from the Big Injun sandstone since early the 1900's. Only Gamma
-preferably, the logs should be zoned, with data sampled from peak Ray, Deep Induction, and Density logs were available for all the
and through extremes, to reduce extraneous errors introduced by eight cored wells. Because of space considerations, we will present
transitional curve features. the data and the results for only one well, 1107. Data and results
for the other seven wells are available upon request. Figure 2
After these steps are completed, the correlation matrix of all shows the log and core data for this well. All the well logs are
independent and dependent variables should be analyzed, to compatible in terms of depth and resolution and are corrected for
establish if there is a dominant X-variable, or if the X’s are different effects.
essentially uncorrelated with each other. This gives to the analyst
some guidelines for selecting the variables and the order in which Empirical models
they should enter the model. However, sensible judgment is still The four latest methods, Tixier, Timur, Coates & Dumanoir, and
required in the initial selection of variables, and also in the critical Coates, were applied to log data from the eight chosen wells.
examination of the model through analysis of residuals. Before that, several steps were performed:
3. Virtual Measurement Technique 1. Determination of porosity from the density log. Figure 3 shows
There are many examples of neural network applications in the a good agreement between log-determined and core-determined
petroleum industry16, from exploration, reservoir and production porosity for this well.
engineering, drilling operations, to neural network-based models
of chemical plants. These systems are in various stages of develop- 2. Estimation of formation factor. Knowing that permeability is
ment, from prototypes to production systems. very sensitive to the cementation factor, m, we have tried to find a
consistent value for this parameter. This can be done by establish-
Neural networks, unlike conventional programs, are general ing a “water line” in a 100% water saturated zone (Figure 4b). The
purpose systems that attempt to achieve good performance by slope of this line is 0.504247, which yields m = 1.98316. The
dense interconnection of simple computational elements. For this resulting formation factor is plotted against porosity in Figure 4c.
reason they are also called connectionist models. The solution to a
problem is not explicitly encoded in the program, but is “learned” 3. Determination of water saturation. We assumed a value of n = 2
by supplying examples of previously solved problems to the for the saturation exponent. Cross plot of porosity against water
network. After the network has learned how to solve the example saturation for the well 1107 is presented in Figure 4d.
problems, it is said to be “trained.” This is called “supervised
training.” New data from the same knowledge domain can then be Having done these steps, now we can determine permeability as a
input to the trained neural network that then outputs a solution. function of computed porosity and water saturation. Figure 5
There are also neural networks that “learn” unsupervised, like presents the computed permeability along with core-determined
Kohonen’s self-organizing map network. permeability versus depth. Figure 6 shows the cross plots of log
and core permeability for each method. From these figures it is
Virtual measurement technique was applied successfully in clear that permeability is overestimated by all empirical models.
determination of permeability from well log data17-20. In this study The best method seems to be Coates & Dumanoir (equation 17).
we are applying this technique to a real data set and compare the This model provides a correction to account for formations that are
results with those obtained by means of previously discussed not at irreducible water saturation. The presence of the G factor
methods. (which is related to lithological classification) in the denominator
of equation 26 has a great influence on the values of permeability.
The major advantage of neural network solution for this problem This factor can subjectively be adjusted by the analyst. Thus, the
is that it does not need all the parameters and the relationship bigger the G, the smaller the permeability.
between these parameters to be explicitly specified. Since the
neural networks “learn” to solve problems through examples, they The fact that permeability is log normal does not mean that if we
are especially suited for subjective and interpretative processes that obtain a good match in log scale we have achieved our objective.
humans can easily perform intuitively, but which we cannot Figure 7 is an example. Looking at the graph with log scale one can
describe in terms of an algorithm or set of equations. be quite enthusiastic. Unfortunatly, the distance D on this graph
turns to be inacceptable (more than 70 md) when we look at the
Application to Heterogeneous Formation graph with normal scale. Therefore, logarithmic representation of
To apply all three previously discussed methods we have chosen the results (as it is done in many papers) can be deceiving.
core and log data from eight wells in Granny Creek Field in West
6 BALAN, B., MOHAGHEGH, S., AMERI, S. SPE 30978
8. Kapadia, S.P. and Menzie, U.: “Determination of Permeability Technical Conference, New Orleans, LA, (September 1994).
Variation Factor V From Log Analysis,” SPE 14402, Annual
Technical Conference, Las Vegas, NV, (September 1985). 21. Aminian, K.: “Characterization of a Complex Resrvoir in West
Virginia,” SPE 26939, Eastern Regional Conference, Pittsburgh,
9. Bloch, S.: “Empirical Prediction of Porosity and Permeability in PA, (November 1993).
Sandstones,” AAPG Bulletin, Vol. 75, No. 7, (July 1991), pp.
1145. 22. Molnar, D., Aminian, K., Ameri, S.: “The Use of Well Log
Data for Permeability Estimation in a Heterogeneous Reservoir,”
10. Ahmed, U., Crary, S.F., Coates, G.R.: “Permeability Estima- SPE 29175, Eastern Regional Conference, Charleston, WV,
tion: The Various Sources and Their Interrelationships,” JPT, (May (November 1994)
1991), pp. 578.
23. Donaldson, A., et al.: “The Fluvial-Deltaic Big Injun Sandstone
11. Yao, C.Y. and Holditch, S.A.: “Estimating Permeability in West Virginia,” Final Report, DOE/BC/14657-15, Bartlesville
Profiles Using Core and Log Data,” SPE 26921, Eastern Regional Project Office, USDOE, 1992.
Conference, Pittsburgh, PA, (November 1993).
19. Mohaghegh, S., Reza, A., Ameri, S., Rose, D.: “Design and
Development of An Artificial Neural Network for Estimation of
Formation Permeability,” SPE 28237, SPE Petroleum Computer
Conference, Dallas, Texas,(August 1994).
Table 3.
Arithmatic Geometric
Mean Mean
Figure 2. Log and core data for well #1107 Figure 3. Log and core porosity for well #1107
SPE 30978 STATE-OF-THE-ART IN PERMEABILITY DETERMINATION FROM WELL LOG DATA: PART 1- A 9
COMPARATIVE STUDY, MODEL DEVELOPMENT
Figure 4. Characteristics of well #1107 Figure 5. Core and computed permeability for well #1107
Figure 6. Computed Permeability versus core permeability for well Figure 7. Logarithmic and normal scales for permeability represen-
#1107. tation.
10 BALAN, B., MOHAGHEGH, S., AMERI, S. SPE 30978
Figure 8. Core and computed permeability for well #1107. Figure 10. Core and computed permeability for well #1107.
Figure 9. Core permeability versus computed permeability for well Figure 11. Core permeability versus computed permeability for
#1107. well #1107.