SPE

International

Society of Petroleum Engineers

SPE 105989
Prediction of Directional Grain Size Distribution: An Integrated Approach
Gbenga Oluyemi, SPE; Babs Oyeneyin, SPE; & Chris Macleod; The Robert Gordon University
Copyright 2006, Society of Petroleum Engineers Inc.
This paper was prepared for presentation at the 30th Annual SPE International
Technical Conference and Exhibition in Abuja, Nigeria, July31st - August 2nd, 2006.
This paper was selected for presentation by an SPE Program 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 Society of Petroleum
Engineers and are subject to correction by the author(s). The material, as
presented, does not necessarily reflect any position of the Society of Petroleum
Engineers, its officers, or members. Papers presented at SPE meetings are subject
to publication review by Editorial Committees of the Society of Petroleum Engineers.
Electronic reproduction, distribution, or storage of any part of this paper for
commercial purposes without the written consent of the Society of Petroleum
Engineers 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 acknowledgement of where and by whom the paper was
presented. Write Librarian, SPE, P.O. Box 833836, Richardson, TX 75083-3836,
U.S.A., fax 01-972-952-9435.

Abstract
Recent advances in grain size distribution
modelling have tended to de-emphasize the
conventional method of carrying out grain size
distribution analysis. This is a direct result of the
difficulties
surrounding
core
acquisition,
preparation and analysis. Huge rig time spent
acquiring cores and additional time spent in the
laboratory is also a major discouraging factor.
In response to this trend, researchers have come
up with models capable of grain size distribution
analysis without any need to acquire cores and
carry out analysis on them. This work is a sort of
response to this new trend and a contribution to
the research drive in this core area.
A new technique for grain size distribution
estimation has been developed. The new
technique integrates both statistical and neural
network methods. Statistical methods were used
to develop sets of equations for estimating some
percentile grain sizes such as d1, d2, d16, d84, d98
and d99. Each equation relates one of these
percentile sizes with d50 and standard deviation
or sorting. Neural network method was used to
predict the median grain size (d50). Krumbein and
Monk and Bergs equations were transposed and
used to estimate the standard deviation.

The technique also has the functionality of

resolving grain size distribution into horizontal and
vertical components for sand failure analysis
The technique is simple and is dependent on only
petrophysical and textural information of porosity,
permeability, irreducible water saturation and pack
structure coefficient which are readily available
from well logs and/or empirical models.
Introduction
Grain size profiles have a wide application in the
petroleum industry. They are useful in the design
and implementation of gravel packing for sand
control; they have been used by geologists for
lithofacie analysis and paleoenvironmental
reconstruction; they can also be of immense
importance for delineating hydraulic flow units in
reservoir rocks as it is believed that sorting will
affect permeability which is a very important flow
property.
Grain size profiling is conventionally done by
performing sieve analysis on core sample and
then plotting the cumulative weight percent
passing through against the sieve sizes on either
an arithmetic or log probability scale. Grain size
distribution parameters such as mean or median
grain size, sorting, coefficient of sorting etc are
then deduced from the plots using various
formulae and relations. However, coring is an
expensive operation; cores are therefore not
always acquired. Even when they are acquired,
they are acquired from only certain sections of the
reservoir, making grain size distribution analysis
impossible in uncored reservoir sections.
Empirical models such as the ones by Slichter
(1899)1, Krumbein and Monk (1942)2 and Bergs
(1970)3 have also been used to estimate grain
size. Most of these models suggest a simple
relationship between grain size and a handful of
other textural and petrophysical parameters. It has

Gbenga Oluyemi, Babs Oyeneyin, & Chris Macleod

however been established that the relationship

between these parameters is far from being
simple but rather complex5. Also most of these
models are only able to output either mean or
median grain size except perhaps Krumbein and
Monk2 and Bergs3 equations which have an
element of sorting in them.
Another method of grain size estimation which has
been reported but is yet to gain wider acceptance
in the industry is the core image analysis4. This
entails taking continuous images along a slabbed
core, subjecting the images to textural feature
extraction and using the extracted features in
partial-least squares regression analysis with core
data to arrive at a prediction model. This method
also suffers from the difficulty surrounding core
acquisition.
Recently, Oyeneyin and Faga (1999)5 reportedly
used a variety of wireline logs as inputs into and
core derived grain size distribution as output from
a commercial feed-forward back-propagation
neural network to develop a model for continuous
grain size distribution prediction. However, the
commercial neural network used as the platform
for the model development is known to be very
slow and non-flexible; it also has fixed training
algorithms and its inner working is difficult to
understand. These seeming disadvantages may
affect predictions by any commercial neural
network. Also the presentation of seven percentile
sizes to the network as target was capable of
increasing the network size and hence its inner
complexity. This may make learning very slow and
prediction error margin very high.
Permeability Anisotropy
As stated in the previous section, a plethora of
models abounds that suggest an existence of
correlation between grain size and a number of
textural/petrophysical properties of rock. Some of
the most common of these properties are
permeability,
packing
coefficient,
porosity,
irreducible
water
saturation,
etc.
Except
permeability which has a directional attribute, all
other properties are scalar quantities with no
directional attributes.
The directional attributes of permeability is caused
by stratigraphic, lithologic, sedimentologic or
depositional nature of the reservoir rocks.
Permeability anisotropy, k , a dimensionless
ratio, is often used to express the directional
nature of permeability. It is mathematically
represented as

k =

kH
kv

SPE

where kH is horizontal permeability and kV is

vertical permeability.
Techniques for estimating permeability anisotropy
are numerous and well documented in the
7,
8,
9
.
literature6,
In this work, the concept of directional permeability
has been used to predict grain size distribution
within the reservoir in both horizontal and vertical
direction. In calculating the horizontal and vertical
permeabilities the models proposed by Klein et
al10 (eq. 2a and 2b) were used.

kV =

10 4 6
RH
Rw

2a

10 4 6
RV
2b
Rw
where is porosity, Rw is water resistivity, RH is
kH =

horizontal resistivity and RV is vertical resistivity.

The directional resistivities, horizontal and vertical
can be obtained from Baker Hughes 3DEX, the
multicomponent induction tool11, 12.
Median Grain Size Prediction Using Neural
Network
In order to remove the shortcomings of Oyeneyin
and Faga (1999)5 neural network model, a simple
two-layer feed-forward back-propagation neural
network was coded in C++ and presented with
log-derived petrophysical and textural parameters
of porosity, permeability, irreducible water
saturation and pack structure co-efficient as
inputs, and core-derived median grain size (d50) as
target output. The inputs were based on the
constitutive parameters of the modified analytical
model presented by Oyeneyin et al. (2005)13.
Figure 1 shows the architecture of the neural
model.
The network, after training and validation, was
presented with the log derived parameters without
any target output. The network was able to predict
d50 with low error margin as shown in figure 2 and
table 1.
The network was later presented with directional
permeability data (horizontal and vertical
permeability) to predict horizontal and vertical
median grain size as shown in figure 3.
Determination Of Other Percentile Sizes From
The Median Grain Size (D50)

Prediction of Directional Grain Size Distribution: An Integrated Approach

SPE
_____________________________________________________________________________________________________________

Median grain size is one of the characterising

parameters of sandstone formation in formation
evaluation and engineering analysis work for oil
and gas operations and grain size distribution
analysis.
Other
important
characterising
parameters are the other percentile sizes and the
standard deviation or sorting. There is therefore
need to develop a procedure for estimating the
values of these other percentiles and standard
deviation.
Derivation Of Equations Relating D50 And
Other Percentile Sizes With Standard Deviation
The equation describing a normally distributed
grain size is given as14:

integrals of relative frequency values for ranges of

values of t from transposed equation 4.

d d 50
t

For -1 < t > +1 (one standard deviation from the

mean), the integrals (as given in statistic tables) of
the relative frequency are approximately 16% and
84% respectively; equation (6) can then be written
as:

d 50 d 16 =

and

d 84 d 50 =

y=

(d d 50 )2
exp

2 2
(2 )

d is the grain size of interest, d50 is mean; y is

2
frequency, is variance and is standard
deviation.
In a normally distributed grain size, mean grain
size = d50 = median grain size.
Let t =

d d 50

Substituting (4) in (3) and integrating the result,

the resulting standard normal equation in which
d50 = 0 and = 1 is

dy =

t2
exp
dt

2
2
1

5
A plot of

dy
versus t gives the normal probability
dt

curve. A typical curve for -4 < t > 4 is shown in

figure 4.
Equations for calculating the major percentile
distribution of grain size can be obtained from the

Similar equations can be written for a number of

other percentiles as shown in table 2.
These equations can be used to estimate d1, d2,
d16, d84, d98, d99 of any normally distributed grain
size once the median or mean grain size and
standard deviation or sorting are known. Other
percentile size distribution can then be found by
interpolation and extrapolation when these values
are plotted on either an arithmetic or log
probability scale.
Equations
Deviation

For

Estimating

The

Standard

In the sets of equations developed for the

estimation of other percentile sizes, standard
deviation is a recurring unknown. It is therefore
necessary to establish a procedure for estimating
standard deviation.
Krumbein and Monk2 gave an equation relating
permeability to textural parameters of geometric
mean grain size and standard deviation, which is a
measure of sorting. This equation was developed
empirically using very well sorted sediment
samples ranging from -0.75 to 1.25phi in mean
grain size, and with standard deviations ranging
from 0.04 to 0.80phi. The equation is expressed
as
2 1.31

k = 760d g e

where k = permeability (d)

dg = geometric mean grain size (phi)
= standard deviation (phi)

Gbenga Oluyemi, Babs Oyeneyin, & Chris Macleod

Transposing equation (9), it becomes

1
k
log e
2
1.3
760 * d g

10

The geometric mean grain size used by Krumbein

and Monk2 is equal to the arithmetic median grain
size for normally distributed sand sizes15. They
can therefore be used interchangeably in equation
(10)
Similarly, Bergs3 gave an equation relating
permeability of sands to the median grain size,
porosity and a term he called phi percentile
deviation of sand grain size distribution, which
equals ( phi84.1 phi16.5 ).
The equation is expressed as

k = 5.1 *10 6 * 5.1 d 502 e

1.385

11

where k = permeability (md)

d50 = median grain size (mm)
= porosity (percent)
size distribution (phi84.1-phi16.5)
The equation has been reported to perform well
when used for 30% - 40% porosity sand16
Transposing equation (11), it becomes

1
k
log e
6
5.1
2
1.385
5.1 *10 * * d 50

results show some degree of agreement with the

measured data. The 31% mean error and the 0.66
correlation coefficient shown in table 3 and figure
5 respectively may be due to the twin factor of
extremely poor sorting of the sands whose data
was used for the analysis and the error inherent in
the data. The sorting of the sand ranges from
3.3phi to 8.6phi; this is outside the range krumbein
2
and Monk employed in his derivation experiment.
Bergs Equation
Data from Natal Group sandstone in South Africa
published by Bell and Lindsay17 has been used to
analyse the transposed Bergs3 equation and to
calculate standard deviation or sorting and various
percentile sizes. The results have been plotted
and shown in figures 8 and 9 and table 4. The
results (figure 8) show very poor correlation
coefficient (0.14) between predicted and
measured
standard
deviation.
The
poor
performance may have been as a result of the
porosity effect on transposed Bergs equation;
porosity is not known to have any effect on grain
sorting.
Directional Grain Size Distribution Prediction

= phi percentile deviation of sand grain

SPE

12

Analysis
Field data from Niger Delta and South Africa have
been used to analyse and compare both Krumbein
and Monk2 and Bergs3 equations respectively. The
results are presented and analysed in the
following sections
Krumbein And Monk Equation
Data from a Niger Delta field has been used to
analyse the Krumbein and Monk2 equation and to
calculate standard deviation and various
percentile sizes. The results have been plotted
and shown in figures 5, 6 and 7; and table 3. The

An excel-based tool has been packaged based on

all the models reported in this work for predicting
grain size distributions in both horizontal and
vertical directions. The models on which the tool is
based are
The neural network model for predicting
horizontal and vertical median grain size
based on horizontal and vertical
permeability
The two models for predicting standard
deviation or sorting
The models for percentile sizes
Figures 10 and 11 are plots of horizontal and
vertical grain size distribution from the tool.
The directional grain size distribution information
in sand producing reservoirs represents a valuable
tool for detecting sand source i.e. where the
produced sand is coming from (whether from near
wellbore cavity or within the reservoir sand matrix)
and the quantity and size distribution of sand
entering the wellbore. This will no doubt be very
useful in sand failure analysis and sand control
design.
Conclusion
This paper has attempted to put forward and
demonstrate the use of a new technique which
integrates both statistical and neural network
methods for the prediction of grain size

Prediction of Directional Grain Size Distribution: An Integrated Approach

SPE
_____________________________________________________________________________________________________________

distribution. The technique has a great potential to

perform better than the age-long empirical and
recent neural network models in making
continuous prediction of grain size distribution
especially if the neural network generalisation
capability is improved by use of large data sets for
training and validation.
The directional grain size distribution (horizontal
and vertical grain size distribution) which the
technique also predicts could be very useful in
detecting sand source in a reservoir producing
sand for failure analysis and sand control design.
Another conclusion that may be drawn from this
study is that transposed Krumbein and Monk
equation performs better than transposed Bergs
equation in calculating standard deviation or
sorting. This is due to the porosity effect on the
transposed Bergs equation.
References
1 Slichter, S. C. 1889, Theoretical Investigation of
the Motion of Groundwater. US Geol. Surv.,
NTIS, Springfield, VA, 19th Annual Report, pp 295384
2 Krumbien, W. C. and Monk, C. D. 1942,
Permeability as a Function of the Size
Parameters of Unconsolidated Sand. Trans.
AIME, Vol. 151, pp 153-162.
3 Berg, R. R., 1970, Method for Determining
Permeability from Reservoir Rock Properties
Trans. Gulf Coast Ass. Geol. Soc., Vol. XX,
pp303-317
4 Oyno, L. et al., 1998,: Prediction of
Petrophysical Parameters Based on Digital Video
Core Images SPE Reservoir Evaluation &
Engineering, Vol. 1, Feb. 1998.
5 Oyeneyin, M. B. and Faga, A. T., 1999,
Formation-Grain-Size Prediction Whilst Drilling: A
key Factor in Intelligent Sand Control
Completions, SPE 56626
6 Schon, J. H. 2003 et al., Imparting Directional
Dependence on Log-Derived Permeability, SPE
Reservoir Evaluation & Engineering 2003
7 You-Sheng, X & Xian-Zhi, X. 2001, A New
Numerical Method For the Analysis of the
Permeability Anisotropy Ratio, Chin. Phys. Soc.,
Vol 11, No 6, June 2002.
8 Lock, P. A. et al. 2001, Estimating the
Directional Permeability of Reservoir Sandstones

Using Image Analysis, Proceedings of the 6th

Nomadic Symposium on Petrophysics, 15-16 May
2001, Trondheim, Norway.
9 Angeles, R. et al., 2005, Efficient and Accurate
Estimation of Permeability and Permeability
Anisotropy from Straddle-Packer Formation Tester
Measurements Using the Physics of Two-Phase
Immiscible Flow and Invasion, SPE 95897,
presented at the 2005 SPE Technical Conference
and Exhibition held in Dallas, Texas, USA, 9-12
Oct., 2005.
10 Klein, J. D. et al., 1995 The Petrophysics of
Electrically Anisotropic Reservoirs, Paper HH
Presented at the 1995 SPWLA Annual Technical
Logging Symposium, Paris, June 26-29
11 Georgi, D. T. et al., 2000, Extrapolation of
Core Permeability Data with Wireline Logs to
Uncored Intervals, Paper KK presented at the
2000 SPWLA Annual Logging Symposium, Dallas,
June 4-7.
12 Kriegshauser, B. et al., 2000, A New
Multicomponent Induction Logging Tool to Resolve
Anisotropic Formations, Paper D presented at the
2000 SPWLA Annual Logging Symposium, Dallas,
June 4-7.
13 Oyeneyin, M. B., Macleod C., Oluyemi, G. &
Onukwu, A., 2005, Intelligent Sand Management,
SPE 98818, paper prepared for presentation at the
29th SPE International Conference and Exhibition
in Abuja, Nigeria, Aug. 1-3 2005.
14 Washington, C., 1992, Particle Size Analysis in
Pharmaceutics and Other Industries Theory and
Practice, Ellis Horwood Ltd. Chichester, 1992
15 Stockham, J. D., 1977 What is Particle Size:
The Relationship Among Statistical Diameters, In
Particle Size Analysis 1977, Ann Arbor Sci Pub,
Ann Arbor, Michigan (Ed. John D. Stockham &
Edward G. Fochtman)
16 Sparlin, D. D., 1974, Sand and Gravel A
Study of Their Permeabilities, SPE 4772, paper
prepared for the SPE of AIME Symposium on
Formation Damage Control, New Orleans, La.,
Feb. 7-8, 1974.
17 Bell, F. G. & Lindsay, P. 1999, The
Petrographic and Geomechanical Properties of
Some Sandstones from the Newspaper Member
of the Natal Group near Durban, South Africa
Engineering Geology 53 (1999) 57-81.

Gbenga Oluyemi, Babs Oyeneyin, & Chris Macleod

SPE

Nomenclature
d
d50
dg
d1...d99
k
kH
kV
Rw
RH
RV

=
=
=
=
=
=
=
=
=
=
=

grain size (mm)

median grain size (mm)
geometric mean grain size (mm)
percentile sizes (mm)
permeability (Darcy)
horizontal permeability
vertical permeability
water resistivity
horizontal resistivity
vertical resistivity
permeability anisotropy

= porosity (percent)
= standard deviation (mm)

Figure 2: Comparison of neural network predicted

d50 with measured d50

Poro
Permeab

d50

Horizontal versus vertical median grain size

Irr.
10

Grain size (microns)

100

1000

1800

Pack
str. co

1805

Input
nodes

Hidden
layer

Output
layer node

1810

Figure 1: The neural network architecture

Depth (ft)

1815
Horizontal median
grain size

1820

Vertical median
grain size

1825

1830

1835

1840

Figure 3: Horizontal and vertical median grain size

Prediction of Directional Grain Size Distribution: An Integrated Approach

SPE
_____________________________________________________________________________________________________________
Predicted versus measured at 9968ft

0.2

120
W eig h t %

d/dt

100
0.1

80
60
40
20
0

Grain size (mm)

1
-4

-3

-2

-1

Predicted

Comparison of standard deviation predicted using

Krumbein equation with measured standard deviation

Com parison of standard deviation predicted using

Bergs equation w ith m easured standard deviation
1.6

R = 0.6605
Standard deviation
(predicted)

S ta n d a rd d e v ia t io n
( p re d ic t e d )

10

6
4
2
0
0

R2 = 0.1385

1.4
1.2
1
0.8
0

Standard deviation (measured)

Figure 5: Comparison of predicted standard

deviation (sorting) using Krumbein and Monk
Equation with the measured standard deviation.

Weight %

80
60
40
20

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

Standard deviation (m easured)

2.2 2.4

Figure 8: Comparison of predicted standard

deviation using transposed Bergs equation with
measured the standard deviation

predicted versus measured at 9962ft

100

Weig h t %

Measured

Figure 7: Comparison of predicted grain

distribution with measured grain distribution using
Krumbein & Monk standard deviation at 9968ft

Figure 4: Example of a normal distribution

10

100
90
80
70
60
50
40
30
20
10
0
0.001

0.010

0.100

1.000

10.000

Grain size (mm)

0
1

Grain size (mm)

Predicted

10

Measured

Figure 6: Comparison of predicted grain

distribution with measured grain distribution using
Krumbein and Monk standard deviation at 9962ft

Figure 9: Predicted grain size distribution Bergs

standard deviation

Gbenga Oluyemi, Babs Oyeneyin, & Chris Macleod

9601

0.127

0.125

0.002

1.574803

1.574803

9606

0.219

0.214

0.005

2.283105

2.283105

9711

0.12

0.123

-0.003

-2.5

2.5

80

9716

0.23

0.22

0.01

4.347826

4.347826

60

9721

0.14

0.56

-0.42

-300

300

9726

0.15

0.12

0.03

20

20

9731

0.249

0.256

-0.007

-2.81124

2.811245

9736

0.251

0.36

-0.109

-43.4263

43.42629

9741

0.44

0.44

9746

0.4

0.25

0.15

37.5

37.5

9751

0.34

0.24

29.41176

29.41176

-6.09024

6.090238

corre
co

0.654932462

0.1
Mean
error
Abs mean
error

Horizontal and vertical grain size distribution at 1805ft

Weight %

100

40
20
0
0.01

0.1

Grain size (mm)

Horizontal GSD

Vertical GSD

Figure 10: Example of directional grain size

distribution at 1805ft

Horizontal and vertical grain size distribution at 1825ft

Weight %

SPE

100
90
80
70
60
50
40
30
20
10
0
0.001

Table 2 : Equations for 1,2,98 and 99 percentiles

Integrals

Equation

-2 < t > +2

2%

d 50 d 2 = 2

and

0.01

0.1

-3 < t > +3

98%
1%

Grain size (mm )

Horizontal GSD

6.090238

and

d 98 d 50 = 2
d 50 d1 = 3

and

Vertical GSD

99%

Figure 11: Example of directional grain size

distribution at 1825ft

Table 1: Error analysis for neural network model

and

d 99 d 50 = 3

Table 3: Error analysis for krumbein & Monk equation

Abs_ error

Std (predicted)

Std
(measured)

Abs_error

% error

Abs_value
(% error)

-9.09091

9.090909

6.361990962

5.351074441

-1.0109165

-18.89184186

18.89184186

56.42202

56.42202

3.497304584

5.083141235

1.5858367

31.19796554

31.19796554

16.66667

4.446930847

4.189680297

-0.2572506

-6.140099763

6.140099763

Depth
(ft)

Measured
(mm)

Predicted
(mm)

Abs_error

9399

0.121

0.132

-0.011

9415

0.436

0.19

0.246

9429

0.048

0.04

0.008

16.66667

% error

9439

0.31

0.4

-0.09

-29.0323

29.03226

5.56706212

5.956795501

0.3897334

6.542668469

6.542668469

9443

0.049

0.03

0.019

38.77551

38.77551

3.285375001

7.097887821

3.8125128

53.71334284

53.71334284

9450

0.236

0.06

0.176

74.57627

74.57627

3.367409645

3.979373349

0.6119637

15.37839379

15.37839379

9455

0.213

0.2

0.013

6.103286

6.103286

1.522756791

3.319045586

1.7962888

54.12064247

54.12064247

9558

0.01

0.011

-0.001

-10

10
3.638075252

5.006941609

1.3688664

27.33937129

27.33937129

2.95771698

4.122805453

1.1650885

28.25960347

28.25960347
47.61235552

9563
9574

0.034
0.029

0.04
0.032

-0.006

-17.6471

17.64706

-0.003

-10.3448

10.34483

9580

0.026

0.024

0.002

7.692308

7.692308

2.975818362

5.680382066

2.7045637

47.61235552

9585

0.033

0.035

-0.002

-6.06061

6.060606

2.207096245

4.658355759

2.4512595

52.62070226

52.62070226

9596

0.066

0.073

-0.007

-10.6061

10.60606

6.434116729

6.200249538

-0.2338672

-3.771899656

3.771899656

Prediction of Directional Grain Size Distribution: An Integrated Approach

SPE
_____________________________________________________________________________________________________________
5.061777468

5.117787378

0.0560099

1.09441651

1.09441651

3.909863101

5.844768884

1.9349058

33.10491521

33.10491521

2.499619277

4.556393349

2.0567741

45.14039756

45.14039756

2.298394037

6.168771307

3.8703773

62.74146143

62.74146143

1.283183273

5.51292532

4.229742

76.72409477

76.72409477

5.330594363

4.997693533

-0.3329008

-6.661089317

6.661089317

6.097719649

8.64385619

2.5461365

29.45602617

29.45602617

Average error

31.60585726

Table 4: Error analysis for Bergs equation

Std (predicted)

Std(measured)

Abs_error

% error

Abs_value (% error)

0.712248662

1.28

0.56775134

44.35557328

44.35557328

1.25667821

1.39

0.13332179

9.591495683

9.591495683

2.123702145

1.12

-1.0037021

-89.61626295

89.61626295

1.762382751

1.36

-0.4023828

-29.58696699

29.58696699

2.109197931

1.11

-0.9991979

-90.01783162

90.01783162

1.640516372

1.09

-0.5505164

-50.50608917

50.50608917

1.995220797

1.18

-0.8152208

-69.08650822

69.08650822

0.348042163

1.11

0.76195784

68.64485018

68.64485018

1.988620712

1.2

-0.7886207

-65.71839267

65.71839267

1.437216881

1.33

-0.1072169

-8.061419624

8.061419624

1.414323358

0.91

-0.5043234

-55.42014923

55.42014923

0.82177843

1.04

0.21822157

20.98284327

20.98284327

2.029022462

1.5

-0.5290225

-35.26816413

35.26816413

1.59764547

0.97

-0.6276455

-64.70571856

64.70571856

Average error

50.1115904