Вы находитесь на странице: 1из 7

Reservoir Characterisation 2012

Consider the spatial interpolation problem for a reservoir sand given in


the table below. The thickness of the reservoir sand d is given at 4
locations A, B, C, and D
A
B
C
D
E

X [m]
250
200
450
400
400

Y [m]
500
300
200
400
300

d [m]
20
30
10
25

The structural model can be described by the nested variogram given in


Figure 1.

a) Describe the variogram in Figure 1 as a sum of two variogram


models; write down the structural model, define all quantities, and
estimate relevant parameters using Figure 1.
The key parameters of any variogram are the following:
Nugget, 0: the height of the jump at the origin, attributed to measurement
errors or spatial sources of variation at distances smaller than the sampling
interval or both.
Sill, i: The value that the semivariogram model attains at the range (the value
on the y-axis) is called the sill. The partial sill is the sill minus the nugget.
Range, : The distance in which the difference of the variogram from the sill
becomes negligible. Note that in models with a fixed sill, it is the distance at
which this is first reached; for models with an asymptotic sill, it is conventionally
taken to be the distance when the semi variance first reaches 95% of the sill.
From the diagram we can say that our experimental semi variogram is best
modelled by a nested structural model; a combination of two Gaussian semivariograms. This is seen from the asymptotic nature of the curve as it
approaches the sill (firstly when h1 and secondly at h10)
The equation for the nested model is given by

( h )= 1 ( h ) + 2 (h)

The general equation for a Gaussian is:

[ ( )]

i ( h )=( 2i i20 ) 1exp

3 h2
+ 2i 0 i=1,2
2
L

The actual equations are given below (for both equations the nugget0):

[ ( )]
2

1 ( h )=12.5 1exp

3 h
2
0.5

Where the sill=12.5 & the range=0.5

[ ( )]

2 ( h )=2.5 1exp

3 h
2
15

Where the sill = 2.5 & the range=15

More specifically we may refer to the range as the practical range for the reasons
stated above (the model has an asymptotic sill)
b) Write down the simple Kriging system in matrix form (values) using
the structural model given for estimation of data position E.

0
C BA
C CA
C DA
C EA

][ ] [ ]

C AB C AC C AD C AE 1
CoA
0 C BC C BD C BE 2
C oB
CCB
0 C CD C CE 3 = C oC
C oD
C DB C DC 0 C DE 4
C oE
C EB C EC C ED 0 5

C) How would the Kriging system change when introducing the extra
condition that the sum of weights equals to 1?
If we wish to constrain the sum of weights to 1 we would use ordinary kriging as
opposed to simple kriging. This involves manipulating our matrix into the form
below:

C AA C AB
C BA

C BB

C CA CCB
C DA C DB
C EA C EB
1
1

][ ] [ ]

C AC C AD C AE 1
C oA
1
C BC C BD C BE 1 2
CoB
C CC CCD C CE 1 3 = C oC
C oD
C DC C DD C DE 1 4
CoE
C EC C ED C EE 1 5
1
1
1
1 0

d) What are the consequences for stochastic simulations for the cases of:
(i) 02=2: The variogram will become a constant equal to the value of the nugget
(ii) : The exponential component will tend towards 1 &

i(h )

will tend

towards the nugget value.


(iii) Using a variogram of a variable exhibiting a linear trend: geostatistical
methods often cannot be used with the presence of a strong trend as they give
erroneous or biased answers.
2. A fine-scale reservoir model has been developed after extensive and
careful analysis of input data. Consider a particular scenario of this
model
a) Describe how you would derive an error estimate for a property at
each location in the context of Kriging and stochastic simulation.
The advantage of using kriging is that we can further calculate an estimation
error called the kriging variance. The availability of an estimation error provides
the basis for stochastic simulation of the possible realisations of Z(u). The kriging
variance is given by the following equation:

b) Consider the case of upscaling from the fine grid. How can you assess the
quality of the upscaling approach?

c) Give three examples where upscaling of a quantity is more error-prone.


d) Consider the data in table 1 (units of mD). Calculate the upper and
lower limits on the effective horizontal and vertical permeabilities and
detail your approach.

We assume that all cells have the same dimensions. The upper and lower limits
on horizontal can be calculated using successive arithmetic and harmonic
averages. The harmonic mean can be expressed as the mean of the inverse of
the values of interest. One solution to the problem is then to past above table as
matrix into matlab (all values are in milli-Darcy):

>>

10
20
1000 400 40
20
10
10 40 20
A= 1000 1000 1000 40
10
400 400 400 10
10
10
10
10 400 1000

This allows us to perform the needed averages along the first or second
dimension of the matrix. A mean along dimension 1 can be taken by:
>> ave_column_arithmetic = mean (A, 1)
ave_column_arithmetic =

288 288 484 178 216

It follows:
>> upper_limit_kH = 1./mean(1./mean(A,1))
upper_limit_kH = 259.6301
Here we make use of the element-wise division operator (the. in front of / makes
it an element-wise operation), resulting in taking the average of the inverses, of
which we take the inverse. The other limits follow accordingly. Note, that the
second argument to the mean function specifies the dimension which is to be
reduced (the dimension disappearing, works similar for sums etc.):
>> lower_limit_kH = mean(1./mean(1./A,2))

lower_limit_kH = 24.6058
>> upper_limit_kV = 1./mean(1./mean(A,2))
upper_limit_kV = 79.8268
>> lower_limit_kV = mean(1./mean(1./A,1))
lower_limit_kV = 22.8543
In summary, we have 24.6mD and 260mD as lower and upper limits of the
horizontal permeability, and 22.9mD and 79.8mD as lower and upper limits of
the vertical permeability.
A seismic survey has been carried out and resulted in defining two
horizons which encase a prospective sandstone reservoir. It has been
verified that on average there is a good anti-correlation between
porosity and impedance. One can assume to first order that a low
resolution porosity measurement is available for that horizon.
Given high resolution porosity and permeability data from logging tools
and core analysis in both horizontal and vertical wells intersecting the
formation, detail how you could derive a high-resolution permeability
model. Assume that the relationship between porosity and permeability
can be established with good accuracy.
The anti-correlation between porosity and impedance is to be expected given
that compaction of rock at greater depths leads to increased acoustic impedance
and lower porosity values. We can also safely assume that there is some level of
positive correlation between porosity and permeability. A high-resolution
permeability model involves the process of permeability upscaling to derive
effective permeabilities for the zone of interest. It is important to note that the
correct upscaling method will depend most heavily on the fine layers in which
there are open completions. This is not always known at the time of upscaling
and as such it is impossible to create an upscaled model that is representative of
all well configurations and operating pressures. Nevertheless we can still build
decent models if we conserve net-to-gross ratios. Our three methods involve
averaging methods; arithmetic, harmonic and geometric averaging. All three
methods can generate directional permeabilities although it is encouraged and
often beneficial to use different methods for the horizontal permeabilities
(PERM_I and PERM_J) and vertical permeabilities (PERM_K). Different scenarios
call for different averaging (for example the arithmetic mean is most effective if
it is employed for situations where there is no coarsening in the direction of
interest)

Reservoir Characterisation 2010


1. The table below lists a set of core permeabilities k measured from a
large scale coring program as a function of depth d.

1.9
1.8
1.7
1.6
1.5
1.4
1.3
1.2
1.1
1
2331

2332

2333

2334

2335

2336

2337

2338

2339

2340

a)
Carry out a data quality check on the permeability data. It is assumed
that the permeability data is log-normally distributed. Give a reason for
your approach.

By plotting the log(Perm) vs depth we see that there should exist a relationship
between the depth and permeability. This is mostly true except of a few points.
We can observe these points and decide if they require removal by a simple QC
involving box-plots & inter quartile ranges.
The quality check of the permeability data of each facies is determined by using
the Inter Quartile Range (IQR): IQR = Q3 Q1
Where Q3 = the top 75% percentile
Q1 = the top 25% percentile
For this quality check of the data, any values that lie outside the extreme values
are eliminated. There are two extreme values maximum and minimum extreme
values, given as below:
Minimum extreme values = Q1 3IQR
Maximum extreme values = Q3 +3IQR

We furthermore consider our two facies separately as they have been segregated
to illustrate they have distinct properties:
For Facies A:
Q1
Q3
IQR
Minimum Extreme
Maximum Extreme

KCOR [mD]
43.3
50.68
7.375
21.18
72.8

For Facies B:
Q1
Q3
IQR
Minimum Extreme
Maximum Extreme

KCOR [mD]
25.85
33.88
8.025
1.78
57.95

We find that all values for Facies A are valid whilst in Facies B we exclude the
60.2mD point as it is greater than our maximum extreme value.
b) What upscaled permeability tensor would you measure for the
layered system from a well pressure test in a well deviating in xdirection from the vertical direction by an angle of 20 degree, assuming
the whole formation contributes equally. Give a reason for your
approach.

k cos +k y sin ( k x k y ) sin cos


Permeability tensor K= x
2
2
( k x k y ) sin cos k x sin + k y cos
Where:

ky

kx

is the horizontal permeability in the x-axis,

is the vertical permeability in the y-axis,

mD

mD

is the angle of dip

2. A Jurassic river formed on top of a thick and extensive and constantly


slowly dipping shale layer and cut wide channels into the shale. The
river transported sediment downstream at moderate velocity, leading
to ripples forming on the river bed. Eventually, the river bed and shale
were covered by a blanket sand. You successfully intersected the riverbed in the centre of the channel and are carrying out a fine-scale
structural analysis of the ripples. Assume that the ripples are
sinusoidal, occur periodically at 40cm intervals, and are about 10cm
thick.
a) What qualitative variogram features do you expect when you follow
the river bed parallel and perpendicular to the ripple direction? Give a
reason for your answer, using the variability in the sand ripples r and
the variability in the covering sand c.
b) How would the variogram look qualitatively in the vertical direction
on the ripple scale? Give a reason for your answer.

Вам также может понравиться