Professional Documents
Culture Documents
SPE 68235
Quality Assurance Tool for PVT Simulator Predictions
Varotsis N., SPE, Consultant-Oilphase Division of Schlumberger, Gaganis V., Consultant, Nighswander J., SPE, Oilphase
Division of Schlumberger
Abstract
The currently available PVT simulators predict the physical
properties of reservoir fluids with varying degrees of accuracy
depending on the type of the model utilised, the nature of the
fluid and the prevailing conditions. Nevertheless, they all
exhibit the significant drawback of lacking the ability to
estimate the quality of their answers.
Artificial Neural Networks (ANNs), trained by large PVT
databases, are increasingly utilized to provide accurate
predictions of physical properties mainly due to their ability to
learn from experience. The utilization of such models offers
the unique capability of estimating the quality of their
predictions as their degree of competence can be evaluated for
each unknown test case1-2. The accuracy of the ANN based
PVT simulators depends heavily on the density of the database
compositional mapping around the coordinates of the
unknown reservoir fluid. Unknown test cases found outside
the available training space may lead to poor predictions.
In this work, a quality assurance tool is presented that is
integrated to the ANN based PVT Expert model. This tool,
for an unknown fluid, qualifies the predictions of the PVT
simulator based on the evaluation of the affinity of the test
case with the training data sets contained in the utilised
database. Subsequently, the competence with which the ANN
model has learned the general trend in the area around any
new test case is assessed numerically.
This innovative approach was successfully tested against a
large set of studies "unseen" by the PVT Expert. The ability of
providing confidence for the accuracy of the PVT predictions
and of assessing their quality, significantly upgrades the
Trademark of Schlumberger
SPE 68235
{ }
i x = std x j ,
j = , ... , C
2
n+
........................................(1)
y new
j
x new
.........................(2)
y new y train
y new
........................................................(3)
i = 101 i j , j = x , c, r , o .............................................(4)
j
tj
composition match between the new test case and the database
one, while i < 0 implies that the difference of their outputs is
o
SPE 68235
max{y i } min{y i }
............................................(6)
avg{y i }
max{ y i avg{y i } }
avg{y i }
............................................(7)
1
1
y i = y + b + e i ..........................................(9)
n i
n i
SPE 68235
Results
For validating the method described above, the Quality
Assurance Tool was tested against a genuine set of 161 oil
fluids of worldwide origin. The results obtained for the Bo and
GOR values at the bubble point are illustrated in Figures 7
and 8 respectively.
The success rate of the PVT Expert outputs for the Bo
compared to the reported lab values is shown in the first
column of Figure 7. For 96.2% of the new fluids tested, the Bo
prediction deviated less than 5% from the lab value. The
percentage of the fluids which were successfully qualified by
the Quality Assurance Evaluator was 95.6%. Three of the
fluids (1.8% of the total) for which the prediction deviation
was more than 5% were qualified. Nevertheless, even for these
cases, the utilized average prediction provided accurate
answers for two out of the three.
The success rate of the PVT Expert outputs for the GOR is
shown in Figure 8. The GOR prediction deviated less than 5%
from the lab value for 85.1% (137 cases) of the new fluids
tested. The percentage of the fluids which were successfully
qualified by the Quality Assurance Evaluator was 73.9%.
Only eight of the fluids (5% of the total) for which the
prediction deviation was more than 5% were qualified. But
even for these cases, the utilized average prediction provided
accurate answers for five out of the eight fluids.
Conclusions
An innovative approach for qualifying the predictions of a
PVT simulator based on artificial neural networks was
developed. The principle upon which this approach is based
on can be applied to all kind of artificial neural network based
prediction models utilized in the petroleum industry. Firstly,
the Quality Assurance Tool estimates the closeness of the
reservoir fluid under test to the characteristics of the fluids
available in the training database by calculating the similarity
indices.
Secondly, the Quality Assurance Tool assesses the degree
of competence by which each ANN learned, during its
training, the underlying trend of the required PVT properties
in the neighborhood of the unknown test case. The prediction
of the PVT Expert for a given property can be confidently
trusted if the new fluid lies within the confidence region
and/or the maximun span and the maximum deviation of the
ANNs predictions do not exceed the predefined thresholds.
The Evaluator was successfully tested against a genuine
set of 161 oil fluids of worldwide origin and in nearly 96% of
the cases it had successully qualified the accuracy of the ANN
models predictions. Such a tool can help promoting several
field and laboratory applications based on predictions. Among
them, is the generation of rapid and reliable PVT studies at the
wellsite within hours enhancing thus the quality of onsite fluid
properties evaluation services such as the FPE8 and widely
extending fluid analysis to reservoir fluid samples recovered
by Formation Testers.
SPE 68235
Nomenclature
Bo = Formation oil volume factor
GOR = Gas to Oil Ratio
RLD = Retrograde liquid deposit
pb = Bubble point
Tres = Reservoir temperature
b = Reservoir oil density
STO = Stock tank oil density
MWRF = Reservoir fluid molecular weight
MWSTO = Stock tank oil molecular weight
MWFG = Flash gas molecular weight
Subscripts
b= bubble point
STO = stock tank oil
FG= flash gas
RF = reservoir fluid
Acknowledgments
The authors would like to thank the management of
Oilphase, Division of Schlumberger EPS, for their permission
to publish this paper.
References
1. Liu Y. : "Unbiased Estimate of Generalization Error and Model
Selection in Neural Networks", Neural Networks, (1995), Vol.8,
No.2, pp.215-219.
2. Courieu P.: "Three Algorithms for Estimating the Domain of
Validity of Feedforward Neural Networks", Neural Networks,
(1994), Vol.7, No.1, pp.169-174.
3. Haykin S., Neural Networks - A Comprehensive Foundation,
MacMillan College Publ. Company, USA, (1994).
4. Varotsis N., et al: "A Novel Non-Iterative Method for the
Prediction of the PVT Behavior of Reservoir Fluids", SPE
56745, (1999).
5. Cho S., Kim J. : "Rapid Backpropagation Learning Algorithms",
Circuits Systems and Signal Processing, (1993), Vol.12, No.2,
pp. 155-175.
6. Choi J., Choi C.: "Sensitivity Analysis of Multilayer Perceptron
with Differentiable Activation Functions", IEEE Trans. on
Neural Networks, (1992), Vo.3, No.1, pp. 101-107.
7. Anderson T., An Introduction to Multivariate Statistical
Analysis, John Willey & Sons, NY, USA, 1984.
8. Varotsis N. and Guieze P.: Onsite Reservoir Fluid Properties
Evaluation, SPE 18317, (1988).
New fluid
mole%
N2
CO2
H2S
C1
C2
C3
iC4
nC4
iC5
nC5
C6
C7
C8
C9
C10
C11
C12+
0.17
0.24
0.00
45.94
7.62
3.37
0.83
2.98
1.65
2.10
3.05
4.21
4.41
3.32
2.40
1.72
15.99
SPE 68235
0.13
0.23
0.00
45.74
8.39
4.41
1.23
2.35
1.30
1.70
3.03
4.11
4.68
3.64
2.76
2.09
14.23
MPa
gr/cm
New
Training
10.01
10.76
385
383
731
732
Bo
GOR
ANN 1
1.577
174
ANN 2
1.592
174
ANN 3
1.575
174
Bo
GOR
ic
8.73
7.53
Bo
GOR
Similarity Test
ix
ir
5.23
8.05
5.23
6.77
PVT Lab
1.581
173
gr/mole
gr/cm
130
123
864
850
Stability Test
ANN 4 ANN 5
1.582
1.595
173
171
io
9.01
6.22
s
7.58
6.38
PVT Results
PVT Expert
1.577
174
Err %
0.3
0.6
MWFG
gr/mole
gr/mole
215
188
30.2
31.0
ANN 6
1.582
173
Avg
1.584
173
Avg
1.584
173
Err %
0.2
0.1
Dev %
0.71
1.25
Span %
1.26
1.73
mole%
mole%
1.40
4.70
0.00
17.28
5.14
9.48
1.83
6.05
2.30
2.67
3.94
6.30
6.15
4.77
3.96
2.55
21.47
0.63
0.48
0.00
20.77
7.81
8.52
1.36
5.15
1.79
3.02
4.32
5.84
6.61
4.81
4.40
3.43
21.05
New fluid
mole%
N2
CO2
H2S
C1
C2
C3
iC4
nC4
iC5
nC5
C6
C7
C8
C9
C10
C11
C12+
0.12
5.08
3.70
81.62
2.31
1.54
0.35
0.84
0.40
0.52
0.68
0.63
0.57
0.45
0.36
0.23
0.62
MPa
gr/cm
New
Training
10.01
10.76
385
383
731
732
Bo
GOR
ANN 1
1.433
99
ANN 2
1.422
104
ANN 3
1.448
106
Bo
GOR
ic
6.05
5.14
Bo
GOR
Similarity Test
ix
ir
-1.14
9.39
-1.14
4.52
PVT Lab
1.469
107
gr/mole
gr/cm
130
123
864
850
Stability Test
ANN 4 ANN 5
1.433
1.441
109
96
io
9.29
5.68
s
-4.95
-3.5
PVT Results
PVT Expert
1.433
99
Err %
2.5
7.5
gr/mole
gr/mole
215
188
30.2
31.0
ANN 6
1.436
109
Avg
1.436
104
Avg
1.436
104
Err %
2.2
3.1
Dev %
0.97
7.69
Span %
1.81
12.5
Dev %
4.18
Span %
7.74
0.36
4.79
2.11
83.81
2.55
1.52
0.35
0.81
0.35
0.45
0.57
0.54
0.45
0.36
0.29
0.20
0.49
MPa
gr/mole
gr/cm
23.7
22.6
755
757
New
Training
27.68
30.16
399
397
0.95
0.97
RLD
ANN 1
2.76
ANN 2
2.62
ANN 3
2.60
RLD
ic
6.38
RLD
Similarity Test
ix
ir
2.65
3.84
PVT Lab
2.76
Stability Test
ANN 4 ANN 5
2.76
2.73
io
7.95
s
4.77
PVT Results
PVT Expert
2.75
Err%
0.43
ANN 6
2.81
Avg
2.71
Avg
2.71
Err %
1.3
SPE 68235
2.2
Predicted
Experimental
2.0
Competence region
1.8
Bo (vol/vol)
Outputs
1.6
1.4
Prediction point
1.2
Training point
1.0
0
15
30
45
Inputs
Pressure (MPa)
Fig. 1-Comparison of the predicted reservoir oil Bo curve to the
PVT Lab one.
1.45
1.2
Predicted
Span
Experimental
1.35
0.8
Dev
1.40
1.0
Average
1.30
0.6
CVD RLD %
0.4
Lab value
1.25
1.20
ANN 1
0.2
0.0
5
10
15
20
25
Pressure (MPa)
30
35
ANN 2
ANN 3
ANN 4
ANN 5
ANN 6
PVT Expert
Confidence region
Limited confidence
region
No confidence
region
s=10
s=0
s=-2
s=-
Similarity
check
Training
Database
Stability
check
Establishment of
similarity
Prediction
qualification
Establishment
of stability
100
80
SPE 68235
100
96.2%
155
95.6%
154
96.2%
155
80
60
60
40
40
20
20
85.1%
137
73.9%
119
85.1%
137
0
PVT Expert
Qualified
Averaging
PVT Expert
Qualified
Averaging