Vous êtes sur la page 1sur 5

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/315800144

A Neural Network approach for hydraulic flow


unit-based reservoir characterization

Conference Paper · October 2009

CITATIONS READS

0 20

1 author:

An Hai Le
Hanoi University of Mining and Geology
6 PUBLICATIONS 5 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Environmental protection and sustainable development View project

All content following this page was uploaded by An Hai Le on 06 April 2017.

The user has requested enhancement of the downloaded file.


A Neural Network approach for hydraulic flow unit-based reservoir
characterization
Le Hai An (1)
(1)
Hanoi University of Mining and Geology, Vietnam (lehaian@humg.edu.vn)

ABSTRACT permeability. However, low resistivity is one special


Recently, neural network techniques have emerged as characteristic of these Lower Miocene reservoirs in the
robust tools to tackle complicated problems in the basin. The presence of low resistivity zones makes
petroleum industry, especially in reservoir difficulties for the log analyst to accurately determine
characterization. This paper demonstrates the water saturation Sw and differentiate between pay
application of neural networks in predicting global zones and non-pay zones. These summation results are
hydraulic elements and hydraulic flow units through really sensitive and have a huge impact on
core - log integration. A new approach to effectively hydrocarbon initial in place (HIIP) assessments. A
use the predicted global hydraulic element resulting in proper approach in these LRLC reservoirs is therefore
netpay calculation is proposed. The derived GHE essential.
outcome has potentially resulted in calculating a more The main causes of LRLC include (i) thin
precise result of netpay in the reservoir. Several shale/sand lamination, (ii) clay minerals, (iii)
approaches in the wireline log interpretation have been conductive minerals such as pyrite, (iv) high salinity of
attempted and it is proved that the hydraulic flow formation water, (v) deep invasion of drilling fluid into
unit-based approach has far more advantages formation and (vi) grain size. To date, several
compared to the conventional versions. Case studies approaches have been attempted, most rely on the
from several low resistivity low contrast reservoirs in environmental correction of resistivity as well as
the offshore Vietnam have been presented in this paper conductive minerals. However, there is no single
to illustrate the new proposed approach. approach proposed for how to precisely define net pay.
All are based on conventional methods that use single
KEYWORDS: hydraulic flow unit, global hydraulic cut off value of porosity and water saturation to
element, neural network, reservoir characterization. calculate the net pay.

INTRODUCTION HYDRAULIC FLOW UNIT-BASED RESERVOIR


To characterize a reservoir, information on lithofacies CHARACTERIZATION
and petrophysical parameters such as shale volume, Ebanks first introduced the concept of the hydraulic
porosity, permeability, water saturation always play flow unit in 1987 and since then, many researchers
important roles. Among them, the top most three key have applied the method in reservoir characterization
parameters for assessing hydrocarbon in place is and description, and reservoir modeling has been done
porosity, water saturation and net to gross ratio. The with great success. Researchers from Heriot-Watt
hydraulic flow unit concept is used here to maximize University proposed and applied not only for
the geological information from core data in order to permeability prediction from wireline logs but also for
integrate with wireline logs using neural networks. It selecting and sampling of representative core plugs for
consequently has increased the efficiency of wireline special core analysis (SCAL) as well as prediction of
log interpretation in reservoir characterization, other petrophysical, geochemical parameters and
especially in the net pay calculation. The resulting dynamic reservoir modeling [Corbett, 2003 and An,
outcome allows more precise integration of G&G data 2004].
in the subsequent reservoir modeling [Amaefule et al,
1993; Abbaszadeh et al, 1995; Svirsky et al, 2004)] Hydraulic Unit and Global Hydraulic Element
concepts
LOW RESISTIVITY LOW CONTRAST A reservoir is characterised by its flow units. A flow
RESERVOIRS unit is defined as a volume of the total reservoir rock
The low resistivity low contrast (LRLC) reservoirs are within which geological and petrophysical properties
rather common in a northeastern part of Cuu Long that affect fluid flow are internally consistent and
basin, where most of the crude oil in Vietnam is predictably different from properties of other rock
currently produced. Lower Miocene sandstone volumes (Ebanks, 1987). A flow unit is a reservoir
reservoir is one of the most important plays in Cuu zone that is continuous laterally and vertically and has
Long basin. Drilling and testing results show that similar flow and bedding characteristics. Within each
reservoirs are very good with high porosity and flow unit, there is a strong relation between porosity

Proceedings of the 9th SEGJ International Symposium –Imaging and Interpretation- , Sapporo, Japan 12-14 October 2009
and permeability i.e. each flow unit should have texture and mineralogy in the discrimination of distinct
distinct petrophysical characteristics. pore geometry facies (hydraulic units) [Amaefule et
Permeability depends on the porosity, and its al., 1993]
geometric distribution and connectivity. The most Corbett et al. have introduced the term global
commonly used porosity - permeability equation is the hydraulic elements (GHEs). A systematic series of a
Kozeny-Carman equation [Kozeny, 1927; Carman, priori FZI values was arbitrarily chosen to define 10
1937]. Amaefule proposed a generalised formula of porosity-permeability elements (global hydraulic
Kozeny-Carman which included the pore shape (Fs) elements). Only 10 were chosen in order to split the
parameter as follows: wide range of porosity and permeability parameter
Φe space into a manageable number of GHEs. The series
3
1
K= (1) of FZI values chosen was as follows: 0.0938; 0.1875;
Fsτ S Vgr (1 − Φ e ) 2
2 2
0.375; 0.75; 1.5; 3; 6; 12; 24; 48 which corresponded
respectively to the lower boundary of GHE1; GHE2;
Where K is permeability, Fs is shape factor, τ is
GHE3; GHE4; GHE5; GHE6; GHE7; GHE8; GHE9;
tortuiosity, SVgr is surface area per unit grain volume,
and GHE10. The GHE template was then constructed
Φe is effective porosity.
on the porosity-permeability crossplot. The GHE
The Flow Zone Indicator (FZI) parameter was
template allows any core plug to be rapidly classified
defined by Amaefule et al. (1993) as follows:
in terms of GHEs merely by plotting its porosity and
1
FZI = (2) permeability values on the template. There is no need
Fs τ SVgr to calculate FZI values. The advantage of this
approach is that it allows data from any reservoir to be
The Reservoir Quality Index (RQI) can be defined plotted on exactly the same reference frame - unlike
as: the Amaefule et al., 1993, conventional approach
K [Corbett et al., 2003].
RQI = 0.0314 (3)
Φe GHE/HU recognition from core data
The constant 0.0314 occurs because the The database consists of wireline logs and core data
permeability K in Equation (1) is in µm2 whereas the from 5 wells located in three different oilfields within
permeability K in Equation (3) is in mD [Amaefule et Cuu Long basin. The LRLC reservoir of Lower
al., 1993]. Miocene in age is a common feature in these fields.
Φz is defined as the pore volume to grain volume To study the reservoir quality through its
ratio: petrophysical properties, porosity and permeability
measured on core plugs and wireline logs from 5 wells
 Φe 
Φ z =   (4) were used. The porosity permeability crossplot of all
 1 − Φ e  core plugs taken from these wells, in Lower Miocene
The generalised equation (3) can be re-written as reservoirs, is shown in Figure 1.
follows:
K
0.0314
RQI Φe
FZI = = (5)
Φz  Φe 
 
1− Φe 
and

log RQI = log Φ z + log FZI (6)

From Equation (6), if RQI were plotted versus Φz


on a log-log plot, the data from a given flow unit or a
similar FZI value will lie on a straight line with a slope
of 1.0. Other flow units will fall on adjacent parallel
lines and each flow unit will have a separate FZI value.
The FZI value will be the same for a given flow unit
having similar pore throat characteristics.
Furthermore the value of FZI can be determined from
the intercept at Φz = 1.0 (or logΦ z = 0). The FZI has Figure 1. Porosity-permeability crossplot (solid line
high geological significance since it is a unique shows regression line of porosity-permeability
parameter that incorporates the geological attributes of transform).

Proceedings of the 9th SEGJ International Symposium –Imaging and Interpretation- , Sapporo, Japan 12-14 October 2009
learning algorithm is a time-consuming process. For
In this study, the porosity and permeability data simplicity and time saving, the conventional
were also plotted on GHE template (Figure 3); it’s back-propagation neural network was used in this
obvious that the porosity permeability relationships are study to build the model based on core derived FZI and
clearly defined within individual GHE. In these 5 wireline logs as input (Figure 3).
wells, 8 GHE are defined, from GHE1 to GHE8. It Six conventional wireline logs: gamma ray (GR),
means that the hydraulic properties of these reservoirs sonic transit time (DT), deep resistivity (LLD), shallow
are very complex and vary in a great range. It is resistivity (LLS), density (RHOB) and neutron (NPHI)
therefore essential to be able to know well in advance were used as input and calculated FZI from core was
what the formation is from wireline log if core data is used as output to the neural network model.
not available.
Since the variation in porosity-permeability
relationship is large, the unique cutoff values for pay
reservoir is no longer valid. Each GHE / rocktype
should have distinguished porosity cutoff values based
on the 1mD cutoff value of permeability, which is the
industry standard for an oil reservoir and 0.1mD for a
gas reservoir. For instance, in this case, from the
porosity-permeability relationship, the porosity cutoff
for the entire reservoir is 0.13 for oil and 0.08 for gas
(Figure 1). If the GHE concept is used, the porosity
cutoff values vary from as high as 0.25 for GHE1 to as
low as 0.03 for GHE8 (oil reservoir) and 0.2 for GHE1
to 0.02 for GHE8 (Figure 2). Similarly, different cutoff
values of water saturation would also be applied for
individual GHE.
Figure 3. Schematic of Neural Network prediction.

Figure 4 shows the neural network predicted FZI and


FZI calculated from core data of the training dataset.
There is a very good agreement between them. The
correlation coefficient between them is R2=0.91.

Figure 2. Core porosity-permeability on GHE template. Figure 4. Core and Neural Network predicted FZI in
the training dataset
GHE prediction model using a neural network
approach The trained neural network was then used to calculate
One of the purposes of this study is to build an FZI in uncored intervals and wells in the entire
empirical model to directly predict GHE by calculating Miocene reservoirs. Consequently, predicted FZI was
FZI from a different combination of wireline logs. used to determine GHE.
The Artificial Neural Network approach, which is
robust and able to reveal the nonlinear relationship RESULTS
between variables, was employed in this study to The results obtained from conventional wireline log
accomplish this task. GHE would then be classified interpretation (shale volume, porosity and water
based on predicted FZI values. saturation) were then used to calculate the net pay for
The construction of a Neural Network as well as a entire reservoir based on two approaches: (i)

Proceedings of the 9th SEGJ International Symposium –Imaging and Interpretation- , Sapporo, Japan 12-14 October 2009
conventional approach: one cutoff values for entire interpretation is limited in terms of net pay calculation
reservoir; (ii) new proposed GHE approach: different if a single cutoff value is being used.
cutoff values for different GHE. In the second The new proposed hydraulic flow unit-based
approach, the predicted neural network GHE was used approach using GHE/HU and Artificial Neural
as an extra input for the reservoir summation. Network is certainly a better and more precise
For the sake of simplicity and when compared with approach.
petrographic analysis from core, it was decided that in In future, a reservoir characterization in LRLC
this Miocene reservoir, GHE should be grouped. The reservoirs based on the integration of data from diverse
first GHE group (HU1) is GHE1, GHE2 and GHE2, disciplines such as core, testing, wireline data, with use
the second group (HU2) is GHE4 and GHE5, the third of the GHE approach, as shown in this study, is
group (HU3) is GHE5 and GHE6, and the fourth group strongly recommended.
(HU4) is GHE7 and GHE8. For the second approach, The results of this study will be tested for
different cutoff values of porosity and water saturation reservoir modeling and simulation of this Miocene
for these 4 groups of GHE were used. reservoir.
The summation results show that there is a
difference between the two approaches, the second REFERENCES
approach, which is based on GHE has far more
advantage and was more precise than that of the first Abbaszadeh, M., Fujii, H., and Fujimoto, F., 1995.
approach when verified with well testing results (Table Permeability prediction by hydraulic flow units –
1). Several pay zones were missed out when using the theory and applications. SPE paper 30158.
first conventional approach. Amaefule, J. O., Altunbay, M., Tiab, D., Kersey, D.
G., and Keelan, D. K., 1993. Enhanced reservoir
Table 1. Reservoir summation in 5 wells descriptions: using core and log data to identify
Hydraulic (flow) Units and predict permeability in
Well GHE approach Conventional uncored intervals / wells. SPE paper 26436.
Approach An, L. H., 2004. Innovative neural network approaches
1 39.5m 34.9m for petrophysical parameter prediction.
2 13.4m 6.4m Unpublished PhD thesis, Heriot-Watt University,
3 35.6m 24.5m 193 p.
4 9.4m 7.2m Carman, P.C, 1937. Fluid Flow through Granular
5 11.2m 11m Beds. Trans. AIChE , v. 15, 150-166.
Corbett, P. W. M., Ellabad, Y., Mohamed, K., and
When using the outcomes in the subsequent Pososyaev, A., 2003. Global Hydraulic Elements
reservoir modeling process, once the GHE has been – elementary petrophysics for reduced reservoir
determined, one can easily relate porosity and modeling. European Association of Geoscientists
permeability through Equation (7), which is derived and Engineers 65th Conference, Paper F-26 EAGE
from Equation (5). The larger the number of GHE in meeting, Stavanger, June 2 - June 5.
the model, the more complex the reservoir model. Ebanks, W. J., 1987. Flow unit concept - integrated
2 approach for engineering projects. Abstract
  Φ  presented June 8, during the roundtable sessions at
 FZI  1 − Φ   the 1987 American Association of Petroleum
K = Φ   (7) Geologists Annual Convention.
 0.0314  Kozeny. J., 1927. Uber Kapillare Leitung des Wassers
  im Boden. Sitzungsberichte. Royal Academy of
  Science, Vienna, Proc. Class 1, v. 136, pp.
For reservoir modeling, to reduce the complexity 271-306.
of the model, two or more GHE could be grouped into
Svirsky D., A. Ryazanov, M. Pankov, P. W. M.
one HU.
Corbett, A. Posysoev, 2004. Hydraulic Flow Units
Resolve Reservoir Description Challenges in a
CONCLUSIONS Siberian Oil. SPE paper 87056
In the LRLC reservoir, the conventional wireline log

Proceedings of the 9th SEGJ International Symposium –Imaging and Interpretation- , Sapporo, Japan 12-14 October 2009

View publication stats

Vous aimerez peut-être aussi