Upload
others
View
7
Download
0
Embed Size (px)
Citation preview
Are fractures present in the reservoirs?
1 of 12
Reservoir Characterization (Task 2.2):
Fracture Mapping of the Clinton sandstone and Copper Ridge dolomite
Objectives
The goal of this subtask is to investigate natural fractures in two formations of interest: (a)
Clinton sandstone near the East Canton Consolidated (ECOF) and Gore Consolidated (GCOF)
Oilfields in eastern Ohio, and (b) Copper Ridge dolomite near the Morrow Consolidated Oilfield
(MCOF) in central Ohio. Natural fractures play a potentially important role in the production of
oil and gas and the storage of CO2 in a geologic reservoir. An understanding of natural fractures
is also crucial when conducting CO2-Enhanced Oil Recovery (CO2-EOR). The results from this
task ultimately supported modeling of reservoirs with natural fractures (Task 4.2) and field
injectivity testing (Task 7.0) in subsequent tasks.
Study Area and Data Sources
logs (i.e., image logs and fracture identification logs) obtained from the Ohio Department of
Natural Resources (ODNR), Division of the Geological Survey. A simplified spreadsheet was
created to capture information on a foot-by-foot basis. Fractures were identified on a foot-by-foot
basis, and a spreadsheet identifying fracture presence, orientation, and whether the fracture was
filled was created using the codes listed in Table 1.
Table 1. Codes used to identify fracture presence, confidence in picks, fracture orientation, and fracture filling.
Presence Confidence Orientation Frac filled
Code Meaning Code Meaning Code Meaning Code Meaning
1 Present 1 Confident 1 Vertical (65-90) 1 Yes
0 Not Present 0 Possible 0.5 Subvertical (25-65) 0.5 Partial
-1 No Data -1 No Data 0 Horizontal (0-25) 0 No
-0.5 Uncertain -0.5 Uncertain
-1 No Data -1 No Data
Once the fracture database was generated, it was used to complete the following work:
• Fracture density maps and cross-sections were created to show the spatial and vertical
distribution of fractures in the formations of interest.
• The prevalence of fractures was compared to oil production in the formations of interest.
• A machine learning process was used to identify presence of fractures using commonly
acquired wireline logs.
Reservoir Characterization (Task 2.2): Fracture Mapping of the Clinton sandstone and Copper Ridge dolomite
2 of 12
Fracture Mapping Results
The fracture density map of the Clinton sandstone is shown in Figure 1. In addition, the fractured
columns and log signatures are shown in the southwest-to-northeast and northwest-to-southeast
trending cross-sections in Figure 2. In general, the wells in the southwestern portion of the study
area, near the Gore Consolidated Oilfield (GCOF) are less fractured than those in the central and
eastern portions of the study area, particularly near the East Canton Consolidated Oilfield
(ECOF). The boreholes in the Clinton sandstone in the western part of the study area contain
relatively thin and isolated zones of fractures. The boreholes in the eastern part of the study area,
however, have more complex fracture systems.
The fracture density map of the Copper Ridge dolomite is shown in Figure 3. In addition, the
fractured columns and log signatures are shown in the west-to-east and north-to-south trending
cross-sections in Figure 4. There is a high prevalence of fractures in the Copper Ridge dolomite
at most of the wells near the Morrow Consolidated Oilfield (MCOF). The percentage of the
interval that is fractured in each well is variable and appears to be somewhat random, ranging
from less than 5% of the interval with data to more than 40% of the interval with data within a
few miles.
Table 2. Total footage of fractured intervals, non-fractured intervals, and intervals without data for the Clinton
sandstone in eastern Ohio and the Copper Ridge dolomite near the MCOF in central Ohio. Fractured intervals are
also indicated as open, partially open, closed, or not indicated as closed or open.
Formation
Fractured Not
Fractured
Total with Data
No Data Open
Partially Open
Closed Not
Indicated ALL
Clinton 29 12 43 30 114 1357 1471 246
Copper Ridge 71 10 67 177 325 1751 2076 2069
The area around the ECOF, in the northeastern corner of the study area, has the largest
concentration of high-producing wells, and is the area with the highest prevalence of fractures
indicated within the study area. The area near the GCOF, in the southwestern corner of the study
area, does not have as large of a concentration of high producing wells. Fewer wells with high oil
production may be because the field was discovered in the early 20th Century, meaning the
highest producing wells may not be captured in the database. The cumulative oil production of
the major oilfields producing from the Clinton sandstone provide further evidence to this point;
the GCOF has the second highest cumulative oil production of the major oilfields producing
from the Clinton sandstone in Ohio, second only to the ECOF.
The oil production is highly variable across the oilfield, possibly because production in the
MCOF is from disconnected erosional remnants. While most wells in the dataset have some
amount of fracturing, the wells with high density of fractures in the northwestern portion of the
field are near areas with some of the highest production in the MCOF. In addition, the four wells
without fractures near the field are in areas with low production.
Reservoir Characterization (Task 2.2): Fracture Mapping of the Clinton sandstone and Copper Ridge dolomite
3 of 12
Additional areas with high producing wells are found in the southern and eastern portions of the
MCOF. These discrete areas are separated by relatively unproductive areas.
Figure 1. Clinton sandstone fracture density map.
Reservoir Characterization (Task 2.2): Fracture Mapping of the Clinton sandstone and Copper Ridge dolomite
4 of 12
Figure 2A. Cross-section A-A’, which covers a horizontal distance of 178 miles (vertical exaggeration – 2,835x).
Reservoir Characterization (Task 2.2): Fracture Mapping of the Clinton sandstone and Copper Ridge dolomite
5 of 12
Figure 2B. Cross-section B-B’, which covers a horizontal distance of 52 miles (vertical exaggeration – 830x).
Reservoir Characterization (Task 2.2): Fracture Mapping of the Clinton sandstone and Copper Ridge dolomite
6 of 12
Figure 3. Copper Ridge dolomite fracture density map.
Reservoir Characterization (Task 2.2): Fracture Mapping of the Clinton sandstone and Copper Ridge dolomite
7 of 12
Figure 4A. Cross-section A-A’, which covers a horizontal distance of 178 miles (vertical exaggeration – 2,835x).
Reservoir Characterization (Task 2.2): Fracture Mapping of the Clinton sandstone and Copper Ridge dolomite
8 of 12
Figure 4B. Cross-section B-B’, which covers a horizontal distance of 52 miles (vertical exaggeration – 830x).
Reservoir Characterization (Task 2.2): Fracture Mapping of the Clinton sandstone and Copper Ridge dolomite
9 of 12
Fracture Prediction by Machine Learning
Fractures were predicted based on common logs using machine learning. As explained in
Bhattacharya and Mishra (2018)1 the motivation is to utilize the more readily available common
log data when more detailed information such as core samples and/or advanced logs are not
available. The methodology for fracture prediction involves three steps:
1. Identify fractures from core and high-resolution sonic and image logs (this provides the
training data)
2. Train the predictive model using different machine learning algorithms
3. Test the model using either blind-well test or cross-validation techniques
After core and image-log-based fracture identification, different machine learning algorithms,
such as Bayesian Network (BN), Random Forest (RF), Support Vector Machine (SVM), and
Artificial Neural Network (ANN) are applied to learn the petrophysical data pattern associated
with the presence of fractures and predict their distribution using Delta_CALI, RHOB, and GR
logs. All the models are cross-validated ten-fold, to assess quality of the results. Numerous
experiments were designed to investigate the optimal parameters for each of the techniques. Six
wells from the Clinton sandstone and ten wells from the Copper Ridge dolomite contained
commonly available logs, such as caliper (CALI), gamma (GR), and bulk density (RHOB).
Other common logs, such as resistivity and neutron porosity were not available in all wells, so
they were not used.
Application of BN, RF, SVM, and ANN with optimal parameters produced different results. The
BN method was found to be the best classifier for the Clinton Formation (accuracy 82%),
whereas the RF method was the best classifier for the Copper Ridge Formation (88%), followed
by SVM and ANN (Figure 5). This may be due to different data patterns in two geologically
different formations, as one of them is sandstone and the other one is carbonate. The BN method
shows accuracy of 82% and 87% for fracture prediction for the Clinton and Copper Ridge
formations, respectively. In addition, BN shows causality of input-output relationships in the
form of a Directed Acyclic Graph (DAG) (Figure 6). It shows that Delta_CALI and RHOB logs
are directly connected via a DAG, which indicates Delta_CALI and RHOB logs are strongly
interrelated for fracture prediction. This observation makes geologic sense, because Delta_CALI
and RHOB logs are highly sensitive to hole size, compared to the GR log. Any change in hole
size (due to either fractures or porous formations) will cause change in Delta_CALI and RHOB
log signatures.
The correlation between the presence of fractures and hydrocarbon production was investigated
using a limited number of wells with the required data. Ten wells producing from the Clinton
sandstone and five wells producing from the Copper Ridge dolomite have both fracture and
1 Bhattacharya, S. Mishra, S. 2018. Applications of machine learning for facies and fracture prediction using Bayesian
Network Theory and Random Forest: Case studies from the Appalachian basin, USA. Journal of Petroleum Science and Engineering 170, pp. 1005-1017.
Reservoir Characterization (Task 2.2): Fracture Mapping of the Clinton sandstone and Copper Ridge dolomite
10 of 12
production data. The number of fractures within the formation of interest was calculated from
each well. Cumulative hydrocarbon production (through 2014) for each formation was plotted
against the footage of fractures for each well (Figure 7).
No direct relation between total number of fractures and hydrocarbon production was found. The
high hydrocarbon production found in wells with less fracturing may be attributed to other
geologic and petrophysical parameters.
Figure 5. Accuracy of fracture prediction using different machine learning algorithms for the Clinton sandstone and
Copper Ridge carbonate formation.
Figure 6. Bayesian Network-derived input-output relationships, indicating that fractures are dependent on the
caliper log deltas, bulk density, and gamma ray logs, bulk density is dependent on the caliper log deltas, and gamma
ray is dependent on the bulk density.
BN RF SVM ANN
Clinton 82 79 76 74
Copper Ridge 87 88 86 86
0
10
20
30
40
50
60
70
80
90
100
Acc
ura
cy o
f Fr
actu
re P
red
icti
on
(%
)
Clinton Copper Ridge
Reservoir Characterization (Task 2.2): Fracture Mapping of the Clinton sandstone and Copper Ridge dolomite
11 of 12
Figure 7. Scatter plot between total number of fractures and cumulative hydrocarbon production for the Clinton
sandstone (left) and Copper Ridge dolomite (right).
Significance
The significance of the work includes the following:
• In general, fractures in the Clinton sandstone were more prevalent in the eastern part of
the study area where the formation was found at depths between 4,000 and 6,000 feet
below ground surface; however, the density of fractures may be attributable to known
faults and structures in these areas rather than depth. Oil production from the Clinton
sandstone was highest near the ECOF, the area with the highest density of fractures.
• Fractures in the Copper Ridge dolomite near the MCOF in central Ohio do not follow as
discernible of a pattern as those in the Clinton sandstone. The total thickness of fractured
intervals in each well was highly variable throughout the field. Most wells (all but four)
in the dataset had at least some fractures. The wells with the highest density of fractures
were found in the northwestern portion of the MCOF where oil production was also high.
• The machine learning process developed to determine if basic wireline logs could be used
to predict fractures proved successful. One method, Bayesian Network, predicted the
fracture presence/absence at a rate of 82% for the Clinton sandstone and 87% for the
Copper Ridge dolomite. Additional data could help to refine and increase the accuracy of
the method; however, the accuracy of the current model is a good initial step. The
investigation into the relationship between fracture density and hydrocarbon production
did not show a strong correlation. This may be because only a few wells (eight producing
from the Clinton sandstone and five producing from the Copper Ridge dolomite) could be
included in the analysis because the necessary data (annual hydrocarbon production data
and fracture identification data) were relatively rare.
0
10000
20000
30000
40000
50000
60000
70000
80000
0 20 40
Hyd
roca
rbo
n P
rod
uct
ion
, bb
ls
Total Number of Fractures
Cum Oil
Cum Gas
0
50000
100000
150000
200000
250000
0 20 40
Hyd
roca
rbo
n P
rod
uct
ion
, bb
ls
Total Number of Fractures
Cum OilCum Gas
Reservoir Characterization (Task 2.2): Fracture Mapping of the Clinton sandstone and Copper Ridge dolomite
12 of 12
For more information, please refer to "CO2 Utilization for Enhanced Oil Recovery and
Geologic Storage in Ohio, Task 2: Reservoir Characterization Topical Report.," OCDO
Grant/Agreement OER-CDO-D-15-08, Columbus, 2017.