Beruflich Dokumente
Kultur Dokumente
BACHELOR THESIS
Abstract
Characterizing and predicting reservoir performance need to be done in order to improve reservoir management
decision. Time consuming and data uncertainty make the numerical simulators less preferable for a quick reservoir
evaluation. Capacitance-resistance model (CRM) proved to be a quick and reliable tool to evaluate waterflood
performance using just production and injection historical data to perform history matching. The CRM
characterize reservoir by quantifying the interwell connectivity and response delay that constitute the CRM
unknown parameters. In this study, the CRM was used to characterize and predict waterflooded reservoir
performance. The CRM was applied to four synthetic reservoir models with different complexities to investigate
the CRM responses toward the reservoir heterogeneity. The result showed that the CRM was able to infer the
reservoir heterogeneity and match the synthetic historical data within more than 0.9 R-squared. The calibrated
CRM model then coupled with fractional flow models to match the oil production performance. Once the oil
production matched, the model then used to predict the production performance and maximize the amount of oil
produced by reallocating water injection rates. To validate the CRM prediction, the results were tested against
numerical simulation results. The result showed that the CRM was able to perform performance prediction and
maximize the amount of oil produced by reallocating the injection rates.
Sari
Karakterisasi dan prediksi performa reservoir perlu dilakukan untuk meningkatkan keputusan dalam manajemen
reservoir. Lamanya waktu dan ketidakpastian data lapangan membuat simulasi reservoir kurang menarik
dilakukan untuk mengevaluasi kinerja reservoir secara cepat. Capacitance-resistance model terbukti menjadi
model yang cepat dan terpercaya untuk memprediksi karakteristik dan performa reservoir hanya dengan
menggunakan data historis injeki dan produksi untuk melakukan history matching. CRM mengkarakterisasi
reservoir dengan menghitung konektivitas dan respon delay antara pasangan sumur injeksi yang menjadi
parameter yang tidak diketahui. Pada studi kali ini CRM digunakan untuk mengkarakterisasi dan melakukan
prediksi performa reservoir injeksi air. Model CRM diterapkan pada empat model reservoir buatan dengan
kompleksitas yang berbeda untuk melihat respon dari CRM terhadap heterogenitas reservoir. Hasilnya
memperlihatkan bahwa CRM mampu memprediksi karakeristik dan mencocokkan data historis dengan lebih dari
0.9 R-squared. Model CRM yang telah terkalibrasi kemudian dikombinasikan dengan model fractional flow untuk
mencocokkan performa laju alir minyak. Setelah laju alir minyak cocok, model digunakan untuk memprediksi
performa dan memaksimalkan produksi minyak dengan melakukan realokasi laju sumur injeksi. Untuk
memvalidasi prediksi model CRM, hasilnya dicocokkan dengan simulator reservoir. Hasilnya menunjukkan
bahwa CRM mampu melakukan prediksi dan memaksimalkan produksi minyak dengan melakukan realokasi pada
laju alir sumur injeksi.
*) Student of Petroleum Engineering Study Program, Institut Teknologi Bandung, 2015 batch
**) Thesis Adviser in Petroleum Engineering Study Program, Institut Teknologi Bandung
Some of the main assumptions in derivation of the
1. Introduction basic CRM differential equation and solutions are
(Sayarpour, 2008): 1) constant temperature, 2) slightly
Characterizing reservoir and predicting reservoir
compressible fluids, 3) negligible capillary pressure
performance have always been the most challenging
effects, 4) constant volume with instantaneous
tasks for petroleum engineers. The engineer must be
pressure equilibrium, 5) constant productivity index.
able to choose the right method, while considering the
In this work, CRM based on the producer (CRMP)
time, resources, and the availability of the data.
is used as the predictive model.
Numerical simulation is the standard approach for
modelling fluid flow in reservoir, providing insight 2.1 CRMP: One time constant for each producer
about reservoir behavior and characteristics, therefore
use to forecast the reservoir performance. But For a control volume around a producer. Liang et
numerical simulations are complex and time al. (2007) presented the governing differential
consuming. Therefore, there was a trigger for equation for this capacitance model by
petroleum engineers to develop a simple predictive
𝑑𝑞𝑗 (𝑡) 1 1 𝑁𝑖𝑛𝑗
model, which usually use material balance on a + 𝑞𝑗 (𝑡) = ∑𝑖=1 𝑓𝑖𝑗 𝑖𝑖 (𝑡) −
𝑑𝑡 𝜏𝑗 𝜏𝑗
reservoir to evaluate its performance. These simple 𝑑𝑃𝑤𝑓,𝑗 (1)
models could be a preliminary estimation of reservoir 𝐽𝑗
𝑑𝑡
characteristics and performance with just a minimum
Where 𝜏𝑗 , is producer j’s time constant,
amount of data.
𝑐𝑡 𝑉𝑝
Capacitance resistance model (CRM) is a quick tool to 𝜏𝑗 = ( ) (2)
evaluate reservoir performance without complex and 𝐽 𝑗
time-consuming reservoir simulations and requires
only historical rate data that is collected during water And pore volume, 𝑉𝑝 , total compressibility, 𝑐𝑡 , and
flooding. productivity index, 𝐽, are producer 𝑗 parameters in its
effective area; the 𝑓𝑖𝑗 term, connectivity, represent the
steady-state fraction of the rate of injector 𝑖 flowing
2. Capacitance-Resistance Model
toward producer 𝑗.
Yousef et al. (2006) introduced capacitance-
resistance model (CRM), a nonlinear data-driven 𝑞𝑖𝑗 (𝑡)
𝑓𝑖𝑗 = (3)
model that measures the communication between well 𝑖𝑖 (𝑡)
in waterflooded reservoir. This essential reservoir For a case of series of step variations of injection rate
information is obtained from the analysis of the well (𝑘)
(SVIR) during time interval 𝛥𝑡𝑘 , 𝑖𝑖 (𝛥𝑡𝑘 ) = 𝐼𝑖 , for
production and injection history without the need for all the injectors, and a constant productivity index, and
the detailed geological information. CRM considers series of linear variations of bottom hole pressure
the effect of capacitance (compressibility) and
(LVBHP) for producer 𝑗. One-time interval solution
resistance (transmissibility), which correspond to two
for a step change in injection rate
parameters, respectively: The degree of fluid storage
(time constant, 𝜏 ) and the degree of connectivity ∆𝑡
(− 𝑘 )
𝜏𝑗
(interwell connectivity, 𝑓 ) between wells. Interwell 𝑞𝑗 (𝑡𝑘 ) = 𝑞𝑗 (𝑡𝑘−1 )𝑒 + (1 −
connectivity ( 𝑓𝑖𝑗 ) also known as gain or allocation (4)
∆𝑡 (𝑘)
factor, is defined by the fraction of injected fluid from −( 𝑘 ) 𝑁𝑖𝑛𝑗 (𝑘) ∆𝑝𝑤𝑓,𝑗
𝜏𝑗
𝑒 ) [∑𝑖=1 [𝑓𝑖𝑗 𝐼𝑖 ] − 𝐽𝑗 𝜏𝑗 ]
injector 𝑖 to the production 𝑗. On the other hand, time ∆𝑡𝑘
𝑷𝟏 𝑷𝟐 𝑷𝟑 𝑷𝟒
𝒇𝟏𝒋 0.946 0.031 0.017 0.005
𝒇𝟐𝒋 0.565 0.051 0.210 0.173
𝒇𝟑𝒋 0.185 0.061 0.046 0.708
𝒇𝟒𝒋 0.206 0.120 0.000 0.674
𝒇𝟓𝒋 0.184 0.043 0.147 0.626
𝝉𝒋 , days 0.836 3.495 4.588 3.086
𝑷𝟏 𝑷𝟐 𝑷𝟑 𝑷𝟒 𝑷𝟏 𝑷𝟐 𝑷𝟑 𝑷𝟒
𝜶𝒋 5.56E-03 1.25E-12 3.02E-12 4.32E-10 𝑲𝒗𝒂𝒍 43.845 2.543 2.561 6.899
𝜷𝒋 0.435 2.018 1.910 1.482 𝑽𝒑 (𝑹𝑩) 1.04E+ 9.88E+ 1.37E+ 4.02E+
07 05 06 06
Figure 12. Total injection rate for Complete sealing barrier case
5.2 Synfield-3: Complete Sealing Barrier
𝑷𝟏 𝑷𝟐 𝑷𝟑 𝑷𝟒
𝒇𝟏𝒋 0.987 0.011 0.000 0.001
𝒇𝟐𝒋 0.000 0.002 0.814 0.184
𝒇𝟑𝒋 0.001 0.985 0.014 0.000
𝒇𝟒𝒋 0.000 0.999 0.000 0.001
Figure 11. Complete sealing barrier model 𝒇𝟓𝒋 0.014 0.000 0.444 0.542
This reservoir was produced for 8 years. The 𝝉𝒋 , days 7.308 29.384 10.019 13.239
bottomhole pressure at the producers is kept constant
at 2500 psia. Fig. 12. shows the individual well
injection rate, large perturbation in injection rate was
created in order to mimic the field conditions.
connectivity to both of producer 𝑃3 and 𝑃4 . From
these results we can draw conclusion that the reservoir
was compartmentalize by three regions and these
conclusions is corresponding to the actual reservoir
model.
Figure 26. Schematic of connectivity map which has the value of fij
greater or equal than 0.1
Figure 24. Total production match and validation of CRMP for
heterogeneous case
Eliminating the value of 𝑓𝑖𝑗 which is lower than 0.1
The results shown a good fit both in history matching resulting the connectivity map shown in Fig. 26. We
and validation with R-squared is 0.999 and 0.998 can say that the region inside the pattern representing
a strong connectivity between well pairs, it is also Using Koval model, the fitting period shown a good fit
quite relatable with the high value of the horizontal with R-squares of 0.991, but on the validation period
permeability (red color). the Koval model did not accurately predict the oil
production rate as shown in Fig. 28, this is quite
relatable since the accuracy of the Koval model to
predict the oil production rates lies in the early stage
of the waterflood. Table 8. shows the results of the
Koval fractional flow fitting parameters.
Table 8. Koval model parameters for heterogeneous case
𝑲𝒗𝒂𝒍 𝑽𝑷 (𝑹𝑩)
𝑷𝟏 3.532 1.87E+05
𝑷𝟐 6.067 5.88E+05
𝑷𝟑 4.211 1.84E+05
𝑷𝟒 4.043 2.47E+05
𝑷𝟓 6.423 1.64E+06
Figure 27. Oil production match and validation using semi- 𝑷𝟔 5.039 4.74E+05
empirical fractional flow model
𝑷𝟕 4.460 2.37E+05
After matching the total production, the calibrated 𝑷𝟖 4.738 3.10E+05
CRM model then coupled with fractional flow model 𝑷𝟗 4.531 2.69E+05
to predict the oil production rate. Fig. 27. shows the
results of matching and validation using semi- Both the semi-empirical model and the Koval model
empirical fractional flow model. The result shown a complete each other, the semi-empirical model can
good fit between the predicted value and the simulated accurately predict the mature waterflood, on the other
value with R-squares value of 0.966 for the history hand the Koval model can accurately predict the early
matching and 0.982 for the validation. Table 7. shows stage of waterflooding.
the results of the semi-empirical fractional flow fitting
parameters. 5.4.3 Optimization
Table 7. Semi-empirical fractional flow model parameters for After obtaining the oil production performance from
heterogeneous case
the CRM and fractional flow model. We used the
calibrated model to perform an optimization by
𝜶𝒋 𝜷𝒋
reallocating the injection rates between injector. The
𝑷𝟏 3.6E-07 1.277 objective function for this optimization is to maximize
𝑷𝟐 2.0E-06 1.071
the amount of oil produced by constraining the same
𝑷𝟑 1.5E-05 0.995
amount of total field injection rates. The optimization
𝑷𝟒 2.6E-07 1.285
is based on 1 years of production from days 2904 until
𝑷𝟓 3.4E-06 0.958
𝑷𝟔 6.7E-07 1.157 3287. Table 9. Shows the injection rate for the five
𝑷𝟕 5.0E-07 1.238 injectors before the optimization, the total injection
𝑷𝟖 4.6E-06 1.054 rate was at 8000 STB/D.
𝑷𝟗 1.3E-06 1.165 Table 9. Well injection rate before optimization
Injection rate, STB/D Figure 30. Oil saturation map after 8 years of production
Time, Days
I1 I2 I3 I4 Fig. 30 shows the simulator results regarding to the
2904 0 3000 2000 3000 remaining oil saturation after 8 years of production.
2931 0 3000 2000 3000 Some areas (Red) still shows very high remaining oil
saturation meaning that there are still many unsweep
2991 0 3000 2000 3000
oil. Counterintuitive activation of injector 𝐼3 ,
3091 0 3000 2000 3000 associated with low value of connectivity, is explained
3101 0 3000 2000 3000 by existence of remaining oil around 𝐼3 after 8 years
3114 0 3000 2000 3000 of production.
3133 0 3000 2000 3000
3162 0 3000 2000 3000
3191 0 3000 2000 3000
3239 0 3000 2000 3000
3287 0 3000 2000 3000
In this study we have seen the CRM results regarding 𝑓 = Interwell connectivity
the four synthetic case studies. CRM was able to 𝑞(𝑡) = Total liquid production, STB/D
qualitatively characterizes reservoir from the 𝑞𝑜 (𝑡) = Oil production rate, STB/D
parameter of interwell connectivity and response delay. 𝐼(𝑡) = Total injection production, STB/D
From the streak case we have seen that CRM 𝑐𝑡 = Total reservoir compressibility, psi-1
accurately predict the presence of permeability streak 𝑉𝑝 = Pore volume, RB
in the reservoir given by the high value of connectivity 𝐽 = Productivity index, STB/psi
between I1 (or I3) and P1 (or P4) and a relatively 𝜏 = Time constant, days
small value of time constant indicating a quick 𝜇 = viscosity, cp
response between the injector signal to the producer 𝐸 = Effective oil-solvent viscosity ratio
𝐻 = Heterogeneity factor
production. For the complete sealing barrier case, the
𝐾𝑣𝑎𝑙 = Koval factor
reservoir compartment was inferred from the zero
𝑆 = Normalized average water saturation
value of 𝑓𝑖𝑗 indicating no communication between the 𝑓𝑤 = Fractional flow of water
well pairs. Same results for the partially sealing barrier 𝑡𝐷 = Dimensionless water
case, the low transmissibility of the well pairs because 𝑥𝐷 = Dimensionless distance
of the presence of the partially sealing barrier could be obs = Observed data
inferred from the value of 𝑓𝑖𝑗 . Combining CRM with est = Estiated data
the fractional flow model gives us the oil production 𝑁𝑑𝑎𝑡𝑎 = Number of data points
performance which is later can be used to perform MSE = Mean square error
prediction and optimization. The optimization 𝛼 = power-law coefficient for semi-
resulting in a better swept efficiency and increase in empirical fractional flow model
𝛽 = power-law coefficient for semi-
oil production by reallocating the injection rate.
empirical fractional flow model
CWI = Effective cumulative water injected in
7. Conclusion the control volume, bbl/D
WOR = Water-oil ratio
1. Different case studies showed that CRM are
capable of inferring reservoir heterogeneity.
2. CRM are capable to perform history
11. References
matching and predict reservoir performance.
3. CRM are capable to maximize the amount of
oil produced by reallocating the injection Cao, F., Luo, H., & Lake, L. W. (2015). Oil Rate
rates. Forecast by Inferring Fractional Flow Models
from Field Data, (February), 23–25.
https://doi.org/10.2118/173315-ms
8. Recommendation
Sayarpour, M., Zuluaga, E., Kabir, C. S., & Lake, L.
1. The applicability of CRM in this work is W. (2009). The use of capacitance-resistance
limited by synthetic case study, furthermore models for rapid estimation of waterflood
it can be tested on the real field data. performance and optimization. Journal of
2. Different optimization techniques may be Petroleum Science and Engineering, 69(3–4),
applied to improve the current solutions. 227–238.
3. Mapping the connectivity map using more https://doi.org/10.1016/j.petrol.2009.09.006
sophisticated method rather than manually
creating line and arrow as representative. De Holanda, R. W., Gildin, E., Jensen, J. L., Lake, L.
W., & Shah Kabir, C. (2018). A state-of-the-art
literature review on capacitance resistance
9. Acknowledgement models for reservoir characterization and
The author is extremely indebted to Dr. Amega performance forecasting. Energies, 11(12).
https://doi.org/10.3390/en11123368
Yasutra, one of the distinguished lecturers in the
Department of Petroleum Engineering, Bandung
Institute of Technology, for his patience and guidance Eshraghi, S. E., Rasaei, M. R., Pourafshary, P., &
Salar, A. (2016). Characterization of Reservoir
throughout the work of this project. Without his
Heterogeneity by Capacitance-resistance Model
valuable assistance, this work would not have been
in Water-flooding Projects, 5(2), 1–13.
completed.
Sayarpour, M. (2008). Development and Application
of Capacitance-Resistive Models to Water / CO
2 Floods. Dissertation, 237.
Figure 33. Log (CWI) and Log (WOR) plot for streak case
Figure 34. Individual oil production match for streak case using semi-empirical fractional flow model
Figure 35. Watercut match for streak case using Koval model
Figure 36. Individual oil production match for streak case using Koval model
Figure 37. Log (CWI) and Log (WOR) plot for heterogeneous case
Figure 38. Oil production match for heterogeneous case using semi-empirical fractional flow model
Figure 39. Watercut match for heterogeneous case using Koval model
Figure 40. Oil production match for heterogeneous case using Koval model