Beruflich Dokumente
Kultur Dokumente
net/publication/315890262
CITATIONS READS
27 184
7 authors, including:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Zhenxue Dai on 28 July 2018.
Applied Energy
journal homepage: www.elsevier.com/locate/apenergy
h i g h l i g h t s
a r t i c l e i n f o a b s t r a c t
Article history: This paper presents an integrated numerical framework to co-optimize EOR and CO2 storage performance
Received 26 December 2016 under uncertainty in the Farnsworth Unit (FWU) oil field in Ochiltree County, Texas. The framework
Received in revised form 11 February 2017 includes a field-scale compositional reservoir multiphase flow model, an uncertainty quantification
Accepted 2 March 2017
model and a neural network optimization process. The reservoir flow model has been constructed based
on the field geophysical, geological, and engineering data. Equation of state parameters were tuned to
achieve field measured fluid properties and subsequently used to predict the minimum miscible pressure
Keywords:
(MMP).
Enhance oil recovery
Co-optimization
A history match of primary and secondary recovery processes was conducted to estimate the reservoir
Uncertainty and multiphase flow parameters as the base case for analyzing the effect of recycling produced gas, infill
CO2 storage drilling and water alternating gas (WAG) cycles on oil recovery and CO2 storage. A multi-objective opti-
Response surface method mization model was defined for maximizing both oil recovery and CO2 storage. The uncertainty quantifi-
Neural network cation model comprising the Latin Hypercube sampling, Monte Carlo simulation, and sensitivity analysis,
was used to study the effects of uncertain variables on the defined objective functions. Uncertain vari-
ables include bottom hole injection pressure, WAG cycle, injection and production group rates, and
gas-oil ratio. The most significant variables were chosen as control variables to be used for the optimiza-
tion process. A neural network optimization algorithm was utilized to optimize the objective function
both with and without geological uncertainty. The vertical permeability anisotropy (Kv/Kh) was selected
as one of the uncertain parameters in the optimization process.
The simulation results were compared to a scenario baseline case that predicted CO2 storage of 74%.
The results showed an improved approach for optimizing oil recovery and CO2 storage in the FWU.
The optimization model predicted that about 94% of CO2 would be stored and most importantly, that this
increased storage could result in about 25% of incremental oil recovery. The sensitivity analysis reduced
the number of control variables to decrease computational time. A risk aversion factor was used to rep-
resent results at various confidence levels to assist management in the decision-making process. The
defined objective functions were shown to be a robust approach to co-optimize oil recovery and CO2
storage.
Published by Elsevier Ltd.
⇑ Corresponding author.
E-mail address: wampomah04@gmail.com (W. Ampomah).
http://dx.doi.org/10.1016/j.apenergy.2017.03.017
0306-2619/Published by Elsevier Ltd.
W. Ampomah et al. / Applied Energy 195 (2017) 80–92 81
1. Introduction addition of new patterns and/or infill drilling on oil recovery and
carbon sequestration. The study illustrated the significance of dif-
CO2 enhanced oil recovery (EOR) is a long proven technology for ferent injection strategies in achieving higher oil recovery and
improving production from mature oil fields. In recent years, there CO2 storage, but examined a case with fixed well bottomhole pres-
has been an increasing global demand for the mitigation of atmo- sures and WAG cycles. The model was scenario-based; it did not
spheric greenhouse gases such as CO2 [1–10]. The CO2-EOR process include uncertainty in operational parameters that might have sig-
has the potential of both improving oil recovery and sequestering nificant effect on both oil recovery and CO2 storage.
large volumes of injected CO2 thereby reducing greenhouse gas This paper presents a full field numerical based optimization
emissions [11–13]. Using anthropogenic CO2 captured from other model under geological uncertainty with a multi-objective func-
industrial process can decrease the amount of CO2 taken from nat- tion to improve oil recovery and CO2 storage within FWU. Fluid
urally occurring reservoirs for EOR processes [14]. This type of pro- analysis, an initial history matched model, and developmental
ject is often referred to as CCUS, or Carbon Capture, Utilization, and strategies adapted from the SWP improved dynamic model are
Storage. The Southwest Regional Partnership on Carbon Sequestra- used in this study [16,40,41]. Initial sensitivity simulation runs
tion (SWP) is one of seven US Department of Energy sponsored delineate between more and less sensitive control variables within
regional partnerships charged with helping to develop technolo- several CO2-EOR operational parameters prior to optimization. A
gies to implement large scale CO2 storage. SWP is currently partic- design of experiment approach using a Latin hypercube sampling
ipating in a project in partnership with Chaparral Energy LLC to algorithm is used to construct a surrogate model to reduce compu-
study a relatively recent commercial-scale CO2 EOR project at tational time. A quadratic polynomial reduced order model is opti-
Farnsworth Unit, Texas. Overall project goals include monitoring mized under geological uncertainty using the NN optimizer. The
and verifying injection and storage of 1,000,000 metric tonnes of novel approach presented in this paper illustrates the quantifica-
CO2 during a 5-year injection period and subsequent 3 year post- tion of incremental oil recovery and CO2 storage in the presence
injection period. of uncertainty.
In CCUS projects, there are two objectives that must be bal-
anced; flood performance to enhance the profitability of the EOR 2. Methodology
project, and storage performance to optimize the long term reduc-
tion of CO2 emissions [15,16]. Numerical simulation results are In this paper a numerical experiment is performed to co-
valuable tools for evaluating the performance of CO2-EOR. How- optimize operational variables and maximize incremental oil
ever, static parameter (e.g. porosity and permeability) and dynamic recovery and CO2 storage volume. The operational variables exam-
parameter (e.g. relative permeability and capillary pressure curves) ined include WAG cycles, WAG ratio, injector and producer bot-
uncertainties strongly affect the simulation results. tomhole pressures, and CO2 compressor capacity. Some geologic
Uncertainties in static parameters can be reduced by history uncertainty is also included. The geologic model was described
matching, however the sensitivity of operational parameters should previously by Ampomah et al. [42].
be incorporated into reservoir performance modelling. Slight The objective function is defined to maximize both oil recovery
changes in flood design parameters can cause dramatic changes in and CO2 storage. The oil production vector is defined as cumulative
predicted oil recovery and overall project net present value (NPV). oil production (Eq. (1)). The CO2 storage is defined as the amount of
To decrease these uncertainties, finding optimized numerical results CO2 purchased minus the amount of CO2 produced plus the
using production optimization could be an effective way [17]. amount of CO2 recycled (Eq. (2)). In this work, an equal unit weight
EOR Optimization process under uncertainty has been used in was assigned to both vectors. A total of 37 possible uncertain vari-
several studies [18–20]. Recent studies have mainly used EOR ables were initially selected and assigned upper and lower bounds.
optimization based on Markovitz’ classical theory [21]. Other Variables included operational parameters such as WAG cycle, bot-
optimization approaches have also been presented in literature tomhole pressure, gas-oil ratio, and production rate target. Sensi-
[22–24]. Al-Sofi and Blunt [25] used a parallel methodology to tivity analysis with a Latin-hypercube sampling technique was
optimize polymer floods under uncertainty. Gharbi [26] presented conducted to study the uncertain variables’ effects on the objective
an EOR expert approach to optimize reservoir management and CO2 function. Each variable was sampled three times (111 simulations
floods in carbonate reservoirs. Several studies have focused on in total) in order to reduce the number of control variables to 12
optimizing NPV of EOR projects under uncertainty [27–29]. These (including the geological uncertainty) and to ensure the optimum
studies only focused on optimizing oil recovery rather than CO2 solution with fewer simulation runs. The outcomes were ranked
storage. and eleven control variables were selected for proxy modelling
There have been a number of studies addressing co- and/or the optimization process. The distribution ranges for
optimization of oil recovery and CO2 sequestration [30–33]. Some selected control variables were adjusted based on sensitivity anal-
of these studies used limited data and simple cases. Most of them ysis outcomes. A quadratic response surface (proxy or surrogate)
used a sector model to present their methodology. Forooghi et al. model was constructed based on the training data from numerical
[32] used numerical modelling on a sector model to co-optimize simulations. The resulting response surface is a second-order poly-
CO2-EOR and sequestration. The study focused on sensitivity anal- nomial model determined by a least-square fit between the poly-
ysis of operational parameters and how their interactions affected nomial response and training data. A typical quadratic response
the co-optimization process. Several studies in the petroleum surface is shown in Eq. (3). The quadratic model requires a mini-
industry have focused on the use of neural network (NN) concept mum of (N + 1) (N + 2)/2 training data where N represents the
in optimization processes [34–39]. This approach is based on trial number of independent variables. A total of 92 training samples
and error. Mathematical details and the robustness of NN opti- were selected using the Latin hypercube sampling technique. Ten
mization technology were discussed in the literature [23,37]. additional runs were simulated to validate the constructed proxy
Few numerical simulation studies of CO2-EOR performance model. Once a valid proxy model was accepted, it was subse-
have focused on the use of purchased vs. recycled CO2 in their quently simulated in lieu of full reservoir simulations for the opti-
simulations. Ampomah et al. [16] presented a compositional mization process.
scenario-based prediction model of the CO2-EOR process which The neural network (NN) optimizer algorithm was used to
incorporated the effects of CO2 purchase, CO2 recycling and the optimize the objective function with uncertainty. NN is a parallel
82 W. Ampomah et al. / Applied Energy 195 (2017) 80–92
network of interconnected neurons within which each accepts a A history matched study of the FWU suggested that combinations
set of inputs from other neurons and computes an output that is of Kv/Kh values resulted in a successful calibration. Vertical aniso-
propagated to the output nodes. This is known as a multilayer tropy ratios (Kv/Kh) of 0.1 and 0.5 were used to construct individ-
feedforward network, wherein processors are connected in a way ual geological realizations. Note that all other grid properties were
suggestive of connections between neurons (Fig. 1). The neural kept the same (Fig. 2). There were two realizations within each
network is initially trained against N + 1 (where N is the number loop in the optimization run. A maximum of 1200 realizations
of control variables) simulation outputs and once a satisfactory were run with the proxy model for each risk factor and a combined
outcome is realized, the NN objective function proxy is optimized objective function approach introduced by Couet et al. [22] was
and compared with the original objective. An adaptive procedure used. The Couet et al. [17] method incorporates a risk aversion fac-
is used where new control variables and/or objective functions tor (k) illustrating the level of confidence in optimization outcomes
are added to the initial training set to fit the NN system whenever as shown in Eq. (4). For example, a risk aversion factor of k = 0
there is a mismatch. The error is minimized by adjusting the implies that there is 50% probability that the computed objective
weights associated with various interconnections. function is greater than the mean, and a k = 2 represents a 98%
For the optimization with geological uncertainty, vertical per- probability that the computed objective function is higher than
meability anisotropy ratio was selected as an uncertain parameter. the vector mean within a loop. These confidence levels assume
Fig. 2. Two possible vertical permeability scenarios for stochastic reservoir modelling with uncertainty. Left represents a vertical permeability distribution with Kv/Kh = 0.5;
right represents vertical permeability with Kv/Kh = 0.1.
W. Ampomah et al. / Applied Energy 195 (2017) 80–92 83
The field was discovered in 1955 with initial oil in place estimated
Base Model (History Match) at 120 MMStb [42,43]. The field was unitized in 1963 with Unocal
as operator. Water injection for secondary recovery started in
Define Objective Function (OF) 1964. Chaparral Energy, LLC (CELLC) acquired FWU in November
2009 and tertiary CO2 enhanced oil recovery began in December
2010. CO2 for the project is 100% anthropogenically-sourced, and
Select Risk aversion factor (λ)
is purchased from the Arkalon ethanol plant in Liberal, Kansas
and the Agrium fertilizer plant in Borger, Texas [14]. Fig. 4 shows
Select initial control variables the location of FWU, CO2 sources and Chaparral’s pipelines. The
geological description of the field is presented in prior studies
Conduct Sensitivity analysis [43–49]. The target reservoir is Morrow B sandstone at a depth
of 7550–7950 ft with an average dip of less than one degree. The
Update control variables productive limit of the FWU extends laterally to about 8300 acres.
The maximum net pay thickness is 54 ft with an average of 22 ft.
There are no recorded gas-oil and/or oil-water contacts. FWU
Proxy Modeling
reservoir initial pressure prior to commencement of production
was 2203 psi with a bubble point of 2059 psi and temperature of
Select Uncertain Cases (N=am)
168°F measured at a depth of about 7900 ft [16]. The initial oil sat-
uration was 69% and the formation volume factor was 1.192 RB/
Run N simulations and STB. The reservoir was slightly undersaturated with an initial
Compute N# of “OF” gas-oil ratio of 345 Scf/stb.
Ampomah et al. [45] presented the initial SWP geological static
Compute Combined OF(F) No model for FWU in 2015. The static model has been used by various
Optimizer researchers to study flow and transport simulation in different
routine context (e.g., hydrogeology, reactive transport, and risk assessment
Is F optimal? of the project) [50–55]. This previous model did not include newly
interpreted faults which was previously unknown. A geostatistical
Yes tools were used in populating reservoir properties without much
constraints. An improved FWU static model subsequently incorpo-
Stop rated a more accurate interpretation of reservoir structure and
stratigraphy using 3-D seismic data integrated with well log and
Fig. 3. Flow chart showing the detailed procedure for optimization under uncer- core data [42,44,47]. Interpretation of 3D seismic data collected
tainty with proxy modelling. for this project identified several previously unknown faults that
were included in the second-generation model. This model also
incorporates results from a hydraulic flow unit (HFU) study to
outcomes to be normally distributed. A flow chart summarizing
describe and characterize the heterogeneity within the target
the optimization under uncertainty procedure is shown in Fig. 3.
reservoir [44]. Eight distinct flow units were identified (Fig. 5)
There are two main constraints that guide the optimization defini-
within the reservoir. The porosity and permeability relationships
tion. In order for a solution to be classified as optimum, it has to
seen in the HFU study were related back to depositional and diage-
meet these two criteria including: (i) reservoir pressure, P has to
netic features noted in core, log and thin section data and descrip-
be more than 3700 psia and less than 5600 psia
tions [42,43,46,47] and a constructed facies model was used to
(3700 < P < 5600), and (ii) the injection target (recycled CO2 vol-
condition and propagate porosity and permeability as a function
ume plus purchased CO2) at any time must be no greater than
of HFU [44].
20,000 MScf/d, the maximum compressor capacity.
An upscaled model extracted from the west half of SWP’s FWU
Objective function ðFÞ ¼ w1 Cumulative Oil Production model has a grid size of 176 163 8 with block dimensions of
100 ft 100 ft. The total number of cells within the model is
þ w2 CO2 Storage ð1Þ
229,504. The Morrow B is the primary reservoir sand and is subdi-
where vided into four (4) layers. The sand is overlain and underlain by
low permeable Morrowan shales. The Morrow shale above the
CO2 storage ¼ CO2 purchased CO2 produced þ Recycle ð2Þ Morrow B sand is overlain by Thirteen Finger Limestone. Together,
X XX X these form an excellent caprock to prevent upward CO2 migration
yðxÞ ¼ b0 þ bi xi þ bij xi xj þ bii x2i ð3Þ [56]. Some preliminary caprock analysis studies indicate that these
i i j i sealing formations are excellent seals capable of providing storage
of supercritical CO2 up to a column height of 10,000 ft. A type log
where y is the output response, b is the regression coefficients, and x
for FWU is shown in Fig. 5. Porosity distribution in the reservoir
is a vector of N inputs. Note that the first, second, and third term
has a range of 9.2–24% with a mean of 14.6%. Permeability ranges
represents linear, interaction and quadratic terms respectively.
from 0.01 mD to 181 mD with an average value of 58 mD.
Combined Objective Function ðCFÞ ¼ l k r ð4Þ Gunda et al. [40] provided a comprehensive reservoir fluid anal-
ysis on a FWU oil sample which has been used in various SWP
k is the risk aversion factor, r is the standard deviation, and l is the reservoir simulation models such as Ampomah et al. [16]. Reser-
mean. voir fluid was tuned to Peng Robinson EOS [57] with a Peneloux
volume correction [58]. The Lohrenz-Bray-Clark (LBC) correlation
3. Model description was used for the calculations of viscosities [59]. After tuning PVT
properties, a slim tube simulation experiment conducted on FWU
Farnsworth Unit field (FWU), located in Ochiltree County Texas, fluids resulted in a minimum miscible pressure (MMP) of about
is a mature hydrocarbon reservoir undergoing tertiary production. 4009 psia compared to experimental value of 4200 psia, represent-
84 W. Ampomah et al. / Applied Energy 195 (2017) 80–92
Fig. 4. Shows location of FWU as well two anthropogenic CO2 sources and various pipelines for transportation.
Fig. 5. (Left) Porosity-permeability relationships constructed for property modelling as a function of eight (8) distinct hydraulic flow units. (Right) Type log of FWU depicting
reservoir and caprock that were used in constructing structural model used in this study.
ing an error of 5% [40]. Sensitivity analyses were conducted to interpreted faults within the reservoir are open to flow; however,
study the effect of using recycled CO2 with or without impurities they are closed within the Shale layers.
on MMP predictions [40]. These analyses enabled modelling of The primary and secondary recovery processes lasted for
CO2-EOR performance incorporating the real world use of both 55 years, from 1955 until December 2010. The simulated tertiary
purchased and recycled CO2. CO2 flood was performed between December 2010 and July 2016.
The current dynamic history matched model used in this work From the simulation results, nearly 6 MMstb of oil was produced
is based on the improved static model that incorporates faults and during primary depletion. A total of 25.65 MMstb of cumulative
reservoir heterogeneity. It includes more primary production wells oil was produced by the end of secondary recovery from the west
within the field than were used in the initial history matched half of FWU, representing nearly 40% of OOIP. Various waterflood
model for FWU [16], as well as three of the main faults. A Sperrevik patterns swept the reservoir to a residual oil saturation of about
model was used to model fault transmissibilities which is based on 27% which could be considered as a successful flood. These results
shale gouge computations [60]. Results indicate that most of the are not unexpected given that there is enough mobility ratio (1.6)
between water and oil, and high injection throughout of at least 1.7
W. Ampomah et al. / Applied Energy 195 (2017) 80–92 85
displaceable pore volumes. Model parameters (mainly permeabil- 4.2. Sensitivity analysis and reduced order modelling
ity) were adjusted to history matched primary and secondary
recovery processes within an error margin of less than 10%. During Sensitivity analysis is used to better understand the effect of
the CO2 flood history match, initial preliminary analysis showed a each parameter on the objective function and reduce the number
good calibration of simulation results to historical data until CO2 of control variables in the optimization process. The main uncer-
breakthrough when there was misfit between simulated and his- tain variables in this work include: gas-oil ratio constraints
torical results. This could result from a potential alteration in wet- (GOR), group production target rates, bottomhole pressures for
tability and changes in fluid interfacial tension due to CO2 producer and injector wells, and WAG cycles. A tornado plot show-
injection. Therefore, it was necessary to adjust Corey parameters ing the top twelve sensitive uncertain parameters including verti-
[61]. Additional 2.6 MMstb of oil has been recovered since tertiary cal permeability anisotropy is shown in Fig. 6. Important control
stage at FWU. variables are listed and defined in Table 2.
The most important aspect of work from the SWP’s perspective
is the amount of CO2 stored within the Morrow B reservoir. A total 4.2.1. Gas–oil ratio (GOR)
of 30 BScf of CO2 was injected, including both purchased and recy- GOR is an important parameter in CO2 flood predictions because
cled gas. About 93% of the purchased CO2 was stored as of July CO2 has higher mobility than oil at reservoir condition. The com-
2016. These results suggest that CO2 sequestration (one of the pri- pressor recycling capacity for any field study must be considered
mary SWP objective) is increased by optimization of the WAG in setting GOR constraints. The sampling space limits were selected
process. based on compressor capacity. Sensitivity analysis shows that GOR
prior to year 2020 was the least important parameter because of
lower gas production during the earlier years of a simulation run.
4. Results GOR after year 2020 also had no significant effect on cumulative
oil production. Simulation results show an overall GOR less than
This section presents results for the baseline model, sensitivity 30 Mscf/Stb thus GOR was not influential enough to be selected
analysis, and optimization under uncertainty. as control variable.
Fig. 6. Shows CO2 volume profile for the baseline case. It is evident that not all available CO2 was injected. This could be due to failure to fully optimize control variables. The
gas injection rate is approximately the same as the CO2 injection rate.
Fig. 7. Tornado plot ranking of twelve control parameters including vertical permeability anisotropy (Kv) from the response surface model as a function of cumulative gas
injection and oil production volume. The x represents uncertain parameters and y represents cumulative gas injection volume.
Fig. 8. Cross plot of simulated/trained experiments in addition to validation points for various responses such as oil production volume (upper left), water production volume
(upper right), amount of CO2 stored within the reservoir (lower left) and CO2 production volume (lower right). The R2 values for all experiments are 1. The clustering of the
points around the equiline (trend line) signifies proxy accuracy.
88 W. Ampomah et al. / Applied Energy 195 (2017) 80–92
approach with different risk factors to estimate confidence levels of probability of storing more than 93% of purchased CO2 within
CO2 storage and cumulative oil production. The confidence levels the reservoir. This is more than the 19% storage that is predicted
assumed a normal distribution for all output vectors including by the manual scenario-based model. There is a 98% certainty that
the multi-objective function. The objective function computed for more than 81% of purchased CO2 would be sequestered within the
each case within a loop was combined into a single representative reservoir during the 20 years of the prediction. The CO2 storage
outcome based on the risk aversion factor. It was evident that for potential in both loops was identical since beyond year 2030 all
each risk aversion factor the mean objective function for Kv/ CO2 injected is recycled CO2. Fig. 13 shows mole fractions of CO2
Kh = 0.5 always has a greater value when compared to Kv/Kh = 0.1. component within the reservoir at the end of CO2 history matching
Results from the optimized case predict a 50% chance that (August 31, 2016) and distribution at the last time step of numer-
incremental oil recovery at end of 20 years could be 17.40 MMStb ical simulation (January 1, 2036). At the end of the optimization
or higher as shown in Table 3. The 25% incremental oil recovery process, the CO2 pattern flood area was saturated with a high per-
shows about 4% increase over the baseline simulation model. There centage of CO2 in each grid block.
is a 98% chance of recovering an additional 15 MMstb of oil beyond
primary and secondary recovery process irrespective of the 5. Discussion
changes in the vertical permeability anisotropy. This increment
shows a slight gain of 0.13% over the baseline case. Fig. 9 shows In the future, oilfield operators may receive incentives for their
two realizations of oil production rate profiles from the optimized storage of CO2 within reservoir while using principles of miscibility
case. The uncertainty associated with the oil production forecast to improve oil recovery. There is certainly a trade-off between
was very low until 2030. The effect of vertical permeability aniso- maximizing oil recovery and CO2 storage. Incorporating risk into
tropy on oil recovery is more predominant after 2030. Fig. 10 predictions gives management more informed options to use when
shows cumulative oil production for the two optimum loops. The making decisions about CO2 field operations. Manual optimization
amount of CO2 production in the Kv/Kh = 0.5 case increased to is computationally expensive and tedious, making the process of
the maximum compressor capacity of 20,000 MScf/d. This was limited value in optimization of multiple parameters. It is difficult
due to the increase in cross flow between layers in the simulation to forecast performance when there are so many uncertain opera-
model. Conversely, because of the lowered vertical permeability in tional variables that are almost impossible to consider in manual
the Kv/Kh = 0.1 case, there was a suppression in the amount of CO2 engineering designs. Manual optimization usually results in
production which reduced the amount of available CO2 for injec- assigning constant operating parameters for the duration of the
tion. The later phenomenon resulted in lower oil production com- prediction, as demonstrated in the baseline case described in this
pared to Kv/Kh = 0.5 case. Fig. 11 illustrates oil saturation at the paper. In the current work, a state-of-the-art NN optimization
last time step of the optimization process (January 1, 2036). The algorithm was utilized to improve forecasting performance of a
left figure showing the case with higher vertical permeability CO2–EOR field-scale project.
shows an improved sweep efficiency over the right figure (Kv/ In this work, numerical simulations were used to predict co-
Kh = 0.1) within the active CO2 pattern area. Within the active optimization of recovery and sequestration by constructing a
flood area, the left figure shows extremely low oil saturation com- unique multi-objective function for each prospective scenario. A
pared to the right figure. Potential fluid flow across some faults is surrogate model using experimental design methodology was con-
seen in this figure. The higher saturation areas show extended structed to reduce computational time. The proxy model as com-
boundaries away from the active flood vicinity. These extended pared to full confirmation simulation run using optimum control
areas contributed to the estimated higher oil in place which subse- variables resulted in an error less than 2%. This confirms the valid-
quently lowered percentage of recovery compared to other predic- ity of the proxy modelling approach used in this work. A statistical
tion studies conducted at FWU. One of the important constraints approach assuming a normal distribution of a risk aversion factor
utilized in this study was average reservoir pressure. An optimum was used to compute a combined objective function. Important
case has to meet the pressure criteria which must always be above vectors such as cumulative oil production and CO2 storage were
the experimental MMP 3700 psia to improve oil recovery as well derived from the best optimum solution, also assuming a normal
as been below the maximum fracture pressure of 5600 psia. distribution.
Fig. 12 shows the pressure profile for both optimized cases. The An initial sensitivity analysis required a total of 111 simula-
lower pressure profile for loop Kv/Kh = 0.5 beyond year 2030 was tions. A single realization required about 1.5 h in a PC having a
expected due to the increase in oil production from this period. CPU of 3.7 GHz. There are limitations in using fewer sampling
However, the least pressure of 4100 psia was above the points within a sample space. A typical Monte Carlo simulation
pre-defined constraint. The pressure was nearly constant for loop run requires hundreds to thousands of simulations to accurately
Kv/Kh = 0.1 case. compute effects on objective functions. For a full field-scale run
CO2 storage is an important objective of the SWP Phase III pro- there must be a compromise between accuracy and runtime. The
ject. The scenario-based model, that served as a baseline model Monte Carlo with Latin hypercube sampler sampled three times
predicted that 74% of purchased CO2 could be stored. Table 4 within the ranges of operationally uncertain variables to ascertain
shows detailed decisions at various confidence levels. The confi- their effect on overall objective function. This process can intro-
dence levels can guide management in weighing risk associated duce some errors in selecting highly uncertain parameters as con-
with each operational strategy to ensure co-optimization of trol variables for optimization design. Experience gained from
oil recovery and CO2 storage. From the table, there is a 50–84% previous manual optimization designs gave some insights in
Table 3
Summary of uncertainty associated with oil production at different confidence levels compared to baseline case.
Risk Factor, k Confidence level Combined Vector, MMSTB Incremental Recovery, % Gains %
0 50 17.40 24.51 3.51
1 84 17.06 24.03 3.03
2 98 15.00 21.13 0.13
W. Ampomah et al. / Applied Energy 195 (2017) 80–92 89
Fig. 9. Oil production profile for two realizations based on optimum solution with different vertical anisotropic permeability.
Fig. 10. Oil production volume profile for two realizations based on optimum solution with different vertical permeability anisotropy.
Fig. 11. Left figure shows oil saturation distribution at the end of 20 years of the numerical simulation for loop Kv/Kh = 0.5. The figure shows internal area which constitutes
the flood pattern locations has been swept to the minimum oil saturation. This complements the high sweep efficiency and increased production recorded for this case. Right
figure shows oil production for loop Kv/Kh = 0.1. The interior flood area had a slight lower sweep efficiency compared to left figure.
90 W. Ampomah et al. / Applied Energy 195 (2017) 80–92
Fig. 12. Pressure profile for two realizations based on optimum solution with different vertical permeability anisotropy. The pressure profile depicts a high response from the
WAG cycles as well as production.
Table 4
Summary of uncertainty associated with CO2 storage at different confidence levels compared to baseline case.
Risk factor, k Confidence level Combined vector, BScf CO2 Stored, % Gains %
0 50 57.63 93.76 19.45
1 84 54.5 88.65 14.34
2 98 49.84 81.09 6.78
Fig. 13. Left figure shows distribution of CO2 fractions within the Morrow B sandstone reservoir at the end of CO2-WAG history match. This was part of data used to initialize
the optimization process. The red regions with the highest concentration of CO2 are near the location of CO2 injection wells during the history matching efforts. Right figure
shows the CO2 mole fraction distribution at the last time step of optimization process. This plot was identical for both loops.
understanding the potential effects of each parameter on the tain parameter. In order to reduce expensive computational time,
objective. This procedure helped us in choosing possible ranges two geological realizations were used. Each of the chosen points
of each uncertain parameter to reduce errors in the sensitivity was initially tested during a history-matching process, resulting
study. The control variables were reduced to twelve to make com- in successful calibration of the simulation model. The objective
putational cost feasible and to insure convergence. function changes approximated a linear behavior within the neigh-
During the optimization design, it is imperative to consider the borhood of the uncertain parameter’s mean. The optimization out-
source of uncertainty in order not to invalidate a history matched come yielded optimum operational constraints to ensure a high
model. A large number of uncertain variables can reduce the accu- percentage of oil recovery without compromising on CO2 storage.
racy of prediction models and may require a long computational One important aspect of a successful CO2-EOR project is the
time for solutions to converge. The possible ranges for chosen maintenance of reservoir pressure above MMP and below the
uncertain parameters also are critical. Uncertain parameters reservoir fracture pressure. It is therefore necessary to keep track
should be chosen with experience derived from a history matching of pressure changes to always operate within this target range.
process to avoid inconsistencies. In this work, knowledge from the At the early stages of prediction one will argue for increasing the
history matching process was used in selecting Kv/Kh as the uncer- oil production group target, however, this should not be done at
W. Ampomah et al. / Applied Energy 195 (2017) 80–92 91
the expense of maintaining MMP. This requires attention to opti- recommendation, or favoring by the United States Government
mizing BHP and production rate. An additional aspect of CCUS pro- or any agency thereof. The views and opinions of authors
jects is maximizing available CO2 usage to ensure both a high expressed herein do not necessarily state or reflect those of the
percentage of CO2 storage and optimal oil recovery. This work United States Government or any agency thereof.
showed that incorporating new patterns, reducing CO2 purchase
over time and increasing the recycled CO2 percentage and above
all optimizing a few key operational variables resulted in more Unit conversions
than 94% CO2 storage and 25% incremental oil recovery beyond
waterflood period at FWU. 1 Mscf = 2.635 lb-mol.
The work applied a framework to co-optimize CO2 storage and 1 lb-mol = 1.977E-05 Metric tonnes.
oil recovery using the multi objective function. This methodology 1 psi = 6894.76 Pa.
can be adapted and used to improve CO2 storage within other large 1 scf = 0.0283168 m3.
scale field projects without compromising on oil recovery. The 1 bbl = 0.1589873 m3.
approach can also be used in different facets of EOR projects and
as well as with other engineering and/or science disciplines.
Acknowledgements
[19] Raghuraman B, Couët B, Savundararaj S, et al. Valuation of technology and [41] White MD, McPherson BJ, Grigg RB, Ampomah W, Appold MS. Numerical
infor-mation for reservoir risk management. SPE Res Eval Eng 2003;6 simulation of carbon dioxide injection in the western section of the
(5):307–16. http://dx.doi.org/10.2118/86568-PA. SPE 86568-PA. Farnsworth unit. Energy Procedia 2014;63:7891–912.
[20] Welkenhuysen K, Rupert J, Compernolle T, Ramirez A, Swennen R, Piessens K. [42] Ampomah W, Balch RS, Ross-Coss D, Hutton A, Will R. An Integrated Approach
Considering economic and geological uncertainty in the simulation of realistic for characterizing a sandstone reservoir in the anadarko basin. Paper OTC-
investment decisions for CO2-EOR projects in the North Sea. Appl Energy 26952-MS presented at offshore technology conference held in Houston-Texas
2017;185:745–61. USA, May 2–5 2016; 2016b. http://dx.doi.org/10.2118/180375-MS.
[21] Markowitz HM. Portfolio selection. J Finance 1952;7(1):77–91. http://dx.doi. [43] Munson T. Depositional, diagenetic, and production history of the Upper
org/10.1111/j.1540-6261.1952.tb01525.x. This work was later refined by H.M. Morrowan Buckhaults sandstone, Farnsworth Field, Ochiltree County, Texas
Markowitz in Portfolio Selection: Efficient Diversification of Investments, New unpub. MS thesis. Canyon, TX: West Texas State University; 1988. 354. pp.
York: John Wiley & Sons; 1959. [44] Rose-Coss D, Ampomah W, Cather M, Balch RS, Mozley P. An improved
[22] Couët B, Bailey WJ, Wilkinson D. Reservoir optimization tool for risk and approach for sandstone reservoir characterization. Paper SPE-180375-MS
decision analysis. In: Proceedings of the 9th European conference on the presented at SPE Western Regional Meeting held in Anchorage, Alaska, May
mathematics of oil recovery, Cannes, France, August 30 – September 2. 23–26; 2016. http://dx.doi.org/10.2118/180375-MS.
[23] Couët B, Djikpesse H, Tonkin T, Wilkinson D. Production enhancement t [45] Ampomah W, Balch R, Grigg R. Analysis of upscaling algorithms in
hrough integrated asset modeling optimization. In: SPE production and heterogeneous reservoirs with different recovery processes. Paper SPE-
operations conference and exhibition, Tunis, Tunisia (June). 173588-MS presented at the SPE production operations symposium held in
[24] Bailey WJ, Couët B, Wilkinson D. Framework for field optimization to Oklahoma City, Oklahoma, USA, March 1–5; 2015a. http://dx.doi.org/10.2118/
maximize asset value, SPE 87026-PA. SPE Reservoir Eng J 2005;8(1):7–21. 173588-MS.
[25] Al-Sofi AM, Blunt MJ. The design and optimization of polymer flooding under [46] Ampomah W, Balch RS, Grigg RB, Dai Z, Pan F. Compositional simulation of CO2
uncertainty. In: SPE Enhanced Oil Recovery Conference, (July). p. 1–12. http:// storage capacity in depleted oil reservoirs. Paper CMTC-439476-MS presented
dx.doi.org/10.2118/145110-MS. at carbon management technology conference held in Sugarland, Houston-
[26] Gharbi RB. Economic optimization of EOR processes using knowledge-based Texas USA, November 16–19 2015; 2015b. http://dx.doi.org/10.7122/439476-
system: case studies. Pet Sci Technol 2001;19(7&8):797–823. http://dx.doi. MS.
org/10.1081/LFT-100106901. [47] Hutton A. Geophysical Modeling and Structural Interpretation of A 3D
[27] Guevara JL, Ortega A, Canelón JI, Nava E, Queipo NV. Model-based adaptive- Reflection Seismic Survey in Farnsworth Unit, TX M.S. Thesis. Socorro NM,
predictive control and optimization of SAGD under uncertainty. In: SPE Latin USA: New Mexico Tech; 2015.
America and caribbean petroleum engineering conference, Quito, Ecaudor, [48] Gallagher SR. Depositional and diagenetic controls on reservoir heterogeneity:
November 18–20. Upper Morrow Sandstone, Farnsworth Unit, Ochiltree County, Texas M.S.
[28] Alhuthali AH. Optimal waterflood management under geologic uncertainty Thesis. Socorro NM, USA: New Mexico Tech; 2014.
using rate control: theory and field applications. In: SPE Annual technical [49] Czoski P. Geologic characterization of the Morrow B reservoir in Farnsworth
conference and exhibition, (May). http://dx.doi.org/10.2118/129511-stu. Unit, TX using 3D VSP seismic, seismic attributes, and well logs M.S.
[29] Chugunov N, Ramakrishnan TS, Lukyanov A, Bailey WJ, Wilkinson D. Method Thesis. New Mexico Tech; 2014.
for adaptive optimization of EOR performance under uncertainty optimization [50] Ahmmed B, Appold MS, Fan T, McPherson BJ, Grigg RB, White MD. Chemical
under uncertainty and global sensitivity analysis. In: SPE reservoir simulation effects of carbon dioxide sequestration in the Upper Morrow Sandstone in the
symposium, 23–25 February, Houston, Texas, USA. Farnsworth, Texas, hydrocarbon unit. Environ Geosci 2016;23(2):81–93.
[30] Van t Veld K, Mason CF, Leach A. The economics of CO2 sequestration through [51] Dai Z, Viswanathan H, Middleton R, Pan F, Ampomah W, Yang C, et al. CO2
Enhanced Oil recovery. Energy Procedia 2013;37:6909–19. Accounting and risk analysis for CO2 sequestration at enhanced oil recovery
[31] Leach A, Mason CF, Van’t Veld K. Co-optimization of enhanced oil recovery and sites. Environ Sci Technol 2016;50(14):7546–54. http://dx.doi.org/10.1021/
Carbon sequestration. Resource Energy Econ 2011;33:893–912. http://dx.doi. acs.est.6b01744.
org/10.1016/j.rese- neeco.2010.11.002. [52] Dai Z, Viswanathan H, Fessenden-Rahn J, Middleton R, Pan F, Jia W, et al.
[32] Forooghi A, Hamouda A, Eilertsen T. Co-optimization of CO2 EOR and Uncertainty quantification for CO2 sequestration and enhanced oil recovery.
sequestration in a North Sea chalk reservoir. Paper SPE 125550 presented at Energy Procedia 2014;63:7685–93.
the 2009 SPE/EAGE reservoir characterization and simulation conference, Abu [53] Pan F, McPherson BJ, Dai Z, Jia W, Lee SY, Ampomah W, et al. Uncertainty
Dhabi, UAE, 19–21 October; 2009. http://dx.doi.org/10.2118/125550-MS. analysis of carbon sequestration in an active CO2-EOR field. Int J Greenh Gas
[33] Ettehadtavakkol A, Lake LW, Bryant SL. CO2-EOR and storage design Control 2016;51:18–28.
optimization. Int J Greenhouse Gas Control 2014;25:79–92. http://dx.doi.org/ [54] Pan F, McPherson BJ, Esser R, Xiao T, Appold MS, Jia W, et al. Forecasting
10.1016/j.i- jggc.2014.04.006. evolution of formation water chemistry and long-term mineral alteration for
[34] Kim Y, Jang H, Kim J, Lee J. Prediction of storage efficiency on CO2 sequestration GCS in a typical clastic reservoir of the Southwestern United States. Int J
in deep saline aquifers using artificial neural network. Appl Energy Greenhouse Gas Control 2016. http://dx.doi.org/10.1016/j.ijggc.2016.07.035.
2017;185:916–28. [55] Ampomah W, Balch RS, Grigg RB, McPherson B, Will RA, Lee S-Y, et al. Co-
[35] Priezzhev I, Scollard A, Lu Z. Regional production prediction technology based optimization of CO2-EOR and storage processes in mature oil reservoirs.
on gravity and magnetic data from the Eagle Ford Formation, Texas, USA: 84th Greenhouse Gas Sci Technol 2016. http://dx.doi.org/10.1002/ghg.1618.
Annual international meeting, SEG, Expanded Abstracts. p. 1354–58. http:// [56] Ampomah W, Balch R, Cather M, Rose-Coss D, Dai Z, Heath J, et al. Evaluation
dx.doi.org/10.1190/segam2014-0289.1. of CO2 storage mechanisms in CO2 enhanced oil recovery sites: application to
[36] Bishop CM. Improving the generalization properties of radial basis function morrow sandstone reservoir. Energy Fuels Article ASAP 2016. http://dx.doi.
neural networks. Neural Comput 1991;3(4):579–88. http://dx.doi.org/ org/10.1021/acs.energyfuels.6b01888.
10.1162/neco.1991.3.4.579. [57] Peng DY, Robinson DB. A new two-constant equation of state. Ind Eng Chem
[37] Kobrunov A, State U, Priezzhev I. Stable nonlinear predictive operator based on 1976.
neural network, genetic algorithm and controlled gradient method, no. 1; [58] Peneloux A, Rauzy E, Freze R. A consistent correction for redlich-kwong-soave
2015. p. 2941–46. volumes. Fluid Phase Equilib 1982.
[38] Dai Z, Stauffer PH, Carey JW, Middleton RS, Lu Z, Jacobs JF, et al. Pre-site [59] Lohrenz J, Bray BG, Clark CR. Calculating Viscosities of Reservoir Fluids from
characterization risk analysis for commercial-scale carbon sequestration. their compositions SPE Paper 915. J Petrol Technol 1964:1171–6.
Environ Sci Technol 2014;48:3908–15. [60] Sperrevik S, Gillespie PA, Fisher QJ, Halverson T, Knipe RJ. Empirical estimation
[39] Dai Z, Middleton R, Viswanathan H, Fessenden-Rahn J, Bauman J, Pawar R, et al. of fault rock properties. In: Koestler AG, Hunsdale R, editors. Hydrocarbon seal
An integrated framework for optimizing CO2 sequestration and enhanced oil quantification: papers presented at the norwegian petroleum society
recovery. Environ Sci Technol Lett 2014;1:49–54. conference, 16–18 October 2000, vol. 11. Stavanger, Norway: Elsevier, Norsk
[40] Gunda D, Ampomah W, Grigg RB, Balch RS. Reservoir fluid characterization for Petroleumsforening Special Publication; 2002. p. 109–25.
miscible enhanced oil recovery. In: Carbon Management Technology [61] Corey AT, Rathjens CH, Henderson JH, Wyllie MRJ. Three-phase relative
Conference November 16-19 2015, Sugarland, Houston-Texas USA. http://dx. permeability. J Petrol Technol 1956;8(11):63–5.
doi.org/10.7122/440176-MS.