SPE-183191-MS

Optimizing CO2 Floods Using Rate Control with Smart Wells under Geologic
Uncertainty

M. Sharma, University of Stavanger; S. V. Taware and A. Datta-Gupta, Texas A&M University

Copyright 2016, Society of Petroleum Engineers

This paper was prepared for presentation at the Abu Dhabi International Petroleum Exhibition & Conference held in Abu Dhabi, UAE, 7-10 November 2016.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents
of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect
any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written
consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may
not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
Carbon dioxide flooding (continuous or WAG) is a proven and effective EOR technique resulting in
improved oil recovery in various types of reservoirs. However, the complex displacement mechanisms in
CO2 flooding, in particular adverse mobility and gravity segregation can reduce its effectiveness. It has been
demonstrated earlier that by adjusting the well rates, we can manage the flood front and improve the sweep
during waterflooding. In this paper, we present the application of rate optimization to CO2 flooding in order
to maximize areal and vertical sweep, and minimize CO2 recycling by delaying CO2 breakthrough.
Field-scale rate optimization problem for CO2 EOR processes involve complex reservoir models,
production and facility constraints and geological uncertainty. We propose an efficient approach for
computing optimal rates for CO2 flooding with application to smart wells completed with inflow control
valves (ICV). The approach is based on equalizing the arrival time of flood front at the producers within
a selected sub-region to maximize the areal sweep efficiency under hierarchy of production and facility
constraints. An additional ‘norm’ constraint on the arrival times is also included to achieve high viscous-
to-gravity ratio (VGR) to minimize gravity segregation. The ‘optimal’ strategy is decided based upon
the trade-off between maximizing sweep and accelerating production. We use streamlines to compute the
analytical sensitivity of arrival times with respect to injection/production rates. Analytical gradient and
Hessian of the objective function are also derived, making the approach computationally efficient for field-
scale applications.
The approach has been demonstrated using a synthetic 2D example and a 3D benchmark field case. The
results clearly demonstrate that rate optimization can improve sweep, delay CO2 breakthrough and increase
oil recovery for the same amount of CO2 injection. In addition, we demonstrate the potential of our approach
for WAG flooding in the 3D benchmark field case. We incorporate geological uncertainty via a stochastic
optimization framework based on the combination of the expected value and variance of a performance
measure from multiple realizations.

Introduction
Carbon dioxide has been successfully used as EOR agent for decades because of its favorable properties of
swelling oil, reducing oil viscosity and reducing residual oil saturation (Holm and Josendal 1974; Stalkup
2 SPE-183191-MS

1983). However, despite its high local displacement efficiency, the process has poor sweep efficiency
due to viscous fingering caused by unfavorable mobility ratio and gravity segregation caused by density
difference between injected and displaced phases (Christie and Bond 1987). Caudle and Dyes (1958)
proposed the control of this fingering and poor conformance by injecting water along with gas to reduce gas
mobility. Strategies like water-alternating-gas (WAG) injection have been thus applied in field to mitigate
this problem.
The sweep efficiency during gasflood depends upon mobility ratio (Habermann 1960), viscous-to-gravity
ratio (Craig et al. 1957; Spivak 1974) and reservoir heterogeneity (Koval 1963; Fayers et al. 1992). Several
authors (Green and Willhite 1998; Jarrell et al. 2002) have discussed the design parameters influencing the
sweep efficiency during CO2 flood. Moreover, modification to uniform coinjection of water and gas (Stone
1982; Jenkins 1984) have been suggested by Rossen et al. (2010) to maximize the distance to the point of
segregation of injected gas/water mixture. However, guidelines for selection of optimal well production/
injection rates during CO2 flood remain to be established.
Waterflood optimization via rate control has received increased attention over last decade because of
rapid deployment of smart well technology. It has been demonstrated through numerical modelling that
the flood front can be managed to improve sweep efficiency during waterflooding by adjusting the well
rates. The field scale optimization problem, however, involves highly complex reservoir model, geologic
uncertainty and facility related constraints. Adressing these without efficient optimization algorithms makes
reservoir management via rate control challenging. The optimization algorithms which have been developed
to address the problem can be split into two broad categories, namely gradient-based algorithms and
stochastic algorithms (Brouwer and Jansen 2004; Tavakkolian et al. 2004). The gradient-based algorithms
involve estimating the gradient of the objective function with respect to the control variables in an efficient
manner. In contrast, the stochastic algorithms such as the genetic algorithm do not require estimation of the
gradient but typically require multiple forward simulations for evaluations of the objective function or an
appropriately defined fitness function. The stochastic optimization approach has the ability to search for
a global solution, and so have an advandage over the gradient-based optimization approach that typically
yields a local solution. However, stochastic optimization approach is computationally demanding, especially
if the number of control variables is large.
Sudaryanto and Yortsos (2001), and Brouwer and Jansen (2004) presented gradient-based optimization
for waterflooding using optimal control theory to maximize the net present value (NPV) or displacement
efficiency at water breakthrough. van Essen et al. (2009) extended the work done by Brouwer and Jansen
(2004) to address geological uncertainty, and used an objective function in terms of the expected value of
NPV obtained from multiple realizations. The previous works have used an adjoint method to compute the
gradient of the objective function and the steepest ascent algorithm to maximize it.
Alhuthali et al. (2007; 2008; 2010) proposed a streamline-based approach to maximize waterflood
sweep efficiency. The approach was based on equalizing arrival time of the waterfront at all producers
within selected sub-regions of a project. This resulted in delayed water breakthrough and reduced water
recycling after water breakthrough. The optimization was implemented under constraints inline with fluid
handling capacity of reservoir, wells and facilities using a sequential quadratic programming approach. The
sensitivities of the waterfront arrival times at the producers to well rates and the gradient and hessian of
the objective function was computed analytically, making the approach efficient and suitable for field-scale
application. Alhuthali et al. (2008) incorporated geological uncertainty in the approach via a stochastic
optimization framework based on the combination of the expected value and variance of a performance
measure from multiple realizations. However, the approach solely focused on maximizing sweep during
waterflood, and not necesaarily the NPV. To address this limitation, Taware et al. (2010) modified the
objective function by adding ‘norm’ penalty term to the objective function. The objective function thus now
consists of two terms. The first term aims to maximize sweep while the second term attempts to accelerate
SPE-183191-MS 3

production subject to operational constraints. The tradeoff between maximizing sweep by equalizing arrival
time and maximizing NPV via production acceleration can be evaluated by adjusting the weight on the
‘norm’ penalty term. The optimal strategy is thus a compromise between the two, and can be arrived at by
using a trade-off curve between NPV versus norm weight.
In this paper, we extend the approach for CO2 flooding to maximize sweep by delaying CO2 breakthrough
and minimize CO2 recycling post breakthrough by improving injection efficiency. We first outline the details
of CO2 flood modeling considered for numerical studies. We discuss the key steps of the approach followed
by underlying mathematical formulation. We, then, demonstrate the robustness of the approach using a
heterogeneous cross section and a synthetic field application for CO2 flood and WAG.

CO2 Flood Modeling

Depending upon the properties of the fluids at reservoir condition, the displacement of the oil by CO2
can be classified as immiscible and miscible (Holm and Josendal 1974). The immiscible displacement
occurs at pressures below minimum miscibility process (MMP) during which incremental oil is recovered
because of oil swelling, reduced oil viscosity and lower interfacial tension between CO2 and oil. However,
the interfacial tension is not zero resulting in some residual oil saturation. On the other hand, miscible
displacement occurs at pressures above MMP, in which over 95% of oil contacted can be displaced because
of zero interfacial tension, in addition to factors mentioned earlier.
We used a three-component modified black oil model (ECLIPSE©) consisting of reservoir oil, injection
gas (solvent) and water to describe miscible displacement of oil by CO2 as suggested by Todd and Longstaff
(1972). The Todd-Longstaff empirical model account for viscous fingering by using mixing parameter ω,
the value of which lies between 0 and 1, to represent the size of the dispersed zone in each grid cell. The
solvent and reservoir oil components are assumed to be first contact miscible (FCM) in all proportions
and consequently only one hydrocarbon phase exists in the reservoir. The model then computes effective
viscosity and density for oil-solvent system based on ω.
We considered ω = 0.67 for our studies with a corresponding residual oil saturation of 12%. Field pressure
during CO2 flood was maintained above MMP for all cases studied to maintain miscibility.

Rate Optimization Approach

In this section, we discuss the steps involved in our approach. The primary objective is to obtain production
strategies from optimal to accelerated production for CO2 flooding. The approach can be employed using
both finite-difference and streamline simulators. The approach can also be implemented on an ensemble of
realizations to account for geologic uncertainty using a stochastic framework.
We run CO2 flood simulation for a time interval of interest for every geologic realization using a finite-
difference simulator. The fluid fluxes thus obtained are used to trace streamlines (Jimenez et al. 2008), and
compute time-of-flight from injectors to producers (Datta-Gupta and King 2007). The travel time residuals
are computed which quantify the misfit between the desired and computed arrival times at each producer
for a specific realization. This forms the first term of the objective function. The second term corresponds
to the arrival time itself.
The analytical sensitivities, which are the partial derivatives of arrival time with respect to well rates, are
computed using streamlines. Only one simulation run per realization is required to compute sensitivitites
thus making our approach applicable to field-scale rate optimization. The Jacobian, which is the gradient
of the residuals, is then computed using the sensitivities. The gradient and hessian of objective function is
computed using residuals and the analytical Jacobian.
The objective function is minimized using Sequential Quadratic Programming (SQP) technique (Nocedal
and Wright 2006) subject to production/injection constraints to generate updated rates. These steps are
4 SPE-183191-MS

repeated for a particular timestep, until a pre-defined stopping criterion on the objective function or number
of iterations satisfied. We then move to next timestep for optimization.

Mathematical Formulation
In this section, we discuss the mathematical formulation underlying the approach to derive optimal
production/injection rates during CO2 flood.

Objective Function Formulation

The objective function consists of two terms (Taware 2012):

(1)

which can be expressed as:

(2)
where the variable ti,m represents the calculated arrival time at well i, belonging to group m. td,m represents
the desired arrival time for the well group m, and is given by the arithmetic average of ti,m, during each
optimization iteration (Alhuthali et al. 2010). The vector q contains the control variables and has a dimension
of n, the number of well rates to be optimized including production and injection.
The first term of the objective function is expressed as the square of the l2-norm of the arrival time
residuals for a single geologic realization g. Minimizing this term ensures that the flood front arrives nearly
at the same time at all producers in a given group of wells within the field. This results in maximizing areal
sweep efficiency for CO2 flood. The second term of the objective function, which is the ‘norm’ penalty term,
ensures that the magnitude of the arrival time is also reduced along with their variance. The reduction in
arrival time leads to acceleration of production/injection. This improves the viscous-to-gravity ratio (VGR),
defined by Fayers and Muggeridge (1990) and written by Tchelepi and Orr (1994) for a cross section of
thickness h and length L as:

(3)

where is the average Darcy velocity, Δμ is viscosity difference between oil and CO2, Δρ is density difference
between oil and CO2 and kv is the geometric mean of permeability. Since increase in VGR corresponds to
lower gravity segregation, it is used as a measure of vertical sweep efficiency for FCM floods (Stalkup
1983; Tchelepi and Orr 1994).
By adjusting the weight η on the penalty term, we can decide on the trade-off between equalizing arrival
time and accelerating production/injection. The derivation of the analytic gradient and Hessian follows the
steps outlined by Alhuthali et al. (2008) and are not repeated here.

Optimization after breakthrough

Since the optimization problem has been formulated in terms of equalizing arrival time of the CO2 front at
the producers, we have incorporated the well gas/oil ratio (GOR) into the objective function. This prevents
allocating high production rates to wells producing large amount of gas after breakthrough, and thus diverts
the CO2 front. This is accomplished by modifying the arrival time to a well by including the GOR at the
well as follows:

(4)
SPE-183191-MS 5

where

(5)

where fGOR i,m is the well produced GOR at well i belonging to group m. The extent of rescaling is controlled
by the exponent term, α. The modified arrival time is the same as the original arrival time if GOR at reservoir
conditions is zero. When the well GOR is greater than zero, the original arrival time will be rescaled based
on the GOR level w.r.t the maximum allowable limit. The rate allocation to the well will thus be lowered in
relation to the wells with less GOR. Moreover, the well is shut when the GOR reaches the maximum limit.

Accounting for Geological Uncertainty

Geologic uncertainity is addressed by considering an expected value of the misfit for an ensemble of
realizations penalized by its standard deviation within the decision analysis framework (Alhuthali et al.
2008) as follows:
(6)
where r is the risk coefficient, which is positive if the decision maker is risk averse, and negative if the
decision maker is risk prone.

Objective Function Minimization

Minimizing the objective function, which is the penalized misfit function as defined in Eq. 6, results
in optimal rates for producers/injectors. For field application, the minimization is carried out subject to
constraints. This problem can be mathematically expressed as:

(7)

where the equality and inequality constraints corresponding to reservoir, wells and facilities capacities are
captured via h(q) and g(q) respectively.
The objective function is then minimized using the Sequential Quadratic Programming (SQP) algorithm
for nonlinear constrained optimization (Nocedal and Wright 2006). We use the SQP algorithm as
implemented in MATLAB®, and the details on formulating the optimization problem can be found in
Alhuthali et al. (2007).

Applications and Discussion

In this section, we first illustrate our approach using a 2D heterogeneous cross section with an injector-
producer pair. Next, we demonstrate the practical feasibility of the approach using a 3D synthetic example,
the benchmark Brugge field case (Peters et al. 2010). Appropriate hierarchical constraints based on ICV,
well and field levels have been considered for rate optimization.

Illustration of the Approach: A 2D Heterogeneous Cross Section

We illustrate our approach for CO2 flooding using a 2D heterogeneous cross section (Fig. 1). The cross
section (200×1×180 grid) has a fixed porosity of 0.10 and spatially heterogeneous permeability. A producer
and an injector with 45 ICVs each have been completed on either sides of the cross section. The large
6 SPE-183191-MS

number of ICVs, though off from a practical scenario, is simply used to demonstrate the stability of the
workflow under large number of constraints. Following constraints were imposed during rate optimization:

Figure 1—2D cross-section: Permeability distribution and well location

• Well production/injection rate ≤ 9000 RB/D

• ICV production/injection rate ≤ 450 RB/D

• Production FBHP ≥ 2000 psia

• Injection FBHP ≤ 6000 psia

• ICV GOR limit: 100 Mscf/STB

The base case considers reactive control where production/injection ICV rate equals to 150 RB/D. The
production ICVs are shut when their GOR exceeds 100 Mscf/STB. The combined effects of heterogeneity,
viscous fingering and gravity segregation lead to preferential CO2 movement towards the top production
ICVs in the base case (Fig. 2a). This results in early CO2 breakthrough and reduced sweep efficiency. Figure
2b shows the time-of-flight map after 2 years for the base case, which suggest that there is sufficient scope
for improving sweep and delaying CO2 breakthrough by equalizing CO2 front arrival time at production
ICVs through rate control.

Figure 2—(a) Gas saturation after 2 years of CO2 injection for the
base case. (b) Corresponding time-of-flight map for the base case

We obtained optimal rates for norm weights of 0, 10, 100 and 1000 for 20 years in timesteps of 6 month
for the base case mentioned above. A norm weight of 0 corresponds to maximizing sweep efficiency while
norm weights of 10, 100 and 1000 result in accelerated production strategy.
Figure 3 shows the gas saturation maps at four different times for base case, sweep efficiency
maximization (norm weight–0) and acceleration (norm weight–1000). For norm weight of 0, the injection
rate is sufficiently low so that the resulting sweep is dominated by the effects of gravitational segregation,
thereby creating gravity tongue. The transition zone between oil and gas is relatively thin compared to
reservoir thickness as mentioned by Fayers and Muggeridge (1990). However, as the norm weight increases,
the production rates and CO2 injection rates approach the upper limit for ICVs and well. This increases
SPE-183191-MS 7

VGR as shown in Fig. 4 and, suppresses the gravity tongue. The VGR has been calculated using geometric
mean of permeability for the heterogeneous cross section. Because of low displacement rate, optimization
corresponding to norm weight of 0 results in low VGR.

Figure 3—Gas saturation maps for base case, sweep efficiency maximization (norm
weight–0) and production acceleration (norm weight–1000) at four different times

Figure 4—VGR for base case, sweep efficiency maximization (norm weight–
0) and production acceleration (norm weight–1000) at different times

The cumulative voidage production for ICV-1 of the producer is shown in Fig. 5a for base case, sweep
efficiency maximization (norm weight–0) and acceleration (norm weight–1000). ICV-1 gets shut when its
GOR exceeds 100 Mscf/STB during base case and sweep efficiency maximization. Increasing the norm
weight reduces gravity segregation, thus allowing ICV-1 to produce for a longer duration without CO2
breakthrough. Fig. 5b shows cumulative voidage production for ICV-45 of the producer for respective cases.
In order to divert CO2 towards lower ICVs, their production rate approaches the maximum limit for norm
weight of 0. This results in high cumulative production for sweep efficiency maximization as compared to
base case. However, increasing norm weight reduces gravity segregation and lower rates are allocated to
producer's lower ICVs.
8 SPE-183191-MS

Figure 5—(a) Cumulative voidage production for ICV-1 (PROD1) for base case, sweep efficiency maximization (norm weight–0)
and production acceleration (norm weight–1000). (b) Cumulative voidage production for ICV-45 (PROD1) for respective cases

Figure 6a shows that the cumulative voidage injection for ICV-1 of injector for sweep efficiency
maximization (norm weight–0) is less than that for base case. The cumulative voidage injection for ICV-45
of injector for respective cases is shown in Fig. 6b. ICV-45 gets shut when its injection FBHP exceeds 6000
psia during base case and acceleration. In order to divert CO2 towards lower ICVs, the injection rate from
lower ICVs approaches the maximum limit for maximizing sweep efficiency.

Figure 6—(a) Cumulative voidage injection for ICV-1 (INJ1) for base case, sweep efficiency maximization (norm weight–
0) and production acceleration (norm weight–1000). (b) Cumulative voidage injection for ICV-45 (INJ1) for respective cases

The effect of rate optimization on field GOR is in accordance to the ICV performance discussed earlier.
The optimal rates corresponding to sweep efficiency maximization (norm weight–0) result in delayed CO2
breakthrough (Fig. 7a). However, the injection and production rates increase during acceleration (norm
weight–1000), resulting in early CO2 breakthrough and more CO2 recycling (Fig. 7b).

Figure 7—(a) Field GOR for base case, sweep efficiency maximization (norm weight–0) and
production acceleration (norm weight–1000). (b) Cumulative field gas production for respective cases
SPE-183191-MS 9

The increase in oil recovery with maximizing sweep efficiency (norm weight–0) is shown in Fig. 8a. It
should be noted that cumulative CO2 injected for maximizing sweep efficiency is significantly less compared
to that for base case (Fig. 8b). Acceleration (norm weight–10, 100 and 1000) results in increase in oil
recovery owing to improved VGR caused by higher CO2 injection rates.

Figure 8—(a) Oil recovery (w.r.t OOIP) after 20 years of CO2 flood for base case, sweep efficiency maximization (norm
weight–0) and production acceleration (norm weight–10, 100 and 1000). (b) Cumulative gas injection for respective cases

It is interesting to express the results for rate optimization in terms of field-wide injection efficiency
which is defined as ratio of cumulative oil produced to cumulative CO2 injected at reservoir conditions
(Datta-Gupta and King 2007). Injection efficiency thus indicates how efficiently injected CO2 is used, a
lower value of which corresponds to higher CO2 recycling. Figure 9 illustrates the injection efficiency for
various norm weights. The optimal rates corresponding to norm weight of 0 increase injection efficiency
significantly over base case. The results clearly demonstrate the trade-off between improving recovery by
accelerating production and improving injection efficiency by reducing CO2 recycling.

Figure 9—Injection efficiency for base case, sweep efficiency maximization

(norm weight–0) and production acceleration (norm weight–10, 100 and 1000)

3D Synthetic Field Example: The Benchmark Brugge Field Case

In this section we illustrate the application of the rate optimization approach for CO2 and WAG flooding
using a 3D synthetic case, the Brugge field. The field case was set up as part of an SPE Applied Technology
Workshop (ATW) to test various history matching and production optimization algorithms (Peters et al.
2010).
The model framework consists of an E-W elongated half-dome with a large boundary fault at its north
edge, and one internal fault with a modest throw. The model has been set up to mimic properties of a
North Sea Brent-type field. An ensemble of 104 realizations was generated based on well logs and reservoir
10 SPE-183191-MS

properties extracted from a high-resolution geologic model. The corase-scale simulation model consists of
60000 grid cells with 9 layers. The model has 20 vertical producers and 10 peripheral injectors as shown
in Fig. 10.

Figure 10—Permeability distribution and well location for Brugge field

The production history for the first 10 years was available to calibrate the realizations. The production
history was based on a ‘true model’ response with added noise. The objective of the exercise was to
implement closed loop proactive reservoir management approach consisiting of two main steps: (i) Model
calibration using the field production history for the first 10 years, and (ii) Production optimization over the
next 20 years to maximize recovery. In this study, we focus on the rate optimization part for the Brugge field.
Alhuthali et al. (2010) covered the details of history matching approach that was used to obtain an ensemble
of calibrated realizations. The optimization has been carried out for following three scenarios in this work:

• Rate optimization for CO2 flooding using a single history matched model (reference realization).

• Rate optimization for CO2 flooding using 10 history matched models (including reference
realization) and application of the optimized rates on a blind realization.
• Rate optimization for water-alternating-gas (WAG) flooding using reference realization.

Based on the location of fault, the wells within the model were divided into two groups (Fig. 10) to
calculate analytical sensitivities of the CO2 front arrival times with respect to well rates. Most of the wells in
the model are equipped with three inflow control valves (ICVs) and so the optimization considers obtaining
optimal rates for ICVs. Optimization was carried out in time steps of 6 months using norm weights of 0
and 10000. A norm weight of 0 corresponds to maximizing sweep efficiency while a norm weight of 10000
results in accelerated production strategy.
The three different levels of constraints used during optimization are shown in Fig. 11. The model has
53 production ICVs and 30 injection ICVs in total. Comprehensive constraint matrices were set up for the
field-scale optimization under following operational and facility constraints:

• Field production/injection rate ≤ 40000 RB/D

• Well and ICV production rate ≤ 3000 RB/D

• Well and ICV injection rate ≤ 4000 RB/D

• Production FBHP ≥ 740 Psia

• Injection FBHP ≤ 2626 Psia

• Water cut limit: 90%

• GOR limit: 100 Mscf/STB

SPE-183191-MS 11

Figure 11—Hierarchical production and facility related constraints considered for rate optimization

The constraint matrices were updated dynamically to ensure that the above-mentioned constraints are
satisfied at each hierarchical level during all time intervals.
CO2 Flood Optimization Using Single Geologic Model. In this section we limit optimization to a single
geologic model. The impact of geological uncertainty will be discussed in the next example.
The base case considers reactive control where producers/injectors are operated at their maximum
capacity subject to constraints mentioned earlier. The producers are shut as their water-cut exceeds 90% or
GOR exceeds 100 Mscf/STB. The second row in Fig. 12 shows the residual oil saturation for layer 1, 4 and
8 at the end of 30 years of field production.

Figure 12—Oil saturation after 20 years of CO2 flood for base case, sweep efficiency
maximization (norm weight–0) and production acceleration (norm weight–10000)

We carried out rate optimization using norm weights of 0 and 10000 for the base case mentioned above.
As shown by third row of Fig. 12, the optimal rates corresponding to norm weight of 0 result in reduction
of residual oil saturation at the end of 30 years, owing to better sweep in all of the layers. The last row of
12 SPE-183191-MS

Fig. 12 shows further reduction in oil saturation for optimal rates corresponding to norm weight of 10000,
owing to improved VGR and reduced gravity segregation.
Figure 13a shows the increasing trend of ‘incremental oil recovery’ with rate optimization for norm
weights of 0 and 10000. The recovery shown in the figure corresponds to production during later 20 years
w.r.t. OIIP. It should be noted that cumulative CO2 injected for sweep efficiency maximization (norm
weight–0) is less compared to that for the base case (Fig. 13b). Acceleration (norm weight–10000) results
in increase in oil recovery owing to improved VGR caused by higher CO2 injection rates. The increase in
injection efficiency with sweep efficiency maximization, as shown in Fig. 14, demonstrates the practical
viability of our approach. Production acceleration, however, results in higher recovery at the expense of
injection efficiency.

Figure 13—(a) Incremental oil recovery (w.r.t OOIP) for 20 years of CO2 flood for base case, sweep efficiency maximization
(norm weight–0) and production acceleration (norm weight–10000). (b) Cumulative gas injection for respective cases

Figure 14—Injection efficiency for base case, sweep efficiency maximization

(norm weight–0) and production acceleration (norm weight–10000)

CO2 Flood Optimization Using Multiple Models: Accounting for Geologic Uncertainty. In this section
we discuss optimization under geologic uncertainty by incorporating multiple realizations.
The realization from the previous case for single realization optimization was considered as reference
realization. We considered 10 history matched realizations (including reference realization) of the model
for rate optimization. The optimization was carried out using the formulation discussed earlier, under risk
neutral conditions (r–0). Figure 15 shows the permeability in the top layer for each of those 10 realizations
to illustrate the spread in property distribution amongst the realizations.
SPE-183191-MS 13

Figure 15—Permeability (layer 1) for 10 history matched realizations used in stochastic optimization

A blind realization, which was not included in the ensemble for optimization, was used to test the response
to rates obtained from the optimization process. To illustrate the robustness of our approach, we also applied
to the same blind realization the optimal rates obtained from the single realization optimization, which was
discussed in the previous section. The perameability (top layer) for the blind realizarion used to test the
optimization approach is shown in Fig. 16.

Figure 16—Permeability (layer 1) for blind realization

Figure 17 compares the field GOR and oil recovery response of the blind realization under following
two scenarios:

• Scenario 1: Use the rates derived from optimization of single realization (SR)

• Scenario 2: Use the rates derived from optimization of multiple realizations (MR)

Figure 17—(a) Field GOR for blind realization for base case rates, and optimized rates obtained from
sweep efficicency maximization (norm weight–0) from single realization (SR) and multiple realizations
(MR) scenarios (b) Incremental oil recovery (w.r.t OOIP) for 20 years of CO2 flood for respective cases
14 SPE-183191-MS

We used norm weight of 0 for both the scenarios. Fig. 18 shows that the optimal rates corresponding
to scenario 2 resulted in higher oil recovery and higher injection efficiency over scenario 1. The results
clearly demonstrate the ability of our approach to address geologic uncertainty during field-scale flood
optimization.

Figure 18—Injection efficiency for blind realization for base case rates, and optimized rates obtained from sweep
efficicency maximization (norm weight–0) from single realization (SR) and multiple realizations (MR) scenarios

WAG Flood Optimization Using Single Geologic Model. In this section we illustrate the application of
our approach for WAG flooding using the reference realization.
The base case considers alternate injection of water and CO2 for 4 months and 2 months respectively
under reactive control described earlier. This results in WAG ratio (volume water/volume CO2 at reservoir
condition) of 2:1 during each cycle (Fig. 19a). The optimal production/injection rates for maximizing sweep
efficiency were obtained using norm weight of 0. The average WAG ratio for 20 years of production for
sweep efficiency maximization is close to that for the base case. However, as shown in Fig. 19b, the WAG
ratio for each cycle varies from 1.2 RB/RB to 3.9 RB/RB. Acceleration corresponding to norm weight of
10000 causes the injection rate, for both water and CO2, to approach the upper limit. This results in WAG
ratio of 2:1 for most of the WAG cycles during acceleration.

Figure 19—(a) Average WAG ratio for 20 years of WAG flood for base case, sweep efficiency maximization (norm
weight–0) and production acceleration (norm weight–10000). (b) WAG ratio for 6 months cycle for respective cases

The effects of rate optimization on field water cut and GOR are shown in Figs. 20a and 20b. The optimal
rates for maximizing sweep efficiency result in delayed CO2 breakthrough and reduced water and CO2
recycling. However, the production/injection rates increase with acceleration resulting in higher field water
cut and GOR.
SPE-183191-MS 15

Figure 20—(a) Field water cut during WAG flood for base case, sweep efficicency maximization (norm
weight–0) and production acceleration (norm weight–10000). (b) Field GOR for respective cases

Figure 21 shows the increasing trend of ‘incremental oil recovery’ with rate optimization for norm
weights of 0 and 10000. It should be noted that cumulative water and CO2 injected for maximizing sweep
efficiency is less compared to that for the base case (Figs. 22a and 22b). Acceleration results in increase
in oil recovery owing to improved VGR caused by higher water and CO2 injection rates. Finally, from Fig.
23 we can see that sweep efficiency maximization improves injection efficiency, which further decreases
with increase in norm weight.

Figure 21—Incremental oil recovery (w.r.t OOIP) for 20 years of WAG flood for base case, sweep
efficiency maximization (norm weight–0) and production acceleration (norm weight–10000)

Figure 22—(a) Cumulative water injected during WAG flood for base case, sweep efficiency maximization (norm
weight–0) and production acceleration (norm weight–10000). (b) Cumulative gas injection for respective cases
16 SPE-183191-MS

Figure 23—Injection efficiency during WAG flood for base case, sweep efficiency
maximization (norm weight–0) and production acceleration (norm weight–10000)

Conclusions
In this paper, we discussed an approach for optimizing CO2 flood via rate control, which relies on making
an optimal decision based on a compromise between maximizing sweep and accelerating production.
Following conclusions can be made from this study:

• The approach is computationally efficient because the sensitivities, and the gradient and hessian of
the objective function are computed analytically. It requires only one simulation per iteration for a
realization, thus making it suitable for large field application for CO2 flood.
• The constraints for phase rates and pressure at different levels of hieraracy can be incorporated
during optimization for real field applications through comprehensive constraint matrices.
• Multiple realizations can be incorporated in the workflow to obtain optimal rates under geologic
uncertainty.
• Optimal production/injection rates can be obtained to improve sweep and increase oil recovery for
water-alternate-CO2 flood.
The applicability of the approach has been demonstrated using 3D synthetic example, the Brugge field.
The optimal rates resulted in significant increase in oil recovery and decrease in CO2 recycling over base/
reactive case. Ten history matched realizations were used to obtain optimal rates which were applied to a
blind realization to demonstrate the robustness of the approach in addressing geologic uncertainty during
optimization. Finally, the approach was used to improve sweep during WAG flood resulting in improvement
in oil recovery and reduction in water and CO2 recycling.

Nomenclature
e = arrival time residual vector, t, day(s)
E = expected value
fgas = ratio of produced GOR to maximum allowable GOR limit, dimensionless, fraction
f(q) = scalar objective function, t2, sq day(s)
g(q) = inequality constraints, L3/t, B/D [m3/d]
h(q) = equality constraints, L3/t, B/D [m3/d]
g = geologic realization index
i and j = well index
kv = geometric mean of permeability, L2, md
m = group index
Nprod,m = number of production well(s) in group m
SPE-183191-MS 17

Ngroup = number of group(s)

p(q) = scalar objective function, t2, sq day
q = total fluid rate vector, L3/t, B/D [m3/d]
r = risk coefficient, dimensionless
t = arrival time vector, t, day(s)
ti,m = arrival time at producer i which belongs to group m, t, day(s)
td,m = desired arrival time for group m, t, day(s)
= average Darcy velocity, L/t, ft/day [m/d]
α = exponent term, dimensionless
Δμ = viscosity difference between oil and CO2, m/Lt, cp [Pa.s]
Δρ = density difference between oil and CO2, m/L3, lbm/ft3 [kg/m3]
η = norm weight, dimensionless
λK = Karush-Kuhn-Tucker multipliers for inequality constraints
λL = Lagrange multipliers for equality constraints
σ = standard deviation

References
Alhuthali, A.H., Oyerinde, D. and Datta-Gupta, A. 2007. Optimal Waterflood Management Using Rate Control. SPE Res
Eval & Eng 10 (5): 539–551. SPE-102478-PA. http://dx.doi.org/10.2118/102478-PA.
Alhuthali, A.H., Datta-Gupta, A., Yuen, B. and Fontanilla, J.P. 2008. Optimal Rate Control Under Geologic Uncertainty.
Paper SPE 113628 presented at the SPE/DOE Symposium on Improved Oil Recovery, Tulsa, Oklahoma, USA, 20–
23 April. http://dx.doi.org/10.2118/113628-MS.
Alhuthali, A.H., Datta-Gupta, A., Yuen, B. and Fontanilla, J.P. 2010. Field Applications of Waterflood Optimization
via Optimal Rate Control With Smart Wells. SPE Res Eval & Eng 13 (3): 406–422. SPE-118948-PA. http://
dx.doi.org/10.2118/118948-PA.
Brouwer, D.R. and Jansen, J.D. 2004. Dynamic Optimization of Water Flooding with Smart Wells Using Optimal Control
Theory. SPE J. 9 (4): 391–402. SPE-78278-PA. http://dx.doi.org/10.2118/78278-PA.
Caudle, B.H. and Dyes, A.B. 1958. Improving Miscible Displacement by Gas-Water Injection. Trans., AIME, 213:
281–283.
Christie, M.A. and Bond, D.J. 1987. Detailed Simulation of Unstable Processes in Miscible Flooding. SPE Res Eng 2 (4):
514–522. SPE-14896-PA. http://dx.doi.org/10.2118/14896-PA.
Craig, F.F. Jr., Sanderlin, J.L., Moore, D.W. and Geffen, T.M. 1957. A Laboratory Study of Gravity Segregation in Frontal
Drives. Trans., AIME, 210: 275–282.
Datta-Gupta, A. and King, M.J. 2007. Streamline Simulation: Theory and Practice. Textbook Series, SPE, Richardson,
Texas 11.
ECLIPSE©, Version 2010.2. 2010. Schlumberger Information Solutions.
Fayers, F.J. and Muggeridge, A.H. 1990. Extensions to Dietz Theory and Behavior of Gravity Tongues in Slightly Tilted
Reservoirs. SPE Res Eng 5 (4): 487–494. SPE-18438-PA. http://dx.doi.org/10.2118/18438-PA.
Fayers, F.J., Blunt, M.J. and Christie, M.A. 1992. Comparisons of Empirical Viscous-Fingering Models and
Their Calibration for Heterogeneous Problems. SPE Res Eng 7 (2): 195–203. SPE-22184-PA. http://
dx.doi.org/10.2118/22184-PA.
Green, D.W. and Willhite, G.P. 1998. Enhanced Oil Recovery. Textbook Series, SPE, Richardson, Texas 6.
Habermann, B. 1960. The Efficiency of Miscible Displacement as a Function of Mobility Ratio. Trans., AIME, 219:
264–272.
Holm, L.W. and Josendal, V.A. 1974. Mechanisms of Oil Displacement By Carbon Dioxide. J Pet Technol 26 (12):
1427–1438. SPE-4736-PA. http://dx.doi.org/10.2118/4736-PA.
Jarrell, P.M., Fox, C., Stein, M. and Webb, S. 2002. Practical Aspects of CO2 Flooding. Monograph Series, SPE,
Richardson, Texas 22.
Jenkins, M.K. 1984. An Analytical Model for Water/Gas Miscible Displacements. Paper SPE 12632 presented at the SPE
Enhanced Oil Recovery Symposium, Tulsa, Oklahoma, 15–18 April. http://dx.doi.org/10.2118/12632-MS.
Jimenez, E.A., Datta-Gupta, A. and King, M.J. 2008. Full Field Streamline Tracing in Complex Faulted Systems With
Non-Neighbor Connections. Paper SPE 113425 presented at the SPE/DOE Symposium on Improved Oil Recovery,
Tulsa, Oklahoma, USA, 20–23 April. http://dx.doi.org/10.2118/113425-MS.
18 SPE-183191-MS

Koval, E.J. 1963. A Method for Predicting the Performance of Unstable Miscible Displacement in Heterogeneous Media.
SPE J. 3 (2): 145–154. SPE-450-PA. http://dx.doi.org/10.2118/450-PA.
MATLAB®, Version 7.11.0. 2010. Natick, Massachusetts: The MathWorks, Inc.
Nocedal, J. and Wright, S.J. 2006. Numerical Optimization, second edition. New York: Springer Science + Buisness Media.
Peters, L., Arts, R.J., Brouwer, G.K., Geel, C.R., Cullick, S., Lorentzen, R.J., Chen, Y. et al. 2010. Results of the
Brugge Benchmark Study for Flooding Optimization and History Matching. SPE Res Eval & Eng 13 (3): 391–405.
SPE-119094-PA. http://dx.doi.org/10.2118/119094-PA.
Rossen, W.R., van Duijn, C.J., Nguyen, Q.P., Shen, C. and Vikingstad, A.K. 2010. Injection Strategies To Overcome
Gravity Segregation in Simultaneous Gas and Water Injection Into Homogeneous Reservoirs. SPE J. 15 (1): 76–90.
SPE-99794-PA. http://dx.doi.org/10.2118/99794-PA.
Spivak, A. 1974. Gravity Segregation in Two-Phase Displacement Processes. SPE J. 14 (6): 619–632. SPE-4630-PA.
http://dx.doi.org/10.2118/4630-PA.
Stalkup, F.I. 1983. Miscible Displacement. Monograph Series, SPE, Richardson, Texas 8.
Stone, H.L. 1982. Vertical Conformance in an Alternating Water-Miscible Gas Flood. Paper SPE 11130 presented
at the SPE Annual Technical Conference and Exhibition, New Orleans, Louisiana, 26–29 September. http://
dx.doi.org/10.2118/11130-MS.
Sudaryanto, B. and Yortsos, Y.C. 2001. Optimization of Displacements in Porous Media Using Rate Control. Paper SPE
71509 presented at the SPE Annual Technical Conference and Exhibition, New Orleans, Louisiana, 30 September–
3 October. http://dx.doi.org/10.2118/71509-MS.
Tavakkolian, M., Jalali F., F., and Emadi, M.A. 2004. Production Optimization using Genetic Algorithm Approach. Paper
SPE 88901 presented at the Nigeria Annual International Conference and Exhibition, Abuja, Nigeria, 2–4 August.
http://dx.doi.org/10.2118/88901-MS.
Taware, S., Sharma, M. and Datta-Gupta, A. 2010. Optimal Water Flood Management Under Geological Uncertainty
Using Accelerated Production Strategy. Paper SPE 133882 presented at the SPE Annual Technical Conference and
Exhibition, Florence, Italy, 19–22 September. http://dx.doi.org/10.2118/133882-MS.
Taware, Satyajit Vijay (2012). Optimal Reservoir Management and Well Placement Under Geologic Uncertainty.
Doctoral dissertation, Texas A&M University. Available electronically from http://hdl.handle.net/1969.1/ETD-
TAMU-2012-08-11670.
Tchelepi, H.A. and Orr, F.M. Jr. 1994. Interaction of Viscous Fingering, Permeability Heterogeneity, and Gravity
Segregation in Three Dimensions. SPE Res Eng 9 (4): 266–271. SPE-25235-PA. http://dx.doi.org/10.2118/25235-PA.
Todd, M.R. and Longstaff, W.J. 1972. The Development, Testing, and Application Of a Numerical Simulator for Predicting
Miscible Flood Performance. J Pet Technol 24 (7): 874–882. SPE-3484-PA. http://dx.doi.org/10.2118/3484-PA.
van Essen, G.M., Zandvliet, M.J., Van den Hof, P.M.J., Bosgra, O.H. and Jansen, J.D. 2009. Robust
Waterflooding Optimization of Multiple Geological Scenarios. SPE J. 14 (1): 202–210. SPE-102913-PA. http://
dx.doi.org/10.2118/102913-PA.