You are on page 1of 30

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

net/publication/316948321

New Trends in Waterflooding Project Optimization

Conference Paper · January 2017


DOI: 10.2118/185472-MS

CITATIONS READS
5 246

3 authors:

Jose Luis Mogollon Taher Lokhandwala


Halliburton Rice University
50 PUBLICATIONS   454 CITATIONS    7 PUBLICATIONS   16 CITATIONS   

SEE PROFILE SEE PROFILE

Edwin Tillero
Loughborough University
18 PUBLICATIONS   33 CITATIONS   

SEE PROFILE

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

Waterflood Assesstment View project

Waterflood Optimization View project

All content following this page was uploaded by Edwin Tillero on 16 June 2017.

The user has requested enhancement of the downloaded file.


SPE-185472-MS

New Trends in Waterflooding Project Optimization

J. L. Mogollón, T. M. Lokhandwala, and E. Tillero, Halliburton

Copyright 2017, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE Latin America and Caribbean Petroleum Engineering Conference held in Buenos Aires, Argentina, 18-19 May 2017.

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
Water is the most commonly used injection fluid for flooding/energizing oil reservoirs. Despite oil price
fluctuations, water use has continued because of its wide availability, relatively low cost, and ease of
handling. Decades of research and field application experiences have yielded a sound theoretical approach
and practical knowledge of the subject. Nevertheless, water injection deployment and operations can still
benefit from optimization. This paper discusses the state-of-the-art use of numerical optimizers based on
smart algorithms and stochastic machines that couple subsurface, surface, and economic models.
During planning and operations of waterflooding projects, many decisions are made, such as the number,
location, and drilling sequence of new injector and producer wells, total and per well injection rates, well
conversion, and fluid withdrawal rates. In addition, each decision variable has multiple options, which
combined can generate hundreds or thousands of scenarios, raising the key question of how the optimum
scenario can be determined in a timely manner. Furthermore, the optimum scenario selection process should
consider uncertainty (e.g., reservoir properties and oil prices) as well as operational constrains.
Based on previous experience, a general workflow was developed and fine-tuned to help identify
optimum scenarios. The workflow begins by defining the scenario matrix using available validated history-
match models. Models are coupled with an automatic optimizer/stochastic machine. The study cases
considered reservoirs with heavy-to-medium oil, injection by pattern and flank, large variations in original
oil in place (OOIP), and number of wells for waterflooding implementation and reactivation planning.
Optimization runs typically require hundreds of iterations to approach the maximum or minimum
objective business function. Each iteration corresponds to a scenario. To identify the optimal scenario
quickly, various strategies were tested: parallel computing and new methodologies of sequential
optimization with reduced number of decision variables, initial exploratory runs with a shortened economic
horizon time, and stochastic analysis of selected scenarios of the optimization run. All of these strategies
proved successful, depending on the specific situation.
The workflow application in three case studies yielded approximately 30% cumulative production and
net present value (NPV) increments, with less economic risk than the traditional deterministic simulation
approach and reduced water cut up to 40%; compared to base scenarios, Np and NPV increases higher than
200% were obtained. Furthermore, the workflow application generated a large number of scenarios that
provided flexibility to modify operations during unexpected events.
2 SPE-185472-MS

Optimizers/stochastic machines were determined to be a valid means to quantitatively estimate the


economy and risks and are a fundamental tool for managing waterflooding projects, resulting in better
scenarios than the traditional deterministic approach. The approach is also applicable to all types of enhanced
oil recovery (EOR) projects.

Introduction
To extend the life of mature fields, inputting energy economically is a key requirement. Water is the most
commonly injected fluid, and more than half of oil globally is obtained through waterflooding.
It is generally recognized that the first water injection occurred in the Pithole City area of Pennsylvania
in 1865. The understanding of waterflooding evolved in the late 1940s; in the 1950s, waterflooding use
increased. As a result of decades of practice and research, the quality and quantity of current waterflooding
knowledge is significantly improved. Nevertheless, the oil industry still experiences many challenges in
terms of both further improving performance and avoiding project failure.
Planning and re-planning a flooding project are of paramount importance; during this time, decisions are
made that will affect the entire life cycle of the project. Some decisions are irreversible once implemented.
A large number of key decisions must be made (e.g., injection rate; slug size; number, location, and drilling
sequence of new injector and producer wells; injection-production facilities size and characteristics).
A typical field development design approach involves a small number of stand-alone solutions that are
not designed to achieve a common objective. Decisions are often made reactively on a day to day basis,
rather than on a proactive planning basis. This approach must evolve to take into account the complexity
of upstream or downstream consequences of operational decisions.
For every planning decision, there are two or more options to be determined. The combination of decision
options generates hundreds to thousands of scenarios, raising the key question of how the optimum scenario
in the planning or replanning stages can be determined.
This paper focuses primarily on the above question. The aim is to generate a state-of-the-art approach
to fundamental topics for supporting the design, execution, and redevelopment of waterflooding projects
using numerical optimization. The efficient use of resources is extremely important, even more so when oil
prices are low. To achieve efficiency, numerical optimization is one of the most valuable available tools.
Once the reservoir model is built and properly history matched, the traditional deterministic/manual
numerical simulation approach allows, in a reasonable time, examination of only a few scenarios on
a trial and error basis. The transition from a deterministic to a probabilistic/optimized plan is enabled
by optimization program systems that automatically couple the reservoir and the economic models and
use numerical algorithms to efficiently determine the maximum of the objective business function (e.g.,
NPV, internal rate of return, and cumulative oil production). This paper discusses the fundamentals of the
optimizers used, the application methodology, case studies, and benefits.
When a flooding project begins, new data are acquired and the planned operation timing can be altered.
Because of these two elements, periodic updates of the reservoir model are important. Updates can be
manually accomplished in a reactive fashion; but, significant manpower efforts and time are necessary.
Consequently, the flooding project is not actually optimized except for the short period immediately after
the update. It still might be too late for even the simplest, least-costly corrective actions, for example, if
injection fluids are already channeled to the producer well, or a bottom aquifer began to interfere with the
flooding process.
A proactive approach includes monitoring coupled with model updates and forecasts in real time.
However, surveillance and optimization of waterfloods pose many challenges. Updating reservoir
models is tedious and time-consuming, involving multiple data sources, model updates, and simulations.
Technological challenges include large simulation models and limited real-time data. All of these
SPE-185472-MS 3

issues affect the speed at which the best actions to optimize project execution are determined. Current
developments in this topic are addressed in the last section of this paper.

Optimizer Function
Difficulties applying classical optimization techniques to waterflooding projects, and to well location
problems during field development plans (FDPs) in general, include working with a nonlinear solution in
an "n" dimensional space with many potential local optima (Fig. 1), a set of equations that are often not
differentiable, and a mix of continuous and discrete variables. These issues are intrinsically difficult to solve,
and the time necessary increases rapidly with the number of variables and constraints.

Figure 1—Example of multiple solutions topography of nonlinear problems.

Classical linear and nonlinear programming techniques can become focused on the first good solution,
whether it is a local optimal solution or an actual global optimal solution. Integer programming methods,
which are presumed to have a global solution capability, cannot easily handle important nonlinear
relationships or uncertainties. Solutions for optimization issues can be determined through an examination
of many location combinations with the nonlinear effects of fluid drive, water injection, physical constraints,
and economics.
The optimization cases analyzed in this paper were conducted using a software system comprising
a reservoir simulator, economics analysis package, and optimizer (i.e., DMS™ software). The reservoir
simulator includes commercially available Nexus®, ECLIPSE, and IMEX software.
The optimization solver used here employs standard and heuristic global search methods (i.e.,
"metaheuristics"), which sample a set of solutions that is too large to be completely sampled. These methods
are tabu search, scatter search, linear programming, and neural networks. The optimizer uses values for the
objective and for any specific requirements and then updates the decision variables to be used as input to the
next computation of the objective function. With this architecture, the optimizer does not require specifics
of the complex physical model; thus, formulations can be fairly straightforward and alternative objectives,
constraints, and requirements are easily specified. The flow simulator handles the nonlinear physical model
production flow and constraints.
The software system comprises two main calculation loops—the outer optimization loop and the inner
uncertainty loop, both of which have been described in previous works (Cullick et al. 2003, 2005; Narayan
et al. 2003) (Fig. 2). The outer loop is composed of the optimizer, computation of the objective function,
summary statistics, and connections to databases. The inner loop can be executed as an uncertainty engine
with Monte Carlo or quasi-Monte Carlo sampling, such as Latin hypercube sampling.
4 SPE-185472-MS

Figure 2—Optimization loops of the implemented software system.

Hundreds of simulations are often necessary to improve an objective value. The optimizer system
submits job instructions to a server dispatcher, which distributes individual simulations on a network of
central processing units running Linux or Windows® operating systems. The dispatcher checks processor
availability and speed and collects the results to send back to the user. Typically, in this work, the outer
optimization loop had more than 200 and up to 500 simulator executions (Fig. 3). A typical optimization
would run on more than 20 Windows-based system processors, but there is no limit on how many can be
accessed by the dispatcher. These Windows-based system processors consisted of dual- and single-processor
personal computers, ranging from 3.2 to 800 GHz.

Figure 3—Optimization run; each iteration corresponds to the forecasted NPV of a Case 1 waterflooding scenario.

The general optimization problem can be posed as maximized or minimized:


F(x,y)
Subject to
Ax ≤ b (constraints)
gl≤ G(x) ≤gu (requirements)
SPE-185472-MS 5

l ≤ x ≤ u (bounds)
where x is a vector of decision variables and y includes uncertainty variables and external constraints.
As with any optimizer that uses a global, evolutionary approach, the optimizer does not necessarily
converge to a single mathematical or "provable" optimal. It progressively improves the objective by finding
"better" solutions as the procedure progresses (Fig. 2). Thus, stopping criteria are imposed as a specified
number of iterations or as some improvement in objective.
If no uncertainty is posed by the parameters, then the inner loop simulation is executed only once and the
calculated single objective function value is passed to the outer loop. When there is uncertainty, a stochastic
simulation (Monte Carlo or quasi-Monte Carlo) is conducted in the inner loop to generate statistics for the
objective. Production profiles are computed using the sampled values of the uncertain parameters and the
values of the decision variables obtained from the optimizer. Because a finite difference flow simulator was
used, the production system was modeled rigorously, with as much detail as appropriate in the geologic
model, the tubing hydraulics and constraints, and the depletion strategy.
Fig. 4 shows key components in the search schematically:

• Construct a reference set of candidate solutions of decision variables obtained from previous
iterations or initial guesses.
• Generate linear combinations of reference solutions to create new solutions within the
combinations (i.e., link the reference set of solutions and update solution candidates through linear
combinations of previous solutions that improved the objective value in a process called "path
relinking"),
• Map the new points onto feasible solutions; in other words, honor the constraints using a linear
or mixed integer program with the ability to "tunnel" through infeasible regions that violate
constraints.
• Compare the objective function and constraints values output from the simulator and update the
reference set for the next iteration.

Figure 4—Optimization loops of the implemented software system (after Cullick et al. 2003).

If the solution set is deemed to lack sufficient quality, a diversification process is invoked to rebuild the
reference set. The optimizer tracks multiple "stacks" of reference sets and previous solutions that improved
the objective. Thus, the algorithm is efficient at executing in parallel and takes advantage of the grid
computing dispatcher.
6 SPE-185472-MS

Methodology: Numerical Optimization Workflows


Based on previous experience, a general flooding optimization workflow was developed and fine-tuned
(Fig. 5). It helps standardize the identification of optimum scenarios. This workflow is applicable in any
exploitation plan optimization, such as asset development with limited data (Tillero et al. 2011), secondary
polymer flooding (Mogollon et al. 2016), and surfactant flooding (Temizel et al. 2015). Nevertheless, this
paper focuses on using the workflow in waterflooding projects optimization.

Figure 5—General workflow for numerical optimization.

The workflow assumes a validated history-match model exists. Otherwise, an analytical model can be
used with known less-predictive capability and reliability. Before defining the scenario matrix, available
models are reviewed and updated.
The definition of the parameters to be used in the optimization run includes identification of the objective
function to be optimized, which is usually an economic parameter, such as NPV, to be maximized. For NPV,
the following equations apply.

(1)

(2)

(3)
Where Np1, Gp1, and Wp1are the cumulative production of oil, gas, and water, respectively, for each
year. We1 is the cumulative, N is the number of wells, Cwell, Cpipe, and Cpump are the initial (i = 1) necessary
investments for the producer and/or injector wells, pipe, pump, where Cpipe and Cpump are selected based on
each iteration's pipeline length and diameter, and each pump inlet pressure. Poil and Pgas are oil and gas prices.
Coil, Cgas, Cwater, and Cwater_inj are the variable operating expenses for each unit of oil, gas, water produced,
and water injected, respectively.
SPE-185472-MS 7

Elaboration of the scenario matrix consists of combining the decision variables and options for each
variable, as well as identification of the operational constrains, such as rig number, minimum drilling
time, maximum injection capability, injection rate-bottomhole pressure (BHP), abandonment pressure, and
economic limits.
All organizations involved in flooding decisions and operations should be aligned, and the most
important aspects should be considered from the beginning. The business objective function is agreed to
at the managerial-strategy levels, and the decision variables/options, uncertainty variables, and constraints
are initially identified in brain-storming exercises attended by representatives of all disciplines [(e.g.,
earth science; reservoir; production; drilling and completion; surface facilities; and health, safety, and
environment HSE)]. Table 1 shows typical results of such exercises; notice that decision variables have
been shortlisted for sake of simplicity.

Table 1—Brainstorm exercise matrix of decision variables and options.

Decision Variables Option 1 Option 2 More Options

Injection pattern Inverted five-spot Direct line Inverted seven-spot


Injection fluid Water Polymer AS, ASP, AP
Injection capacity Existing plant repairs New plant —
Water source Lake Formation —
Injection pipeline Conventional Flexible —
Pipeline diameter 3 in. 4 in. —
Facilities automation Basic supervision Overall automation —
Artificial lift Gas Rod pump ESP, PCP
Well geometry Vertical Vertical Slanted
9 5/8 in.
Casing 9 5/8 in. with 7 in. —
with 5 1/2 in.
Openhole + liner Cased hole,
Completion Openhole + liner
+ gravel pack cased hole + liner
Tubing size 2 3/8 in. 2 7/8 in. 3 1/2 in.
Tubing conveyed
Perforation Casing gun —
perforating

Such exercises combining decision variables and their options generate hundreds to thousands of
scenarios. Fig. 6 describes how scenarios form for a number of decision variables. Notice that many more
decision variables are to be considered; for example, total and injection rate per well, conversion of producer
wells to injector wells, and workovers.

Figure 6—Example scenarios matrix for waterflooding optimization.


8 SPE-185472-MS

Even when using optimizers with smart algorithms that make searching the objective function maximum
easier, the process of finding the best scenario is computationally intensive and the running time might be
unsuitable, depending on the iteration time. Therefore, in most cases, the number of options is limited by
ascertaining the best options ranges through deterministic runs or successive exploratory optimization runs
with a limited number of parameters. More detail is described in a later section of this paper.
Once decision variables/options and constrains are selected, the subsurface-surface-economy models are
coupled through the stochastic-optimizer machine, and the optimization run begins.
The stochastic runs shown in the next step of the general workflow deliver important information, such as
the risk and probability of a negative value for the objective business function for the scenario considered.
In addition, information is provided about the expected range of a number of operational parameters, such
as injection-production fluid volume with time. Table 2 shows an example of uncertainties variables and
the statistical characteristics usually considered.

Table 2—Multipliers used for probabilistic risk analysis.

Variation
Description Distribution Type Maximum
Coefficient

Crude oil price Normal 15 —


Recompletion costs Log-normal 15 —
Water production Normal 10 —
Crude oil production Normal 30 1.1
Crude oil production unit cost Log-normal 15 —
Water injection unit cost Log-normal 15 —
Gas production Normal 30 1.2
Water injection rate Normal 30 1.6
Injection well investment Log-normal 15 —
Production well investment Log-normal 15 —

Using the two loops described in Fig. 2, the optimization system has the capability of performing
the stochastic run in the same run as the optimization. However, running time can be significant. For
example, for a given model and computing power with yields of 30 minutes for a forecast simulation and
an optimization run that typically forecasts 200 to 300 scenarios (i.e., iterations in Fig. 3), determining the
maximum would require 4.2 to 6.3 days. However, if 500 to 2,000 stochastic runs are to be accomplished for
each scenario, this is equivalent to up to 600,000 forecast runs. As such, the stochastic runs are conducted
for selected scenarios identified in the optimization run, as exemplified in Case Study 1 discussed below.
The stochastic run coupled with the optimization considers, along with economic uncertainties, the
reservoir uncertainties, such as permeability, reservoir volume, and oil saturation. In the case of a posteriori
stochastic run of selected scenarios, the focus is on injection-production fluid volumes uncertainties, costs,
and oil price. The premise is that reservoir uncertainties are included in the fluid volume uncertainties. In
both cases, knowing the benefit-risk of different high-value scenarios allows the selection of the one that
better fits the operating company strategy.
Next, the surveillance and monitoring plan is designed. This is the last step closing the workflow cycle
shown in Fig. 5. With data collected, the models are updated and the optimization cycle is performed again,
this time with limited decision variables, options, and uncertainties.
SPE-185472-MS 9

Running Time Solutions


As reservoir simulation technologies become more advanced, it is not unusual to have models with one
million cells or more, even after upscaling. Because of finely gridded models and increased resolution,
sometimes with hundreds of wells, optimization run times can become prohibitive to some operators. FDPs
can also involve long planning times, usually 15 years or longer, as can be observed in the case studies
discussed in this paper. This means that one forecast run of a simulation can often necessitate several hours
to accomplish.
Typical optimization exercises involve selecting a combination of decision variables that can provide
the maximum (or minimum) value of a specified dependent objective variable. For example, it might
necessary to consider the combination of decisions that would provide the maximum NPV over a 15-year
timeframe. As an example, a scenario can be posited where only the choice of drilling wells over this
period is considered. It might be desirable to drill between one and 10 producers and between one and 10
injectors during the 15-year timeframe. In this scenario, the combination of producers and injectors that
would provide the best NPV needs to be determined. Just knowing exactly which wells to drill to provide
the highest NPV in this simple case yields more than one million possibilities (1,046,529). If decisions are
included regarding injection rate, well scheduling, and other more complicated factors, such as well patterns
and configurations, the possibilities increase significantly. Considering sufficient time to only perform 300
optimization iterations from a solution space of millions of scenarios for a 15-year field development plan,
with 2 hours for each forecast run, requires 3.6 weeks of run time. This simple example shows the extent
of the problem.
To help reduce optimization run times, several solutions are proposed:

• Better hardware

• Simpler models

• Reducing the solution space

Better Hardware
Using better hardware is the simplest solution to resolve long run times. With the advent of better technology
at lower prices and advances in optimization algorithms and parallel computing, this can be an effective
solution. In particular, adding hardware capacity in conjunction with parallel computing methods can reduce
run times dramatically. Benchmarking studies by software simulation groups determined that increasing the
number of computing cores—for example, from 2 to 24—can reduce run times by a factor of 6 to 7. These
studies also show a declining benefit as the number of cores increases. For example, the reduction in run
time from 2 to 4 cores would be greater than the reduction in run time from 4 to 6 cores. In fact, increasing
the number of cores beyond a certain critical run time might increase run times. Therefore, a study of the
proper computing power is recommended to ensure the best achievable run times for any optimization study
are determined.

Simpler Models
Rashid et al. (2013) state that reducing run time in optimization problems often involves using reduced-
order models. Reduced-order models could include techniques such as upscaling and conversion to black oil
models. Adaptive gridding techniques also reduce total simulation time: using fine gridding in most critical
areas and coarse gridding outside those areas. The use of adaptive gridding to reduce run times has been
documented; Perez-Perez et al. (2014) show a reduction in computing time of up to 15 fold for a steam-
solvent simulation study. Simpler models can be beneficial in the reduction of run time, but the tradeoff
between loss of accuracy and increases in run time need to be considered. Typically, the greater the increase
10 SPE-185472-MS

in run time, the larger the risk of accuracy loss. In complex reservoirs, in particular, upscaling can lead to
a loss of fidelity in forecasting.
At minimum, it is recommended that history-matched injection-production responses and volumetric
calculations should maintain a high degree of fidelity. Optimized forecast results should also be cross-
checked with the high-resolution models to determine if results can be mirrored between the high-resolution
and upscaled models.

Reducing the Solution Space


A few additional techniques exist to help reduce run times, which can be broadly classified as making
changes to optimization workflows that lead to reduced potential solution space. This process includes
two categories—breaking up the optimization problem and better constraining the problem; these are not
mutually exclusive. Advantages of using these techniques are that they can avoid additional expenses
associated with costly hardware and have significantly less fidelity loss than using simple models.
Here, two examples of breaking up the optimization problem are presented. First, the number of decision
variables and their options can be reduced by running preliminary optimizations to immediately differentiate
key decision variables that most affect objective outcomes from those having negligible effects (Tillero
et al. 2014). Simultaneously, decision variable options that do not contribute to high-value scenarios can
be identified. For example, the best pattern for injection can be determined based on immediate runs,
while subsequent work on optimization can identify aspects, such as the optimized injection rate, operating
pressures, etc.
A further example of breaking up the optimization problem is provided in Millan et al. (2012), which
shows that reduced computing time is also achieved by making injection-production optimization decisions
based on a reduced forecast horizon. After successive runs with localized objective functions with short time
spans are finished, a final run with the full forecast horizon is conducted to determine the final economic
outcome with the reduced number of decision variables (Fig. 7).

Figure 7—Successive optimization cycle runs to reduce total run time.

Here, decisions based on favorable early outcomes are considered to be set for the remainder of the
optimization period. These decisions then also act as constraints for future runs.
Properly constrained models reduce run time and also might provide quick identification of the global
optima. While the previous example shows one method of constraining problems, a well-constrained
problem from the initiation of the optimization run can dramatically improve results. Fig. 8 shows results
from an exercise to compare the objective function with the selection of wells as decision variables.
The well-constrained objective function on the left, where watercut constraints are added in the reservoir
simulation software, performs significantly better than the less-constrained objective function on the
right, where shutoff constraints are added within the optimization algorithm. Specifically, procedures (or
constraints) included in the numerical model were the shutting of the top perforations as producers reached
a watercut of 90% and the shutting of all perforations as producers reached a watercut of 93%. In the case
where such constraints were given to the optimizer instead of the numerical model, an upper bound of a
maximum of 93% watercut per production well was imposed. The optimizer converged to a much quicker
SPE-185472-MS 11

solution when the constraint on perforation-based watercuts was imposed. As a result, in the limited time
period under consideration, the optimal value was also higher.

Figure 8—Results for imposition of constraints in optimization.

In conclusion, several possible solutions exist to help reduce run times. With improvements in hardware
and workflows, run times can be reduced further to help advance quicker and more accurate optimizations.

Case Studies: Achievements


Study cases for this paper were selected to discuss a wide range of conditions. Table 3 shows significant
differences between the reservoirs in terms of OOIP, API gravity, permeability, number of wells, and
production history. Geology and numerical simulation models are described in later sections.

Table 3—Reservoir characteristics.

Production
OOIP (STB Active Total
Case °API K (D) History
millions) Wells Wells
(years)

Case 1 36 0.01 to 0.48 109.6 40 3 20


Case 2 19 0.5 to 2.5 727.6 80 52 72
Case 3 17.4 0.6 to 0.8 1116 60 72 155

These three cases illustrate different methodological approaches: 1) straight coupling of the reservoir and
economical models by the optimizer system, 2) the implementation of a deductive hypothetic-deductive
method (HDM) using deterministic forecasts as a mean to prove-disprove the hypothesis and the stochastic
optimizer as a supporting tool, and 3) the use of successive approximations by activities cycles to an
optimized numerical solution in which groups of parameters and periods are considered partially instead
of all at once.

Case 1: Pattern Injection Light-Oil Reservoir, Straightforward Coupling of Models-Optimizer


Situation and Complications. To maximize economic benefits, the primary objective of this study was to
optimize the number of new injector and producer wells, the drilling program, as well as fluid injection and
production rates. This case represents a relatively small field, in which optimization runs can be achieved
in a reasonable time with the available computing power.
The reservoir is a complex anticline with a series of structural blocks generated by major tectonic events,
located at a 14,300-ft depth. It was under primary production through rock-fluid expansion mechanisms.
The depositional environment is fluvial-deltaic; the four sand bodies have channels interdigitating. Three
types of rock are present, with a weighted permeability of 110 md. The oil is 35° API, with good mobility
relative to water.
12 SPE-185472-MS

Since 1976, 13% of the OOIP has been produced. During this period, the pressure declined from an initial
6,180 psi to lower than 3,000 psi, causing asphaltene deposition problems. Estimated ultimate recovery
without flooding was 14%.
During the early years of production, wells presented rates up to 1,400 STB/D. The reservoir showed
a production peak of 9,000 STB/D with nine active wells out of 20 drilled in this area. Presently, because
of declining reservoir pressure and mechanical problems, the total production is approximately 400 STB/
D from three active wells.
The available model is three-phase black oil with 200,000 cells. Production was properly history matched.
Current So was estimated to be 0.62, with a residual to water of 0.3. The results of deterministic numerical
forecast of waterflooding showed a possible 8.5% recovery factor. Running time was less than 30 minutes.
Proposed Solution. Using the optimizer machine, the available numerical model was coupled to an
economic model that considers the corporative guidelines of the operating company and the fiscal regime.
Stochastic runs were conducted for scenarios selected from the optimization run.
Fig. 9 shows the workflow used. Notice this was one of the first waterflooding numerical optimization
procedures conducted, and the workflow evolved later to that presented in Fig. 5. Therefore, Fig. 9 is
presented to better understand the work conducted and as a reference for the historical evolution.

Figure 9—Optimization workflow used in Case 1.

Approximately 20 personnel participated in the brainstorming exercise. The qualitative analysis


conducted by a reduced team yielded the different variables, values ranges, and constraints used in the runs,
which were validated by the extended team and managers.
The decision variables and constrains included the following:

• Well number maximum: nine producers and four injectors. This was determined to maintain the
minimum well spacing allowed.
• Maximum wells to be drilled: three after the second year of the project, as per rig availability.

• Location schedule: combination of 13 possible new locations.

• Maximum injection rate: 4,000 STB/D/well without fracturing the formation. Total injection plant
capacity was greater than 16,000 STB/D.
Table 3 shows the coefficients of variation and distribution types used for the stochastic risk analysis.
SPE-185472-MS 13

Results. Once begun, the optimization run was accomplished in fewer than 5 days without further
intervention of the users, as the optimizer automatically coupled the reservoir and economic models. The
run (Fig. 3) showed high NPV; this represents >85 monetary values at 25 iterations. Afterward, a large
number of iterations showed scenarios with NPVs between 85 and 92. The optimizer challenged the results,
determined areas with low values, and turned back to areas of high values. After 250 iterations, it was
considered that the "n" dimensional space was well explored and the run finished.
A key decision before the optimization run was to use injection and surface facilities already available.
For water injection, a nearby plant abandoned by another project was chosen. For production, it was checked
that facilities currently in use had sufficient spare capacity. Consequently, the project's capital expenditures
(CAPEX) was relatively low, with approximately 90% used for well construction. As expected, a good
correlation between NPV and Np was observed (Fig. 10). The maximum value had an NPV 6.28 times
greater than the base case, and the Np was 9.59 times greater.

Figure 10—Correlation of NPV and Np from the Case 1 optimization run; both parameters in relative units.

Fig. 11 shows the oil production forecast for all scenarios, the difference between scenarios is significant.
A production peak of 4,600 STB/D was achieved in 6.5 years by the maximum NPV scenario. After 20
years, for all scenarios with water injection, the forecasted oil production was still higher than current values.
No attempt was made to obtain a production plateau, as surface facilities were already in place.
14 SPE-185472-MS

Figure 11—Oil production forecast for all scenarios of the Case 1


optimization run; the green line corresponds to the maximum NPV scenario.

The injection rate showed a clear effect on NPV, with a maximum at 14,000 STB/D (Fig. 12), which
also has the maximum Np. At any injection rate, a number of scenarios exist below the local maximum that
correspond to insufficient drainage points (i.e., producer wells and unappropriated injector-producer wells
drilling sequence or locations).
With an injection rate of 11,000 STB/D, the reservoir reached 6,500 psi, which was close to the initial
value. Further injection rate increments did not increase pressure but increased water production and
therefore production costs slightly. All water injection scenarios were able to repressurize the reservoir
above the bubble point (i.e., 3,500 psi).

Figure 12—Water injection rate effect on NPV for all scenarios in Case 1 optimization run.

Six scenarios were selected as examples of obtaining more information by stochastic calculations that
support strategic decisions (Table 4 and Fig. 13).
SPE-185472-MS 15

Table 4—Description of scenarios selected for the stochastic runs.

Case Description

1 Base, only maintenance activities


Optimized location of three producers, one workover,
2
and one reperforation
Best previous scenario, four injectors and seven
3
producers
4 Optimized four injectors and seven producers
Maximum NPV of optimization, four injectors and nine
5
producers
6 Optimized two injectors and five producers

Figure 13—Relationship between risk and the incremental Np and NPV of


selected scenarios; numbers inside squares refer to the cases in Table 4.

The incremental factor is the ratio of the scenario mean values of NPV or Np to the base case values.
Uncertainties considered are shown in Table 2. Risk was estimated using the following formula:
(4)
where S is the standard deviation and X is the mean.
The 20-year fluid injection profiles and drilling schedules for each scenario were taken from the
optimization run output; Case 3 information was derived from a previous study. Each stochastic run
sampled randomly 3,000 inputs of the variables in Table 2 and calculated for fluid volumetric and economic
parameters, such as internal rate of return (IRR), investment efficiency (IE), and Np. The means and
standards deviations were calculated using these outputs.
The scenarios selected included those with high and low NPVs. Base Case 1 and Best Previous Scenario
3 are references.
Scenario 2, which considers three new producer wells, one workover and one reperforation, showed
almost twice the NPV and Np than the base case, with a relatively small risk increment and low cost. All
water injection scenarios showed NPV and Np increments greater than 4. In general, the risk is greater for
scenarios having higher Np and NPV, mostly as a consequence of greater drilling activity.
A comparison of Cases 3 and 4 shows how the optimizer can improve traditional scenario outcomes,
even keeping the number of wells constant. Scenario 4 shows an NPV close to Scenario 5 but with lower
risk. Scenario 4 also requires two fewer wells than Scenario 5; releasing the rig for activity in other fields
might be more profitable and should be decided according to the operator's opportunity portfolio.
16 SPE-185472-MS

In summary, this case study demonstrated the significant value added by the numerical optimization
with relatively small effort. Actually, the optimization of the waterflooding plan indicated scenarios with
NPV and Np increases greater than 400%. Compared to previous traditional numerical forecasts, the
optimization increased NPV by 46% and Np by 16%. Even considering the same number of wells as
the best traditional scenario, the optimization showed increases of NPV by 46% and Np by 16%, with
reduced risk. Additionally, the optimization indicated a large number of high-value scenarios, some of which
were not obvious, providing and unprecedented number of options and flexibility for waterflooding project
implementation.

Case 2: Pattern Injection Heavy-Oil Reservoir, Deductive HPM


Situation and Complications. The main objective of this study was to maximize the value of a
waterflooding project to be implemented in the Module V area of the Lower Lagunillas-03 reservoir. Early
calculations showed a marginal economy for a horizontal 1:1 direct-line injector-producer horizontal wells
pattern, which has been used successfully in neighboring areas. This made it necessary to test different well
patterns, which would be difficult using a single numerical optimization run (Tillero et al. 2014).
The Lower Lagunillas-03 reservoir, in Maracaibo Lake, Venezuela, has been producing for more than 80
years from the La Rosa (25° API) and Lower Lagunillas (19° API) formations, with the Lower Lagunillas
formation being the most prolific member. The formations were deposited in fluvial and fluvial-deltaic
sedimentary environments. More than 1,000 wells have been drilled; approximately 300 wells are currently
active, with an average production rate of 60 STB/D.
The average pressure of 350 psi was less than one-third its oil bubble point pressure; whereas, more
than 70% of the OOIP remained. Because of its low energy and large reserves, a plan to maintain reservoir
pressure and improve oil recovery by waterflooding in Lower Lagunillas-03 was implemented during
the last 40 years. In this reservoir, most waterflooding patterns have been tested, such as flank injection,
peripheral, direct line, inverted seven-spot, and direct line with horizontal wells. Nevertheless, different
results have been observed in term of production behavior and recovery factor. Therefore, an optimal
waterflooding scheme has not been identified for Module V (Lower Lagunillas formation) or future
waterflooding operations to help improve oil recovery and revitalize the potential of this mature reservoir.
Module V represents a locally refined model (Fernández et al. 2011) having 34 # 103 cells. OOIP is
approximately 671 million STB (614.7 million STB and 56.6 million STB for the Lower Lagunillas and La
Rosa formations, respectively) after equilibrium and convergence of initial conditions is achieved. History
matching of produced fluids and pressure behavior was achieved so the model could be used for forecasting
injection strategies in the main flow units (FS, LL_FS, and LL-90). For more detail, refer to Tillero et al.
(2014).
Proposed Solution. An adaptation of the HDM was created, from formulation to selection of feasible
scenarios (hypotheses) for hydrocarbon assets development (Fig. 14). HDM research is a typical version
of the scientific method, which describes the proceeding to falsify or corroborate, by experiment, some
predictions of a hypothesis. Popper (1963) established a deductive view on the scientific method, stating
that a hypothesis can only be falsified, meaning that a hypothesis should be affirmed or disproved using
experimentation or prediction.
SPE-185472-MS 17

Figure 14—HDM flowchart applied to asset development.

The sequence for formulating hypotheses, experimenting, analyzing, and formulating new ones (Fig.
14) begins by identifying the problem and stating general objectives about the research. To establish
association between the problem and the variables that affect it, it is necessary to characterize the
subject of inquiry (reservoir) through an analytical or numerical model. The next step is to formulate
hypothetical or theoretical explanations about future behavior of the subject of inquiry, considering field
data, measurements, observations, and expert criteria, among others. Some logical deductions based on
variables to be measured are needed before beginning the experimental process of hypotheses.
Hypotheses are tested by performing forecasts with analytical or numerical integrated models,
considering all aspects of a petroleum production system. Additionally, economic aspects should be
considered. The data obtained from predictions are gathered, and they can be statistically analyzed to
determine if certain relations among variables are important. These analyses require evaluating hypotheses
with the expectation of being proved or disproved or whether some emerging hypotheses might be conducted
and others might not. In turn, some hypotheses could be refined by data acquisition and pilot tests, among
others), if predictions are not accomplished substantially. The process must be cyclic until analysis of
prediction determines the probability of the hypotheses.
It is worth mentioning that the workflow in Fig. 14 represents refined steps to limit options/deterministic
runs and optimal scenario identification with numerical algorithms of the general workflow in Fig. 5.
18 SPE-185472-MS

For this study, in addition to the base scenario, three initial scenarios were tested for waterflooding of
Module V considering the last field experiences from the proper reservoir; in Module IV, a waterflooding
project using direct-line patterns was implemented by drilling horizontal producer and injector wells but
draining from the La Rosa formation. In Module V, the same type of wells and pattern were initially
considered but draining from the Lower Lagunillas formation. These four scenarios were grouped as
"Hypotheses Set 1."
The base scenario comprises minor well activities to maintain operations in the field. Scenario 2 includes
drilling 12 producers and 13 injectors (2,000 STB/D), all horizontal wells in flow unit LL_90. Scenario 3
considers developing Scenario 2 (LL_90), but after 6 years, horizontal producers and injectors wells are
drilled in flow unit LL_FS. Scenario 4 first considers developing the plan in Scenario 3 (LL_90 and LL_FS),
but after 9 years, 12 producers and 13 injectors (4,000 STB/D) are drilled in the FS unit. In all scenarios,
horizontal wells with a 1:1 ratio, direct line, with a horizontal length of 2,000 ft were considered.
Hypothesis Set 2 is composed of Scenarios 5 to 9. It initially tests seven-spot inverted patterns that
proved successful in neighboring areas of the field. Scenario 6 explores the hypothesis of shorter well
spacing between the producer and injector through diminishing heterogeneities effects that would increase
production. Scenarios 7 through 9 were defined taking into account similar premises as Scenario 6 but
increasing the number of well patterns, including sectors not previously drilled.
Results. These hypotheses were tested by numerical modeling, and they were also accompanied with
probabilistic assessment using an economic optimization process.
Scenarios 2 to 4 showed that a drilling campaign of injector and producer horizontal wells, in the main
flow units, substantially increased oil recovery. However, none of the scenarios provided better economic
benefits than Base Scenario 1 (Fig. 15). This was a result of the significant initial investment and large
number of existing vertical wells that would be abandoned during the horizontal drilling campaign because
of the interference between their drainage radii.

Figure 15—NPV vs. time for Base Scenario 1 and horizontal well Scenarios 2 to 4.

The tested initial hypotheses falsified the theory of maximizing asset value (Module V) through
waterflooding using horizontal well patterns. This result led to the formulation of a new group of hypotheses
(Hypotheses Set 2) considering how best to develop Module V's reserves using a new proposal of
waterflooding patterns or types of producer and injection wells. Therefore, Hypotheses Set 2 initially
considered minimizing the CAPEX by converting vertical producers to injector wells and using commingled
injecting-producing in the three main units (i.e., LL_90, LL_FS, and FS). The process began by testing
seven-spot inverted patterns, which proved successful in neighboring areas of the field. Scenario 5 comprises
redrilling a well as an injector, converting 12 existing producing wells to injection wells with one available
rig, and activating three producing wells using platform activities.
SPE-185472-MS 19

Surprisingly, the forecast showed less daily and cumulative production than the base scenario.
Furthermore, a sensitivity analysis conducted coupling the reservoir model with the economic model using
an optimizer machine showed and inverse dependence of NPV with the number of converted wells (Fig. 16).

Figure 16—NPV effect on the producer to injector well conversions, Scenario 5 (after Tillero et al. 2014).

Production losses resulting from the conversion of the producing wells was not replaced through water
injection. Therefore, the tested hypothesis falsified the theory of maximizing asset value through only
converting to inverted seven-spot patterns.
Scenario 6 explores the hypothesis of shorter well spacing between the producer and injector through
diminishing heterogeneities effects that would increase production. It considers drilling 12 injector wells
with one available rig, redrilling one producing well, and reopening one producing well through platform
activities using 11 well patterns.
Despite the greater Np value compared to Base Scenario 1, the NPV of Scenario 6 resulted in a loss of
USD 27 million (Table 5). As a consequence, new conjectures resulted from Scenario 6 regarding increasing
oil recovery through inverted five-spot waterflooding patterns. Such conjectures necessitated formulating
a new hypothesis, taking into account the proposed pattern and other premises:

• Sensitizing the injection rate of each well might lead to greater oil recovery.

• Increasing the number of inverted five-spot patterns might help increase ultimate recovery.

• Optimizing the drilling schedule might help maximize the NPV.


20 SPE-185472-MS

Table 5—Summary of economic and production results from Scenarios 1 to 9, Module V, Case Study 2.

Scenario Np Wp (STB millions) Wi (STB millions) NPV (USD millions) ΔRF (%)

1. Base 18.8 5 0 202 3


2. Drilling campaign of horizontal well in flow unit
22.5 54.8 183 144.7 3
LL_90
3. Drilling campaign of horizontal well in flow units
34.1 90 319 183.4 5
LL_90 and LL_FS
4. Drilling campaign of horizontal well in flow units
50.4 133 506 207 5
LL_90, LL_FS, and FS
5. Inverted seven-spot waterflooding pattern, flow
14.5 13 106 104.4 2
units LL_90, LL_FS, and FS
6. Inverted seven-spot waterflooding pattern (pilot
20.6 21 61 175 3
test), flow units LL_90, LL_FS, and FS
7. Inverted seven-spot waterflooding pattern for
54.3 134 284 386 8
short-term, flow units LL_90, LL_FS, and FS
8. Inverted seven-spot waterflooding pattern for
69.7 194 359 455 10
mid-term, flow units LL_90, LL_FS, and FS
9. Inverted seven-spot waterflooding pattern for
87.1 229 409 533.6 13
long-term, flow units LL_90, LL_FS, and FS

Scenarios 7 to 9 were defined taking into account a similar premise as Scenario 6 but increasing the
number of well patterns.
Scenario 7 considered drilling 33 injection and 16 producing wells with one available rig. In addition,
this scenario included drilling a twin well and activating one producing well. Scenario 8 considered drilling
56 injection wells and 37 producing wells with one available rig, redrilling one well, and activating one
producing well using platform activity. This meant an extended area was considered. Because extending
the area to be flooded proved profitable, Scenario 9 explored increasing the number of inverted five-spot
patterns until the concession area was filled. This scenario considered drilling 79 injection and 53 producing
wells with one available rig.
Concerning the drilling schedule, an optimal sequence was determined based on NPV as the objective
function vs. cumulative oil production (Fig. 17). Injection rate values of 1,500, 3,000, and 4,500 STB/D
were considered for the sensitivity process, with most injection rates set at 3,000 STB/D.

Figure 17—Example of the drilling sequence and injection rate optimization;


each dot is a scenario and colors represent injection rates. Yellow = 3,000 STB/D.

Scenarios 7 to 9 showed increased Np up to 463%, NPV up to 263%, and an additional recovery factor
up to 10%, respectively, compared to the base scenario (Table 3). Scenario 9 comprises Scenarios 6, 7, and
SPE-185472-MS 21

8, allowing development of Module V in stages to reduce uncertainties associated with the asset life cycle
(mature field), meaning that waterflooding would be used to maximize existing wells (Scenario 6) to obtain
experience and collect data for better characterization of patterns productivity and injectivity. If the pattern
productivity was profitable, the next stages of implementation (Scenarios 7 and 8) would begin, with their
respective profitability assessments. Finally, the total number of patterns suggested by Scenario 9 would
be completed.
Fig. 17 shows the risk quantification results. In this figure, Scenarios 2 to 6 are clearly unsuitable because
they have a smaller or similar NPV to Base Scenario 1 but with more risk. Scenarios 7 to 9, while more
risky than Base Scenario 1, also showed substantially higher NPVs.

Figure 17—Mean of NPV (USD millions) vs NPV std. deviation (%) of tested hypotheses; bubble side is proportional to Np.

In summary, HDM was used to corroborate or falsify the proposed hypotheses (expectations from
scenarios) through numerical simulations of hydrocarbon asset development. Using HDM can be beneficial
for the formulation and selection of feasible exploitation plans to maximize hydrocarbon asset values.
For Module V, formulation, testing, analysis, and proposition of new conjectures about injection patterns,
well configuration, and injection rates, among others, were initiated by the results analysis (economic
optimization) of deterministically tested hypotheses (Scenarios 1 through 4).
Contrary to expected results, waterflooding schemes, such as inverted seven-spot patterns and horizontal
1:1 direct-line injector-producer horizontal wells, which were successful in neighboring areas, showed
marginal economic benefit.
The risk analysis provided useful elements for decision making. Scenarios related to horizontal wells
increased risk more than those involving vertical wells, with an acceptable value of 50% expressed as the
ratio of standard deviation to the absolute NPV.
The implementation of HDM provided useful insight and an appropriate framework for defining
scenarios to help maximize hydrocarbon asset values. It differs from a traditional approach in that new
information was generated and it did not follow the initial settings. This can be exacerbated in the presence
of counter-intuitive findings. Flexibility when developing an FDP can help increase the added value.

Case 3: Pattern Injection Heavy-Oil, Successive Approximation Method


Situation and Complications. The primary objectives of numerical optimization in this case was to
determine scenarios with better performance than those from the deterministic scenarios in the previous
runs, maximize the economic benefit, and quantify risks to support managerial decisions concerning the
proposed waterflooding plan (Millan et al. 2012).
22 SPE-185472-MS

This reservoir has four units, ranging in age from Medium- to Late-Miocene, deposited in a fluvial-
lacustrine environment. The main facies are associated with meandering channels with twisted knees, edges
of channels, and floodplains. The reservoir is in a faulted monocline.
In 1956, exploitation began through commingled production (Fig. 18). In 1970, the production peaked
at 68,000 STB/D. The drive mechanisms are an aquifer and gas in solution. In 1964, the injection of water
by flank was begun for a total of 600 million barrels of water accumulated.

Figure 18—Production history.

Drilling and water injection were suspended in 2005 and 2007, respectively, after years of low
daily production. By the time the optimization study was performed, the production was 5,000 STB/D.
Approximately 33% of the OOIP had been produced. However, previous studies indicated that the reservoir
is a candidate for revitalization. Given the high volumes of heavy oil, 17.4° API, and the remaining areas
with little or no sweep, the production potential was considered high.
Because of declining production and the existence of remaining reserves with estimated average oil
saturation of 0.42, the operator began redesigning the operation scheme before conducting the numerical
optimization study. Static and dynamic models were constructed, and five cases were numerically simulated
as possible business plans, focusing on water injection and infill drilling.
The best deterministic case consisted of infill drilling 39 wells, repairing 22 wells, reconditioning/
recompleting two injection wells, converting four producer wells to injector wells, and drilling a horizontal
injector to inject 52,000 BWPD. This case reached a peak production forecast of 31,000 STB/D. By 2050,
recovery is estimated to reach 50.1%, which represents an addition of 17% compared to the base case.
The average production between 2040 and 2050 is projected at 9,000 STB/D. A detailed description of the
reservoir, numerical model, and previous work is discussed in Millan et al. (2012) and references therein.
The construction of injection and producer wells comprised a significant percentage of the total
investment. Because of the current low pressure and energy conditions, the water injection project was
redesigned. It was decided to focus on decision variables associated with those aspects of the project and the
uncertainties of reservoir variables that affect the flow capacity of crude oil to the wells and, consequently,
production.
The main complication was the optimization running time of the model, which had 227 # 103 cells, 155
wells, and 60 years of production history. The operator requested runs be made locally with relatively small
computing capacity and some network instability.
Proposed Solution. To perform optimization runs in just a few days by decreasing the iterations time, it was
proposed and tested to begin with exploratory runs focusing on production and short spans. It is known that
a strong correlation between NPV with Np exists for relatively non-intensive CAPEX projects, and NPV
is strongly determined by the early performance rather than after many years. Accordingly, the exploratory
runs considered 5-year forecasts to maximize Np. Later, a final run considered a 20-year forecast and NPV
maximization.
SPE-185472-MS 23

The optimization procedure comprises four cycles. Cycle 1 was defining the reconditioning/recompletion
and reactivation activities to be performed during 2011 to 2013 using available resources. After 2014, one
drilling rig was assumed to be available. Using the optimizer, the selection of new well locations and types
(i.e., injector or producer) to be drilled each year was optimized to first maximize production (Np) and later
NPV. It should be emphasized that the activities scheduling considered constraints of drill rig availability,
capacity of water injection facilities, and time required to install lines.
The procedure continued as follows: after Cycle 1, activities for 2012 were fixed according to the best
scenario determined, and its conditions were used as input for the Cycle 2 run. In the same way, Cycle 2
had fixed activities for 2013 onward.
Fifty-five possible locations were determined for producer wells and nine possible locations were
identified for injector wells from the Department of Integrated Studies. Three cycles were performed to
maximize Np. In 2017, two rigs are assumed to be available, which is sufficient to complete the drilling
campaign. Therefore, final optimization was conducted as Cycle 4, which considered all decision variables,
with the objective function being maximizing NPV.
Each optimization cycle considered 5-year forecast runs (i.e., the period that most significantly affects
NPV). This was performed to decrease iteration time and hardware needs. After the scenarios with maximum
Np or NPV were identified, 20-year forecasts were performed for each of them.
Based on historical water injection, Cycle 4 was considered to have a low rate of initial water injection
(i.e., < 6,000 B/D). Water injection was also expected to show significant declines over time as a result of
pressurizing the reservoir. Therefore, to reduce the number of variables, fixed rates were assigned according
to previous deterministic simulation runs for Injector Wells I-1, I-2, and I-3 drilled in 2013 and I-5 (Cycle
Optimization 2) in 2015.
To optimize the water injection rate in the remaining five wells that targeted deeper formations, the
injection rate achieved at the maximum BHP (i.e., BHPmax rate) was used as a constraint. It is worth mentioning
that values greater than these should fracture the formation. Variable values were assigned by using the
following multipliers: BHPmax rate × 1, BHPmax rate × 25/40, and BHPmax rate × 1/4.
The sum of the injection rates of the five wells was restricted to 100,000 BWPD; this is the total capacity
of the injection plant was less at 25,000 BWPD. Given that the maximum oil production rate in previous
deterministic runs was 37,000 B/D, the injection of 100,000 BWPD ensured a replacement factor. Therefore,
it was considered unnecessary to build an additional water injection plant.
The economic model was developed using a spreadsheet. The investment efficiency was anticipated to
be high because the investment to be made in new facilities was relatively low; the use of existing facilities
was maximized. Therefore, the ultimate target function was to maximize NPV. The greatest cost would be
associated with installing flow lines, particularly for drilling of new producer and injector wells.
After the optimization cycles, key scenarios were selected and stochastically analyzed.
Results. Best scenarios were identified for each cycle. As expected, a good linear correlation existed
between NPV and Np (Fig. 19), which justified the use of Np in the exploratory runs.
24 SPE-185472-MS

Figure 19—Np and NPV correlation for 330 scenarios of Cycle 4 optimization (after Millan et al. 2012).

NPV and Np showed to be sensitive to the number of injector and producer wells to obtain maximum
output (Fig. 20). The NPV difference between the worst and best case was 20%. Many scenarios showed
high NPV values, which provides flexibility when choosing the scenario to be used and eventually allows
switching between scenarios according to operational situations.

Figure 20—Effect of the number of new wells on NPV.

Table 6 summarizes the results for the best scenario of each cycle (i.e., those with maximum NPV).
Notice the 300 to 400% increments in Np and NPV compared to the base scenario.
SPE-185472-MS 25

Table 6—Summary of 20-year forecast results of selected scenarios for four optimization cycles compared to the base case.

Year 2014, Cycle Year 2015, Cycle Year 2016, Cycle Year 2017, Cycle
1, Optimum 2, Optimum 3, Optimum 4, Optimum
Parameter Base Scenario Scenario 74 Scenario 124 Scenario 257 Scenario 328

Np (STB millions) 33 109 141 160 171


Wp (STB millions) 20 82 150 190 204
Wi (STB millions) — 201 269 456 482
NPV (−) 426 1,199 1,576 1,880 2,008
IE (−) 159 29 26 24 20
Return on investment
— — 254 164 120
(ROI) (%)

The optimizer provides complete information for all scenarios, including number of wells, location and
drilling sequence, number of workovers, location and work sequence, total and per well injection rate, fluid
injection-production, and pressure forecasts. Table 7 shows the optimized number and types of wells to be
drilled for each cycle. This information cannot be easily obtained using other methods.

Table 7—Number of injector and producer wells to be drilled during the proposed 4-year campaign of the exploitation plan.

Year 2014, Cycle 1, Year 2015, Cycle 2, Year 2016, Cycle 3, Year 2017, Cycle 4,
Parameter
Optimum Scenario 74 Optimum Scenario 124 Optimum Scenario 257 Optimum Scenario 328

New injector wells 2 2 1 —


New producer wells 10 10 11 17
Total 12 12 12 17

In summary, the cumulative oil production forecast was maximized with the redesign of the reservoir
exploitation plan; a method that considered four successive cycles of optimization by numerical algorithms
was used. Each cycle considered the following decision variables: location and sequence of injection and
producers wells to be drilled in a 4-year campaign constrained by rig availability. In Cycle 4, NPV was
maximized and included water injection rates constrained by water injection plant capability as a decision
variable.
The value of a methodology focused on better definitions of possible actions during the early years of
the exploitation plan was proven, showing the best result for NPV. This methodology has the advantage of
identifying actions and annual investments, and depending on oil price changes or availability of resources,
the proposed plan can be partially implemented and still provide substantial economic benefits. Other
advantages provided by the methodology that made the study feasible are a reduction in necessary hardware
and run time optimization.
Forecasts of Np and NPV increases were highly positive for all cycles, resulting in values between 300
and 400% higher than the base case. According to the stochastic quantification of risk, the possibility of
negative values is negligible if the plan is operationalized correctly.
One restriction on this type of project is the amount of information available. With complete information,
including knowing the actual reservoir conditions and changes in oil prices, additional phases are not
necessary, and it would be relatively easy to define an exploitation plan to operate indefinitely. In actuality,
however, the commitment to drilling and construction is dependent on receiving new information from
the market (oil prices) and field (oil trapped), so successive stages enable adjusting activities over time.
Therefore, the exploitation plan should be reviewed frequently.
Most studies end with determining the amount of fluid to be injected; this study proposes a dynamic
injection. Future work should include a continuous assessment of the value generated by water injection.
26 SPE-185472-MS

Given the current cost of resources, the efficient use of drilling rigs, injection, personnel, and facilities
can reduce costs and extend the life of the reservoir. Follow-up studies are needed to achieve efficiency;
optimization should be performed continuously. Information generated through surveillance and other
studies can potentially help improve project economics by reducing the number of wells to be drilled or
improving the location of new prospects. Therefore, continued study, including a review program for water
injection optimization, is recommended.

Moving Forward: Intelligent Wells and Integrated Smart Waterflooding


Previous sections showed a general optimization workflow (Fig. 5) and study cases that highlighted three
different methodologies to identify optimum field development scenarios through the use of numerical
algorithms. These methodologies apply when computing capacity is enough (Case study 1), in situations
in which two or more deployment schemes are to be ascertained e.g. line drive, 5 and 7 spots well patterns
(Case study 2) and in situations in which one scheme is selected and optimization focuses on optimizing
its parameters (Case study 3)
The three study cases showed the substantial value added by numerical optimization. However, these
cases are snapshots of waterflood optimization. In reality, operators can often face challenges in field
operations that cannot be accounted for in planning like unexpected production behavior, well shutdowns
and changes in operational and budgetary plans. Any of these challenges can cause existing field
development plans to be changed, either in the short or the long term. To tackle this reality, a dynamic
approach to waterflood optimization is called for.
To this end, several equipment solutions continue to improve waterflooding deployment. For example,
intelligent well completions allow for continuous, real-time monitoring through downhole sensors and
surface control using downhole valves. While these technologies undoubtedly enhance the deployment of
waterflooding, a more holistic solution is the use of integrated workflows that can be used to create a robust
software, network, and engineering infrastructure to leverage new digital technologies in the field to enhance
waterflooding deployment.
Workflows serve to optimize modeling and diagnostic capabilities at the reservoir, well, and surface
levels. Workflows typically consist of mapping real-time input variables from digital instrumentation in the
field, transmitting them to collaboration centers by means of a sophisticated communication infrastructure,
and orchestrating processes necessary to achieve the desired outputs, whether they be problem diagnosis,
updating reservoir, well and network models, or setting optimized operational targets. Each workflow
can operate independently or, ideally, in an integrated fashion. For example, if production targets are not
being met, pump settings for individual wells can be optimized to increase production. Conversely, a fully
integrated optimization run taking into account the latest data from digital instrumentation in the field
can help determine a solution to increase production while ensuring that the NPV for the asset remains
optimized.
Workflows can use a swim-lane based architecture (Fig. 21) to enable them to run individually or in
tandem (i.e., in an integrated fashion).
SPE-185472-MS 27

Figure 21—Smart waterflooding; basic workflows for an integrated approach.

The swim-lane-based approach allows for field updates to be made in real-time as well as on a periodic
basis. Workflows can provide daily diagnosis of problems in injection and production wells through
monitoring of operations with set targets. Well models and network models can be adjusted based on field
data, and decisions can be made to meet field targets set at regular intervals by planning departments. On
a macro scale, the reservoir workflows, either individually or in an integrated fashion, can automatically
history match the reservoir model at regular intervals and set new operational targets through numerical
modeling and optimization exercises.
This approach to waterflooding provides a vastly less error prone operational environment, as well
as mitigates a key issue in waterflood optimization planning—continuous optimal operations. While
optimization exercises presented in the case studies here provide the best plans going forward, they are
static exercises. As new operations are conducted in the field and new information is obtained, these plans
should be continually updated. The integrated, swim-lane-based approach to smart waterflooding helps
ensure different levels of updates occur, from the well to the integrated level, to continuously help ensure
operations achieve the goal of optimized waterflooding.

Conclusions
This paper demonstrates that numerical optimization of waterflooding projects allows determining scenarios
to help improve production, increase recovery factors, and provide economic benefits, such as increased
NPV, compared to using traditional simulation methods.
The optimization runs determine nonobvious scenarios, and the large number of scenarios identified
provides unprecedented flexibility for project deployment and reassessment.
28 SPE-185472-MS

The optimization running time is an important constraint in some cases. However, various solutions based
on using more powerful hardware/parallel computing, simplified models, and reduced solution space have
proven successful. All of these solutions can be incorporated in the general workflow proposed.
Stochastic analysis of all scenarios considered during an optimization run can be time consuming.
Analysis of selected scenarios is valuable for quantifying risks and the most important variables. The
knowledge provided by stochastic optimization—scenarios with maximum benefits, a large number of
scenarios with higher value than those determined using traditional approaches, and risk quantification—
is a powerful tool for improving water project management.
Information gathered during a project is of paramount importance for updating models and ascertaining
the best scenario for an extended period of time. However, manually conducting these processes in a
reactive manner can necessitate significant time and leave possible added value unrealized. The current
trend of automatically capturing and processing data, updating models, and performing optimization runs
that identify necessary actions is helping solve these issues; efforts to fully integrate the injection network,
reservoir, and production network models are expected to advance waterflooding optimization and, in
general, all types of EOR flooding.

Acknowledgments
The authors thank Halliburton for permission to publish this paper. Reviews of the manuscript by F. Peretti
and L. Lujan are gratefully acknowledged.
DMS™ is a registered trademark of Landmark Oil & Gas Ltd., Houston, Texas.
Nexus® is a registered trademark of Landmark Oil & Gas Ltd., Houston, Texas.
Windows® is a registered trademark of Microsoft Corporation, Redmond, Washington.

Nomenclature
BHP = bottomhole pressure
BWPD = barrels of water per day
DCV = definition, conceptualization, and visualization
EOR = enhanced oil recovery
FDP = field development plan
FEL = front end loading
HDM = hypothetic-deductive method
IE = investment efficiency
IRR = internal rate of return
MMscf/D = million standard cubic feet per day
Np = cumulative oil
NPV = net present value
OOIP = original oil in place
ROI = return of investment
So = oil saturation
STB/D = standard barrels per day
VCD = visualization, conceptualization, and definition
VFP = vertical flow performance
Wi = cumulative injected water
Wp = cumulative produced water
ΔRF = incremental recovery factor
SPE-185472-MS 29

References
Cullick, S.A., Heath, D., Narayanan, K. et al 2003. Optimizing Multiple-field Scheduling and Production Strategy with
Reduced Risk. Presented at the SPE Annual Technical Conference and Exhibition, Denver, Colorado, USA, 4-6
October. SPE-84239-MS. http://dx.doi.org/10.2118/84239-MS.
Cullick, S.A., Narayanan, K., and Gorell, S. 2005. Optimal Field Development Planning of Well Locations with Reservoir
Uncertainty. Presented at the SPE Annual Technical Conference and Exhibition, Dallas, Texas, USA, 9-12 October.
SPE-96986-MS. http://dx.doi.org/10.2118/96986-MS.
Fernández, E., Acosta, A., and Merhdad, M. 2011. Modelo Geoestadístico y Evaluación de Inyección de Agua mediante
Simulación Numérica en el Yacimiento LL-03, Módulo V, Formación Lagunillas. Internal Report PDVSA: 57.
Millan, M., Mogollón, J.L., Lujan, L. et al 2012. Optimization of a Heavy Oil Reservoir Exploitation Plan by Using a
Numerical Algorithm and Focusing on Short Time Span. Presented at the World Heavy Oil Conference, Aberdeen,
Scotland, 29 May. WHOC12-357.
Mogollón, J.L., Tillero, E., Gutiérrez, I. et al 2016. Numerical Maximization of the Secondary Polymer Flooding Value in
a Mature, Offshore, Heavy Oil Reservoir. Presented at the Offshore Technology Conference, Houston, Texas, USA,
2-5 May. OTC-27189-MS. http://dx.doi.org/10.4043/27189-MS.
Narayanan, K., Cullick, A.S., and Bennett M. 2003. Better Field Development Decisions from Multiscenario,
Interdependent Reservoir, Well, and Facility Simulations. Presented at the SPE Reservoir Simulation Symposium,
Houston, Texas, USA, 3-5 February. SPE-79703-MS. http://dx.doi.org/10.2118/79703-MS.
Perez-Perez, A., Mujica, M., Bogdanov, I. et al 2014. A Methodological Analysis of the Mechanisms Associated to Steam-
Solvent Co-injection Processes using Dynamic-Gridding. Presented at the SPE Improved Oil Recovery Symposium,
Tulsa, Oklahoma, USA, 12-16 April. SPE-169075-MS. http://dx.doi.org/10.2118/169075-MS.
Popper, K. 1963. The Logic of Scientific Discovery. Abingdon-on-Thames, UK: Routledge.
Rashid, K., Bailey, W., Couet, B. et al 2013. An Efficient Procedure for Expensive Reservoir-Simulation Optimization
Under Uncertainty. SPE Economics & Management 5 (4): 21–33. SPE-167261-PA. http://dx.doi.org/10.2118/167261-
PA.
Temizel, C., Purwar, S., Urrutia, K. et al 2015. An Analysis of the Effect of Surfactant Flood on Residual Oil under
Composition-Dependent Multi-Contact Water-Oil Miscibility in a North Sea Reservoir. Presented at the Offshore
Technology Conference, Houston, Texas, USA, 4-7 May. OTC-25657-MS. http://dx.doi.org/10.4043/25657-MS.
Tillero, E., Perez, E., Gonzalez, J. et al 2011. Optimization, Uncertainty, and Risk Analysis for Asset Development
with Limited Data and Multiple Geographical Environments: A Heavy-Oil Field Case. Presented at the
SPE Heavy Oil Conference and Exhibition, Kuwait City, Kuwait, 12-14 December. SPE-148903-MS. http://
dx.doi.org/10.2118/148903-MS.
Tillero, E., Mogollón, J.L., Gutierrez I. et al 2014. Maximizing Asset Value by Waterflooding Based on a Hypothetic-
Deductive Method for Formulation, Evaluation, and Selection of Exploitation Scenarios. Case Study Lower
Lagunillas-03 Reservoir. Presented at the SPE Heavy and Extra Heavy Oil Conference: Latin America, Medellín,
Colombia, 24-26 September. SPE-171137-MS. http://dx.doi.org/10.2118/171137-MS.

View publication stats

You might also like