You are on page 1of 21

SPE-206300-MS

Selection of Representative Scenarios Using Multiple Simulation Outputs for


Robust Well Placement Optimization in Greenfields

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


Seyed Kourosh Mahjour, Antonio Alberto Souza Santos, Susana Margarida da Graca Santos, and Denis Jose
Schiozer, University of Campinas

Copyright 2021, Society of Petroleum Engineers

This paper was prepared for presentation at the 2021 SPE Annual Technical Conference and Exhibition held in Dubai, UAE, 21 - 23 September 2021.

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
In greenfield projects, robust well placement optimization under different scenarios of uncertainty
technically requires hundreds to thousands of evaluations to be processed by a flow simulator. However,
the simulation process for so many evaluations can be computationally expensive. Hence, simulation runs
are generally applied over a small subset of scenarios called representative scenarios (RS) approximately
showing the statistical features of the full ensemble. In this work, we evaluated two workflows for robust
well placement optimization using the selection of (1) representative geostatistical realizations (RGR)
under geological uncertainties (Workflow A), and (2) representative (simulation) models (RM) under the
combination of geological and reservoir (dynamic) uncertainties (Workflow B). In both workflows, an
existing RS selection technique was used by measuring the mismatches between the cumulative distribution
of multiple simulation outputs from the subset and the full ensemble. We applied the Iterative Discretized
Latin Hypercube (IDLHC) to optimize the well placements using the RS sets selected from each workflow
and maximizing the expected monetary value (EMV) as the objective function. We evaluated the workflows
in terms of (1) representativeness of the RS in different production strategies, (2) quality of the defined
robust strategies, and (3) computational costs. To obtain and validate the results, we employed the
synthetic UNISIM-II-D-BO benchmark case with uncertain variables and the reference fine- grid model,
UNISIM-II-R, which works as a real case. This work investigated the overall impacts of the robust well
placement optimization workflows considering uncertain scenarios and application on the reference model.
Additionally, we highlighted and evaluated the importance of geological and dynamic uncertainties in the
RS selection for efficient robust well placement optimization.

Introduction
The location of the wells in petroleum reservoirs plays an essential role in the performance of oil production.
However, determining the right place of wells can be challenging due to the interaction of rock and fluid
and the effects of adjacent wells on each other (Janiga et al., 2019). Thus, the well placement optimization
is required to find the best well positions within the reservoir. The location of the wells is determined
by maximizing the targeted objective functions, such as the net present value (NPV) or cumulative oil
2 SPE-206300-MS

production (NP), obtained by running a numerical simulator. The simulation process can be complicate
significantly when incorporating uncertainty related to the reservoir properties (Almeida et al., 2014).
Hence, instead of considering a deterministic reservoir model, the robust well placement optimization is
applied over the ensemble of reservoir models to capture the reservoir uncertainty space (Badru and Kabir,
2003; van Essen et al., 2009; Yang et al., 2011, and Chang et al., 2015). During the robust well placement
optimization, the user looks for an optimal solution that has high efficiency on average for all reservoir
models under reservoir uncertainty (Yang et al., 2011).

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


The desirable course for the robust well placement optimization is to use all possible scenarios but often
the computational and time constraints of users render this impossible, and it becomes humanly difficult
to handle all scenarios. Therefore, simplification techniques are needed to accelerate the optimization
process (Schiozer et al., 2019). The techniques can usually be divided into three main groups: (1) building
proxy models, (2) generating low-fidelity models and, (3) selecting the representative scenarios (RS). It
is noteworthy that inappropriate simplification techniques can lead to defining an unsuitable production
strategy during the robust optimization, which can decrease the project's financial return. Hence, it is
important to carefully select a proper technique according to the type of available data and interpretations.
Proxy models have been used in risk analysis, history matching, and robust optimization in different studies
(Scheidt et al., 2007; Feraille and Marrel, 2012; Douarche et al., 2014; Panjalizadeh et al., 2014, and Imrie
and Macrae, 2016). These models are applied to bypass the numerical simulation using analytical functions.
Although this method is quick to evaluate the thousands of possible solutions, defining different assumptions
and approximations that are not consistent with the physical rules of fluid flow in the reservoir can reduce
the accuracy of the production forecast (Trehan et al., 2017; Schiozer et al., 2019).
Generating low-fidelity models is another method of simplification in rock and fluid features (Wilson and
Durlofsky, 2013; Aliyev and Durlofsky, 2015; Santos et al., 2020). In this technique, the fine grid models
are upscaled to coarse grid models to reduce the simulation runtime. Generating the low-fidelity models is
a fairly simple task and, to an extent, these models also follow the physical rules of the reservoir However,
the resolution of the models is low and some sub-grid heterogeneity is missed during the upscaling process
(Durlofsky, 1998, and Zabalza-Mezghani et al., 2004).
Nowadays, RS selection methods are widely applied to decrease the number of scenarios and accelerate
the robust optimization in greenfields (Schiozer et al., 2004; Scheidt and Caers, 2009; Shirangi and
Durlofsky, 2016; Kang et al., 2019, and Mahjour et al., 2020a). In these methods, a subset of scenarios
is normally selected, which roughly represents the features of the full ensemble. It can, however, be
challenging to ensure the uncertainty space of the reservoir using the RS.
A variety of RS selection techniques exist and these depend on the usage of petrophysical properties
(static data) or simulation outputs (dynamic data), which are referred to as the features of the scenarios (Lee
et al., 2013; Haghighat Sefat et al., 2016; Shirangi and Durlofsky, 2016; Sarma and Chen, 2013, Mahjour et
al., 2020a, and Meira et al., 2020). The feature can be anything to consider as the input used in RS selection
(Mahjour et al., 2020b). To apply the petrophysical properties, there is no need to run numerical simulation
but, in order to employ the dynamic data, running numerical simulation is required (Mahjour et al., 2021).
The petrophysical properties can be referred to as reservoir features that are not changed over time, such as
facies, porosity, and permeability. In contrast, the simulation outputs, such as NPV and oil recovery factor
(ORF), are normally changed during the production period.
Note that the RS selection methods differ given the different purposes. For instance, the method used to
choose the RS for well placement optimization is not the most suitable one to be applied in well control
optimization (Shirangi and Durlofsky, 2016). Additionally, two main issues regarding the RS selection need
to be considered: (1) the number of RS should be sufficiently large to cover the reservoir uncertainty, and
(2) the number of RS should be limited to increase computational efficiency (Mahjour et al., 2020b). These
are the reasons why additional studies are still needed to investigate the best method for RS selection.In
this study, we evaluate two workflows to select the RS for efficient robust well placement optimization. We
SPE-206300-MS 3

test an existing RS selection technique named RMFinder (Meira et al., 2020) in different ways to find how
representative the decisions are by using different RS sets.
Recently, RMFinder has been widely used to select the RS in different field development applications
(e.g., Schiozer et al., 2019; Santos et al., 2020). In all works, a base production strategy was defined
to obtain the simulation outputs. The main base of the RMFinder is the integration of a mathematical
function with a metaheuristic optimization algorithm to ensure that the simulation outputs obtained from
the subset and the full ensemble are statistically close to each other. This method allows the user to

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


consider the relative importance of each specific simulation output during the scenario reduction process.
Furthermore, RMFinder considers cross-plots and cumulative distribution function (CDF) curves plotted
by the simulation outputs, and the importance weight of each one can be defined by the user to select the
RS. Details of the RMFinder method can be found in Meira et al. (2020).
To obtain and validate the results, we employed the UNISIM-II-D-BO benchmark case as a synthetic
fractured reservoir model and the reference fine-grid model, UNISIM-II-R, which works as a real case
(Correia et al. 2015). While the robust well placement optimization workflows have been well studied for
uncertain scenarios, their overall impact on the reference models working as the real reservoirs has not
been investigated yet. This paper also highlights and evaluates the importance of geological and dynamic
uncertainties in the RS selection for efficient robust well placement optimization.

Methodology
We investigate two workflows for robust well placement optimization using the RS selection (Figure 1).
Each workflow contains four main steps: (1) scenario generation (blue boxes), (2) scenario reduction or RS
selection (green boxes), (3) filtering using the available historical data (peach boxes), and (4) optimization
(yellow boxes). The focus of this study is on the scenario reduction step to select the RS shown in green:
1. Workflow A: selecting the representative geostatistical realizations (RGR) considering geological
uncertainties only, which are then combined with all dynamic uncertainties without further scenario
reduction;
2. Workflow B: selecting the representative (simulation) models (RM) considering the combination of
geological and dynamic uncertainties simultaneously.
4 SPE-206300-MS

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


Figure 1—Robust well placement optimization using representative scenarios selected
based on two approaches: RGR selection (Workflow A) and RM selection (Workflow B).

RGR selection is normally performed under geological uncertainties, which means that the focus of the
scenario reduction is on the geological realizations. Geological uncertainties are related to the distribution
of petrophysical properties during geological modeling (Santos et al., 2018, and Mahjour et al., 2019).
Instead, the RM set is generally selected under the combination of geological and dynamic uncertainties.
Dynamic uncertainties can be related to the flow properties, production system availability, and economic
investments/expenditures (Santos et al., 2018).
In both workflows, we apply and test RMFinder to select the RS based on multiple simulation outputs
(dynamic data). The simulation outputs considered in this study are divided into two main groups:
1. Field indicators: net present value (NPV), cumulative oil production (NP), oil recovery factor (ORF),
oil in place (OIP), water in place (WIP), cumulative water production (WP), and cumulative Water
Injection (WI);
2. Well indicators: well economic indicator (WEI) for production and injection wells (Botechia et al.,
2013).
The purpose of calculating WEI is to determine the revenue, costs, and investments of each well during
the time.
We investigate each workflow in terms of:

• The representativeness of the RS in different recovery mechanisms (water injection and water
alternating gas (WAG) injection) and well positions;
• The quality of the defined robust strategies;

• The computational time and the number of flow simulation runs.

Workflow A: RGR selection


In Workflow A, we first build nnumber of simulation models under geological uncertainties in a way that
each model has a unique geological realization (GR). According to Figure 2, the dynamic uncertainties
related to the fluid properties, rock-fluid properties, operational data, and economic data are not considered
SPE-206300-MS 5

in the simulation process. Hence, the number of simulation models is the same as the number of GR. We
then simulate the models based on a base production plan to obtain the simulation outputs.

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


Figure 2—Generating and simulating "n" simulation models under geological uncertainty.

The simulation outputs are then used in RMFinder to select a subset of simulation models. Since a unique
GR is employed in each simulation model, the selected simulation models by RMFinder show the RGR
set as well.
The RGR set is incorporated with the discretized dynamic uncertainties using a sampling method named
the Discretized Latin Hypercube with Geostatistical realizations (DLHG) (Schiozer et al., 2017) to describe
and characterize all kinds of uncertainties into the number of simulation models. Then, a filtering indicator
called normalized quadratic deviation with sign (NQDS) (Avansi and Schiozer, 2015) is applied to reduce
uncertainty and choose a small number of simulation models that are close to the past reservoir behavior.
We highlight that the uncertainty combination and model filtering steps are not the main goals of this study
and are independent of the technique used to select the RS. More details about these steps are described by
Schiozer et al. (2019). Finally, we use the subset in robust well placement optimization.

Workflow B_RM selection


In Workflow B, the scenario reduction process is applied under the combination of geological and dynamic
uncertainties. First, we combine the full ensemble of GR with the discretized dynamic uncertainties to
generate 7' number of simulation models (Figure 3) and then filter the simulation models using the same
techniques used in Workflow A.

Figure 3—Generating and simulating "l" simulation models under geological and dynamic uncertainties (l>n).

We simulate all approved (filtered) models to estimate the outputs using the base production plan. Finally,
the outputs are used in RMFinder to select the RM set for robust well placement optimization.

Validation
To check the efficiency and consistency of the defined workflows, the results obtained from Workflow A
and Workflow B are evaluated in terms of:
6 SPE-206300-MS

1. Representativeness of the selected RS in different production plans;


2. Quality of the optimized production strategy defined in each workflow;
3. Computational costs.

Representativeness of the selected RS in different production strategies


The process of representative scenario selection applied in this study is based on measuring the mismatch
between the cumulative distribution of simulation outputs obtained from the RS and the full sets; therefore,

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


defining and simulating a base production plan is required to obtain the outputs. To do so, we consider
an unfeasible production plan, called production evaluation zero (PE0), defined with the sole purpose of
obtaining a fine and wide response of reservoir behavior for RS selection. PE0 is based on a five-spot well
pattern and under water flooding to cover the entire reservoir without any constraints in the production
system. The selection of the five-spot well pattern is to ensure the unbiased coverage of the reservoir model.
We then apply other production plans based on different recovery mechanisms and well positions over
the RS selected from the PE0 simulation outputs. The goal is to investigate the effect of changing production
plans on the representativeness of the RS. Hence, we defined two production plans to achieve this objective:
(1) production evaluation one (PE1) and, (2) production strategy zero (PS0). PE1 is defined by the five-
spot well patterns with the same well position and well number as the PE0, but under the water- alternating
gas (WAG) injection. The PS0 is a feasible production strategy under water flooding defined as the best
solution obtained for a base (most likely) case.
To assess the representativeness of the RS, we compare the CDF curves of simulation outputs for the
RS and full sets. We use a non-parametric test named two-sample Kolmogorov-Smirnov (KS) (used, for
example, by Ferreira et al., 2017 and Mahjour et al., 2020c) to measure the maximum absolute distance
(Dmaxbetween the CDF curves. The Dmaxvalue is defined by Equation 1:
(1)
where Ffull,n(x)is the CDF of simulation outputs for the full set with sample size n,and FRS,m(x)is the CDF
of those for the RS set with sample size m.
If Dmaxis less than Dcriticai, the samples come from a same distribution and uncertainty space. The Dcriticalfor
the 5% significance level is defined by Equation 2:
(2)
However, the Dcriticaimay not provide an accurate indicator for the correspondence between the RS and
full samples if the sample sizes are very small (Jensen et al., 1997). Hence, Ferreira et al. (2017) proposed
a second indicator called Dlimitto provide a more precise comparison between the CDF curves of simulation
outputs for the samples. Dlimitis the acceptable distance between two points on the CDF curves for the set of
RS and full ensemble. The acceptable value of Dlimitis defined by the user given the required acceptance level
for the scenario reduction process. In this study, a quantitative analysis supported by the visual inspection of
CDF curves found the acceptable level of the Dlimitvalue to be around 0.15. Accordingly, the RS set shows
good representativeness of the full set if the Dmaxvalue is less than both Dcriticaland Dlimitvalues. Figure 4
shows the comparison process between the RS set with its full ensemble (reference set) in each workflow.
SPE-206300-MS 7

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


Figure 4—Comparing the results obtained from the RS set and its full set in each workflow.

Quality of the optimized strategy defined in each workflow


In this section, the subset of simulation models selected from each workflow is separately used in finding the
optimal well positions based on the initial production evaluation (PE0). We employ a robust optimization
technique called iterative discrete Latin hypercube (IDLHC) sampling (von Hohendorff et al., 2016)
to maximize the NPV. The IDLHG selects the best samples by gradually reducing the search space
at each iteration. During the robust optimization process, all selected simulation models are evaluated
simultaneously so that, at the end of the process, one optimized strategy is chosen considering the
probabilistic indicator expected monetary value (EMV).
We evaluate the quality of the optimized strategies in terms of profitability and reliability. To do so, the
defined strategies are applied to a reference fine-grid model working as a real reservoir with the determined
results. We first evaluate the profitability of the strategies by comparing the NPV obtained from each strategy
in the reference model. We then check the reliability of the strategies by using the CDF curves of field and
well simulation outputs. In this case, the outputs determined by the reference model must be within the
range of those obtained from the filtered simulation models in each workflow.

Computational costs
We compare Workflows A and B in terms of the computational time and the number of flow simulations.
In order to speed up the simulation process, a parallel computing machine is performed in this study.

Application
Testing and evaluating the defined workflows requires having a set of uncertain scenarios and a reference
model working as the real case. In this study, we use a synthetic fractured model named UNISIM-II-D-
BO (Correia et al., 2015). The case was created by the UNSIM research group to test different techniques
during field development and is available online at https://www.unisim.cepetro.unicamp.br/benchmarks/en/
unisim-ii/unisim-ii-d. The reference model called UNISIM-II-R was built by Correia et al. (2015) based on
the features of Brazilian pre-salt reservoirs, Ghawar field, and a real carbonate reservoir (Filed A). UNISIM-
II-D-BO was created based on pseudodata collected from UNISIM-II-R.
8 SPE-206300-MS

UNISIM-II-R
UNISIM-II-R model was built as a fine-grid geological model with known features to obtain the pseudodata
to construct UNISIM-II-D-BO, including well logs from three pseudo-wells and history data for 1.5 years
of production from one pseudo-well (the Wildcat). The dimension of the model is 5000x 5000x150m
defined by a corner point grid in which the size of each grid cell is 50mx50mx1m. Four types of facies
were characterized using a real carbonate geological model including grainstone, packstone, nonreservoir,

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


and Super-K (high permeable zone). The sequential indicator simulation (SIS) and sequential Gaussian
simulation (SGS) were employed to distribute the facies and geological features conditioned by well logs
data obtained from the three wells. According to the core data analysis, a correlation between the porosity
and permeability was determined. The correlation was used for porosity and permeability modeling. The net-
to-gross (NTG) ratio is defined as 1 for all permeable zones and 0 for the non-reservoir zone. The percentage
of the facies in the reservoir volume and the petrophysical characterization in each facies, including mean
and standard deviation (SD), are described in Table 1. Figure 5 indicates the fine- grid porosity map for
the matrix.

Table 1—Percentage of facies and matrix petrophysical characterization (in Correia et al., 2015).

Facies Facies percentage (%) Porosity (fraction) Permeability


(mean, SD) (mD) (mean, SD)

Grainstone 45 (0.20, 0.05) (200, 100)

Packstone 40 (0.15, 0.05) (50, 50)

Super-K 5 (0.25, 0.05) (7000, 1000)

non-reservoir 10 (0,-) (0,--)

Figure 5—Matrix porosity map for UNISIM-II-R.

The discrete fracture network modeling (DFN) and the Oda method (Oda, 1985) were applied to distribute
the fractures into two sets and convert the DFN into effective properties, respectively. The DFN modeling
and conversion to effective properties were applied directly to the UNISIM-II-R grid for the numerical
simulation. Table 2 describes the statistical features of the DFN.

Table 2—DFN statistical features (in Correia et al., 2015).

Fracture set Aperture (m) (mean, SD) Orientation/Dip Length(m) (mean, SD)

1 (0.0005, 0.0003) (80/160) (150, 50)

2 (0.0005, 0.0003) (80/70) (150, 50)


SPE-206300-MS 9

UNISIM-II-D-BO
UNISIM-II-D-BO includes a set of geological and dynamic uncertainties used in building an ensemble of
scenarios. By using a sampling technique, 500 unique GR were generated under geological uncertainties.
To accelerate the numerical simulation, the upscaling process was applied over the GR. The grid size of the
upscaled realizations is 100x100x8, totaling 95220 grid cells. During the upscaling process, the arithmetic

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


average was first used to upscale the porosity models weighted by NTG. The absolute permeability models
were then upscaled using the flow-based upscaling method to maintain the Super-K flow properties and
build three effective permeability models in three directions: i, j, and k. The NTG models were then upscaled
by arithmetic average to be the same as the porosity models. Eventually, to upscale the fracture models,
the properties obtained from the converted DFN were used. The geological uncertain variables are fully
described in the work of Correia et al. (2015). Furthermore, the dynamic uncertainties with continuous
variation were discretized into three probability levels. Table 3describes the dynamic uncertain attributes
with their probability levels. In this study, level 0 is considered as the most likely occur; however, defining
the probability level is a subjective task.

Table 3—Discretized dynamic uncertainties (in Correia et al., 2015).

Uncertain parameters Level (probability)

-1 0 1

Reservoir Water-relative permeability (KR) KR-1(0.3) KR0(0.4) KR1(0.3)

Pressure-volume-temperature (PVT) PVT-1(0.3) PVT0(0.4) PVT1(0.3)

Bubble point pressure (PB) PB-1(0.3) PB0(0.4) PB1(0.3)

Rock compressibility (CP) CP-1(0.3) CP0(0.4) CP1(0.3)

Operational Well index multiplier (dWi) 0.70 (0.33) 1.00 (0.34) 1.40 (0.33)

Group availability 0.91(0.33) 0.96(0.34) 1.00(0.33)

Platform availability 0.90(0.33) 0.95(0.34) 1.00(0.33)

Production well availability 0.91(0.33) 0.96(0.34) 1.00(0.33)

Injection well availability 0.92 (0.33) 0.98 (0.34) 1.00 (0.33)

Production plans
We considered three production plans to generate the simulation outputs: PE0, PE1, and PS0. The PE0
and PE1 were defined based on 28 vertical producers and 28 vertical injectors, as shown in Figure 6a. The
28 injection wells in PE1 inject water and gas in 6-month cycles. All gas produced is reinjected without
the separation of gas streams. Water injection rates were controlled to maintain the hydrocarbon volume-
weighted average at around initial reservoir pressure (450 kgf/cm2). PS0 is a production strategy optimized
nominally for the base (most likely) model, including 18 vertical wells (10 producers and 8 injectors). The
location of the wells in PS0 is shown in Figure 6b.
10 SPE-206300-MS

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


Figure 6—Maps showing the well placement for (a) PE0 and PE1, (b) PS0.

A black-oil system and dual-permeability approach are used to simulate the defined production plans
during 10957 production days. In PE0 and PE1, the wells simultaneously open from the first date of
production forecast and remain open until the maximum date for field abandonment. PS0 considers a feasible
drilling and opening schedule of wells. The required information for the simulation process and calculating
the simulation outputs are described in Tables 4-6.

Table 4—Well data.

Type Well producer Well injector Unit

Water rate - Max 5000 m3/day

Liquid rate Max 3000 - m3/day

Gas lift rate Max 200000 - m3/day

BHP Min 275 Max 480 Kgf/cm2

Well radius 0.108 0.108 m

Geometric factor 0.37 0.37 -

Angular well fraction 1 1 -

Well skin factor 0 0 -

Table 5—Platform data.

Type Platform production Platform injection Unit *103

Max water rate 120 240 bbl/day

Max liquid rate 180 - bbl/day

Max oil rate 180 - bbl/day

Max gas rate 8000 8000 m3/day

Table 6—Deterministic economic parameters

Parameters Value Unit

Price of oil 257.9 USD/m3

Price of gas 0.026 USD/m3

Cost of oil production 48.57 USD/m3


SPE-206300-MS 11

Parameters Value Unit

Cost of gas production 0.013 USD/m3

Cost of water production 4.86 USD/m3

Cost of water injection 4.86 USD/m3

Vertical well drilling 23.40 106USD

Vertical well completing 26.94 106USD

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


Vertical well connection (well-platform) 13.30 106USD

Vertical well recompletion 10.97 106USD

Well conversion 11.02 106USD

1st Inflow Control Valve (ICV) (for each well) 1.00 106USD

2nd or more ICV (for each well) 0.30 106USD/ICV

Percentage of investment in drilling and completion 8.2%

Annual discount rate 9%

Results and discussion


The results and discussion are described in the following subsections, according to the methodology and
validation steps.

RS selection
In both workflows, we used the simulation outputs obtained for PE0 to select the RS using the RMFinder.
These provided a total of 63 simulation outputs, with seven field indicators and 56 well indicators (one
WEI for each well). Since a probabilistic optimization algorithm with a random component was used in
RMFinder, we ran the method ten times, generating ten RS sets, to ensure that the representativeness of the
RS was consistent within each workflow. We also considered the relative importance of each simulation
output to select the RS, which is represented by a weight attributed to each field and well indicator.

Workflow A
We simulated 500 GR under geological uncertainties to calculate the simulation outputs and then select the
RGR. Although the number of RGR can be selected by the user, we propose a method to select the optimal
number of RGR based on the Dmax. In this technique, we first selected 5, 7, 10, 12, 15, 17, 20, 22, and
25 RGR out of 500 GR (reference set) using RMFinder. Next, the CDF curves for all simulation outputs
were obtained from the selected RGR sets and the full GR ensemble was plotted. Finally, we calculated the
Dmaxvalues for each case and drew the interval plots of the Dmaxusing the individual standard deviation. Table
7 describes the Dmaxvalues obtained from all 63 simulation outputs based on 5 RGR and 500 GR. According
to Figure 7, it is noticeable that if the number of RGR is higher than 20, there are no significant changes
in the interval of Dmaxvalues, meaning that there is no significant improvement in RS representativeness.
Furthermore, the average of Dmaxvalues for all simulation outputs obtained from 20, 22, and 25 RGR were
less than 0.15, which is below the DUmitvalue of 0.15 for this study. Hence, to select the number of RGR that
has the best cost-benefit relation, we considered two features jointly: (1) the value of Dmaxshould not change
significantly by adding RGR, (2) the average values of Dmaxobtained from the field and well indicators
should not be more than DUmit.Accordingly, and considering our computational infrastructure, we selected
20 RGR (four percent of the full set) as the optimal number of RGR for this case.
12 SPE-206300-MS

Table 7—Dmax values between the CDF curves of field and well data obtained from 5 RGR and 500 GR.

Dynamic parameter Dmax Dynamic parameter Dmax

OIP 0.313 WEI-i26 0.18

WIP 0.157 WEI-i27 0.252

NPV 0.153 WEI-i28 0.34

ORF 0.245 WEI-p01 0.244

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


NP 0.129 WEI-p02 0.212

WP 0.177 WEI-p03 0.366

WI 0.14 WEI-p04 0.252

WEI-i01 0.384 WEI-p05 0.27

WEI-i02 0.132 WEI-p06 0.292

WEI-i03 0.276 WEI-p07 0.308

WEI-i04 0.188 WEI-p08 0.546

WEI-i05 0.17 WEI-p09 0.322

WEI-i06 0.102 WEI-p10 0.416

WEI-i07 0.128 WEI-p11 0.29

WEI-i08 0.262 WEI-p12 0.306

WEI-i09 0.364 WEI-p13 0.31

WEI-i10 0.206 WEI-p14 0.26

WEI-i11 0.338 WEI-p15 0.26

WEI-i12 0.132 WEI-p16 0.306

WEI-i13 0.128 WEI-p17 0.218

WEI-i14 0.234 WEI-p18 0.248

WEI-i15 0.196 WEI-p19 0.294

WEI-i16 0.156 WEI-p20 0.346

WEI-i17 0.276 WEI-p21 0.404

WEI-i18 0.072 WEI-p22 0.166

WEI-i19 0.308 WEI-p23 0.19

WEI-i20 0.102 WEI-p24 0.266

WEI-i21 0.238 WEI-p25 0.274

WEI-i22 0.124 WEI-p26 0.338

WEI-i23 0.286 WEI-p27 0.372

WEI-i24 0.292 WEI-p28 0.348

WEI-i25 0.146

Key: WEI-i01 : WEI calculated from the injection well 1. WEI-p01 : WEI calculated from the production well 1.
SPE-206300-MS 13

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


Figure 7—Interval plot of Dmaxvalues obtained from the simulation outputs of RGR and GR sets.

The Dmaxvalues between 20 RGR and 500 full samples were calculated for 63 outputs and compared to
the Dcriticaiand Dlimitvalues. Table 8 indicates the number of indicators that have higher Dmaxthan Dcriticaland
Dlimitfor all three production plans: PE0, PE1, and PS0. The results show that the representativeness of RGR
was affected as the production plan changed. Accordingly, the representativeness of the RGR in Workflow
A could not be maintained by changing the production plan due to the high number of indicators obtained
from PE1 and PS0 having high Dmaxvalues.

Table 8—Number of unacceptable indicators for each production plan in Workflow A.

PE0 PE1 PS0

Dcritical 0 out of 63 1 out of 63 0 out of 25

Dlimit 3 out of 63 35 out of 63 18 out of 25

According to Workflow A, we used the DLHG sampling to combine the 20 RGR selected from 500
GR with other dynamic uncertainties and generated 60 simulation models. There is a 1:3 ratio between the
number of RGR and simulation models. This ratio has been estimated by Mahjour et al. (2020b) as a suitable
ratio to cover the uncertainty space of the reservoir. We then measured the misfit between the simulation
models and the observed data obtained from the past performance of the reservoir using the NQDS indicator.
The observed data was provided by the Wildcat well during 516 production days considering the oil rate
(Qo), gas rate (Qg), water rate (Qw), and bottom-hole pressure (BHP). The result shows that 24 simulation
models out of 60 simulation models were matched with the observed data based on a defined tolerance range.

Workflow B
Workflow B began by combining 500 GR to other dynamic uncertainties and building 1500 simulation
models using the DLHG sampling. The ratio between the number of GR and simulation models was 1:3. We
then applied the NQDS to select a set of well-matched simulation models with the observed data. According
to the defined tolerance range, we selected 594 filtered simulation models out of 1500. We simulated the
filtered models using the PE0 to obtain the simulation outputs. The outputs were then used in RMFinder
to select the RM. For better comparison and evaluation of both workflows, we selected 4% of the full set,
the same as Workflow A, to be considered as the RM in Workflow B. Hence, 24 RM were selected out
of 594 filtered simulation models. Table 9 describes the number of indicators that have higher Dmaxthan
Dcriticaiand Dlimitfor all production plans. Although the results show that the representativeness of the high
number of field indicators obtained from PE1 and PS0 was acceptable, the representativeness of some well
14 SPE-206300-MS

indicators was not preserved considering both Dcriticaiand Dlimitvalues. Hence, according to Workflow B, the
representativeness of the RM decreased as the production plan changed.

Table 9—Number of unacceptable indicators for each production plan in Workflow B.

PE0 with 63 indicators PE1 with 63 indicators PS0 with 25 indicators

Dcritical 0 1 2

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


Dlimit 4 19 12

Figure 8 compares the percentage of unacceptable field and well indicators in different strategies based
on the Dlimitvalues between 20 RGR and its reference set for Workflow A, and 24 RM and its reference set
for Workflow B. From the bar chart, we observe that the impact of changing production plans on the loss
of the representativeness of the RS in Workflow A is higher than Workflow B. It is therefore recommended
to consider both geological and dynamic uncertainties to select the RS due to the lower percentage of lack
of representativeness in Workflow B.

Figure 8—Percentage of unacceptable indicators in different strategies based on the Dlimitvalues


between 20 RGR and its reference set for Workflow A, and 24 RM and its reference set for Workflow B.

Robust well placement optimization


In this section, we apply robust well placement optimization using the IDLHG method on the two subsets
of 24 simulation models obtained from each workflow to find the optimal well positions based on the PE0.
In this study, the IDLHC is adjusted based on 20 iterations considering 10000 samples.
Figure 9a shows the well map of optimized production strategy from Workflow A (OPSA1), including
nine vertical producers and seven vertical injectors. The EMV of filtered simulation models obtained from
the OPSA1 in Workflow A is USD 2.52 billion.
SPE-206300-MS 15

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


Figure 9—Well map of the optimized strategies defined in: (a) Workflow A, called OPSA1, and (b) Workflow B, called OPSB1.

Figure 9b indicates the well map of optimized production strategy from Workflow B (OPSB1), including
nine vertical producers and seven vertical injectors. The EMV of filtered simulation models obtained from
the OPSB1 in Workflow B is USD 2.44 billion.
We also compared the NPV distribution for the two ensembles of filtered models of both workflows.
We applied the OPSA1 and the OPSB1 on 24 filtered models generated by Workflow A and 567 filtered
simulation models obtained by Workflow B. Figure 10a shows the CDF curves of NPV determined by
OPSA1 and OPSB1, based on the 24 filtered models of Workflow A. Figure 10b indicates the CDF curves
of NPV for the same strategies considering the 567 filtered models of Workflow B. In both figures, the
CDF of NPV plotted by OPSA1 is slightly on the right side of the CDF of NPV obtained from OPSB1.
This proves that both workflows defined the optimized strategies, which have almost the same profitability
considering NPV of the reference model and the distribution of NPV obtained from the filtered models.

Figure 10—CDF curves plotted by the NPV from OPSA1 and OPSB1 based on
(a) 24 filtered models of Workflow A, and (b) 567 filtered models of Workflow B.

Figure 11 compares the filtered model sets generated from Workflows A and B considering the
distribution of NPV determined by the OPSA1 and OPSB1. The results show that the 24 filtered models
from Workflow A and 567 filtered models from Workflow B have roughly the same behavior in different
optimized strategies defined by the workflows.
16 SPE-206300-MS

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


Figure 11—CDF curves plotted by the NPV from 24 filtered models from Workflow
A and 567 filtered models from Workflow B based on (a) OPSA1, and (b) OPSB1.

Validation and application in UNISIM-II-R


The quality of optimized strategies defined by Workflows A and B is evaluated in terms of profitability and
reliability. To do so, we apply the OPSA1 and OPSB1 in UNISIM-II-R (Figure 12).

Figure 12—Transfer of OPSA1 and OPSB1 from UNISIM-II-D-BO (coarse-


grid model) to UNISIM-II-R (reference fine-grid model) with known results.

One of the advantages of synthetically carrying out field development studies is the possibility of
determining the real (reference) NPV by applying these strategies to the reference case and running the
production strategies until the end of the reservoir life. The reference NPV for OPSA1 and OPSB1 is
USD 2.39 billion and USD 2.38 billion, respectively. Accordingly, by comparing the reference NPV of the
optimized strategies, we found that is the performance of OPSA1 was slightly more profitable than OPSB1,
but the difference is negligible (a difference of less than 0.4%).
To check the reliability of the workflows, we plotted the CDF curves of the field and well simulation
outputs obtained from the filtered models in each workflow. Figure 13 depicts the CDF curves of simulation
outputs for the 24 filtered models from Workflow A and based on the OPSA1, compared with the reference
answer (black vertical line). We noticed that the field indicators obtained from the filtered models adequately
captured the reference model. However, we also observed that in some well indicators there are inferior data
SPE-206300-MS 17

matches between the filtered models and the reference response that can be explained due to the different
convergence of the uncertainty values during the filtering process.

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


Figure 13—CDF curves of 24 filtered models obtained from Workflow A and based on
OPSA1. The black vertical lines represent the reference values obtained from UNISIM-II-R.

Figure 14 shows the CDF curves of simulation outputs for the 567 filtered models from Workflow B and
based on OPSB1, compared with the reference response (black vertical line). The same results as Workflow
A were observed here, so that the field indicators obtained from the filtered models adequately captured the
reference model while, in some well indicators, there were some mismatches between the filtered models
and the reference response.
18 SPE-206300-MS

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


Figure 14—CDF curves of 567 filtered models obtained from Workflow B and based on
OPSB1. The black vertical lines represent the reference values obtained from UNISIM-II-R.

According to the results of this section, if we consider the geological uncertainties to select the RS in
workflow A, we have nearly the same results in terms of defining profitable and reliable optimized strategy
as we consider both geological and dynamic uncertainties to select the RS in Workflow B. Therefore, the
geological uncertainties play a significant role in robust well placement optimization.

Computational Cost
In this section, we evaluate Workflows A and B in terms of the computational costs used in the scenario
generation, scenario reduction, and filtering steps. Figure 15 compares the elapsed time used in each
workflow based on PE0. Workflow A is 19 % faster than Workflow B due to the fewer simulation runs. The
difference of computational elapsed time between Workflows A and B can change due to: (1) the complexity
of geological realizations in each case study, (2) the number of wells used in the initial production plan to
select the RS, and (3) the CPU rate. Therefore, excessive computation can be avoided during the simulation
step if we first select the RS under geological uncertainties alone. The optimization step in both workflows
took around two months using 24 simulation models.
SPE-206300-MS 19

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


Figure 15—The computational elapsed time in each workflow for
the scenario generation, scenario reduction, and filtering steps

Conclusions
This work investigated two workflows for robust well placement optimization using the RS selection
in greenfield: (A) selecting the RGR under geological uncertainties and (B) selecting the RM under a
combination of geological and dynamic uncertainties.
The workflows were tested in the UNISIM-II-D-BO benchmark case, a synthetic fractured reservoir
model under uncertainty, and validated in the reference fine-grid model, UNISIM-II-R, taken as the real
reservoir with known results. We highlight the following specific findings:
1. The application of the RS selection method led to a 96% decrease in the number of scenarios in both
workflows. Changing the production strategy impacts the representativeness of the RS sets obtained
from both workflows (in Workflow A, 37% of all simulation outputs lost their representativeness
when strategy changed, while, in Workflow B, this loss was 23%). It is therefore recommended to
consider both geological and dynamic uncertainties to select the RS if the production plan changes
during field development.
2. The production strategies obtained from both workflows have similar solutions (virtually the same
profitability). Furthermore, the workflows are reliable considering the comparison of the NPV ranges
obtained from the filtered models and the reference NPV from UNISIM-II-R. We therefore found
that the role of geological uncertainties is more significant than dynamic uncertainties in the selection
of RS for well placement optimization. However, in some well indicators, we observed mismatches
between the filtered models and the reference response, which can be explained by the different
convergence of the uncertainty values during the filtering process.
3. Due to the fewer filtered models obtained from Workflow A, this workflow can be superior in reducing
the computational cost for field development optimization processes. Hence, we can use the RMFinder
methodology to slightly simplify RS selection (saving some additional time) by previously selecting
the geological realizations (second method).
Because the applicability of the robust well placement optimization workflows has been verified in
this study, real field applications are important in future research. Furthermore, the RMFinder method can
be coupled with other RS selection techniques, such as cluster-based and machine learning approaches.
Therefore, it is important to use an optimal RGR selection by considering the features of each technique
and model parameters in each method.

Acknowledgments
We gratefully acknowledge the support of EPIC - Energy Production Innovation Center, hosted by the
University of Campinas (UNICAMP) and sponsored by Equinor Brazil and by FAPESP - Sao Paulo
20 SPE-206300-MS

Research Foundation (grant number 2017/15736-3). We acknowledge the support of ANP (Brazil's National
Agency of Oil, Natural Gas, and Biofuels) through the R&D levy regulation. Acknowledgments are
extended to the Center for Petroleum Studies (CEPETRO) and School of Mechanical Engineering (FEM) at
UNICAMP. We also acknowledge the funding from Energi Simulation and thank the Computer Modelling
Group Ltd. (CMG) and Mathworks for software licenses and technical support.

References

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


Aliyev, E., and Durlofsky, L. J.(2015). Multilevel Field Development Optimization under Uncertainty Using a Sequence
of Upscaled Models.Mathematical Geosciences, 49(3), 1–33. 10.1007/s11004-016-9643-0.
Almeida, F. L. R., Davolio, A., and Schiozer, D. J.(2014). A New Approach to Perform a Probabilistic and Multi-
Objective History Matching.SPE Annual Technical Conference and Exhibition held in, Amsterdam, The Netherlands.
10.2118/170623 -MS
Avansi, G. D.and Schiozer, D. J.(2015). A New Approach to History Matching using Reservoir Characterization
and Reservoir Simulation Integrated Studies.Offshore Technology Conference held inHouston, Texas, USA, 04-07
May.10.4043/26038-MS
Badru, 0., and Kabir, C. S.(2003). Well Placement Optimization in Field Development. SPE Annual Technical Conference
and Exhibition held in Denver, Colorado. 10.2118/84191-MS.
Botechia, V. E., Gasper, A. T., and Schiozer, D, J, (2013).Use of Well Indicators in the Production Strategy Optimization
Production Process. EAGE Annual Conference & Exhibition incorporating SPE Europec, 10-13 June, London,
UK.10.2118/164874-MS.
Chang, Y., Bouzarkouna, Z., and Devegowda, D.(2015). Multi-objective Optimization for Rapid and Robust Optimal
Oilfield Development under Geological Uncertainty. Computational Geosciences, 19(2015), 933–950. 10.1007/
s10596-015-9507-6.
Correia, M. G., Hohendorff, J., Gaspar, A. T. F. S., and Schiozer, D.(2015). UNISIM-II-D: Benchmark Case Proposal
Based on a Carbonate Reservoir. SPE Latin American and Caribbean Petroleum Engineering Conference held in Quito,
Ecuador, 18-20 November. 10.2118/177140-MS.
Douarche, F., Da Veiga, S., Feraille, M., Enchéry, G., Touzani, S., and BarsalouR.(2014), Sensitivity Analysis and
Optimization of Surfactant-polymer Flooding under Uncertainties.Oil & Gas Science and Technology Revue dIFP
Energies nouvelles, 69(4), 603–617. 10.2516/ogst/2013166.
Durlofsky, L. J.(1998). Coarse Scale Models of Two-Phase Flow in Heterogeneous Reservoirs: Volume Averaged
Equations and Their Relationship to Existing Upscaling Techniques. Computers & Geosciences, 2(2), 73–92. 10.1023/
A:1011593901771.
Feraille, M., and Marrel, A.(2012). Prediction under Uncertainty on a Mature Field. Oil & Gas Science and Technology
- Revue dIFP Energies nouvelles,67(2) 193–206. 10.2516/ogst/2011172.
Ferreira, C. J., Davolio, A., and Schiozer, D. J.(2017). Evaluation of the Discrete Latin Hypercube with Geostatistical
Realizations Sampling for History Matching Under Uncertainties for the Nome Benchmark Case. Offshore Technology
Conference Brasil, 24-26 October,Rio de Janeiro, Brazil.10.4043/28073-MS.
Haghighat Sefat, M., Elsheikh, A. H., Muradov, K. M., and Davies, D. R.(2016). Reservoir Uncertainty Tolerant, Proactive
Control of Intelligent Wells. Computational Geosciences, 20(3), 655–676.10.1007/s10596-015-9513-8.
Imrie, C. E., and Macrae, E.J.(2016). Application of Experimental Design to Estimate Hydrocarbons Initially in Place.
Petroleum Geoscience,22(1), 11–19. 10.1144/petgeo2014-071.
Janiga, D., Czarnota, R., Stopa, J., and Wojnarowski, P.(2019). Self-adapt Reservoir Clusterization Method to
Enhance Robustness of Well Placement Optimization. Journal of Petroleum Science and Engineering,173(2018),
37–52.10.1016/j.petro1.2018.10.005.
Jensen, J. L., Lake, L. W., Corbett, P. W. M., & Goggin,D. J.(1997).Statistics for Petroleum Engineers and Geoscientists.
New Jersey: Prentice Hall.
Kong, B., Kim, S., Jung, H., Choe, J., and Lee, K.(2019). Efficient Assessment of Reservoir Uncertainty Using Distance-
Based Clustering: A Review. Energies, 12(10), 1859.10.3390/en12101859.
Lee, K., Jeong, H., Jung, S.and Choe, J.(2013) Characterization of Channelized Reservoir Using Ensemble Kalman Filter
with Clustered Covariance. Energy Exploration & Exploitation, 31(1), 17–29. 10.1260/0144-5987.31.1.17
Mahjour, S. K., Correia, M. G., Santos, A. A. S., and Schiozer, D. J.(2019). Developing a Workflow to Represent Fractured
Carbonate Reservoirs for Simulation Models Under Uncertainties Based on Flow Unit Concept. Oil & Gas Science
and Technology - Rev. IFP Energies Nouvelles, 74, 15.10.2516/ogst/2018096.
Mahjour, S. K., Correia, M. G., de Souza dos Santos, A. A., and Schiozer, D. J.(2020a). Using an Integrated
Multidimensional Scaling and Clustering Method to Reduce the Number of Scenarios Based on Flow-Unit Models
under Geological Uncertainties. ASME. Journal of Energy Resources Technology,142(6), 063005. 10.1115/1.4045736.
SPE-206300-MS 21

Mahjour, S. K., Santos, A. A. S., Correia, M. G., and Schiozer, D. J.(2020b). Developing a Workflow to Select
Representative Reservoir Models Combining Distance Based Clustering and Data Assimilation for Decision Making
Process.Journal of Petroleum Scienceand Engineering.190(2020),107078.10.1016/j.petro1.2020.107078.
Mahjour, S. K., Santos, A. A. S., Correia, M. G., and Schiozer,D. J.(2020c). Two-stage Scenario Reduction Process for
an Efficient Robust Optimization. ECMOR XVII - 17th European Conference on the Mathematics of Oil Recovery,
Online Event, 14-17 September.10.3997/2214-4609.202035105.
Mahjour, S. K., Santos, A. A. S., Correia, M. G., and Schiozer, D. J.(2021). Scenario Reduction Methodologies under
Uncertainties for Reservoir Development Purposes: Distance-based Clustering and Metaheuristic Algorithm. Journal

Downloaded from http://onepetro.org/SPEATCE/proceedings-pdf/21ATCE/1-21ATCE/D011S020R003/2490359/spe-206300-ms.pdf by Seyed Kourosh Mahjour on 19 September 2021


of Petroleum Exploration and Production Technology.10.1007/s13202-021-01210-5.
Meira, L. A., Coelho, G. P., Silva, C. G., Abreu, J. A. L., Santos, A. A. S., and Schiozer, D.J.(2020). Improving
Representativeness in a Scenario Reduction Process to Aid Decision Making in Petroleum Fields. Journal of Petroleum
Science and Engineering,184, 106398.10.1016/j.petrol.2019.106398.
OdaM.(1985). Permeability Tensor for Discontinuous Rock Mass. Geotechnique, 35(4), 483–495. 10.1680/
geot.1985.35.4.483.
Panjalizadeh, H., Alizadeh, N., and Mashhadi, H.(2014). A Workflow for Risk Analysis and Optimization of Steam
Flooding Scenario using Static and Dynamic Proxy Models. Journal of Petroleum Science and Engineering,121(2104),
78–86. 10.1016/j.petro1.2014.06.010.
Santos, S. M. G., Gaspar, A. T. F. S., and Schiozer, D. J.(2018). Managing Reservoir Uncertainty in Petroleum Field
Development: Defining a Flexible Production Strategy from a Set of Rigid Candidate Strategies. Journal of Petroleum
Science and Engineering, 171(2018), 516–528. 10.1016/j.petro1.2018.07.048.
Santos, S. M. G., Santos, A. A. S., and Schiozer, D. J.(2020). Selecting Representative Models for
Ensemble-Based Production Optimization in Carbonate Reservoirs with Intelligent Wells and WAG Injection.
ECMOR XVII - 17th European Conference on the Mathematics of Oil Recovery, Online Event, 14-17
September.10.3997/2214-4609.202035041.
Sarma, P., and Chen, W. H., Xie, J.(2013). Selecting Representative Models from a Large Set of Models. SPE Reservoir
Simulation Symposium held in The Woodlands, Texas, USA, 18-20 February. 10.2118/163671-MS
Scheidt, C., and Caers, J.(2009). Uncertainty Quantification in Reservoir Performance Using Distances and Kernel
Methods - Application to a West Africa Deepwater Turbidite Reservoir. Society of Petroleum Engineers,Volume 4,
Issue 04. 10.2118/118740-PA.
Scheidt, C., Zabalza-Mezghani, I., Feraille, M., andCollombier, D.(2007). Toward a Reliable Quantification of Uncertainty
on Production Forecasts: Adaptive Experimental Designs. Oil & Gas Science and Technology - Rev. IFP Energies
nouvelles, 62(2), 207–224. 10.2516/ogst.2007018.
Schiozer, D. J., Ligero, E., Suslick, S., Costa, A., and Santos, J.(2004). Use of Representative Models in the Integration of
Risk Analysis and Production Strategy Definition. Journal of Petroleum Science and Engineering, 44(1-2), 131–141.
10.1016/j.petro1.2004.02.010.
Schiozer, D. J., Santos, A. A. S., Santos, S. M. G., and Filho, J. C. H.(2019). Model-Based Decision Analysis Applied to
Petroleum Field Development and Management. Oil & Gas Science and Technology - Rev. IFP Energies Nouvelles,
74, 46. 10.2516/ogst/2019019.
Shirangi, M. G., and Durlofsky, L. J.(2016). A General Method to Select Representative Models for Decision Making and
Optimization under Uncertainty. Computers&Geosciences,96,109–123.10.1016/j.cageo.2016.08.0023
TrehanS., CarlbergK. T., and DurlofskyL. J.(2017). Error Modeling for Surrogates of Dynamical Systems using Machine
Learning. International Journal for Numerical Methods in Engineering, 112, 1801–1827. 10.1002/nme.5583.
van Essen, G., Zandvliet, M., van den Hof, P., Bosgra, 0., and JansenJ. D.(2009). Robust Waterflooding Optimization of
Multiple Geological Scenarios, SPE Journal, 14(1), 24–27. 10.2118/102913-PA.
von Hohendorff FilhoJ. C., MaschioC., and SchiozerD. J.(2016) Production Strategy Optimization Based on Iterative
Discrete Latin Hypercube. Journal of the Brazilian Society of Mechanical Sciences and Engineering,38(8),
2473–2480. 10.1007/s40430-016-0511-0.
Wilson, K. C.and Durlofsky, L. J.(2013). Optimization of Shale Gas Field Development using Direct Search
Techniques and Reduced-physics Models. Journal of Petroleum Science and Engineering, 108, 304–975. 10.1016/
j.petro1.2013.04.019.
Yang, C., CardC., Nghiem, L., and Fedutenko, E.(2011). Robust Optimization of SAGD Operations under Geological
Uncertainties. PE Reservoir Simulation Symposium held in The Woodlands, Texas, 21-23 February.10.2118/141676-
MS.
Zabalza-Mezghani, I., ManceauE., FerailleM., JourdanA.(2004). Uncertainty Management: From Geological Scenarios
to Production Scheme Optimization, Journal of Petroleum Science and Engineering, 44(1-2), 11–25. 10.1016/
j.petro1.2004.02.002.

You might also like