Sunteți pe pagina 1din 16

SPE-187595-MS

Uncertainty Assessment on Short-Term Production Forecasts Using


Integrated Asset Modeling and Experimental Design

Xiaopeng Li, Colorado School of Mines; Suryansh Purwar, Sergio Sousa, Cesar Bravo, and David Vasconcelos,
Halliburton

Copyright 2017, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE Kuwait Oil & Gas Show and Conference held in Kuwait City, Kuwait, 15-18 October 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
In field development, many decision-make processes are associated with uncertainties, which sometimes
can expose the project to significant engineering and economic risks. Accurate uncertain assessment on
field-scale production forecast requires multidisciplinary approach, such as Integrated Asset Modeling
(IAM), to accommodate subsurface response, well performance, and surface facilities simultaneously.
However, the added complexity can make IAM prohibitively and computationally expensive and time-
consuming, particularly for uncertainty analysis and optimization studies, which normally demand
thousands of simulation runs.
This paper presents a workflow that effectively integrates all the production components to an IAM then
generates its proxy model for faster deployment. First, to properly configure the model, some practical
sensitivity analyses were performed on how reservoir model selection and forecast period affect the
uncertainty assessments. Second, in order to generate the proxy model, two-level screening design was first
used to reduce the number of variables. Third, simulations through experimental design were conducted.
The resulting response surface was used to fit a multiple regression model to reduce simulation runtime and
maintain prediction accuracy at the same time.
The results suggested that in order to build accurate IAM and avoid misleading uncertainty results,
it is recommended to take careful considerations to select proper tools to model different components
of integrated production system. It was also found that the relative importance or influence of some
parameters will not remain constant but evolve through the life of field. Thus, it is necessary to keep
the production forecast period of uncertainty analysis consistent with the development period associated
with field strategies and economic objectives. To reduce the computational cost and accelerate decision-
making process, a proxy model was generated to approximate this physical IAM model with high correlation
coefficient of 99.65%. In case study, the high impact factors that can explain most variations in response
were identified from totally 29 uncertain variables through screening design, including water-oil contact,
thickness, oil density, initial reservoir pressure, and porosity. The validation results showed that proxy
model predictions matched closely with simulation results from actual IAM with correlation coefficient of
95.70%. Based on a synthetic production forecast uncertainty study using Monte Carlos simulation, the
2 SPE-187595-MS

proxy model can reduce the runtime for five-thousand simulations to seconds from hundreds of hours if
using fully physical IAMs.
Although using proxy model generated through experimental design for reservoir simulation studies is
common approach, it has not been applied to integrated asset modeling which couples reservoir simulator
with well and surface facilities. This workflow allows efficient and accurate uncertainty estimation for
production forecast and field-scale optimizations.

Introduction
Uncertainty exists in reserve estimation and production forecast because of the unavailability of accurate
information, limitations of measurements, inherent assumptions, etc. Traditionally, there are six techniques
for uncertainty assessment, including analog, volumetrics, decline curve, material balance, rate transient,
and reservoir simulations (Purvis and Kuzma, 2016). Techniques of analog and volumetrics are most likely
to be applied in the exploration stage of field development where there is not sufficient information for
proper applications of other techniques. Decline curve has dominated the reserve estimation techniques
for decades, but their accuracy depends on the quality of data and the interpretations often suffer from
heavy subjectivity even with the aids of sophisticated statistical methods or machine learning techniques
(Gong et al. 2011; Purvis and Kuzma, 2016). Material balance method uses a tank model to describe the
entire reservoir and neglects the heterogeneity of reservoir properties and non-uniform fluid distribution.
Similarly, rate transient analysis (RTA) surffers from simplistic assumptions, subjectivity, non-unique
solutions, etc. Fully physical reservoir simulator produces more reliable and accurate results, however, can
be computationally expensive and time consuming. Furthermore, production history and seismic data can
be utilized as constraints to achieve better production forecast uncertainty estimation using reservoir models
through history matching (Litvak et al., 2005).
Uncertainty analysis is usually carried out in a probabilistic way where information about the probabilities
of the uncertain variables is used to produce a range of outcomes with relative likelihoods instead of a
single answer. Risk and uncertainty analysis are required to identify opportunities to optimize production
operations and make robust planning through hesteogeneous information and complex interactions between
different production system components (Saputelli et al., 2007). The assessment results can facilitate the
decision making process for the development strategy maker and field operation management teams to
identify the most sensible parameters and critical issues and maximize hydrocarbon recovery under existing
constraints and conditions.

Integrated Asset Modeling (IAM)


Generally, the entire production system can be divided into three segments: reservoir, well, and surface
facilities. In terms of field-scale study, the decision-making processes on different segments are often
broken-down and disconnected (Rodriguez, et al., 2007). Optimization and design for each unit are
traditionally carried out separately: reservoir engineer only models the reservoir response; production
engineer does the well model; process engineer simulates the surface pipeline network. Optimum solution
to a sub-system does not necessarily fit the constraints of other segments. In order to obtain a global solution
or optimum operating conditions, the engineers need to go back and forth and iterate the processes until a
solution is achieved that satisfies all the requirements. Therefore, an integrated approach that couples all the
production components from reservoir all the way to surface separator is preferred by the operator, such as
Integrated Reservoir and Surface-Pipeline-Network Models, Integrated Production/Asset Modeling (IPM/
IAM) (Litvak and Wang, 2000; Ghorayeb et al., 2003; Ojeke et al., 2011, Roadifer et al., 2012). Silva and
Pinto (2015) showed that production forecast can be significantly improved and flexibility and robustness
of project planning can be greatly increased if uncertainties in reservoir model and operation schedule can
be combined and assessed. The IAM can be a good candidate for uncertainty analysis of the entire field,
SPE-187595-MS 3

which integrates all the subsurface and surface engineering data and sophisticated interactions between the
production components into a holistic view of the field.
The benefits of IAM include time and cost savings by accelerating the decision making process. It not
only ensures proper project planning on subsurface development and surface infrastructure construction in
an economic sense, but also allows any variation from uncertainties associated the execution and operation
processes over the entire life cycle of the field.

Proxy Model through Experimental Design


IAM produces efficient decision-making process, offers insight of component interactions, identifies
significant opportunities, and avoids time-consuming iterations between separate solutions of subsurface
and surface network. However, by coupling multiple components, the computational cost and simulation run
time for IAMs can increase significantly compared to a reservoir-simulation alone approach. To accelerate
the process and ensure some degree of accuracy at the same time, the Experimental Design technique has
been applied to quantification of uncertainties and risk analysis in the oil and gas production forecasts and
economic evaluations (Vanegas Prada et al. 2006; Vanegas et al. 2005).
Experimental design can be used to design simulation study and aims to extract the maximum information
at minimum cost (less simulation runs) (Oehlert, 2000). To study the effects of various uncertain parameters
and their interactions, several parameters are varied simultaneously according to a predefined pattern. The
purpose of the experimental design in uncertainty studies is to generate a response surface that relates
multiple explanatory variables to one or more response variables (Purwar et al., 2010). This methodology
has been extensive applied on reservoir simulation studies for probabilistic production forecast. Thus, a
complex reservoir simulator can be approximated by a proxy model, which can be a simple regression model
generated from the simulation results (response surface) of experimental design (Kalla and White, 2007).
However, it has never been applied to the more complicated Integrated Asset Models.

Summary
As discussed above, Integrated Asset Model offers many benefits for field-scale uncertainty analysis and
optimizations, but suffers from high computational cost. Using a proxy model to approximate complex
physical models, like reservoir simulator, can be considered as a viable alternative. However, it has not
been tested that whether this approach is reliable or even feasible for complex integrated field study. If this
approach can be implemented, the decision-making processes will be greatly improved for fast deployment
and better quality support for the field. In this paper, a workflow is presented to effectively integrate
all the production components to an Integrated Asset Model, and then generate its proxy model through
Experimental Design. To properly configure the model, some practical sensitivity analyses were performed
on the effects of reservoir model selection and forecast period. The resulting proxy model was used to
perform production forecast uncertainty case studies involving Monte Carlos simulation.

Production Forecast Uncertainty Analysis Workflow


In the life cycle of a field, reservoir engineers are faced with numerious decision-making processes, such
as development strategies, operational plans. Traditionally, reservoir simulation studies provide supports to
forecast production under different operating conditions and identify the best development plan to maximize
economic recovery. However, the results of reservoir simulation are often limited by failing to consider
other factors, such as multiphase flow in wellbores, surface facility capacities (Hepguler et al., 1997).
Thus, in the view of integrated asset management, surface and subsurface teams need to work together
multidisciplinarily, share ideas, and make real time decision on a regular base. This requires a comprehensive
asset management workflow that brings in all the subsurface and surface data and untilizes multidisciplinary
processing tools to integrate different production segments in an IAM model for the production forecast
process. Therefore, the tools and procedures to build an integrated asset model are introduced first, followed
4 SPE-187595-MS

by the procedures of using screening design and experimental design to properly generate a proxy model.
Finally, the workflow is summarized at the end.

IAM Implementation
A unique toolkit (an IPM suite and a reservoir simulator) is utilized to provide necessary resources and
technology to implement this comprehensive model, shown in Figure 1 (Petroleum Experts Ltd., 2016;
Landmark, 2016). Different tools cover different component of the production system and are solved
simultaneously at each time step to converge to a global solution. The IPM suite contains all the necessary
software to model an oil and gas production system as well as connection interface that allows other
programs to access the tools for simulation control. However, the IPM suite is more recognized and applied
by the industry for well modeling and surface network simulation. The reservoir simulator delivers accurate
and quick solutions for subsurface reservoir systems and also provides the ability to couple subsurface
environment with surface facilities. Thus, this workflow utilizes the reservoir simulator to provide the
reservoir responses. The these responses will feed the dynamic IPR (Inflow Performance Relationship)
curves, which later on will be coupled with VLP (Vertical Lifting Performance) curves at the node of well
bottom-hole.

Figure 1—The big picture of field production system. The decision making process should be based on
the production forecast under the constraints from reservoir response, well model, surface facilities.
Full scale field studies using IAM rely on the integration of multiple tools and models to cover the
different segments of the production system instead of a single subsurface reservoir simulator.

The process of implementing IAM using IPM suite and reservoir simulator involves five basic steps:
1. Prepare well models for all the wells;
2. Create the surface network and facility models;
3. Link surface network models with all the well models and generate VLP curves;
4. Create a reservoir model with necessary well connections;
5. Couple reservoir simulation model with well models at bottom hole for all the wells;
6. Setup simulations and obtain the results.
Steps 1 to 3 are completed using the IPM suite for well model and surface facilities. Step 4 utilizes
the reservoir simulator. All the components are integrated in Step 5 and the simulation is triggered in an
integration interface. Detailed procedures on how to implement the steps are given below.
SPE-187595-MS 5

Well Model. The key output of well model for the IAM simulations is the well performance curve or
VLP curve. The input data to generate VLP curves include but not limited to: PVT, reservoir parameters,
deviation survey, downhole equipment data, geothermal gradient, and average heat capacity.
PVT section is critical to the success of IAM, because it involves the matching process to fit existing
correlations to the actual test data. First, what being used in the actual VLP curve calculations are the
calibrated correlations instead of the actual PVT data. It is important to rematch/recalibrate the PVT data
every time any parameters related to PVT change, for example, Oil Gravity, Gas Gravity, Temperature, etc.
Second, the PVT data input in the PVT section must be the same with the PVT table in reservoir simulator
to keep the consistency. Additionally, if initial reservoir temperature is considered as an uncertain variable,
it is suggested to change the geothermal gradient table accordingly because the bottom-hole temperrature
(usually the last row) of the table normally refers to the initial reservoir temperature.
The VLP curves report the flowing bottom-hole pressure for various rates at a given set of flowing
conditions (WHP – wellhead pressure, GOR and Water Cut). To generate VLP curves, a series of variables
with varying values are required. An example is shown in Table 1. One of resulting VLP curves for top
Node pressure (WHP) of 7.9 bara, GOR of 177.2 m3/m3, and 0% Water Cut is shown in Figure 2.

Table 1—Cases setup for VLP calculations

Top Node Pressure (WHP), bara Gas Oil Ratio, m3/m3 Water Cut, %

7.90801 177.2 0
67.6626 253.391 11
127.417 362.341 22
187.172 518.136 33
246.926 740.918 44
306.681 1059.49 55
366.435 1515.04 66
426.19 2166.46 77
485.945 3097.97 88
545.699 4430 99

Figure 2—VLP Curve generated from well model indicating the relationship between liquid rate and tubing
bottom-hole pressure for Top Node pressure (WHP) of 7.9 bara, GOR of 177.2 m3/m3, and 0% Water Cut.
6 SPE-187595-MS

Surface Network Model. The surface network model was used to model the surface facilities including
the pipeline network, joints, and separator. Figure 3 shows an example of production network model.

Figure 3—A Production Network, which includes 12 wells, 5 manifolds, 1 gathering station, and 1 separator.

Totally 12 wells are connected to 5 different manifolds. The gathering station 1_6 combines all the flows
from the manifolds and transports the produced fluids to Separator STAT 1-7. The layouts of the pipeline
network, length, elevation, and diameter of each pipe should be properly entered into the model. A critical
step is to link the well models to the well elements in surface network model. Morover, the number of wells
should be consistent with that in reservoir model. All constraints should be set according to the actual field
conditions.
Reservoir Model. The reservoir model can be properly set up in the reservoir simulator. The input data
comprise all the reservoir and fluids properties, such as Initialization Data, Rock Property Table, Fluids
Property Table, Grid-Related Data, and Aquifer Data, as well as all the well information and production/
injection schedule. In this study, there are totally 32×27×6 = 5184 simulation blocks.
It should be noted that the wells listed in the reservoir model should be consistent with the wells in
surface network model, because these two models are coupled through well connections. In another word,
the reservoir response is coupled with well and surface facility model at the bottom-hole.
Connecting Subsurface and Surface Models. A connection interface will now be used to link and transfer
the data between the surface network and the reservoir model, as shown in Figure 4. This system is the
whole structure of IAM model. Once the model is setup, simulation can be set to forecast production. The
simulation starts from the beginning of 2006 to the beginning of 2008 with fixed time step of 3 weeks in
this study.
SPE-187595-MS 7

Figure 4—Structure map of the IAM model. The pink triangle icon at the top is the separator
and the pink triangles with green border are the wells associated with reservoir model
and the brown triangles with green border are the wells in surface network model.

Proxy Model
Although IAM can increase the production forecast accuracy by removing some simplifying assumptions
of traditional reservoir simulations and enabling cross-discipline collaborations (Torrens et al. 2014), the
added complexity increases the computational time and makes an IAM model computationally expensive
than a multimillion-cell reservoir simulator. A proxy model is an alternative to fully physical model to save
much time and reduce the computational cost.
Once IAM model is ready, a built VBA program can be used to conduct simulation studies for proxy
model generation. The controller manages the communications between subsurface simulators and surface
networkwork simulations via an open message-passing interface and also synchronizes the changing
parameters through time.
First, the previous steps of building an IAM model accomplish the task of setting up a base case for
the uncertainty analysis. Second, a comprehensive search for uncertain variables is carried out. Whether a
variable or a parameter should be considered as uncertain variables depends on the actual cases. However,
which variables from the list of all the uncertain variables to use for the proxy model shall not be determined
only by experience but also by sensitivity studies. The reason is that the engineers tend to select the variables
with relatively higher uncertainties or with relative more importance. This judgement or selection can be
very subjective and jeopardize the efficiency and accuracy of proxy model. Sometimes, some variables with
higher uncertainty do not necessarily have the much influences on the results. Thus, it is recommended to
do a screening design test before actually generating a proxy model. Third, after the high impact factors are
selected, experimental design can be used to set up simulation runs to generate the response surface. Then
the simulation results will be used to fit the proxy model – a regression model.
Screening Design. To choose the variables/factors for experimental design, usually a sensitivity study
for screening purpose is carried out (Friedmann et al., 2003). The most common procedure is to vary one
variable value at a time and keep others fixed for each simulation run. The simulation results of screening
are compared against the values of base case. The difference between the response due to variable value
change and the base value measures the influence of this variable. Then the most sensible variables whose
changes produce more effects on the value of production forecast are chosen for the experimental design.
Through a comprehensive search, totally 29 variables are considered with uncertainty. It is impractical
to use all the uncertain variables for the proxy model, since some variables have little or no effects on the
results. At the same time, a method that can identify the key parameters governing uncertainty in production
8 SPE-187595-MS

and economic forecasts in the early phases of the study will significantly ameliorate the data acquisition
program (Friedmann et al., 2003). Thus, a sensitivity study for screening purpose is conducted.
All the base values of variables were varied by ±10% one at a time. The new forecast cumulative
production then is compared against to the base case. The variables that cause greater variations in the
response are considered as high impact factors.
The value table is shown in Table 2. The results were organized to plot a Tornado chart, shown in Figure
5. To conduct experimental design, one critical criterion is that all the factors should be independent from
each other, thus, the top 5 independent variables selected for proxy model are, WOC (Water-Oil Contact),
Thickness, DENOIL (Oil Density), PINIT (Initial Reservoir Pressure), and Porosity.

Table 2—Value table for screening design.

Variables Description Low Base High

SKIN Skin factor -2 0 2


DENOIL Density of oil 788.5458 876.162 963.7782
PINIT Initial Reservoir Pressure 495 550 605
WOC Water-oil contact 1620 1800 1980
KROW Max oil relative permeability 0.72 0.8 0.88
PERMEABILITY Reservoir permeability 0.9 1 1.1
NG Gas exponent 1.8 2 2.2
DRAIN AREA Drainage area 0.9 1 1.1
THICKNESS Formation thickness 0.9 1 1.1
POROSITY Porosity 0.9 1 1.1
GOC Gas-oil contact 900 1000 1100
TINIT Initial reservoir temperature 126 140 154
KRW Max water relative 0.405 0.45 0.495
permeability
SORG Residual gas saturation 0.18 0.2 0.22
SOW Residual oil saturation 0.27 0.3 0.33
SWC Ireducible water saturation 0.18 0.2 0.22
SGC Ireducible gas saturation 0.0225 0.025 0.0275
NOW Oil exponent (to oil) 1.8 2 2.2
NOG Oil exponent (to gas) 1.8 2 2.2
NW Water exponent 1.8 2 2.2
CR Rock compressibility 0.000009 1.00E-05 0.000011
KRG Max gas relative 0.495 0.55 0.605
permeability
DENGAS Density of gas 0.77238 0.8582 0.94402
PSAT Saturation pressure 348.6416 387.3796 426.1176
BO Oil formation volume factor 0.9 1 1.1
VO Oil viscosity 0.9 1 1.1
RS Solubility 0.9 1 1.1
BG Gas formation volume factor 0.9 1 1.1
VG Gas viscosity 0.9 1 1.1
SPE-187595-MS 9

Figure 5—Tornado Chart for Screening Desgin Results.

Experimental Design. Experimental design aims to extract the maximum information at minimum cost.
For each simulation run, several parameters are varied simultaneously according to a predefined statistically
optimal pattern. Three settings were assigned for each of the 5 selected factors, optimistic, expected, and
pessimistic. Those settings correspond to different levels of -1, 0, 1, as shown in Figure 6.

Figure 6—Factor levels - an example of PINIT (initial reservoir pressure).

The Box-Behnken design was chosen to design the experiments. Note that different experimental design
methods are suitable for different scenarios. Finally part of the design table is shown in Table 3.

Table 3—Experimental design table using Box-Behnken.

DENOIL PINIT WOC THICKNESS POROSITY

0 0 0 0 0
0 0 -1 -1 0
⋮ ⋮ ⋮ ⋮ ⋮
1 0 0 0 1
10 SPE-187595-MS

Each time, some variables values will change during the simulations. It is tedious and time-consuming
to run all the simulation cases manually. A VBA program is built to automatically conduct simulation
design according the values tables and experimental design tables. These speadsheet-based predefined
experimental design routines and advanced mathematical non-linear solver enable the automatical
prediction of the field productions under all the constraints, operation capacities to assess the subsurface
and surface uncertainties from all aspects.
The procedures are summarized in Figure 7. Every time VBA will read experimental table and modify
the values in IAM, conduct simulation, and extract production profile to Excel and calculate cumulative
production and record in an Excel sheet.

Figure 7—Generate response surface procedure.

Multiple Regression. All the simulation results or response surface can be fitted to a simple regression
model through multiple regression method. The full form of quadratic formula was used which has 1
constant, 5 coefficients for linear terms, 5 coefficients for square terms, 10 coefficients for interaction terms.
The coefficient matrix is given in Table 4 with a fitting correlation coefficient of 99.65%.

Table 4—Full Quadratic Coefficient Matrix - R2 = 99.65%

Constant Linear Square Interaction

-6297792 DENOIL 1093.124 SQ DENOIL -0.19854 DENOIL*PINIT -0.04524


PINIT -97.48 SQ PINIT 0.069414 DENOIL*WOC -0.44366
WOC 6380.329 SQ WOC -1.61129 DENOIL*THICKNESS -92.8868
SQ
THICKNESS 79105.2 -24696.1 DENOIL*POROSITY -29.1182
THICKNESS
SQ
POROSITY 32041.83 -12473.7 PINIT*WOC 0.087444
POROSITY
PINIT*THICKNESS -0.26795
PINIT*POROSITY 3.397591
WOC*THICKNESS 50.88075
WOC*POROSITY 19.4416
THICKNESS*POROSITY -6445.45
SPE-187595-MS 11

Moreover, deeper screening on the terms of regression model can be done to further improve the
efficiency. In this approach, full quadratic form was used. But some of terms in the formula didn't show
high impact to the prediction results and thereby can be removed.

Summary
In summary, the workflow can be described in Figure 8, including IAM mode preparation, proxy model
generation, and model application. The model preparation time can vary case by case depending on the
complexity of the problem. But it only takes several hours to generate a proxy model after an IAM model
is prepared and the utilization of proxy model can save significant time as demonstrated in the following
case study.

Figure 8—Proposed production forecast uncertainty analysis workflow.

Results and Discussions


After the framework/workflow was developed, some sensitivity studies were performed to properly
configure the model and design the uncertainty studies. Some important results were generated and
discussed. Then, the proxy model built was put into validation test to examine the accuracy and efficiency
of this workflow.

Selection of Reservoir Model


As discussed earlier, there are six different methods can be used to generate reservoir responses and produce
IPR curves. The question can be asked whether the selection of reservoir tool would affect the uncertainty
results. The first two techniques, analog and volumetrics are usually applied in the very early stage of field
development. The decline curve analysis and rate transient analysis are usually very subjective, lack physical
explanations and rely on high quality data. Moreover, most of the methods cannot accommodate all the
uncertain variables, except material balance and reservoir simulation, which were selected for comparison.
With all the parameters the same, the reservoir model was switched between tank model of material balance
method and reservoir simulation model and sensitivity studies were carried out.
To keep the simplicity, a simpler case than the case used to generate proxy model was designed For each
simulation, one variable value will be increased or decreased by 10% and forecast cumulative production
is recorded. Then, all the forecast results will be compared to the base case values. The variables introduce
greater variations to the cumulative two-year forecast are considered to have more influence to the model.
Finally, all the parameters are ranked in the order of their influences for both IAM models. The results
12 SPE-187595-MS

shown in Figure 9 and Figure 10 shed some insights. It was found that material balance tank model did not
provide reasonable results, for example, all the relative permeability parameters have no or little effects on
production forecast. The reason is that the material balance tank model is oversimplified and not able to
capture the internal dynamic changes inside the reservoir. But reservoir simulation model does and thus is
recommended for the IAM model.

Figure 9—Sensitivity study results for IAM model using tank model.

Figure 10—Sensitivity study results for IAM mdoel using physical reservoir simulator.
SPE-187595-MS 13

Forecast Period Effect


Some of the engineers that perform uncertainty analysis on production forecast tend to think that if
everything is kept the same, while the production forecast period is changed from 1 year, to 2 years or 5
years, the uncertainty results will remain the same. However, the simulation results showed not. Figure 11
illustrates the sensitivity results for three different forecast periods. The longer bar indicates that the variable
has higher ranking and more influence on production forecast. It can be seen that the two-year forecast
ranking results showed some variations compared with the one-year forecast results. Then, the five-year
forecast exhibited larger differences in the ranking results. This finding is important to the operators/decision
makers when they try to identify the critical parameters for their target/objectives of different development
periods. For example, the PINIT, initial reservoir pressure, has more influence for the one-year and two-year
forecasts than the five-year forecast. If short-term production is the focus as for unconventional plays, more
attentions should be given to the uncertainty associated with the estimation of initial reservoir pressure. But
for longer term objectives, it is less critical. Thus, the forecast period can affect the relative importance
of uncertain variables. The variables in red color are found to have fixed relative influence on production
forecast, while the significances of other variables dynamically evolve. To summarize, the uncertainties
associated with production forecast have similar "time" value with currency.

Figure 11—Sensitivity study on varying forecast period.

Proxy Model Validation


To validate the model, new cases were generated by varying the base values by ±5%. Simulation results from
IAM model indicated by the red dots were compred with the predictions from proxy model indicated by the
bar, as shown in Figure 12. A good match was achieved with correlation coefficient of 95.70%. This proves
the validity of this workflow that it is able to generate proxy model that can approximate fully physical IAM
models and produce accurate predictions. Additionally, by closer examination, the largest prediction errors
are associated with WOC and thickness. This is probably due to high non-linear relations between these
two variables and production forecast. Some further steps can be done to improve the forecast accuracy for
14 SPE-187595-MS

these conditions, for example, increase the data point density on response surface for these two variables,
or change the regression model formulation to better capture the non-linearity.

Figure 12—Proxy model validation.

Proxy Model Application on Uncertainty Analysis


To test the performance of proxy model, an uncertainty analysis on production forecast through Monte Carlo
simulation was performed. In this synthetic case, different probability distribution functions are assigned
to the five factors, shown in Figure 13. Monte Carlo Simulations were run for 5000 iterations with Latin
Hypercube sampling. The resulting probability distribution curve and cumulative probability curve for two-
year cumulative production forecast are shown in Figure 14. Associated statistical results are also produced,
including P(50)/Mean = 191.7 MMBBO, 90% confidential interval = [172.0, 210.9] MMBBO, Std Dev
= 12.1.

Figure 13—Define the uncertainty for the variables.


SPE-187595-MS 15

Figure 14—Probablity curve and cumulative probability curve for cumulative production forecast.

The most significant aspect of the proxy model is the simulation run time. For 5000 runs, it only takes
3 seconds using the proxy model compared with 500+ hrs if actual fully physical IAM is used (assume
average fully physical IAM run takes 6 mins). Besides uncertainty analysis, this workflow can also be
applied to other important problems, such as optimization by integrating designing/operating parameters
into the study. Thus, the decision making process can be greatly accelerated using proxy models.

Conclusion
A workflow has been developed to effectively integrate all the production components to an IAM then
generate its proxy model for faster deployment. The conclusions are summarized as follows.
1. To build accurate IAM, it is recommended to take careful considerations to select proper tools to model
different components of integrated production/asset system and avoid misleading uncertainty results.
2. It was also found that the relative importance or influence of some parameters will not remain constant
but evolve through the life of field. Thus, it is necessary to keep the production forecast period
of uncertainty analysis consistent with the development period associated with field strategies and
economic objectives.
3. To reduce the computational cost and accelerate decision-making process, a proxy model was
generated to approximate this physical IAM model with high correlation coefficient of 99.65%.
4. In case study, the high impact factors that can explain most variations in response were identified
from totally 29 uncertain variables through screening design, including water-oil contact, thickness,
oil density, initial reservoir pressure, and porosity.
5. The validation results showed that proxy model predictions matched closely with simulation results
from fully physical IAM with correlation coefficient of 95.70%.
6. Based on a synthetic production forecast uncertainty study using Monte Carlos simulation, the proxy
model can reduce the time for five-thousand simulation runs to seconds from hundreds of hours for
fully physical IAMs.

Acknowledgement
We would like to thank Landmark Software & Services of Halliburton for support in publishing this paper
and providing the licenses for the tools used in this work.
16 SPE-187595-MS

References
Friedmann, F., A. Chawathe, and D.K. Larue. 2003. Assessing Uncertainty in Channelized Reservoirs Using Experimental
Designs. SPE Reservoir Evaluation & Engineering 6 (4): 264–74. 10.2118/85117-PA.
Gong, X., Gonzalez, R.A., McVay, D., Hart, J.D., 2011. Bayesian Probabilistic Decline Curve Analysis Quantifies Shale
Gas Reserves Uncertainty. Society of Petroleum Engineers. 10.2118/147588-MS.
Guirillot, D. 2005. Uncertainty Assessments In Hydrocarbon Forecasts For Mature Fields. Paper WPC-18-0882 presented
at 18th World Petroleum Congress. Johannesburg, South Africa, 25–29 September.
Landmark-Halliburton. 2016. Nexus User Guide. Houston, TX.
Hepguler, G., Barua, S., Bard, W., 1997. Integration of a Field Surface and Production Network with a Reservoir Simulator.
SPE Computer Applications 9, 88–92. 10.2118/38937-PA.
Kalla, S., White, C.D., 2007. Efficient Design of Reservoir Simulation Studies for Development and Optimization. SPE
Reservoir Evaluation & Engineering 10, 629–637. 10.2118/95456-PA.
Litvak, M.L., Wang, C.H., 2000. Simplified Phase-Equilibrium Calculations in Integrated Reservoir and Surface-Pipeline-
Network Models. SPE Journal 5, 236–241. 10.2118/64498-PA.
Litvak, M., Christie, M., Johnson, D., Colbert, J., Sambridge, M., 2005. Uncertainty Estimation in Production Predictions
Constrained by Production History and Time-Lapse Seismic in a GOM Oil Field. Presented at the SPE Reservoir
Simulation Symposium, Society of Petroleum Engineers, The Woodlands, Texas. 10.2118/93146-MS.
Oehlert, G. W. 2000. A First Course in Design and Analysis of Experiments. W. H. Freeman.
Ojeke, A., I. H. Itotoi, D. Nnamdi, J. Umurhohwo, and O. Benjamin. 2011. A Robust Approach to Field Development
Plan Integrating Multi Dynamic Reservoir Models with Surface Network. Abuja, Nigeria, 30 July - 3 August.
10.2118/150735-MS.
Petroleum Experts Ltd. 2016. IPM Manuals. Edinburgh, Scotland.
Purvis, D. C., and H. Kuzma. 2016. Evolution of Uncertainty Methods in Decline Curve Analysis. Paper SPE-179980-
MS presented in SPE/IAEE Hydrocarbon Economics and Evaluation Symposium. Houston, Texas, USA, 17–18 May.
10.2118/179980-MS.
Purwar, S., C. J. Jablonowski, and Q. P. Nguyen. 2010. A Method For Integrating Response Surfaces Into Optimization
Models With Real Options: A Case in Gas Flooding. Paper SPE-129566-MS presented at SPE Hydrocarbon Economics
and Evaluation Symposium. Dallas, Texas, 8–9 March. 10.2118/129566-MS.
Roadifer, R.D., Sauve, R.E., Torrens, R., Mead, H.W., Pysz, N.P., Uldrich, D.O., Eiben, T., 2012. Integrated Asset
Modeling for Reservoir Management of a Miscible WAG Development on Alaska. Presented at the SPE Annual
Technical Conference and Exhibition, Society of Petroleum Engineers, San Antonio, Texas, USA. 10.2118/158497-
MS.
Rodriguez V., R. Angelica, K. Solano, S. Guevara, M. Velezquez, and L. A. Saputelli. 2007. Integration of Subsurface,
Surface, and Economics Under Uncertainty in Orocual Field. Paper SPE-107259-MS presented at Latin American &
Caribbean Petroleum Engineering Conference. Buenos Aires, Argentina, 15–18 April. 10.2118/107259-MS.
Saputelli, L.A., Mijares, G., Rodriguez, J.A., Sankaran, S., Revana, K., Garibaldi, L.A., 2007. Managing Risk and
Uncertainty in the Visualization of Production Scenarios. Presented at the Hydrocarbon Economics and Evaluation
Symposium, Society of Petroleum Engineers, Dallas, Texas, U.S.A. 10.2118/107562-MS.
Silva, V.C., Pinto, J.W., 2015. Reservoir-Schedule Coupled Uncertainty Analysis for PD Projects: Optimization
Opportunities and Improvements for More Robust Production Forecasts. Presented at the Offshore Technology
Conference, Offshore Technology Conference, Rio de Janeiro, Brazil. 10.4043/26309-MS.
Torrens, R., Mohamed, M.E., Al Bairaq, A., Kumar, A., 2014. Integrated Asset Modeling of a Gas Condensate Field
Operating Under Gas Recycling Mode. Presented at the Abu Dhabi International Petroleum Exhibition and Conference,
Society of Petroleum Engineers, Abu Dhabi, UAE. 10.2118/171988-MS.
Vanegas Prada, J. W., J.C. Cunha, and L.B. Cunha. 2006. Uncertainty Assessment of Production Performance for a Heavy
Oil Offshore Field by Using the Experimental Design Technique. Paper PETSOC-2006-125 presented at Canadian
International Petroleum Conference. Calgary, Alberta, 13–15 June. 10.2118/2006-125.

S-ar putea să vă placă și