IPTC 17377

Dynamic Reservoir Uncertainty Evaluation for Production Delivery

Assurance
M. A. Baslaib, A. BenSadok, H. Arii, M. Espinassous, G. Bourdarot, M. Attia, Abu Dhabi Marine Operating Co.
Copyright 2014, International Petroleum Technology Conference

This paper was prepared for presentation at the International Petroleum Technology Conference held in Doha, Qatar, 2022 January 2014.

This paper was selected for presentation by an IPTC Programme Committee following review of information contained in an abstract submi tted by the author(s). Contents of the paper, as
presented, have not been reviewed by the International Petroleum Technology Conference and are subject to correction by the author(s). The material, as presented, does not necessarily
reflect any position of the International Petroleum Technology Conference, its officers, or members. Papers presented at IPTC are subject to publication review by Sponsor Society Commi ttees
of IPTC. Electronic reproduction, distribution, or storage of any part of this paper for commercial purposes without the written consent of the International Petroleum Technology Conference 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
where and by whom the paper was presented. Write Librarian, IPTC, P.O. Box 833836, Richardson, TX 75083-3836, U.S.A., fax +1-972-952-9435


Abstract

A new production development scheme was proposed for one of the major offshore fields in the Gulf. To assist building
assurance of the delivery of this scheme, subsurface uncertainty evaluation was initiated; firstly based on an in-house approach
and later through the use of commercial softwares. The proposed study is primarily focused on dynamic reservoir
uncertainties. The objective of this study is to evaluate the impact of the reservoir uncertainties on plateau duration period i n
the prediction phase.

The workflow, based on experimental design and response surface, advantages and outcomes of this study are
presented and its limitations discussed. This study was completed over a short period of time, thanks to an optimization of the
CPU resources. This was a key advantage obtained while carrying large models with long history exceeding 50 years.

The study resulted in successfully delivering probabilistic profiles (P90, P50, P10) in order to assure the production
delivery of the proposed development scheme and enables to develop risk mitigation plans. A ranking of the most influential
uncertainty parameters with quantification of their interactions is obtained. Consideration is also given to the history match
quality, which results in reducing the parameters distributions and highlights the value of the data acquisition.

Introduction

ADMA launched a study to assist building assurance of the delivery of new production development scheme and to assess the
risk of applying such strategy. The objective of this study is to evaluate the impact of the reservoir uncertainties on plateau
duration period in the prediction phase. This study come up with a probabilistic forecast (P90, P50, P10) instead of a single
deterministic forecast as usual. First, ADMA launched the study in-house in two phases. The fist phase is simple and based on
mono-parameter sensitivity at a time method but requires a large number of runs and doesnt evaluate the impact of
interactions between parameters. 13 parameters are used and 4 runs are generated for each parameter. The second phase is
based on the Monte Carlo analysis. After assigning a probabilistic law to each uncertain parameter, n realizations of the
uncertain parameters are picked to obtain a sample of n reservoir simulations. This method is expensive in terms of number of
reservoir simulation runs (at least 100 runs for 10-15 parameters). Secondly, ADMA called for two companies and their
commercial softwares (X and Y) to tune the study further.

ADMAs optimized methodology is based on experimental design and response surface modelling to model flow
simulations. The same 13 uncertain parameters have been considered for this Field study. Those parameters have been re-
evaluated and constrained by historical data in order to reduce uncertainty range and they were assigned a new distribution
law. Using new ranges and distribution laws, uncertainties are propagated and probabilistic production forecasts are obtained.
The work flow of this study is shown in figure 1.
2 IPTC 17377

Figure 1Study Scope of Work

Study Details and Findings (X software)

Scope of Work.

ADMAs methodology is based on Experimental Design and Response Surface Modelling to model flow simulations
as per the proposed methodology shown in figure 1. This methodology is composed of three main steps:
a. Uncertainty Identification and Evaluation.
b. Uncertainty Constrained by History Data.
c. Uncertainty Impact.

Uncertain Parameters Identification and Evaluation.

13 parameters have been used. Most of the distributions of these parameters are triangle distributions based on the
distribution of the actual data (see table 1). Three uncertain parameters have been defined as discrete parameters; nevertheless
in the study they will have a continuous behavior. Those parameters are oil viscosity, Nw (5 tables) and Ng (7 tables).
Parameter Uncertainty Range Distribution
1
Kx-y (Mult) 0.5- 2.0 Triangular Distribution
2
Kz (Mult) 0.5- 2.0 Triangular Distribution
3
Tz (Mult) 0.001- 100 Log- Normal
4
Krw @Sorw (Mult) 1.0-2.0 Triangular Distribution
5
Krg @Sorg (Mult) 0.8 1.25 Triangular Distribution
6
SGCR (Mult) 0.2 2.0 Triangular Distribution
7
SORW (Mult) 0.4 1.2 Triangular Distribution
8
SORG (Mult) 0.5 1.65 Triangular Distribution
9
Now (Mult) 0.62 1.56 Triangular Distribution
10
Nog (Mult) 0.62 1.56 Triangular Distribution
11
Nw (Mult) 0.23 1.35 Triangular Distribution
12
Ng (Mult) 0.66 1.38 Triangular Distribution
13
Oil Viscosity (cp.) 0.21 - 0.34 Triangular Distribution

Table 1Uncertainties distributions
IPTC 17377 3
Experimental Design.

Latin Hypercube Design is the chosen experimental design for the study. It is the most efficient experimental design
recommended for building the surface response. It has the advantage of taking into account parameter probability distribution,
interaction between parameters, quadratic effect (and not only linear effect) for a minimum of runs number: about 3n runs for
n parameters plus several confirmation runs to test the quality of the surface response. The surface response or proxy is a
mathematical model reproducing the reservoir simulator response. It was planned to generate 30 simulations for the
construction of the response surface plus 6 confirmation runs (blind tests). The commercial software X was used for generating
the 36 data sets, based on Latin Hypercube Design (LHD) theory.

Uncertainty Constrained by Historical Data.

Figures 2, 3, 4 and 5 show the objective functions of the 36 runs for field oil production rate (FOPR), field gas
production rate (FGPR), field water production rate (FWPR) and well static pressure (WBP9). The objective function
(generally the square difference between simulated and measured data) is defined by combining, with possible weight on
controlling parameters, all the responses considered for history matching: oil, water and gas rates and pressure by wells. After
running the 36 simulations, it is clearly seen from figure 2 that FOPR during the historical period is not affected by the 13
uncertain parameters; all 36 simulations give the required FOPR with a maximum value of 0.1 objective function & all models
within closed range. However, it is unmistakable showed that there is a large impact on FWPR, WBP9 and to a lesser extend
in FGPR during the historical period. Because of this, a global objective function is performed for water production, gas
production and static pressure in order to perform a probabilistic optimization in X software. The contributions for the base
case simulation of each property to the global objective function are 39% for FGPR, 29% for FWPR and 32% for the well
static pressure. The acceptance of the runs should have an objective function of 4400 or less which is for the base case as
shown in figure 6. From the analysis based on the global objective function (figure 7), it is showed that 3 parameters have a
strong impact on the objective function which are SOGCR multiplier, KRWM multiplier and SOWCR multiplier which will
lead the optimization & data acquisition.


Figure 2Objective Function for field oil production rate


Figure 3Objective Function for field gas production rate
4 IPTC 17377

Figure 4Objective Function for field water production rate


Figure 5Objective Function for well static pressure (WBP9)


Figure 6Global Objective Function for static pressure + FWPR + FGPR

IPTC 17377 5

Figure 7Quantitative sensitivity analysis for Global Objective Function

Uncertainty Impact.

Figure 8 (Pareto plot) shows a quantitative analysis of the impact of the 13 parameters on the cumulative field oil
production (FOPT) as 01/01/20XX as well as the existence of interactions in between parameters. Two parameters (SOWCR
multiplier and PERMX multiplier) account for more than 70% of the field oil production total variation (dark blue bar). The
difference in between the dark blue bar and light blue bar is providing us a sign of interaction in between the parameters
involved; nevertheless for this particular case interactions are not higher than 5%.

Four parameters, SOGCR multiplier Now multiplier, Nog multiplier and Ng multiplier account for more than 30% of
the field oil production total variation (dark blue bar). As matter of fact, it is evidently seen that by narrowing the ranges of
these 6 uncertain parameters, the uncertainty on prediction field oil production total can be reduced significantly.


Figure 8Pareto Plot for FOPT at 01/01/20XX

Figure 9 (Pareto plot) shows a quantitative sensitivity analysis of the impact of the 13 parameters on the plateau
duration as well as the existence of interactions in between parameters. 5 parameters have the most effect on the plateau
duration which are PERMXY multiplier, Now multiplier, SOWCR multiplier Nog multiplier and viscosity. All of these
parameters are matching with the previous 6 parameters except viscosity. This parameter, viscosity, should be added to the
previous 6 parameters to be considered as the main affecting uncertainty parameters.
6 IPTC 17377


Figure 9Pareto Plot for Plateau duration

Figure 10 shows field oil production rate from the beginning of prediction until 01/01/20XX. The dash black line
shows the P0 and P100 values, the envelope in light grey shows the solution space in between P10 and P90 and the light blue
line shows P50 values. All these lines were calculated from the response surfaces (one response surface per period). In
addition, it has been superimposed 4 simulations out of the 36 simulations; these 4 simulations were selected in order to show:
Base Case, simulation 31 (Magenta)
Best Match, simulation 13 (Blue)
P50, simulation 5 (Black)
P10, simulation 8 (Green)


Figure 10Probabilistic Plateau (FOPR)

It is clearly seen from the figure that the base case and the best match have a similar behavior regarding field oil
production rate, it is also important to notice that the base case is practically following the P80 curve calculated from response
surface which means that we are confidence about the base case. The main different in between P50 (simulation 5) and the
base case is the match in pressure, nearly all wells in the base case show an underestimation for pressure, however pressure in
P50 (simulation 5) has a better match.

Study Details and Findings (Y software)

Scope of work.

In this evaluation study, the same proposed workflow was implemented as outlined in Figure 1. Uncertainty
parameters were defined with given ranges and pre-defined distribution function forms. Mismatch parameters (i.e., FOPR,
FWPR, FGPR and WBP9) were used as analysis parameters. Y software offers different experimental design and sampling
techniques. For this study a Latin-Hypercube technique was used to generate 40 simulation cases (experiments). Each
experiment is defined by new set of input parameter decks to the Eclipse simulator which are automatically prepared by Y

IPTC 17377 7
software. The output data of the all Eclipse runs was post-processed by Y software to calculate the mismatch parameters as
well as other key performance indicators.

Uncertain Parameters Identification and Evaluation.

The first objective of this study was to explore the influence of dynamic reservoir uncertainties on plateau duration
period in the prediction phase. This evaluation was started with 13 pre-determined uncertainty parameters as previously
described. Some refinement in parameters correlation well introduced compared to previous study. Nw & Ng have been
considered as continuous parameter distributions rather than discrete as in previous study.

Experimental Design.

Latin Hypercube was used as a sampling technique. 40 runs were planned to be generated. Two cases were selected
for blind tests and were therefore not part of proxy building process.

For the purpose of this study, the global error is used (Objective function) as a tool to identify the cases in which the
historical data is matched withing an acceptance error. The error contribution in the response parameter definition was initially
tuned based on the accebtable standard deviation. The standard deviations for the mismatch parameters are 200 bbl for oil rate,
200 bbl for water rate, 1000 Mscf for gas rate and 50 psi for the shut-in pressure. A filter criteria was defined for the global
mismatch value with a limit of 4800 (the global mismatch quality value is the summation of partial error contributions from
FOPR/FOPRH, FWPR/FWPRH, FGPR/FGPRH and shut-in pressure of 210 wells). This criteria excluded 12 cases having a
poor match and the ranges on the posteriori distributions have been reduced to limit them within acceptable history match.

Uncertainty Impact.

Y software has the capability to calculate the linear relationship of any uncertainty parameter vs. response parameter.
Negative values reflect negative correlation and positive values reflect positive correlation. Using Pearson method, the
following Pareto chart (figure 11) was generated to see the impact of the uncertainty parameters on the plateau duration. This
plot doesnt show the interaction between the parameters. There are 6 main parameters have the most effect on the plateau
which are Sorw multiplier, viscosity, Tz multiplier, Nw multiplier, Ng multiplier and PERMXY multiplie. Three of these
parameters are matching with the parameters in the previous study. This difference is due to the consideration of the continuos
of Nw & Ng as continuos distribution compared to the previous study.


Figure 11The effect of uncertainty parameters on plateau duration

Different techniques have been used to generate the proxy (auto & manual regression and kriging with and without
reducing the distribution ranges) and different sampling techniques (Monte Carlo and Latin-Hypercube) are used to get (P90,
P50, P10 risks). Table 2 summarizes the results of all approaches. The base case with plateau duration of 11.6 years is very
close to P10 risk which confirms a conservative base case as per previous study.





8 IPTC 17377

Sample Tech.
& space
P10 P50 P90
Auto regression M.C., 10000
12.7 16.6 21
Auto regression LH, 1000
13.6 16.9 20.2
Manual Regression
(without Adj.)
M.C., 10.000
11.3 15.4 19.8
Manual Regression
(without Adj. dist.)
LH, 1000
11.2 15.4 19.4
Manual Regression
(with Adj. dist.)
LH, 1000
12.5 16.6 20.5
Kriging (with Adj.
dist.)
LH, 500
12 17 21
Original simulation
data
40 cases + Base
3.0 15.42 19.42
Original simulation
data
Global<4800
28 cases + Base
11.5 16.6 20
Original simulation
data
Global<9500
35 cases + Base
5.6 15.9 20.4
LH: Latin-Hypercube M.C.: Monte Carlo

Table 2Summary of Y software results of all approaches

Conclusion & Recommendation

1. Deliver insurance of the new production scheme (plateau from 11 to 21 years).
2. Probabilistic forecast (P90, P50, P10) from the dynamic model.
3. By this study, it will be helpful to reduce the impact of the uncertainties by making mitigation plans or requesting
additional studies & acquisition.
4. 13 dynamic uncertainy parameters have been evaluated.
5. Successful identification and ranking of the impact of uncertain parameters with quantitatification of the interactions
between parameters.
6. Considering history match for reducing individual parameter uncertainties.
7. Optimized CPU resources (36 runs for 13 parameters) instead of use 100 runs for Monte Carlo approach.
8. Short study < 1 month (including simulation run time and training).

Nomenclature

CPU = Central Processing Unit
FGPR = Field Gas Production Rate, Mscf/stb
FGPRH = History Field Gas Production Rate, Mscf/stb
FOPR = Filed Oil Production Rate, stb/d
FOPRH = History Field Oil Production Rate, stb/d
FWPR = Field Water Production Rate, stb/d
FWPRH = History Field Water Production Rate, stb/d
K
rg
@ S
org
= Gas Relative Permeability @ Residual Oil Saturation respect to Gas
K
rw
@ S
orw
= Water Relative Permeability @ Residual Oil Saturation respect to Water
K
x-y
= Permeability in x-y direction, mD
K
z
= Permeability in z direction, mD
N
g
= Gas Corey Exponent
N
og
= Oil Corey Exponent Respect to Gas
N
ow
= Oil Corey Exponent Respect to Water
N
w
= Water Corey Exponent
S
GCR
= Critical Gas Saturation
S
org
= Residual Oil Saturation Respect to Gas
S
orw
= Residual Oil Saturation Respect to Water
IPTC 17377 9
T
z
= Transmisibility in z direction
WBP9 = 9 Point Pressure Average, psia


Acknowledgments

The authors wish to thank ADNOC and ADMA-OPCO management for their support to publish this paper. Thanks extend
to Field Development and Subsurface Technology teams for supporting and providing all the data required.

References

C. y. Peng, SPE, Curtin Uni of Tech; R. Gupta, Curtin Uni of Teck. Experimental Design and Analysis Methods in Multiple
Deterministic Modelling for Quantifying Hydrocarbon In-Place Probability Distribution Curve. Paper SPE 87002
presented at the SPE Asia Pacific Conference on Integrated Modelling for Asset Management held in Kuala Lumpur,
Malaysia, 29-30 March 2004.

E. Lisboa and R. Duarte. Uncertainty Analysis Considering the Production History: Evaluation of a Real Field. Paper SPE 136681
presented at SPE Latin American & Caribbean Petroleum Engineering Conference held in Lima, Peru, 1-3 December 2010.

G. G. Becerra; A. P. Modenesi; E. F. A. Lisboa. Uncertainty History Matching and Forecasting, a Field Case Application. Paper
SPE 153176 presented at SPE Latin American & Caribbean Petroleum Engineering Conference held in Mexico City, Mexico,
16-18 April 2012.

L. C. Reis, SPE. Risk Analysis With History Matching Using Experimental Design or Artificial Neural Networks. Paper SPE
100255 presented at SPE Europec/EAGE Annual Conference and Exhibition held in Vienna, Austria, 12-15 June 2006.

M. K. Choudhary, SPE; S. Yoon, SPE, Chevron Energy Technology Co.; B. E. Ludvigsen, Scandpower PT. Application of Global
Optimization Methods for History Matching and Probabilistic Forecasting-Case Studies. Paper SPE 105208 presented at the
15
th
SPE Middle East Oil & Gas Show and Conference held in Bahrain, 11-14 March 2007.