Sei sulla pagina 1di 23

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

net/publication/266670912

A Pattern-Based Approach to Waterflood Performance Prediction Using


Knowledge Management Tools and Classical Reservoir Engineering
Forecasting Methods

Article · April 2014


DOI: 10.2118/169587-MS

CITATIONS READS

0 539

3 authors, including:

Emre Artun
Middle East Technical University, Northern Cyprus Campus
25 PUBLICATIONS   101 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Emre Artun on 02 May 2017.

The user has requested enhancement of the downloaded file.


Int. J. Oil, Gas and Coal Technology, Vol. 13, No. 1, 2016 19

A pattern-based approach to waterflood performance


prediction using knowledge management tools and
classical reservoir engineering forecasting methods

Emre Artun*
Petroleum and Natural Gas Engineering Program,
Northern Cyprus Campus,
Middle East Technical University,
Kalkanli, Guzelyurt, TRNC,
Mersin 10, 99738, Turkey
Email: artun@metu.edu
*Corresponding author

Maurice Vanderhaeghen
QRI International, LLC,
2 Houston Center,
909 Fannin, Suite 2200,
Houston, TX 77010, USA
Email: mlvanderhaeghen@gmail.com

Paul Murray
Upstream Digital Intelligence,
1855 Post Oak Dr,
Houston, TX 77027, USA
Email: paul@upstreamdigitalintelligence.com

Abstract: An efficient and rapid workflow is presented to estimate the


recovery performance of an existing vertical-well, pattern-based waterflood
recovery design using knowledge management and reservoir engineering in a
collaborative manner. The knowledge management tool is used to gather
production data and calculate pattern-based recoveries and injection volumes
by defining pattern boundaries and allocating annual well injection/production
volumes in a systematic manner. Classical reservoir engineering forecasting
methods, namely, a combination of oil cut versus cumulative recovery
performance curves, and decline curve analyses are applied to forecast the
performance of the waterflood pattern of interest. Extrapolating established
trends of oil cut vs. recovery for each pattern quantified future performance
assessments. Time is attached to the performance by introducing liquid
rate constraints. Forecasting using both constant and declining liquid rates
differentiated the impact of deteriorating reservoir pressure and oil-cut trends
on individual pattern oil rate forecasts thus defining current efficiency of each
pattern. [Received: November 20, 2014; Accepted: September 22, 2015]

Copyright © 2016 Inderscience Enterprises Ltd.


20 E. Artun et al.

Keywords: waterflooding; predictive data analytics; performance prediction;


knowledge management; decline curve analysis; carbonate reservoirs; pattern
flood; dashboards; business intelligence; pressure monitoring.

Reference to this paper should be made as follows: Artun, E.,


Vanderhaeghen, M. and Murray, P. (2016) ‘A pattern-based approach to
waterflood performance prediction using knowledge management tools and
classical reservoir engineering forecasting methods’, Int. J. Oil, Gas and Coal
Technology, Vol. 13, No. 1, pp.19–40.

Biographical notes: Emre Artun is an Assistant Professor of Petroleum and


Natural Gas Engineering at Middle East Technical University, Northern Cyprus
Campus. He previously worked for Quantum Reservoir Impact as a Reservoir
Engineering Analyst, and Chevron Energy Technology Company as a
Reservoir Simulation Consultant. He received his PhD from Penn State
University, MS from West Virginia University, BSc from Middle East
Technical University (Ankara, Turkey), all in Petroleum and Natural Gas
Engineering. He also holds a graduate minor in computational science from
Penn State. His research interests include screening and optimisation of
secondary/enhanced oil recovery processes, computational intelligence and
data-driven modelling applications in reservoir engineering problems.

Maurice Vanderhaeghen is a Reservoir Management Consultant. After


graduating from Imperial College, London in 1988 with a Master’s of
Petroleum Engineering, he trained as a Reservoir Engineer with Energy
Resources Consultant (ERC) in England. He then worked for Soekor (South
Africa), Woodside Offshore Petroleum (Australia), ConocoPhillips, Hess
Corporation (Algeria) and Quantum Reservoir Impact (QRI).

Paul Murray has 20 years of experience leveraging knowledge management


and business intelligence for business value in the upstream sector. He is
currently an independent consultant for Upstream Digital Intelligence, LLC. He
has had the pleasure working with clients across the globe including:
Saudi Aramco, Microsoft, SAIC, Kuwait Oil, PEMEX, Quantum Reservoir
Impact (QRI), Kinder Morgan, and Chevron. He is a published author of
numerous papers on the subject of KM/BI/Analytics in oil and gas.

This paper is a revised and expanded version of a paper entitled ‘A


pattern-based approach to waterflood performance prediction using knowledge
management tools and classical reservoir engineering forecasting methods’
presented at the 2014 SPE Western North American and Rocky Mountain Joint
Meeting, Denver, Colorado, 17–18 April 2014.

1 Introduction

Secondary recovery and pressure maintenance by waterflooding is a significant recovery


process in the life of an oil field. This is a critical reservoir management practice for
optimum recovery from undersaturated oil reservoirs. Varying reservoir characteristics
and limited water injection capacities make it critical to have a good understanding of the
reservoir and managing the waterflooding design and operations to maximise the
efficiency.
A pattern-based approach to waterflood performance prediction 21

Good understanding of the reservoir geology and sufficient amount of high-quality


surveillance data typically end up in a robust flow-simulation model that represents the
reservoir flow dynamics reasonably well. These models are used as decision-making
tools to design the water injection process such that the volumetric sweep efficiency is
maximised. However, in the case of highly heterogeneous reservoirs, the challenge
associated with adequate reservoir characterisation often makes fundamental
understanding of the reservoir flow dynamics difficult to predict. It was shown in very
early experimental studies of waterflooding that heterogeneities can affect the sweep
efficiency significantly regardless of the pattern design and result in a wide range of
efficiencies (Crawford, 1960).
Waterflood performance prediction has been a deep interest to the petroleum industry
since the early years of scientific studies related to reservoir engineering. Studies that are
based on the fundamentals of Arps’ (1945) decline curves and experimental studies
(Crawford, 1960) are followed by development of simple analytical tools throughout the
years as a result of continuing interest (Higgins and Leighton, 1963; Higgins et al., 1966;
Hiatt, 1968; Khan, 1971; Craig, 1973; Chapman and Thompson, 1989). With the
advancement of computational technologies and power, numerical simulation has become
a standard tool for waterflood performance prediction (Thomas and Driscoll, 1973;
Tompang and Kelkar, 1988). However, due to the aforementioned challenges related to
reservoir description and representation in the model construction process, interest in
empirical and simple tools has not been altered. Empirical methods are powerful by
incorporating only measured, quantitative data that inheritably accounts for geological
complexity coupled to the chosen recovery process.
A number of empirical and analytical studies that incorporate modern computational
tools to improve prediction of waterflood performance have been presented in the
petroleum engineering literature. These studies all focus on measured quantities related to
reservoir performance. Examples of these studies include the use of fractal geostatistics
(Di Julio, 1993), artificial neural networks (Mohaghegh et al., 1993; Nikravesh et al.,
1996; Aminian et al., 2000), decision trees (Fedenczuk et al., 2002), capacitance-
resistance models (CRMs) (Sayarpour, 2009), associative modelling and ant colony
optimisation (Popa et al., 2012). In addition to these, classical decline-curve analysis-
based methods have been studied comparatively which identified advantages and
disadvantages of various methods (Bondar and Blasingame, 2002; Can and Kabir, 2012).
In this study, a combination of well-known reservoir engineering methods and a
practical knowledge management tool are used collaboratively to predict the performance
of waterflooding patterns. Knowledge management is primarily used as a part of a
short-term forecasting initiative. The main objective was to follow a pattern-based
approach to focus on injection patterns to identify flood performance, group trends and
outliers. This is done by defining pattern boundaries and allocating fraction of well
production accordingly using the knowledge management tool. This has put the focus on
individual patterns rather than individual wells or the field in general. Allocated annual
injection and production volumes of patterns are used to utilise some of the well-known
reservoir engineering forecasting and visualisation methods such as water-oil-ratio
(WOR) extrapolation method, classical decline curve analysis, and bubble maps. The
workflow is utilised for both a synthetic reservoir model and real field data with inverted
waterflooding patterns. Main advantages of this approach can be listed as follows:
22 E. Artun et al.

1 focusing on injection patterns allow capturing the actual control volume flow
dynamics, rather than focusing on the field at a broader scale, or well pairs at a more
localised scale
2 the methodology only requires well locations and production/injection volumes, and
hypothetical pattern definitions, which is typically readily available for any
waterflooded oilfield
3 performance prediction using either constant or declining liquid rates allow
implementing different artificial lift conditions, and/or facility constraints in the
field.
Limitations of the methodology include:
1 There is a challenge associated with characterising the connectivity between
injectors and producers for accurate allocation of production from wells which
are on or very close to pattern boundaries. Although physically located inside the
boundaries of a pre-specified pattern, a well’s production may fully or partially
belong to another pattern due to connectivity of the reservoir.
2 In reservoirs where the recovery mechanism is mostly by expansion due to large
pressure declines, rather optimistic results may be obtained in terms of the water
displacement efficiency. Therefore, understanding the recovery mechanism is very
important for properly interpreting the results.
In the following sections, the methodology followed, results obtained, and main
conclusions drawn from the study are presented.

Figure 1 The workflow followed to forecast pattern performance and calculate contributions of
pressure and water to the overall decline (see online version for colours)

FIELD Sum BOPD & Well Count Avg BOPD (by Lift)

Avg BOPD & Water Cut vs. Time Avg BOPD per Well

2 Methodology

The methodology followed in this study is composed of two principal components:


1 an organised knowledge management through appropriate workflows and tools
2 classical reservoir engineering forecasting methods.
A pattern-based approach to waterflood performance prediction 23

These two groups of methods are explained in this section. The primary objective of this
study is to forecast the performance of water injection. The methodology starts with
utilising proper knowledge management practices that accesses and manipulates data for
the forecast. After obtaining forecasts using constant and declining liquid rates, the
contribution of declining reservoir pressure and oil cut trends to the overall decline of a
given pattern can be estimated. This systematic approach results in a simple workflow
that is shown in Figure 1. The detailed explanations of each step of the workflow are
included in the following subsections.

2.1 Knowledge management tool (AVAILS+)


Knowledge management can be broadly defined as a systematic approach to getting the
right information or data to the right people at the right time (Behounek, 2003). As higher
volumes of data become available in relatively shorter time frames with newer tools,
technologies and processes, the oil industry requires asset teams to analyse the data at the
same pace, and make quick and reliable decisions to manage assets better, increase
production and create value. Therefore, proper knowledge management has become a
necessity in every stage of oil field operations. However, this is a challenging process
especially in large enterprises (>50,000 people, >1 million BOPD) due to communication
and interface issues (Al-Saad et al., 2013). Data challenges exist including disparate
datasets as well as disparate data systems. Production and injection data, while stored in
similar repositories, are metered differently including frequency and variance. Key data
for analysis, such as described in this paper, requires data that may reside within oil and
gas applications, such as FINDER®, GEOLOG®, ECLIPSE®, LANDMARK®, and other
systems that lack clear inter-operability. Looking ahead these challenges are only going
to intensify with the adoption of Big Data technology such as VMonitors and down-hole
sensors coupled with supervisory control and data acquisition (SCADA) systems for real-
time surveillance and monitoring.
A goal of knowledge management over the years has been the ability to integrate
information from multiple perspectives to provide the insights required for valid
decision-making. In order to best manage an asset, field development decisions are made
based on a variety of inputs from production, to costs, to equipment procurement, to
drawdowns, to voidage, to pressure, to facility constraints. The total picture is what
should drive decisions, such as; where to inject, how much to inject, and whether to
re-design the injection strategy or the injection well itself. In order to make sound
decisions regarding the water injection strategy, a holistic picture of field performance is
required, specifically injection-pattern performance to date. However with the volume,
variety, and velocity by which the data was being made available, expecting a
comprehensive, field-wide analysis of injectivity performance would be very difficult
and time-consuming with traditional forecasting and analytics tools – such as
simulation software. The complexity amplifies with the advancement in well
completion/design technologies and field development strategies that include drilling
more non-conventional, maximum reservoir contact (MRC) wells rather than
conventional wells.
The knowledge management tool that is primarily used is called AVAILS+ which
was developed as a part of a short-term forecasting initiative. It is an empirically-driven
tool, deriving its forecast and monitoring capabilities from observed reservoir and
24 E. Artun et al.

well-performance data, moving away from simulation-centric reservoir forecasting


methods (Al-Saad et al., 2013). The initiative is successfully implemented by engineers
to perform a series of analysis and empirical modelling on the diverse collection of data
including injection pattern analysis as seen in the operational indices close-up (Figure 2).

Figure 2 Operational indices close-up: AVAILS+ supports a suite of integrated,


reservoir-focused analysis whose primary goal is to determine the forecast
assumptions and operational indices that will drive the annual production forecast
(see online version for colours)

PGOR Well Test Data


Incremental Barrels of Oil Regional Oil Rate Distribution

Voidage Primary
SW flanks
Oil
(Water injection)
S pattern Water
NW pattern
Drilling Efficiency
NE flanks GOR
Rig Count E pattern BHP
Sum. BOPD & Well Count
Operational ESP Daily Results ESP Trips
Efficiency

Major Shutdowns

Avg. BOPD per Region


Primary
SW flanks
S pattern
NW pattern
NE flanks
E pattern

Source: After Al-Saad et al. (2013)


The knowledge management (KM) tool AVAILS+ envelopes a variety of key data
sources and provides the end-user with statistical outputs in highly visual manner
(dashboards). The empirically-driven datasets can be kept up to date in a timely fashion
while the interface allows for engineers to interact and filter, limit, or augment the data
based on their insights. The tool is developed on TIBCO Spotfire® software, a leading
analytics software package providing a high degree of speed and flexibility in data
aggregation, relation, and transformation coupled with highly visual outputs
(dashboards).
To complement this study, the KM tool enabled engineers to isolate subsets of
injectivity performance by pattern or similar clustered subsets of data (Figure 3). By
grouping and analysing the data within these meaningful subsets allowed for deeper and
more robust analytics to be performed. The results of which could be cross-analysed with
other injection patterns (i.e., data subsets) in order to gain a deeper understanding of the
injection performance fundamentals. The KM tool contains all the field data which
primarily includes the well names, locations and production/injection history for this
study. This tool, while keeping all the required data in an organised manner, also
allocates well production/injection to each injection pattern, based on pre-defined
hypothetical pattern boundaries and location of the wells.
A pattern-based approach to waterflood performance prediction 25

Figure 3 Selecting and reviewing pattern performance within the KM tool (see online version
for colours)

FieldField

Well-01
Well-02
Well-03
Well-04
Well-05
Well-06
Well-07
Well-08
Well-09
Well-10

Note: The mashup of data coupled with logical hierarchies, such as date, enable engineers
to interact with, pivot, and analyse subsets of pattern performance in the field.

Figure 4 Defining pattern boundaries for allocation of well production to injection patterns based
on their location with respect to boundaries

Well-1

Well-2

Pattern 1 Pattern 2

Well-4
Well-3

Legend
Producer
Pattern 3 Pattern 4
Injector

Pattern boundary

To obtain the pattern production volumes, well production volumes are allocated to the
pattern(s) depending on its proximity to the pattern(s). This allocation is rather simplistic
that ignores deviations due to the heterogeneity in the reservoir. If the heterogeneity and
connectivity in the reservoir can be well-defined, a more accurate set of allocation factors
of the produced volumes could be determined through methods such as 3D streamline
simulation (Thiele and Batycky, 2006). Figure 4 illustrates the allocation process for
inverted water-injection patterns:
26 E. Artun et al.

• if a producer is located on a boundary that belongs to only one pattern, %100 of the
well’s produced volume is allocated to that pattern (e.g., Well-1 – Pattern 1)

• if a producer is not located on a boundary but inside the hypothetical boundaries


of a pattern, %100 of the well’s produced volume is allocated to that pattern
(e.g., Well-2 – Pattern 1)

• if a producer is located on a boundary that is shared between two patterns,


50% of the well’s produced volume is allocated to one of the patterns and other 50%
is allocated to the other pattern (e.g., Well-3 – Patterns 1 and 3)

• if a producer is located on a boundary that is shared between four patterns,


25% of the well’s produced volume is allocated each of the four patterns
(e.g., Well-4 – Patterns 1, 2, 3 and 4).

2.2 Classical reservoir engineering forecasting methods


2.2.1 WOR extrapolation method
There are many extrapolation methods that supplement the decline curve analysis used in
the industry. It was shown that when there are no planned operational changes in a
fully-developed waterflood, a plot of fractional water cut (fw) vs. cumulative recovery
(Np) can be used to estimate ultimate recovery at a given economic water-cut (Ershaghi
and Amoregie, 1978). Baker (1998) stated that plotting WOR versus cumulative recovery
for a given pattern or a group of patterns, and comparing it with the average behaviour of
all patterns could give a qualititative indicator of volumetric sweep efficiency. Successful
field applications of the method were presented in the literature (Currier and Sindelar,
1990). Another variation of the same approach is plotting the logarithm of the fractional
flow of oil (fo) versus cumulative oil recovery (Np) which results in a straight line for a
producer under an established water flood (Bondar and Blasingame, 2002). In this study,
latter form of the plot is used to identify established trends for the pattern-based
production. After plotting the logarithm of the fractional flow of oil (fo) vs. cumulative
recovery (Np), established trends are defined with an exponential equation in the form:
f o = aebN p (1)

After identifying the coefficients a and b, the established exponential trend between
log(fo) and Np the ultimate recovery can be calculated by extrapolating the trend to a
desired value of water cut (fw = 1 – fo).

2.2.2 Production forecast with constant or declining liquid rate constraints


Once establishing a reasonable trend using the WOR extrapolation method, it can be used
for forecasting future performance of a given injection pattern. Two possible approaches
include assuming
1 constant liquid rate production
2 declining liquid rate production.
A pattern-based approach to waterflood performance prediction 27

Constant liquid rate is useful when there are certain facility limits for liquids or the
reservoir is produced at the desired off-take rate and operating pressure. Another
important assumption that lies under constant liquid production is that there is efficient
pressure maintenance around the injection pattern. This can be identified by looking at
historical liquid production rate. Stable liquid production indicates that pressure is
maintained with water injection and the decline in oil rate is only due to increase in the
water production. However, if there is an observable decline in the liquid production rate
as well, it shows that the pressure is not maintained and the decline is due to both
deteriorating pressure and increasing water production. Then, it is more realistic to follow
the second approach which takes into account the declining liquid rate in the reservoir.
Both of the approaches could be utilised for forecasting purposes but it is important to
analyse historical performance to have an understanding of the actual reservoir dynamics.
For the constant liquid rate forecast, first the liquid rate to be used for future forecast
should be determined. This could be either the stable historical liquid production rate, ql,
of the pattern, or it could depend on certain facility limits, if there is any. Using the last
observed cumulative oil production value of the pattern, Npn, and the coefficients a and b
of the exponential trendline in equation (1), fractional flow of oil, fo, and water, fw, can be
estimated. Then, the oil and water flow rates can be calculated using the following
equations:
qo = f o ql (2)

qw = f w ql (3)

Finally, the cumulative oil recovery for next time step (e.g., next month), Np,n+1, can be
calculated by adding the incremental production using the estimated oil production rate in
equation (2), and multiplying it with the number of days in the month, m, to account for
the monthly cumulative production:
N p , n +1 = N p , n + mqo (4)

Then, fo, fw, qo, and, qw are again calculated using the cumulative recovery at the new time
step, Np,n+1. The forecast can continue using the same procedure until the desired time of
interest. When the forecast is completed, the oil rate, water rate, and cumulative recovery
vs. time forecasts are obtained. In this case, since the liquid rate is kept constant, the
decline in the oil rate is only due to increasing water production. In the case of the
declining liquid rate, observed decline rate for the liquid production is needed. This
effective decline rate is determined by analysing the historical liquid rate performance.
Remaining procedure is the same as the constant liquid rate forecast except using a
declining liquid rate for each time step. The decline observed in the oil rate is therefore a
function of both the pressure decline and increasing water production.

2.2.3 Calculating contributions to decline


Decline of the oil rate for the pattern under consideration would have two components in
natural water-drive or waterflooded reservoirs:
1 decline due to decrease in pressure
2 decline due to increasing water production.
28 E. Artun et al.

Depending on the conditions contribution of each of the two factors may differ. It is an
important reservoir management practice to characterise these contributions. As indicated
above, two individual forecasts that consider either constant or declining liquid rate,
capture different decline contributions. Constant liquid rate assumes that pressure is
maintained and the decline in the oil rate is only due to increasing water production. On
the other hand, declining liquid rate includes both the pressure decline, and increasing
water production. Therefore, both forecasts can be used to quantify the percentage
contributions from each sources of decline. A simple way to calculate the decline
contributions can be followed by calculating the effective decline rate using the
forecasted oil rates:
qo, n +1
D = 1− (5)
qo,n

If Dc is the effective decline rate of the forecast using constant liquid rate and, Dd is the
effective decline rate of the forecast using declining liquid rate, these two decline rates
can be used to calculate the contributions. Since Dc includes only the decline due to water
contribution, and Dd includes both decline contributions; the percentage contributions of
the water production, Cw, and the pressure, Cp, can be approximated from:
Dc
Cw = × 100 (6)
Dd

⎛ D ⎞
C p = ⎜1 − c ⎟ × 100 (7)
⎝ Dd ⎠

3 Results and discussion

3.1 Synthetic (simulation) case


To test the proposed methodology, a synthetic reservoir model is used in which four
inverted five-spot water injection patterns are placed (nine producers, four injectors). The
model grid structure is obtained from one of the models provided by Computer
Modelling Group (CMG-IMEX® Reservoir Simulator). Some of the reservoir properties,
initialisation, and well definitions are modified for this specific example. The reservoir is
a heterogeneous reservoir and undersaturated with an initial oil saturation of 0.88 and
initial immobile water saturation of 0.12. There are around 4,000 active grid blocks in the
model with 28, 32, and 6 grid blocks in each direction, x, y, and z, respectively. Top of
the structure, location of wells, and hypothetical boundaries of injection patterns are
shown in Figure 5.
Table 1 shows the contribution percentages from each producer to each pattern
(100%, 50%, or 25%). Again, it should be noted that this is a rather simplistic approach
for a heterogeneous reservoir with varying distances between injectors and producers. A
more realistic set of allocation factors could be determined and validated using other
methods. These allocation factors only provide an approximation to the performance of
each pattern.
A pattern-based approach to waterflood performance prediction 29

Figure 5 Structure top (ft), well locations and hypothetical pattern boundaries of the synthetic
reservoir model (see online version for colours)

Table 1 Contributions to the pattern production from each producer in the synthetic case

Pattern 1 (I1) Pattern 2 (I2) Pattern 3 (I3) Pattern 4 (I4)


Producer Contribution Producer Contribution Producer Contribution Producer Contribution
P8 100% P3 100% P2 100% P9 100%
P4, P6 50% P6, P5 50% P4, P7 50% P5, P7 50%
P1 25% P1 25% P1 25% P1 25%

The schedule starts by drilling producers in a specific order to have a primary recovery
period of five years. Producers come online every three months with a maximum oil rate
of 3,000 STB/d and 350 psia bottom-hole pressure constraint. There is also a field facility
constraint of 10,000 STB/d for liquid production. After the primary recovery
period, water injection starts by drilling injectors at the centres of patterns again with a
1-well/3-months schedule. Injection constraints are maximum water injection rate of
5,000 STB/d and maximum bottomhole pressure of 5,000 psia. The recovery with
waterflooding continues for ten years which is also the duration of the analysis for the
proposed methodology.
Table 2 Layer thickness and permeability distributions of the synthetic model for low-,
mid- and high-heterogeneity cases

Average thickness Low-case perm Mid-case perm High-case perm


Layer
(ft) (md) (md) (md)
1 21 10 5 5
2 23 10 10 10
3 45 10 50 50
4 30 10 10 10
5 43 10 5 250
6 19 10 5 5
30 E. Artun et al.

To demonstrate the effect of heterogeneity in permeability in each layer, three cases were
constructed with varying Dykstra-Parsons coefficients of permeability variation of 0, 0.38
and 0.69. These models are defined as low, mid, and high, respectively, in terms of the
permeability heterogeneity. In the low case, permeability is constant in all layers (10 md),
and the variation increases in the mid case, and increases more in the high case with the
Layer-5 having a permeability of 250 md. The variation is applied by adding high-
permeability layers to the reservoir to study the effects. Table 2 shows the horizontal
permeability and thickness distribution for all layers in the model.
As the first step of the methodology outlined in the previous section, pattern-based
recoveries are calculated using the contribution percentages shown in Table 1 after
running the model for 15 years of primary recovery and waterflooding period. Annual oil
and water production volumes are used to calculate the annual average oil-cut values to
construct oil-cut vs. recovery plots. In Figure 6, log(fo) vs. Np plot for an example pattern
in the synthetic case is presented for different heterogeneity cases. As seen in the graph,
variation in the permeability affects the recovery performance for a given pattern in terms
of the ultimate recovery when the trends are extrapolated to an oil-cut of 0.05. The shapes
of trends are different and it can be noted in the plot that an earlier water breakthrough is
observed in the high case, due to the 250 md layer introduced. It is also observed that
while the trend is established immediately after the water breakthrough for the
homogenous (low) case, it takes a while for the appropriate trends to be established for
the mid and high cases. This highlights the importance of characterising heterogeneities
in the reservoir as the degree of heterogeneity may have a significant impact on the
performance of injection patterns.

Figure 6 Log(fo) vs. Np trends established and extrapolated for an injection pattern for low-, mid-,
and high-heterogeneity cases (see online version for colours)

Low Mid High

0.50
fo

High Mid Low


0.05
0 10 20 30 40 50 60 70 80 90 100
Np, MMSTB
A pattern-based approach to waterflood performance prediction 31

Figure 7 shows the trends for all patterns for the mid-case. This kind of analysis provide
with an opportunity to compare patterns with each other and identify underperforming,
problematic patterns. After identifying the exponential trends, the coefficients of
equation (1), a and b are obtained which define the log(fo) vs. Np relationship for each
pattern. For each pattern, and for each case of the synthetic model, these coefficients are
listed in Table 3. After analysing the historical average rate of the liquid rate and its
decline behaviour, forecasts for the constant and declining liquid rates are obtained.
Figure 8 shows the two-year forecasts for Pattern-3 for the mid-heterogeneity case, as an
example. This pattern was characterised with stable liquid rate that results in a very small
difference in the oil rate (~22 STB/d) between constant and declining liquid rate
forecasts. After obtaining forecasts, the decline contributions are calculated using the
declines observed in the oil rate within 12 months of time in both forecasts. Using
equations (5) to (7), the decline contribution percentages shown in Figure 9 are
calculated. By looking at these contribution charts, it is possible to identify patterns that
are affected more by deteriorating reservoir pressure, or more by water that has broken
through.

Figure 7 Log(fo) vs. Np trends established and extrapolated for 4 patterns pattern for the mid case
(see online version for colours)
Pattern-1 Pattern-2 Pattern-3 Pattern-4

0.50
fo

Pattern-3
Pattern-2 Pattern-1 Pattern-4
0.05
0 20 40 60 80 100 120 140
Np, MMSTB

Table 3 Coefficients for the exponential fit obtained for the pattern behaviour in the
synthetic case

Pattern-1 Pattern-2 Pattern-3 Pattern-4


Low a –0.036 –0.047 –0.029 –0.025
b 1.431 1.592 1.315 1.313
Mıd a –0.034 –0.057 –0.026 –0.027
b 1.278 1.537 1.149 1.146
High a –0.023 –0.057 –0.043 –0.065
b 1.020 1.076 1.257 1.440
32 E. Artun et al.

Figure 8 Oil rate and water cut estimations for the duration of two years using constant and
declining liquid rates for Pattern-3 in the mid-case (see online version for colours)
Pattern-3
1700 0.270
Oil Rate (decline due to water)
1690 Oil Rate (decline due to pressure and water)
Water Cut (constant liq. rate) 0.265
1680 Water Cut (declining liq. rate)
Oil rate, STB/d

Water Cut
1670 0.260
1660
0.255
1650
1640 0.250
1630
0.245
1620
1610 0.240
0 4 8 12 16 20 24
Time, months

3.2 Real case


In this part of the study, a region with four inverted injection patterns are considered from
a field in the Middle East that has around 12 years of waterflooding history. The
carbonate reservoir has heterogeneous permeability distribution with thief zones that is
believed to affect the waterflooding performance significantly. Characterising thief zones
in the reservoir model such that the water movement in the reservoir is represented is a
challenging process. Therefore, it is very important to validate the reservoir model as
much as possible with other methods such as rather simplistic reservoir engineering
methods that rely on observed data. In this paper, we focus on the collaborative
knowledge management-reservoir engineering workflow presented above by highlighting
its practicality for a real case.
In Figure 10, the map of the injectors and producers in the patterns with defined
boundaries is shown. Considering the defined pattern boundaries, the production from the
surrounding producers are allocated using the proportional distribution explained
previously. Then, annual pattern oil and water production volumes are calculated
accordingly. Established oil cut vs. cumulative recovery plots for these patterns are
shown in Figure 11. These fitted behaviours are described with equation (1) with
coefficients a and b. These coefficients obtained for each pattern are shown in Table 4.
These fitted behaviours are extrapolated to the water cut of 0.95 (or oil cut of 0.05) to
estimate ultimate recovery (Np) as shown in Figure 11. Another useful application of this
approach is to estimate ultimate recovery factors. Cumulative productions can be also
represented as recovery factors if it is possible to estimate the original oil in place of the
bounded volume of a given pattern.
A pattern-based approach to waterflood performance prediction 33

Figure 9 Contributions of pressure and water to the overall decline of oil rate in each pattern for
synthetic model case (see online version for colours)

Low Mid High


Pattern-1
Pattern-2
Pattern-3
Pattern-4
34 E. Artun et al.

Figure 10 Map of injectors and producers and hypothetical pattern boundaries for the area
considered in the real case (see online version for colours)
5000
Producers
P3
4500
Injectors
4000

3500

3000
P2
Y, m

2500
P4
2000

1500

1000

500
P1
0
0 1000 2000 3000 4000 5000
X, m

Figure 11 Capturing pattern performance using log(fo) vs. Np plot and extrapolating the
observed behaviour to estimate ultimate recovery at water-cut of 95% (see online
version for colours)
a) Pattern-1 b) Pattern-2

0.50 0.50
fo
fo

y = 0.8803e-0.062x y = 1.0079e-0.039x
0.05 0.05
0 20 40 60 80 100 0 20 40 60 80 100
Np, MMSTB Np, MMSTB

c) Pattern-3 d) Pattern-4

0.50 0.50
fo
fo

y = 1.0193e-0.064x y = 0.9396e-0.08x
0.05 0.05
0 20 40 60 80 100 0 20 40 60 80 100
Np, MMSTB Np, MMSTB
A pattern-based approach to waterflood performance prediction 35

Table 4 Coefficients for the exponential fit obtained for the pattern behaviour in the real case

Pattern 1 Pattern 2 Pattern 3 Pattern 4


a –0.062 –0.039 –0.064 –0.080
b 0.880 1.008 1.019 0.940

Figure 12 shows the extrapolation plot for all patterns with the recovery factor on the
x-axis. This gives a clear understanding of the ultimate recovery factors for each pattern
and their comparison with each other. Another useful analysis can be accomplished
when the estimated ultimate recoveries or recovery factors are shown as a bubble map
(Figure 12). This would provide a regional overview of individual pattern performances.
The ultimate recovery values and recovery factors at the water cut of 95% are tabulated
in Table 5. These forecasts highlighted the optimistic nature of the dated full-field,
history-matched simulation model which provides higher recovery factors than
12%–24% range obtained both for the field in general and for the studied patterns.

Figure 12 Capturing pattern performance using log(fo) vs. recovery factor plot and extrapolating
the observed behaviour to estimate ultimate recovery at water cut of 95% (see online
version for colours)

Pattern 1
Pattern 2
Pattern 3
Pattern 4
0.50
fo

Pattern-4 Pattern-1 Pattern-3 Pattern-2


0.05
0% 5% 10% 15% 20% 25% 30%
Recovery Factor

Note: Visualising regional pattern performance using a bubble map.


Table 5 Forecasted cumulative oil production and recovery factors for each pattern

Pattern Np (MMSTB) at fo = 0.05 RF at fo = 0.05


1 45.9 17%
2 76.1 24%
3 47.0 20%
4 36.7 12%

The established trends are then used to estimate the future performance of each pattern.
To achieve this, two options are considered: constant and declining liquid rate. As
explained in the Methodology section, constant liquid-rate forecast is for patterns which
are well maintained in terms of pressure, but suffer from increasing water production. On
the other hand, declining liquid-rate also captures the declining reservoir pressure. In this
kind of patterns, it is important to understand the contributions of increasing water
36 E. Artun et al.

production and pressure decline to the overall decline in the oil production. Therefore,
these contributions are also calculated for the forecasted rates.
In this example, the last record of liquid production rate is assumed to be constant
during the forecast for the constant-decline rate forecast. This value can also be set as the
facility constraint for the handling of the daily liquid volume. For the declining liquid-
rate forecast, recent averaged liquid rate records are considered to characterise the
amount of decline for the pattern. This decline rate gives an indication of the expected
pressure decline in the pattern with current practices. When we analysed the recent
observed performance, it is observed that the produced liquid rate for Pattern-1 has been
constantly declining (Figure 13) and the liquid rate for other three patterns have been
stable. The effective exponential decline rate for Pattern-1 was estimated as 0.12. This
decline rate is used to forecast the performance with declining liquid rate. In other
patterns, due to the stable liquid rates, a decline rate of 0.001 is used as a generic decline
rate to represent the much less amount of pressure decline.

Figure 13 Liquid rate vs. time for Pattern-1 during the last three years (see online version
for colours)
15000
Actual
14500
14000 Decline fit
Liquid rate, STB/d

13500
13000
12500
12000
11500
11000
10500
10000
10 11 12 13
Time, years

Starting with the last production record, future recoveries were estimated by using the
pattern production trend that was obtained using the WOR extrapolation method.
Figure 14 shows these estimations for the duration of two years for the oil rate and the
water cut. It is seen in these plots that in Pattern-1, including the declining liquid rate
changes the forecast significantly. This is due to the observed decline in the liquid
production rate from this pattern. In all other patterns, because of the stable liquid
production rate, constant or declining liquid rates do not differ much from each other.
After obtaining these forecasts, effective annual decline rates are calculated for each
pattern’s forecast using equation (5). Assuming that this decline rate would include only
the decline due to increasing water production for the constant-liquid rate forecast,
and also that it would include the total decline due to water and pressure for the
declining-liquid rate forecast, these individual contributions are calculated using
equations (6) and (7). Figure 15 shows these distributions for the four patterns under
consideration. Because of the much higher decline rate observed in the liquid rate of
Pattern-1, forecasted rates indicate that 90% of the decline would be due to the pressure
decline. Other patterns (Patterns 2, 3, and 4) were not experiencing an observable decline
rate in recent years. Therefore, this was reflected in their declining liquid rate forecast by
using a relatively low decline rate of 0.001. This resulted in pressure-related decline in
the range of 8%–16%, with the majority of the decline is due to increasing water
A pattern-based approach to waterflood performance prediction 37

production. This approach allows to quantify these contributions, and therefore to


compare individual patterns with each other to identify most problematic injection
pattern(s) and source of the problem in a broad manner.

Figure 14 Oil rate and water-cut estimations for the next two years using constant and declining
liquid rates for all patterns in the real case (see online version for colours)
a) Pattern-1
4000 0.75

3500

3000 0.70
Oil rate, bopd

2500

Water cut
2000 0.65

1500

1000 Water Cut (Constant liquid rate) 0.60


Water Cut (Declining liquid rate)
500 Oil Rate (Decline due to water)
Oil Rate (Decline due to water and pressure)
0 0.55
1 6 11 16 21
Time, months
b) Pattern-2
10000 0.57
Water Cut (Constant liquid rate)
9000
Water Cut (Declining liquid rate) 0.56
8000 Oil Rate (Decline due to water)
Oil rate, bopd

7000 Oil Rate (Decline due to water and pressure) 0.55

Water cut
6000
0.54
5000
0.53
4000
3000 0.52
2000
0.51
1000
0 0.50
1 6 11 16 21
Time, months

c) Pattern-3
7000 0.75

6000 0.70

0.65
5000
Oil rate, bopd

0.60
Water cut

4000
0.55
3000
0.50
2000 Water Cut (Constant liquid rate)
Water Cut (Declining liquid rate) 0.45
1000 Oil Rate (Decline due to water) 0.40
Oil Rate (Decline due to water and pressure)
0 0.35
1 6 11 16 21
Time, months

d) Pattern-4
4000 0.75

3500

3000 0.70
Oil rate, bopd

Water cut

2500

2000 0.65

1500

1000 Water Cut (Constant liquid rate) 0.60


Water Cut (Declining liquid rate)
500 Oil Rate (Decline due to water)
Oil Rate (Decline due to water and pressure)
0 0.55
1 6 11 16 21
Time, months
38 E. Artun et al.

Figure 15 Contributions of pressure and water to the overall decline of oil rate in each pattern
for the real case (see online version for colours)
a) Pattern-1 b) Pattern-2
10% 16%

90% 84%

c) Pattern-3 d) Pattern-4

8% 13%

92% 87%

4 Summary and conclusions

A practical methodology is presented to estimate future performance of waterflooding


patterns. The methodology incorporates classical reservoir engineering forecasting
methods with a practical knowledge management tool to analyse performances of
individual patterns. The knowledge management tool allowed grouping and analysing the
patterns so that deeper and more robust analytics were performed. The methodology is
demonstrated for a synthetic reservoir model, and a real case for a reservoir in the Middle
East.
The main characteristic of the study is that it focuses on injection patterns through
hypothetical-boundary definitions, and approximating pattern-based cumulative volumes,
while most of the studies presented in the literature focuses on individual wells,
injector/producer pairs, or the field. Using patterns in the analysis provides the
opportunity to focus on the problem at a scale that is in between broader field-based scale
and localised well-based scale. This helps to identify problems such as injectivity
problems or artificial lift problems related to patterns so that required actions can be
prioritised and taken. Another aspect of the study is its predictive ability using either
constant or declining liquid rates. This allows implementing different artificial lift
conditions, or facility constraints in the field. It is believed that presented workflow can
be utilised efficiently by asset teams in a collaborative manner and result in significant
value creation by optimising strategies related to injection patterns. The methodology
could also serve as a process to set clear objectives for more detailed numerical
simulation studies for the reservoir under consideration.
A pattern-based approach to waterflood performance prediction 39

One of the limitations of this methodology is the challenge associated with


determining the fractions used to allocate production from wells. While an approach like
the one presented in this paper could be followed which distributes the volumes based on
locations of wells, a more accurate distribution of volumes can be estimated using
methods such as 3D streamline simulation. Another limitation is that in reservoirs where
the recovery mechanism is mostly by expansion due to large pressure declines, rather
optimistic results may be obtained in terms of the water displacement efficiency.
Therefore, it very is important to understand the reservoir drive mechanisms before
interpreting the results of the study.

Acknowledgements

This paper was presented as SPE 169587 at the 2014 SPE Western North American and
Rocky Mountain Joint Meeting, held on 17–18 April in Denver, Colorado. Authors thank
Quantum Reservoir Impact (QRI) management for their permission to publish this work.

Nomenclature
Variables Subscripts
C Decline contribution c Constant liquid rate forecast
D Effective decline rate, 1/time d Declining liquid rate forecast
N Oil volume in stock-tank conditions, STB l Liquid
a, b Exponential trend coefficients n Timestep
f Fractional flow o Oil
q Flow rate, STB/d p Pressure (decline rate contribution);
produced (oil volume)
m Number of days in the month w Water

References
Al-Saad, B., Murray, P., Vanderhaeghen, M., Yannimaras, D. and Naime, R.K. (2014)
‘Development and implementation of the AVAILS+ collaborative forecasting tool for
production assurance in the Kuwait Oil Company, North Kuwait (KOC NK)’, Paper SPE
167375, SPE Econ. & Man., Vol. 6, No. 4, pp.191–198.
Aminian, K., Bilgesu, H.I., Ameri, S. and Gil, E. (2000) ‘Improving the simulation of waterflood
performance’, Paper SPE 65630 presented at the SPE Eastern Regional Meeting,
Morgantown, West Virginia, 17–19 October.
Arps, J.J. (1945) ‘Analysis of decline curves’, Paper SPE 945228, Trans. AIME, Vol. 160, No. 1,
pp.228–247.
Baker, R. (1998) ‘Reservoir management for waterfloods – Part II’, J. Can. Pet. Tech., Vol. 37,
No. 1, pp.12–17.
Behounek, M. (2003) ‘Five elements for successful knowledge management’, J. Can. Pet. Tech.,
Vol. 42, No. 4, pp.12–14.
Bondar, V. and Blasingame, T. (2002) ‘Analysis and interpretation of water-oil-ratio performance’,
Paper SPE 77569 presented at the SPE Annual Technical Conference and Exhibition,
San Antonio, Texas, 29 September–2 October.
40 E. Artun et al.

Can, B. and Kabir, C.S. (2012) ‘Simple tools for forecasting reservoir performance’, Paper SPE
156956 presented at the SPE Annual Technical Conference and Exhibition, San Antonio,
Texas, 8–10 October.
Chapman, L. and Thompson, R. (1989) ‘Waterflood surveillance in the Kuparuk River Unit with
computerized pattern analysis’, Paper SPE 17429, J. Pet. Tech., Vol. 41, No. 3, pp.277–282.
Craig, F. (1973) ‘Engineering waterfloods for improved oil recovery’, Paper SPE 4425, SPE
Distinguished Lecture.
Crawford, P. (1960) ‘Factors affecting waterflood pattern performance and selection’, Paper SPE
1503, J. Pet. Tech., Vol. 12, No. 12, pp.11–15.
Currier, J. and Sindelar, S. (1990) ‘Performance analysis in an immature waterflood: the Kuparuk
River Field’, Paper SPE 20775 presented at the SPE Annual Conference and Exhibition,
New Orleans, Louisiana, 23–26 September.
Di Julio, S. (1993) ‘North Vickers East waterflood performance prediction with fractal
geostatistics’, SPE Form. Eval., Vol. 8, No. 2, pp.89–95.
Ershaghi, I. and Omorigie, O. (1978) ‘A method for extrapolation of cut vs. recovery curves’, Paper
SPE 6977, J. Pet. Tech., Vol. 30, No. 2, pp.203–204.
Fedenczuk, L., Hoffmann, K. and Fedenczuk, T. (2002) ‘Predicting waterflood responses with
decision trees’, Paper PETSOC-2002-028 presented at the Canadian International Petroleum
Conference of the Petroleum Society, Calgary, Alberta, 11–13 June.
Hiatt, W.N. (1968) ‘Simplified performance calculation for pattern waterfloods in stratified
reservoirs’, Paper SPE 2007 presented at the SPE AIME 39th California Fall Meeting,
Bakersfield, California, 7–8 November.
Higgins, R.V. and Leighton, A.J. (1963) ‘Waterflood prediction of partially depleted reservoirs’,
Paper SPE 757 presented at the SPE California Regional Meeting, Santa Barbara, California,
24–25 October.
Higgins, R.V., Boley, D.W. and Leighton, A.J. (1966) ‘Prediction of waterfloods by graphs’, Paper
SPE 1456 presented at the SPE AIME Rocky Mountain Regional Meeting, Denver, Colorado,
23–24 May.
Khan, A.M. (1971) ‘An empirical approach to waterflood predictions’, Paper SPE 2931,
J. Pet. Tech., Vol. 23, No. 5, pp.565–573.
Mohaghegh, S., Ameri, S., Aminian, K. and Chatterjee, U. (1993) ‘Performance evaluation of
waterflood project in Southern West Virginia’, Paper SPE 26937 presented at the SPE Eastern
Regional Conference and Exhibition, Pittsburgh, Pennsylvania, 2–4 November.
Nikravesh, M., Kovscek, A.R., Murer, A.S. and Patzek, T.W. (1997) ‘Neural networks for field-
wise waterflood management in low permeability, fractured oil reservoirs’, Paper SPE 35721
presented at the SPE Western Regional Meeting, Anchorage, Alaska, 22–24 May.
Popa, A., Sivakumar, K. and Cassidy, S. (2012) ‘Associative data modeling and ant colony
optimization approach for waterflood analysis’, Paper SPE 154302 presented at the
SPE Western Regional Meeting, Bakersfield, California, 19–23 March.
Sayarpour, M., Zuluaga, E., Kabir, C.S. and Lake, L.W. (2009) ‘The use of capacitance-resistance
models for rapid estimation of waterflood performance and optimization’, J. Pet. Sci. Eng.,
Vol. 69, No. 3, pp.227–238.
Thiele, M.R. and Batycky, R.P. (2006) ‘Using streamline-derived injection efficiencies for
improved waterflood management’, Paper SPE 84080, SPE Res. Eval. Eng., Vol. 9, No. 2,
pp.187–196.
Thomas, J. and Driscoll, V. (1973) ‘A modeling approach for optimizing waterflood performance,
slaughter field chickenwire pattern’, Paper SPE 4070, J. Pet. Tech., Vol. 25, No. 7,
pp.757–763.
Tompang, R. and Kelkar, B. (1988) ‘Prediction of water flood performance in stratified reservoirs’,
Paper SPE 17289 presented at the SPE Permian Basin Oil and Gas Recovery Conference,
Midland, Texas, 10–11 March.

View publication stats

Potrebbero piacerti anche