ARMA 13-177

Hydraulic Fracturing Design OptimizationBakken Case Study

Hadi Jabbari and Steven A. Benson
Petroleum Engineering Department, University of North Dakota, Grand Forks, ND 58202 USA

Copyright 2013 ARMA, American Rock Mechanics Association

This paper was prepared for presentation at the 47th US Rock Mechanics / Geomechanics Symposium held in San Francisco, CA, USA,
23-26 June 2013.

This paper was selected for presentation at the symposium by an ARMA Technical Program Committee based on a technical and critical
review of the paper by a minimum of two technical reviewers. The material, as presented, does not necessarily reflect any position of
ARMA, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper for commercial purposes without
the written consent of ARMA is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 200 words;
illustrations may not be copied. The abstract must contain conspicuous acknowledgement of where and by whom the paper was
presented.

ABSTRACT: Low permeability shale formations, such as the Bakken, require a large fracture network to enhance well
productivity. In this research, through a comprehensive fracture-simulation/reservoir-performance study, we evaluated the
main parameters controlling the fracture stimulation in horizontal wells. The main goal was to investigate opportunities to
optimize hydraulic fracturing and production of horizontal Williston Basin Bakken Formation wells. The project area used in
the investigation was located in Williams County, North Dakota. To design a successful hydraulic fracture treatment, three
tasks were conducted: First, a reservoir simulation to evaluate the response of the reservoir to fracture stimulation and to
calibrate the reservoir model was performed using two steps involving: a) sensitivity analysis (SA) to determine the
significant well/reservoir properties and parameters and b) history matching (HM) simulation results to the production data
from the stimulated horizontal well in the study area. Second, the amount of fracturing materials was estimated and
preliminary pump schedules were developed based on selected design parameters that include: half-length, pump rate, and
maximum proppant concentration. Design parameters screening was conducted using 2D fracture geometry solutions for
fracture treatment parameters. An optimization task was performed to identify optimal stimulation treatment(s) that together
with optimal operating conditions would return a maximum value for the objective function (e.g. Net Present Value). Finally,
fully-3D hydraulic fracture modeling was utilized to perform implicit, coupled, finite difference/finite element solutions to
basic conservation equations. The pump schedule obtained from the scoping design was changed in terms of the pad
volume and proppant schedule for treatment optimization. The overall goal of such a schedule refinement was to place the
right amount of proppant in the right place along the fracture.
Past studies by Breit et al. [1] used reservoir
1. INTRODUCTION simulation to compare multi-well to single-well
The Bakken reservoir of Williston Basin is composed completions in the Bakken Formation. In their
of an upper shale member, a middle member of modeling, they considered a homogeneous layer with
sandstone and siltstone, and a lower shale member. dual porosity and with an anisotropic permeability
The first development of the Bakken Formation ratio of 4 to 1. Sadrpanah et al. [2] also used reservoir
occurred in the 1960s [1]. Early Bakken development simulation to evaluate multiple hydraulic fractures in
with vertical wells resulted in minor production. Later, horizontal wells. Using explicit simulation to
horizontal drilling in the 1980s improved production as implement the hydraulic fractures into the horizontal
compared to previous vertical well completions. wells, they predicted the fracture behavior and post-
Activity in developing the Bakken Formation has frac production profiles to make optimal designs for
increased due to the success of horizontal drilling maximum production enhancement. Lentz et al. [3]
coupled with multi-stage hydraulic fracturing described the benefits of re-fracture treatment in
stimulation. However, production forecast which horizontal wells in the Bakken Formation. They
may be acquired by fracture modeling and reservoir concluded that more perforations and diversion
simulation is quite challenging for these types of techniques would be attributed to the success of the
wells. treatments.
In recent years, there have been several studies on the Besler et al. [4] studied the stimulation and operation
simulation of horizontal wells in the Bakken of horizontal completions in the Middle Bakken
Formation [1, 3, and 4]. The main goal of the Bakken Formation of North Dakota and Montana. They
reservoir simulation past efforts has been to simulate compared the production histories of the fractured
the production performance of the wells and to horizontal wells to offset wells completed with other
develop optimum scenarios for further field techniques to evaluate best industry practices. Cox et
development. al. [5] investigated the production performance of the
Bakken wells, and by using reservoir simulation and

1
pressure transient test analysis, evaluated optimal
economics in the early phase of the Bakken
development. A similar study, with an emphasis on
economics, was conducted on the Marcellus shale by
Schewitzer et al. [6] where several hydraulic fracture
treatments were evaluated with the goal of determining
the most efficient wellbore/fracture configurations.
The economic study for a hydraulic fracturing project
was performed using the average present-day
completion costs and projected sale price of gas used
in the calculation process. Shanqiang et al. [7], on the
other hand, studied more complex fluid flow physics
and stimulation practices in making long-term
production forecasts for unconventional reservoirs.
They proposed a probabilistic reservoir simulation
workflow to provide realistic range of production
forecast with application in the Bakken Formation.
The work presented in this paper provides the results
of an integrated fracture/reservoir simulation together
with a financial analysis to evaluate the production and
economics of the Bakken horizontal wells stimulated
with a variety of fracturing strategies. The main goal
of this research was to investigate the opportunities for
optimizing hydraulic fracture treatments in the Bakken
horizontal wells.
2. RESERVOIR MODEL DESCRIPTION
BAKKEN CASE STUDY
In an effort to adjust the well/reservoir properties and
parameters, history matching of the reservoir model
was performed for the study area to yield the observed
production data. Reservoir simulation was used to
perform three major tasks to provide insights into the
design of hydraulic fracture stimulation. These tasks
include: a) sensitivity analysis (SA), which is used to
determine how sensitive an objective function (an
expression or single quantity) is to different parameters
and their ranges in values, b) history matching (HM)
for calibrating the reservoir parameters conducted by
an automated algorithm to simulation results, and c) an
optimization (OP) stage for developing best case
scenarios for future hydraulic fracture treatments in the
Bakken Formation.
The location of the study area is in section 36-T156N-
R95W in eastern Williams County, North Dakota, on
the eastern flank of the Nesson Anticline (Fig.1). The
matrix porosity values used in the reservoir simulation
were 0.05 (Upper Bakken), 0.06 (Middle Bakken), and
0.05 (Lower Bakken). The matrix permeability values
were 0.00012 md (Upper Bakken), 0.003 md (Middle
Bakken), and 0.00015 (Lower Bakken) [8]. The
observed data of Well X was used for the history
matching process- adjusting the reservoir simulation
parameters that could best represent the actual
reservoir. The cumulative production data for the study
well are shown in Fig.2.
Figure 3 shows the grid setup for the simulation model
where a horizontal well with 3600 ft lateral length is
2
completed in the Middle Bakken member and
stimulated with multi-stage transverse fractures.
2.1. Sensitivity Analysis of Reservoir Properties
and Parameters
A sensitivity analysis was performed to ascertain how
a given model output varies based on the input
parameters. This is an important method for checking
the quality of a given model as well as a powerful tool
for checking the robustness and reliability of the
analyses [9]. The sensitivity analysis helps reservoir
modelers achieve a better understanding of how
different parameters influence their results. The
information from such analysis can later be used in
other tasks, such as history matching, optimization,
and uncertainty assessment as it helps determine which
parameter(s) to vary and to what degree.
2.2. Sampling Method
For a given set of parameters and values, the parameter
space is usually very large, and it would be too
perplexing to select a reliable design (the set of job
patterns). According to the theory of experimental
design (Design of Experiments, or DOE), an efficient
design should have two characteristics to be acceptable
that include [10, 11, 12]: a) the input parameters
should be approximately orthogonal. This means that
in a design matrix, the correlation between the vectors
of any pair of columns should be either zero or very
small, and b) the sampling point (job patterns) should
be evenly distributed in the parameter space. Or, in
other words, the design matrix should represent all
possible job patterns (space-filling). This means that
no two points are close to each other.
One common method used to select a job pattern is by
random design, but its main disadvantage is that the
interpretation of the results is not accurate due to
random confounding and the estimated coefficients
can be biased [10, 12]. To avoid the random design
problem, we have used the Latin hypercube sampling
(LHS) method proposed by Mckay et al. [11], in that
the input variables are considered to be random
variables with known distribution functions. The
parameters whose changes were examined in the
sensitivity analysis are listed in Table 1. Each
parameter is given a range of values over which it can
vary.
The results from 2,300 simulation runs were collected
and analyzed to screen for the parameters (factors)
which appeared to have significant effects on the
response (objective function). The statistical
significance of each effect and interaction was
examined by comparing its signal-to-noise t-ratio, t E
(or tn 1 ), to the critical t-value, denoted by t* / 2 , which
is the inverse of the two-tailed t-distribution. When the
populations standard deviation ( ) is unknown (i.e.
small samples, or small n ), we can replace it by an data, Rwm, p is the measured maximum change for well
estimate, s p (pooled standard deviation), then the
w and production data p , and Ewm, p is measurement
quantity in Eq.1 follows the t-distribution with n 1
error.
degrees of freedom (for n runs) [12].
The global objective function (Eq. 2) was used for
Y o
tn 1, / 2 (1)
history matching only the cumulative oil production
sp n (Cum_OIL) and cumulative water production
(Cum_WTR) data of Well X. The cumulative gas
Note that the significance level ( ), which is left to production data was found to be unreliable. By forcing
the investigator, was considered to be 5% as the well to produce at historic oil rates the reservoir
reasonable accuracy was required. Finally, we used a model was matched to the production history. The
Tornado chart of the effects to determine the final history matching plot in Fig.6 shows the
magnitude and the importance of the effects on the convergence of the objective function after 1467 runs.
cumulative oil and cumulative water production. The
charts in Figs.4 and 5 display the effects. The The relative permeability curves were developed using
significant parameters were obtained from the the Corey correlation which is power-law in the water
sensitivity analysis. Note that the absolute values of saturation [14-16] (Eqs.3-6). Note that since there was
their t-ratios (for Cum_OIL and Cum_WTR) were no experimental data available for the relative
found to be greater than the critical t-value (obtained permeability curves, the end-point saturations were
from t-distribution for the specified degree of estimated from the history matching process.
freedom). nw
S w Swcrit
krw krwiro .(3)
2.3. History Matching 1 S wcrit Soirw
now
History matching is a well-examined method of tuning So Sorw
reservoir characteristics (properties) used in the model krow krocw (4)
to match historical field data. This process is 1 S wcon Sorw
accomplished through an iterative trial-and-error nog
process. The sensitivity analysis described above was Sl Sorg S gcon
krog krogcg (5)
used to identify the significant parameters, shown in 1 S
wcon Sorg S gcon
Table 1. These parameters were then adjusted, over
realistic ranges, in order to achieve a close match ng
between the simulation results and the field data, while S g S gcrit
krg krgcl ...(6)
insignificant parameters were set to constant values 1 S S S
based on available data. gcrit oirg wcon
During the history matching process a global objective Also, the linear relative permeability (X-curves) was
function was used to measure the relative difference considered for the grids containing the hydraulic
between historical data and simulation results. In such fractures. The parameters estimated from the history
a function the well variables are accounted for by matching process are listed in Table 1. As shown in
means of a root-mean-squared error method (RMSE). Figs. 7, 8, and 9, reasonable matches were obtained for
The goal of the history matching is to obtain small OIL_PROD, Cum_OIL and Cum_WTR production
values of objective functions to correspond with small trends.
differences between historical data and simulation The reservoir simulation is followed by hydraulic
results. The global objective function is given by Yang fracture modeling (2D/3D) to simulate the fracturing
et al. [13]: process and evaluate the different stimulation
T ( w, p )
strategies for the study well in the Bakken Formation.
( Rw, p , t Rw, p , t )
s m 2

t 1

1 N ( w) T ( w, p ) (2) 3. FRACTURE TREATMENT DESIGN

Eg Ww, p
Rw, p
N ( w) m m
p 1 4 Ew , p
Ww, p 3.1. Optimal Treatment Materials
p 1
Hydraulic fracture stimulation is required for
Where, subscripts w , p , and t are well, production economic development of low permeability reservoirs,
data, and time, respectively, N ( w) is the total such as the Bakken Formation. This is because a
highly conductive fracture results in a negative skin.
production data from well 1 to w , Ww, p is weight, At the same time, there is no single fracture treatment
T ( w, p) is the total of time step, Rws , p ,t represents design that best fits all possible cases. The amount of
knowledge about the treatment environment shapes the
simulation results while Rwm, p ,t is measured historical design process to a very large degree. When the area to

3
be hydraulically fractured is new, there are generally a
large number of potential uncertainties that may have
effects on the production responses, such as the static
and dynamic parameters.
We have presented hydraulic fracturing designs for a
Bakken horizontal well with 10000-ft lateral drilled
inside a 1280 acre drainage area. The goal of the
design was normally to consider all plausible
combinations of fluids and proppants to find the best
candidates of treatment materials. Improper fracturing
design can result in fractures that are too narrow that
may cause proppant bridging and screenout or too
wide that can allow too much proppant settling. In this
study we have done a comparative study of proppants,
pads, and fracturing fluids to determine the best
combination(s) for carrying out the stimulation job in
the Bakken horizontal wells.
The first step was to use a 2D fracture simulation
method (PKN model in this case) to quickly perform
the sensitivity analysis, preliminary designs, and
calculating the fracturing material sizes prior to
performing fully-3D fracture modeling. Table 2 shows
the results of the scoping calculations that were carried
out for 27 combinations of the candidate fracturing
materials. Note that the average fracture height used in
the 2D modeling was estimated from the microseismic
mapping data of the well, being 120 ft [8].
A financial analysis was used to evaluate the costs and
benefits associated with each hydraulic fracturing
design. The expense considered for fracturing job
economic analysis and gross treatment sizing were
restricted to four major components: a) fluid cost, b)
proppant cost, c) pumping charges, and d) gross
equipment charges. The treatment costs per fracture
stage, the sum of the 4 major components, are shown
in Table 2. Other parameters used in the economic
analysis are shown below in Tables 3 and 4. All cost
values used are general averages based on personal
communications with various operators.
An optimization task was conducted to identify the
optimal stimulation plan (fracture treatment design)
and operating conditions (flowing bottomhole
pressure) that would yield a maximum value for the
objective function, being net present value (NPV) in
this case. In the optimization study the history matched
reservoir model was used to forecast the reservoir
response based on different hydraulic fracturing
strategies. The main goal was to determine the best
fracture treatment(s) that would yield the highest profit
from the well stimulation plan(s). The well was
operated at bottomhole pressures of 1000, 1500, and
2000 psi during the 5-year period. The NPV is made
up of three terms: a) the value of the oil produced
being $96/STB, b) the cost for water
disposal/processing being $3/STB, and c) the capital
expenses (CapEx) (Table 3). The annual interest rate
for calculating the discounted NPV used in this study
was considered as 6%.
4
The method of Latin hypercube plus proxy
optimization was used to find the optimal treatment
scenario(s). This method consists of four main steps: a)
constructing combinations of the input parameter
values to obtain the maximum information from the
minimum number of simulation runs (Latin hypercube
design), b) building an empirical proxy model by using
the data from Latin hypercube design runs (proxy
model), c) conducting a proxy-based optimization, and
d) validation of the optimal solution obtained from the
proxy by iterative simulation runs. The algorithm is
shown in Fig.10.
Figure 11 illustrates the results of such NPV
calculations using the Latin hypercube algorithm. The
iterative procedure, over 324 runs, was able to
optimize the NPV from 27 million to almost 84 million
USD. The NPV was calculated for a standard
horizontal well completed with 12, 18, and 36 fracture
stages. It is clear from this figure that the job-IDs 39,
106, 162, 198, 280, and 318 with the specifications
shown in Table 5, outperformed the other treatment
cases.
Using crosslinked gel plus either resin-coated sand or
ceramic as the fracture treatment materials appeared to
be the best choices due to the higher cumulative oil
produced in 5 years as compared to all the other
combinations of treatment materials. One of the
notable successes with this design strategy would be
the elimination of proppant flowback. Many operators
in the Bakken reported tremendous problems with
production of sand, requiring frequent pump changes
as the pumps were damaged by crushed sand. A
review of wells completed by three operators using
high-strength proppants indicated zero pump failures
due to proppant sand flowback [4].
3.2. Hydraulic Fracture Simulation
The optimal fracture treatment materials were
determined by integrating fracture modeling (2D, PKN
model) with reservoir simulation. As a next step, 3D
fracture simulation was performed to more accurately
model the created fracture in order to find an
optimized pump schedule. Figure 12 shows a log style
illustration of the major data that was input into the
fracture model as compared to depth in the study area.
The geologic model was comprised of five layers with
differing stresses, permeabilities, and moduli. They
were: Lodgepole, Upper Bakken, Middle Bakken,
Lower Bakken, and Three Forks.
Since the estimation of fracture geometry plays a
major role in evaluating the completion plans, a
commercial hydraulic fracture simulator was used to
estimate the created fracture geometry and its
characteristics. To evaluate the success of a fracture
treatment in a horizontal well, the production response
of the well due to the hydraulic fractures should be
investigated. First, we need to know the fracture
geometry and fracture conductivity. This information
comes from fracture geometry models that range from
simple hand calculation procedures (2D models) to
complex 3D models (pseudo-3D or fully 3D).
Fluid injected at the beginning of the fracturing job
(pad) initiates and opens up the fracture. Pad provides
the necessary extra fluid that is leaked off into the
formation during a treatment. The pad also generates
sufficient fracture length and width to place the
proppant. If the pad volume is too small, the treatment
may screen out. If the pad volume is too large, we will
waste money, the fracture height will grow into
unwanted zones (i.e. the Lodgepole), and the fracture
will not close as rapidly as it would with a smaller pad
volume. After pumping the specified volume of pad,
the proppant concentration is ramped up step-by-step
as the slurry is being injected. If proppant prematurely
bridges in the fracture during pumping, the treating
pressure will rise rapidly to the technical constraint, a
situation called screen-out.
For calculating the proppant ramp schedule, a material
balance method was used, such as the power-law
method of Nolte [17]. The power-law method is
explained thoroughly by Economides et al. [18] (Table
6). In an ideal pump schedule, the proppant schedule
should be designed such that a uniform proppant
concentration is obtained in the fracturewith the
value of concentration equal to the maximum proppant
concentration ( ce , which is a design parameter) and
a minimum pad volume is used to achieve the desired
fracture half-length ( X f _ Design ).
In the end, the pump schedule designed for the
selected treatment scenario (optimal treatment case)
was developed as shown in Fig.13 with the details
presented in Table 7. Obviously, a pump schedule may
vary based on the fracturing treatment materials,
and/or based on the desired fracture half-length and
fracture conductivity. Additionally, Tables 8 and 9
present the characteristics of the fluid and proppant
used for the treatment.
Using fully-3D modeling with the developed pump
schedule, the fracture dimensions and fracture
conductivities were calculated. These parameters were
then evaluated to ensure that the fracture would not
grow upwards into the Lodgepole Formation. Several
operators have reported difficulty keeping fractures
contained within the target Bakken horizon, and have
observed radioactive tracer indicating vertical height
growth into the shallower Lodgepole Formation. In
some cases, high H2S or elevated water-cut indicated
fractures had propagated into the Lodgepole Formation
[4]. Table 10 contains the details of the treatment
design and the results from the fully-3D fracture
simulations.
Note that in the 3D modeling we have used a finite
element method (FEM) for calculating fracture width
and fracture propagation. Generally, for cases where
formation layers have differing values for modulus,
5
FEM needs to be used for fracture width/propagation
calculations. FEM should be used in any case where
modulus values in different layers differ by a factor of
2 or more [19, 20].
Furthermore, three main constraints were considered in
the design of the hydraulic fracture treatments: a) a
limit of 1000 psi was considered for the net-pressure
that can influence the surface treating pressure
(#HHP), and can have an effect on the fracture height
growth (unwanted fracture growth into the Lodgepole),
b) a fracture width limit that should be at least 3 times
the proppant diameter in order to prevent the proppant
bridging and involuntary screenout [19], and c) the
injection time must be less than 24 hours. As is clear
from the calculations in Table 10, these constraints are
validated in this case.
4. CONCLUSIONS
A study was conducted on hydraulic fracturing designs
for a Bakken horizontal well. First, in an effort to
adjust the reservoir properties, history matching and
reservoir simulation were performed for the study
case. Next, the calibrated reservoir model was used for
fracture treatment optimization and material selections.
Using the results from reservoir simulation and
hydraulic fracture modeling, NPV cash flow models
were developed for various fracture treatment cases. A
comparison of the oil production and the NPV of
several strategies for well stimulation was the
objective of the simulations and modeling. Finally,
hydraulic fracture modeling was performed to
accurately model the created fracture so as to find an
appropriate pump schedule.
The main goal of this research was to improve
hydraulic fracturing design by improving proppant
coverage along the fracture, optimizing treatment
volumes and pump rates, understanding fracture
geometry, and optimizing proppant selection. The
design of a successful fracturing job in a low
permeability formation (e.g. the Bakken) requires a
treatment that yields a long fracture. For this reason,
no improvement in the NPV values was observed in
cases with low fracture half-lengths (Fig.11). The
results from this study suggest that for a Bakken well
with 10,000-ft lateral, a fracturing strategy utilizing a
high fracture half-length, with a high number of
fractures (36 or more) would provide the most efficient
balance between the operating charges, fracture
treatment costs, drilling expenses, and the benefit
earned from the incremental oil production.
The effective fracture width in an optimal case where
x f 1000 ft is shown in Fig.14. As is clear from this
figure, the fracture height is nearly equal to the
thickness of the pay zone. Generally, it is believed that
in-situ stress differences (in the vertical stress profile)
are the major controlling factor of fracture height
confinement [21]. Figure 15 depicts the fracture tip
position at successive stages of fracture propagation.
Figure 16 depicts the proppant coverage versus natural fracture density in y-
DJ_NatFrac :
fracture penetration. Other results from the fully- direction (used in DualPOR model)
3D/FEM fracture simulations are shown in Figs.17-20. DK_NatFrac : natural fracture density in z-direction
HHP : hydraulic horsepower, hp
The results from this research also imply that using
ratio of vertical to horizontal
crosslinked gel plus either resin-coated sand or KvKhRatio :
permeability of matrix
ceramic as fracturing materials would lead to optimal
krocw : kro at connate water
stimulation in the Bakken horizontal wells. The major
concerns about the use of sand as proppant may krwiro : krw at irreducible oil
include: a) lost conductivity caused by inadequate krogcg : krog at connate gas
proppant strength due to increased closure stress, as krgcl : krg at connate liquid
severe crushing of frac sand from aggressive mp : Proppant mass, lb
drawdown or reservoir depletion would dramatically
decrease well productivity, and b) lost conductivity n : run number
due to proppant flowback, caused by overflushing the nw : exponent for calculating krw
treatment [3, 22, and 23]. Overdisplacing proppant now : exponent for calculating krow
carried in a crosslinked fluid may seriously nog : exponent for calculating krog
compromise the wellbore to fracture conductivity. ng : exponent for calculating krg
With ceramic, no proppant fowback has been observed
in the Bakken Formation. Overall, higher production matrix permeability modifier (used
PermMtrxMultplier :
in HM)
performance has been observed from high quality
ceramic wells in the Bakken Formation, suggesting PermNatFrac : natural fracture permeability
higher ultimate recovery with better proppants [4]. matrix porosity modifier (used in
PorMtrxMultplier :
HM)
Note also that in this research the basic theory of fluid PorFracMultplier :
natural fracture porosity modifier
loss from a hydraulic fracture has been employed (used in HM)
based on 1-D fluid loss (also called Carter fluid loss sp : pooled estimate of standard
deviation of individual responses
[19]), which is valid for matrix fluid loss as long as the
lateral propagation (height and length) is rapid Sw : water saturation
compared to the rate of fluid leakoff normal to the Swcrit : critical water saturation
fracture. If the hydraulic fracture intersects with Soirw : irreducible oil for water-oil table
existing natural fractures, then these natural fissuers Sorw : residual oil for water-oil table
may begin to open under the high pressure associated Swcon : connate water saturation
with the injected fluid. This can dramatically increase
Sorw : residual oil for water-oil table
the rate of fluid loss [19]. The natural fractures will
begin to open when the net pressure inside the fracture Sl : liquid saturation
begins to exceed a critical value (PnetCrit.). As the net Sorg : residual oil for gas-liquid table
pressure rises above this critical level, the rate of fluid S gcon : connate gas saturation
loss then becomes proportional to the product of Swtr : connate water saturation
natural fracture density and (Pnet - PnetCrit.) cubed. The
te : time at end of pumping, min
critical net pressure can be obtained by analyzing the
Nolte-Smith log-log plot of net treating pressure versus ti : time of injection, min
time [24]. Since this type of data for the study well tE : signal-to-noise t-ratio
was not available, the extra fluid loss due to natural
t* : critical t-ratio
fractures was considered by increasing the leak-off
coefficient to some degree. Vi volume of injection into one wing,
: cu ft
Vf : volume of fracturing fluid
ACKNOWLEDGEMENTS w : fracture width, in

The authors would like to thank the Institute for Energy Xf : fracture half-length, ft
Studies and the Department of Petroleum Engineering at the
University of North Dakota. We also appreciate NSI
Y : sample average
: mean
Technology Inc. and CMG for their software supports.
: proppant schedule exponent,
dim.less
NOMENCLATURE e : fluid efficiency at end of pumping,
dim.less

CPor_Mtrx : matrix compressibility

CPor_NatFrac : natural fracture compressibility REFERENCES
DI_NatFrac :
natural fracture density in x- 1. Breit, V.S., Jr., D.H. Stright, and J.A. Dozzo. 1992.
direction (used in DualPOR model) Reservoir Characterization of the Bakken Shale from

6
 Modeling of Horizontal Well Production Interference
Data. Paper SPE 24320 presented at the SPE Rocky
Mountain Regional Meeting, Casper, Wyoming, May
18-21.
2. Sadrpanah, H., T. Charles, and J. Fulton. 2006.
Explicit Simulation of Multiple Hydraulic Fractures
in Horizontal Wells. Paper SPE 99575 presented at
the Europe/EAGE Annual Conference and
Exhibition, Vienna, Austria, June 12-15.
3. Lantz, T., D. Greene, M. Eberhard, S. Norrid, and R.
Pershall. 2007. Refracture Treatments Proving
Successful In Horizontal Bakken Wells. Rocky
Mountain Oil & Gas Tech. Symp., Denver, Colorado,
USA, April 16-18.
4. Besler, M. R., J.W. Steele, and T. Egan. 2007.
Improving Well Productivity and Profitability in the
Bakken A Summary of Our Experiences Drilling,
Stimulating, and Operating Horizontal Wells. Paper
SPE 110679 presented at the SPE Annual Technical
Conference and Exhibition, Anaheim, California,
USA, November 11-14.
5. Cox, X.A., D.M. Cook, K. Dunek, R. Daniels, C.
Jump, and B. Barree. 2008. Unconventional Resource
Play Evaluation: A Look at the Bakken Shale Play of
North Dakota. Paper SPE 114171 presented at the
SPE Unconventional Reservoirs Conference,
Keystone, Colorado, USA, February 10-12.
6. Schweitzer, R. and H.I. Bilgesu. 2009. The Role of
Economics on Well and Fracture Design Completions
of Marcellus Shale Wells. Paper SPE 125975
presented at the SPE Eastern Regional Meeting,
Charleston, West Virginia, USA, September 23-25.
7. Shanqiang, L., M. Wolff, J. Ciosek, F. Rasdi, L. Neal,
P. Arulampalam, and S. Willis. 2011. Probabilistic
Reservoir Simulation Workflow for Unconventional
Resource Play: Bakken Case Study. Paper SPE
142896 presented at the SPE EUROPEC/EAGE
Annual Conference and Exhibition, Vienna, Austria,
May 23-26.
8. BRC report (Bakken Research Consortium). 2008.
Hydraulic fracturing and microseismic monitoring
project, a report submitted by Headington Oil Co. and
XTO Energy Inc. on behalf of the Bakken Research
Consortium: Brigham Exploration, Continental
Reseources, Encore Operating, Hess Corporation,
Petro-Hunt LLC, Whiting Petroleum, Schlumberger
Oilfield Services, MicroSeismic Inc., TerraScience,
DOE National Energy Technology Laboratory, and
Lawrence Berkeley National Laboratory.
9. Saltelli, A., S. Tarantola, F. Campolongo, and M.
Ratto. 2004. Sensitivity analysis practice; A guide to
assessing scientific models. John Wiley & Sons Ltd.,
England.
10. Cioppa, T.M. 2002. Efficient nearly orthogonal and
space-filling experimental designs for high-
dimensional complex models. Naval Postgraduate
School PhD Dissertation, Sep.
11. Mckay, M.D., R.J. Beckman, and W.J. Conover.
1979. A comparison of three methods for selecting
7
values of input variables in the analysis of output
from a computer code, technometrics.
12. Lawson, J. and J. Erjavec. 2001. Modern statistics for
engineering and quality improvements. DUXBURY,
USA.
13. Yang, C., L. Nghiem, , C. Card, and M. Bremeier.
2007. Reservoir model uncertainty quantification
through computer-assisted history matching: SPE
109825 Annual Technical Conference and
Exhibition, Anaheim, California, November 1114.
14. Brooks, R.H. and A.T. Corey. 1964. Hydraulic
properties of porous media. Hydraulic paper No.3,
Colorado State University.
15. Brooks, R.H. and A.T. Corey. 1966. Properties of
porous media affecting fluid flow, J. of the Irrigation
and Drainage Division, Proc. Of ASCE, 92, No.IR2,
61-88.
16. Honarpour, M., L. Koederitz, and A.H. Harvey. 1986.
Relative permeability of petroleum reservoirs. Boca
Raton, Fla. Publishing, C.R.C. Press, 1986.
17. Nolte, K.G. 1986. Determination of proppant and
fluid schedules from fracturing-pressure decline. SPE
Production Engineering, Vol.1, NO.4 (SPE 13278-
PA).
18. Economides, M., Oligney, R. and P. Volka. 2002.
Unified fracture design-- bridging the gap between
theory and practice. Orsa Press, TX. USA.
19. Gidley, J.L., S.A. Holditch, D.E. Nierode, and R.W.
Jr. Veatch. 1990. Recent advances in hydraulic
fracturing. SPE Monograph, Vol. 12, SPE,
Richardson, TX, USA.
20. Perez, N. 2004. Fracture Mechanics. Kluwer
Academic Publishers. Boston.
21. Warpinski, N.R., R.A. Schmidt, and D.A. Northrop.
1982. Insitu stresses: the predominant influence on
hydraulic fracture containment. J. of Pet. Tech., 653-
664, March.
22. Phillips, Z.D., R.J. Halverson, S.R. Strauss, J.M.
Layman, and T.W. Green. 2007. A case study in the
Bakken Formation: changes to hydraulic fracture
stimulation treatments result in improved oil
production and reduced treatment costs, paper SPE
108045 presented at the 2007 Rocky Mountain Oil
and Gas Symposium, Denver, April 16-18.
23. Wiley, C., B. Barree, M. Eberhard, and Tom Lantz.
2004. Improved horizontal well stimulations in the
Bakken Formation, Williston Basin, Montana. SPE
Annual Technical Conference and Exhibition, 26-29
September 2004, Houston, Texas.
24. Kim, G.H. and J.Y Wang. 2011. Interpretation of
hydraulic fracturing pressure in low-permeability gas
formations. SPE 141525, presented at the SPE
Production and Operations Symposium, 27-29 March
2011, Oklahoma City, Oklahoma, USA.
 Fig.1 Project area location in eastern Williams County, ND (Courtesy of BRC 1 [8])

Fig.2 Cum-production dataWell X

Fig.3 Reservoir simulation grid setup grid top

1
Bakken Research Consortium

8
Fig.4 Tornado plot of Cum_OIL (linear model t-ratios)

9
 Fig.5 Tornado plot of Cum_WTR (linear model t-ratios)

Fig.6 Final history-matching iteration of 1467 jobs showing the convergence of

the objective function (red dots represent the cases with the lowest error values)

10
 Fig.7 Oil production rate history match

Fig.8 Cumulative oil production history match-- oil production rate constraint

Fig.9 Cumulative water production history match-- oil production rate constraint

11
Fig.10 The algorithm of Latin hypercube plus Proxy Optimization (courtesy of Computer Modeling Group)

Fig.11 NPV calculations for 324 different scenarios

Fig.12 Input parameters for fracture geometry

modeling in Well X

12
 Fig.13 Developed pump schedule using Nolte method

41X-36H

Stress (psi) Max Width 0.05 in

Propped Width
Profile at Closure
LdgPol

TVD
ft
10300
U-Bkn

10350
M-Bkn

10400
L-Bkn

10450

7000 7500 8000 8500 -0.028-0.021-0.014-0.007-0.000 0.007 0.014 0.021 0.028

Fig.14 Width profile at closure

41X-36H
Stress (psi) Vs. Time ( 5.5, 10.3, 22.3, 49.4 min)
9800
TVD
ft
10000

10200

10400

10600
TFrk

10800

7000 7500 8000 8500 200 400 600 800 1000

Fracture Penetration (ft)

Fig.15 Propagation of fracture tip

13
 41X-36H
10100 59.04 min
TVD
0.000
ft
10200 0.058

LdgPol

Proppant Coverage lb/ft^2

0.116
0.174
10300
0.232
M-Bkn 0.290
10400 0.348
0.406
10500 0.464
0.522
TFrk

0.580
10600
0.151 m/sec

7000 7500 8000 8500 200 400 600 800 1000 1200
Stress (psi) Fracture Penetration (ft)

Fig.16 Proppant placement in the fracture at closure

41X-36H
10100 55.95 min

Width/Critical Bridging Width WIDTH/WIDTHC

TVD
0.000
ft
10200 0.300
LdgPol

0.600
0.900
10300
1.200
M-Bkn

1.500
10400 1.800
2.100
10500 2.400
2.700
TFrk

3.000
10600
0.151 m/sec

7000 7500 8000 8500 200 400 600 800 1000 1200
Stress (psi) Fracture Penetration (ft)

Fig.17 The ratio of fracture width to average proppant particle diameter at closure
41X-36H
10100 55.95 min
TVD
0.000
ft
10200 26.000
LdgPol

52.000
78.000
Temperature F

10300
104.000
M-Bkn

130.000
10400 156.000
182.000
10500 208.000
234.000
TFrk

260.000
10600
0.151 m/sec

7000 7500 8000 8500 200 400 600 800 1000 1200
Stress (psi) Fracture Penetration (ft)

Fig.18 Temperature distribution in the fracture at closure

14
 41X-36H

0.20
0.15
Average Width (in)

0.10
0.05

At Shut-in
Average Width (in)
200 400 600 800
Penetration (ft)

Fig.19 Average width profile vs. penetration at closure

41X-36H
10100 59.04 min
TVD
0.000
ft
10200 0.000
LdgPol

0.000

Cumulative Loss M-Gal

0.001
10300
0.001
M-Bkn

0.001
10400 0.001
0.001
10500 0.002
0.002
TFrk

0.002
10600
0.151 m/sec

7000 7500 8000 8500 200 400 600 800 1000 1200
Stress (psi) Fracture Penetration (ft)

Fig.20 Cumulative fluid loss at closure

15
Table 1 Parameters and properties examined in the Sensitivity Analysis & History Matching
Lower Level Upper Level History-
Parameters Base Value
(-1) (+1) Matched
PorMtrxMultplier 1 0.5 10 10
PermMtrxMultplier 1 0.5 20 10
PorFracMultplier 1 0.5 10 10
PermNatFrac 0.001 0.00001 0.01 0.001
KvKhRatio 0.1 0.01 0.3 0.1
CPor_Mtrx 1.00E-06 2.00E-06 8.00E-06 8.00E-06
CPor_NatFrac 5.00E-06 1.00E-06 1.00E-05 2.00E-06
DI_NatFrac 200 100 1000 1000
DJ_NatFrac 200 100 1000 500
DK_NatFrac 50 20 300 185
Rel. Perm. Table -- Matrix
swcon 0.1 0.05 0.2 0.1
swcrit 0.3 0.25 0.45 0.3
soirw 0.1 0.05 0.25 0.10
sorw 0.25 0.25 0.45 0.25
soirg 0.1 0.05 0.25 0.1
sorg 0.4 0.25 0.45 0.4
sgcon 0 0 0.02 0
sgcrit 0 0.01 0.06 0
krocw 1 0.3 1 1
krwiro 0.3 0.2 0.7 0.6
krgcl 1 0.4 1 0.9
krogcg 1 0.3 1 1
nw 2 1 3 1
no 3 2 5 3
nog 2 1 3 2
ng 2 1 3 2
Rel. Perm. Table Nat. Frac.
sgconf 0 0 0.01 0
sgcritf 0 0.01 0.06 0.06
sorgf 0.4 0.25 0.45 0.35
soirgf 0.1 0.05 0.25 0.15
sorwf 0.25 0.25 0.45 0.3
soirwf 0.1 0.05 0.25 0.1
swcritf 0.3 0.25 0.45 0.40
swconf 0.1 0.05 0.25 0.05
krgclf 1 0.4 1 0.90
krogcgf 1 0.3 1 0.90
krwirof 0.3 0.2 0.7 0.55
krocwf 1 0.3 1 0.9
nwf 1.5 1 2 2
nof 1.5 1 2 1.5
nogf 1.5 1 2 2
ngf 1.5 1 2 1.5
Swtr(i) 0.2 0.3 0.4 0.37
Hyd. Frac. Cond. (kfw), md-ft 22 22 500 22

16
Table 2 Comparison of different fracturing treatmentsper stage
kfw Vf_total mp Material
NO. Cases xf HHP
(md-ft) (M-Gal) (M-lbs) Cost
1 500 261 11.3 32.3 9164 $338,463
Slickwater
2 1000 260 22.6 64.4 9101 $341,038
20/40 Ottawa Sand
3 1500 297 38.2 110.3 9204 $351,700

4 500 1039 11.3 32.3 9168 $344,745

Slickwater
5 1000 1035 22.7 64.3 9104 $353,362
16/30 RC Sand
6 1500 1183 38.4 110.2 9208 $372,789

7 500 824 10.8 32.3 9130 $345,901

Slickwater
8 1000 828 21.9 64.9 9075 $357,612
20/40 Ceramic
9 1500 946 37.1 111.3 9175 $380,716

10 500 157 7.4 19.4 8814 $325,526

50# Linear HPG
11 1000 197 17.9 48.6 8925 $336,430
20/40 Ottawa Sand
12 1500 225 30.2 83.4 9003 $347,381

13 500 623 7.4 19.3 8816 $329,253

50# Linear HPG
14 1000 782 18 48.6 8928 $345,805
16/30 RC Sand
15 1500 894 30.3 83.3 9006 $363,337

16 500 498 7.2 19.5 8798 $330,156

50# Linear HPG
17 1000 626 17.4 49.1 8906 $348,895
20/40 Ceramic
18 1500 715 29.3 84.1 8981 $369,083

19 500 261 11.3 32.3 9164 $342,870

30# XLink Gel
20 1000 319 27 78.9 9339 $362,368
20/40 Ottawa Sand
21 1500 359 45.1 133.2 9459 $381,940

22 500 1039 11.3 32.3 9168 $349,152

30# XLink Gel
23 1000 1271 27.1 78.9 9344 $377,589
16/30 RC Sand
24 1500 1430 45.3 133.2 9465 $407,562

25 500 824 10.8 32.3 9130 $350,113

30# XLink Gel
26 1000 1008 26 79 9298 $381,737
20/40 Ceramic
27 1500 1134 43.4 133.4 9413 $415,491

Table 3. Cost of fracture treatment & completion.

Components Cost
Fixed Cost $400,000
Drilling & Completion
Lateral length=4500 ft $5,500,000
Lateral length=10000 ft $8,500,000
Proppants
Ottawa Sand $0.09-0.12/lb
Resin-Coated Sand $0.25-0.31/lb
Ceramic $0.32-0.39/lb
Fluid
Slickwater $0.09/gal
Linear HPG $0.32/gal
Xlinked Gel $0.48/gal
Pumping charges $36.40/HHP Includes all pump EQP

17
 Table 4. Other parameters used in the economic analysis.
Parameter Value
Oil Price, $/STB 96
Water disposal/process, $/STB -3
Monthly Operational Cost, $ -7,900

Table 5 Optimal fracture treatment cases

Bottomhole Fracture
Xf, Fracturing fluid Proppant Number
Job ID pressure, conductivity,
(ft) Type Type of stages
(psi) (md-ft)
39 1500 1008 Ceramic
106 1000 1271 RC Sand
162 1000 1008 Ceramic
1000 Xlinked gel 36
198 2000 1271 RC Sand
280 1500 1271 RC Sand
318 2000 1008 Ceramic

Table 6 Developing proppant schedule using power-law method [17]

1 e
a) Calculation of the exponent of proppant concentration curve:
1 e

b) Calculation of pad volume and pad pumping time: V pad Vi , t pad ti

t t pad
c) Mass per unit of injected slurry volume: c ce
te t pad

d) Converting the concentration from mass per slurry volume into mass added per unit volume
c
of base fluid ( c a ): ca
1 c
p

Table 7. Pumping schedule using Nolte method [14]

Conc. (PPG) Cum
Slurry Vol Fluid Vol Rate Fluid Prop Pump Time
Proppant
(M-Gal) (M-Gal) Start End (BPM) Type Type (min)
(M-lbs)
6.00 6.00 0.0 0.0 10.00 Slickwater - 0.0 14.3
0.32 0.29 3.0 3.0 10.00 Xlinked Gel 40/70 Sand 0.6 0.8
4.04 3.62 3.5 3.5 10.00 Xlinked Gel 20/40 Ceramic 3.5 9.6
9.38 8.31 4.0 4.0 10.00 Xlinked Gel 20/40 Ceramic 10.6 22.3
Total Slurry, M-Gal 19.7 Total Fluid 18.2
Total Proppant, M-lb 46.8 Avg. Conc 2.6
Total Pump Time, min 47.0 Pad % 30.4
M is a Roman for thousand.

18
Table 8. Proppant data [Courtesy of NSI Tech., 2012]

Ceramic 20/40
Specific Gravity 3.70
Damage Factor (1.0 = No Damage) 0.85
Stress, psi 0 2000 4000 8000 16000
KfW @ 2 lb/sq ft (md-ft) 7300 7100 6400 4800 2200

RC Sand 16/30
Specific Gravity 2.55
Damage Factor (1.0 = No Damage) 0.80
Stress, psi 0 2000 4000 8000 16000
KfW @ 2 lb/sq ft (md-ft) 11800 10800 8900 5500 1500

Table 9. Fracturing fluid data [Courtesy of NSI Tech., 2012]

Slickwater
Specific Gravity 1.04
Data @ Wellbore @ FormTmp 1.0 hr 2.0 hr 4.0 hr 8.0 hr
vis(cp @ 170 1/sec) 188.0 123.0 88.0 67.0 35.0 14.0
non-Newtonian n' 0.46 0.53 0.58 0.61 0.64 0.67
2
K(lb/sec/ft )x1000 0.06 0.03 0.02 0.01 0.00 0.00

30# X-linked gel

Specific Gravity 1.04
Data @ Wellbore @ FormTmp 2.0 hr 3.7 hr 5.2 hr 6.6 hr
vis(cp @ 170 1/sec) 500.0 400.9 226.0 124.8 83.2 54.1
non-Newtonian n' 0.72 0.72 0.72 0.72 0.72 0.72
K(lb/sec/ft2)x1000 0.04 0.03 0.02 0.01 0.01 0.00

Table 10 The results of fully3D/FEM fracture simulation

'Hydraulic' Length (ft) 1091.8
Half Length
Propped length (ft) 870.0
Max Net Pressure (psi) 1213.0
Final Net Pressure (psi) 755.0
Surface Pres-End of Pad (psi) 7116.0
PRESSURE:
Surface Pres-Start of Flush (psi) 5621.1
Surface Pres-End of Job (psi) 6656.9
Maximum Hydraulic orsepower 1779.0
Max Exposure to Form. Temp. (min) 45.3
TIME:
Time to Close 12.0
RATE: Fluid Loss Rate during pad (BPM) 0.17
EFFICIENCY: At end of pumping schedule 0.96
Average In Situ Conc.(lb/ft^2) 0.2
PROPPANT: Average Conductivity (md-ft) 556.3
Fcd (KfW/kXf) 213.15
HEIGHT: Max Fracture Height (ft) 150.2
WIDTH: Avg width at end of pumping (in) 0.14
Total Fluid Volume (M-Gal) 18.2
VOLUMES:
Total Proppant Volume (M-Lbs) 46.8

19