Combining Phy, Stats and Heuristicsw

REE185589 DOI: 10.
2118/185589-PA Date: 5-January-18 Stage: Page: 1 Total Pages: 20
Combining Physics, Statistics, and

Heuristics in the Decline-Curve Analysis
of Large Data Sets In Unconventional
Reservoirs
Rafael Wanderley de Holanda, Eduardo Gildin, and Peter P. Valkó, Texas A&M University
Summary
Analytical single-well models have been particularly useful in forecasting production rates and estimated ultimate recovery (EUR) for
the massive number of wells in unconventional reservoirs. In this work, a physics-based decline-curve model accounting for linear flow
and material balance in horizontal multistage-hydraulically-fractured wells is introduced. The main characteristics of pressure diffusion
in the porous media and the fact that the reservoir is a limited resource are embedded in the functional form, such that there is a transi-
tion from transient to boundary-dominated flow and the EUR is always finite. Analogously to the frequently used Arps (1945) hyper-
bolic model, the new model has only three parameters, where two of them define the decline profile and the third one is a multiplier.
This model is applied to a large data set in a work flow that incorporates heuristic knowledge into the history matching and uncer-
tainty quantification by assigning weights to rate measurements. The heuristic rules aim to lessen the effects of nonreservoir-related var-
iations in the production data (e.g., temporary shut-in caused by fracturing in a neighboring well) and emphasize the reservoir dynamics
to perform reliable predictions. However, there are additional degrees of freedom in the way these rules define the values of the weights;
therefore, a criterion is established that “calibrates” the uncertainty in the probabilistic models by adjusting the parameters in the heuris-
tic rules. Uncertainty quantification and calibration are performed using a Bayesian approach with hindcasts. This methodology is
implemented in an automated framework and applied to 992 gas wells from the Barnett Shale. A comparison with the Arps (1945)
hyperbolic model, the Duong (2011) model, and stretched exponential model for this data set shows that the new model is the most con-
servative in terms of estimated reserves.
Introduction
The so-called “shale revolution” has brought a surge in oil and natural-gas production, especially in North America. At the same time,
forecasting rates and estimating reserves in the emerging shale plays have proved increasingly difficult. It is widely accepted that more-
accurate methods of reserves estimation are necessary to increase awareness during financial forecasts, asset evaluation, and corporate
decision making. However, the industry still relies on the empirical methods of reserves estimation developed in the middle of the pre-
ceding century. These methods lack the proper validation needed to provide a high confidence in their outcomes (Lee and Sidle 2010).
Therefore, there is a need for further development of “reliable technologies” that can provide consistent, repeatable, and reasonably cer-
tain results. Among other things, “reliable technologies” should reflect the dramatic increase of openly available production data and
should be dependent on the application of the scientific method, which includes improving the understanding of the underlying physics
and incorporating it in the models (Sidle and Lee 2010, 2016).
There is a variety of applicable methods for reserves estimation: volumetric and material-balance calculations, decline-curve analy-
sis, analogs, history-matching analytical and/or numerical models, and regional, corporate, or other type curves.
In unconventional reservoirs, decline-curve analysis is probably the most-used method (Lee and Sidle 2010). The basic assumption
of this approach is that the future rates can be inferred by the extrapolation of the trend in the past production history. As reported by
Arps (1945), this practice had been conducted since the beginning of the preceding century. Arps (1945) postulated a differential equa-
tion for the rate decline with time, from which the exponential, harmonic, and hyperbolic models were derived. Even though his work
was primarily empirical, other works have shown that these functional forms can be related to fluid flow under specific circumstances.
Fetkovich (1980) observed that the exponential decline is equivalent to radial-boundary-dominated flow of a single-phase slightly com-
pressible fluid with constant well bottomhole pressure. Camacho-Velázquez (1987) and Camacho-Velázquez and Raghavan (1989)
showed that the Arps (1945) exponential and hyperbolic models can be considered as a valid approximation for boundary-dominated
flow in a solution-gas-drive reservoir. However, in unconventional reservoirs, the onset of boundary-dominated flow happens much
later in time and can be pinpointed only with huge error margins. As a consequence, the Arps decline exponent (b) is often identified as
greater than unity, violating the assumptions imposed by Arps (1945). A suggested remedy from Robertson (1988) is often applied, and
other techniques, such as the transient hyperbolic model of Fulford and Blasingame (2013), can also resolve the contradiction; however,
they result in an increase in the number of parameters to be identified (or assumed a priori).
Acknowledging some of the drawbacks of the Arps (1945) model family, numerous other empirical models have been used in the
decline-curve analysis of unconventional reservoirs. Duong (2011) proposed a model to capture the extended transient flow commonly
observed in these formations. His model can have an initial increase in the production rates, which can last up to 1 month, justified by
fracture reactivation. Duong (2011) defines the production behavior of most unconventional reservoirs as “fracture-dominated flow.”
The often-recognizable one-half slope on log-log plots is often attributed to the drainage of the matrix compartment into the fracture
network, as explained by Bello and Wattenbarger (2008). The Duong (2011) model originally has three parameters, but one of his inter-
esting suggestions is that there is a correlation between two of the parameters in a given resource play. Similarly, Valkó (2009) pro-
posed the stretched exponential model, which also is empirical and has three parameters, one of which is suspected to have a
characteristic value in a given geological setting. One important aspect of these and other recently suggested empirical models is that
Copyright V
C 2018 Society of Petroleum Engineers
This paper (SPE 185589) was accepted for presentation at the SPE Latin America and Caribbean Petroleum Engineering Conference, Buenos Aires, 17–19 May 2017, and revised for
publication. Original manuscript received for review 31 March 2017. Revised manuscript received for review 28 June 2017. Paper peer approved 14 August 2017.
2018 SPE Reservoir Evaluation & Engineering 1
ID: jaganm Time: 14:40 I Path: S:/REE#/Vol00000/170036/Comp/APPFile/SA-REE#170036

REE185589 DOI: 10.2118/185589-PA Date: 5-January-18 Stage: Page: 2 Total Pages: 20
they are more tolerant to a large variety of commonly occurring trends in actual data and result in finite estimate of “contacted hydro-
carbons,” the very property the Arps decline with b > 1 is lacking.
When fitting decline models to production data, it is important to have a reduced number of parameters that can be identified from
the data. This is one of the reasons that three-parameter models [the Arps (1945) hyperbolic, the Duong (2011) model, stretched expo-
nential] have been commonly applied in the industry. If the number of parameters increases in an attempt to better describe the nuances
in the production response caused by a more-complex porous-media-flow phenomena, the data become sparse for history matching and
the parameters’ uncertainty increases. This is known as “the curse of dimensionality” (Burnham and Anderson 2002) and can be more
critical if monthly production is used instead of daily rates.
To be more reliable, models for shale need to incorporate basic physical concepts, such as fluid flow and fracture configuration (Lee
and Sidle 2010). Therefore, it is important to acknowledge the dual-porosity nature of these systems, where the matrix is represented by
a primary porosity with significant contribution to the total pore volume (PV) but very-reduced flow capacity, and the fracture is repre-
sented by a secondary porosity with great flow capacity in a reduced volume. In the oil industry, Warren and Root (1963) were the first
to present a mathematical formulation of dual-porosity systems for naturally fractured reservoirs.
Mathematically, the solutions for production rates of dual-porosity systems typically are more complicated because of the require-
ment of inversion from the Laplace space, which can be computationally expensive—for example, the solutions for different geometries
of shale-gas reservoirs with multistage-hydraulic-fractured horizontal wells (Bello and Wattenbarger 2008; Bello 2009). Shahamat
et al. (2015) provides an alternative procedure that does not require inversion from the Laplace space and is valid for transient and
boundary-dominated flow in linear liquid and gas reservoirs. They coupled the concepts of material balance, distance of investigation,
and boundary-dominated flow, and then discretized this in time assuming a succession of pseudosteady states and updating the size of
the investigated reservoir in the analytical equations. The drawback from their approach is that the equations are not in a closed form,
and for this reason iterations or smaller timesteps are required. Ogunyomi et al. (2016) derived simple material-balance equations for a
double-porosity system from the integration of the diffusivity equation with defined boundary conditions. They suggest a time-domain
approximation to this problem by assuming constant pressure at the fracture/matrix interface, and their solution is expressed in terms of
the complementary error function. However, their model becomes impractical if monthly reported production is used, because the tran-
sition from fracture to matrix transient most likely cannot be identified.
Fuentes-Cruz and Valkó (2015) formulate the dual-porosity problem allowing variable matrix-block size as an increasing function
of the distance from the fracture plane, which is a more-reliable representation of the consequences of the stimulation treatment. Their
model also presents a one-half slope for the linear transient flow observed in unconventional wells. However, they also were able to
quantify the effect on well performance as a result of the distribution of matrix block sizes and matrix/fracture-permeability contrast.
Their solution was also in the Laplace space. For a more-detailed physical description of fluid flow in naturally and hydraulically frac-
tured reservoirs, the reader is also referred to Kuchuk et al. (2016) and Zhao et al. (2013).
According to Lee and Sidle (2010), the application of analytical and numerical models to unconventional reservoirs can be challeng-
ing because of scarce measurements of reservoir properties; reduced understanding of the physical principles controlling gas flow in the
tight formations; and the history matching can be time-consuming when applied to a large number of wells. However, a significant
improvement can be achieved if no reservoir properties are required a priori, and instead a reduced number of parameters are inferred
from the production history; if the identified parameters are implemented in a function with embedded physics; and if the history match-
ing is computationally fast. The model and automated framework presented here represent an effort in this direction.
Uncertainty analysis plays a major role when using reduced-physics models to make production forecasts and economic appraisal
(Weijermars et al. 2017). When investing in a field-development plan, it is essential to be aware of the risks taken and determine a proba-
ble range of reserves volumes. For this reason, uncertainty-quantification algorithms have been widely applied to decline-curve analysis
(Cronquist 1991; Chang and Lin 1999; Cheng et al. 2008; Gong et al. 2014; Fulford et al. 2016; Yu et al. 2016). Purvis and Kuzma (2016)
provides an overview of methods commonly used. However, the pure application of such algorithms still can result in biased estimates
and frequently in overconfidence. Therefore, probabilistic calibration becomes a requirement in the pursuit of a “reliable technology.”
When analyzing publicly available data for a large number of wells, it is noticeable that many production histories present disconti-
nuities in the decline behavior caused by unreported reasons. However, those wells should not be simply excluded from the data set
because it is also necessary to compute their contribution to the total reserves, even if this results in higher uncertainty. So, it is essential
to preprocess the data before obtaining history-matched and probabilistic models. The problem is that data analysis and outlier classifi-
cation can be quite subjective and tedious when performed for hundreds to thousands of wells. Therefore, it is necessary to have an
automatic and consistent way of treating the data, and it must be dependent on clear reasoning. For this reason, heuristic rules are imple-
mented in our approach as a way to treat data points that poorly represent the full productive capacity of the well and capture the last
trend in the production history.
In this context, Chaudhary and Lee (2016) proposed the use of the local outlier-factor method for rate and pressure data, which clas-
sifies outliers using the distances of the k nearest neighbors in a time series. Castineira et al. (2014) applied quantile regression to gener-
ate probabilistic models as an alternative method that is less sensitive to outliers. The method proposed here assigns a weight to each
data point. These weights are incorporated in the history matching and the Bayesian approach (for uncertainty quantification). They
control the effect of each data point in the forecasts. An automatic procedure dependent on heuristic rules defines the value of these
weights. There are some degrees of freedom in these heuristic rules that allow us to probabilistically calibrate the full data set.
The objective of this work is to provide a more-robust framework for automated decline-curve analysis for large portfolios in unconven-
tional reservoirs by proposing a new three-parameter physics-based model and a data-treatment algorithm dependent on heuristic rules, and
adapting such rules to a Bayesian approach with probabilistic calibration. At this moment, some years of production history from unconven-
tional reservoirs are available, so 992 gas wells from the Barnett Shale are taken as a case study. This allows one to look back and compare
the predictability of empirical and physics-based models, as well as to evaluate the performance of this automated framework.
Physics: Jacobi h Function No. 2 as a Decline-Curve Model

Model Derivation. The model proposed here is obtained by coupling the material-balance equation with the pressure solution for the
homogeneous linear 1D reservoir depicted in Fig. 1. The governing equations and boundary and initial conditions are similar to those
presented in Ogunyomi et al. (2016), but the model is simplified by assuming an infinitely conductive fracture.
For a reservoir under primary production, the material-balance equation for the drainage volume of each well can be written as
dp
V p ct ¼ q; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð1Þ
dt
2 2018 SPE Reservoir Evaluation & Engineering

where p is the average reservoir pressure, q is the production rate, Vp is the drainage PV, and ct is the total compressibility of
the system.
x=0 x = xi x=L
Linear flow No-flow boundary

Fracture face
∂ 2p φμct ∂p
= ∂p (L,t )
∂x 2 k ∂t =0
p (0,t ) = pwf ∂x
Initial condition: p (x,0) = pi
Horizontal well
Fig. 1—Representation of the physics of the proposed decline model. The pressure distribution is defined by the linear flow with
constant pressure on the fracture face and no-flow boundary on the other end of the drainage volume; pressure is initially in
equilibrium.
The definition of average reservoir pressure is modified to

ð
1 L
pðtÞ ¼ pðx; tÞdx; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð2Þ
L xi
where 0 xi L instead of xi ¼ 0. The reason for this modification is that it allows the model to have an initial delay and buildup in
the production response, which is often observed in field data because of several physical or operational reasons, as will be further dis-
cussed. This is an empirical aspect introduced to the previous model of Wattenbarger et al. (1998), and thus it is emphasized that xi
does not have an explicit physical meaning.
The derivation of the solution for the pressure distribution in the matrix domain over time, p(x, t), is presented in Appendix A. Sub-
stituting Eq. A-30 into Eq. 2 and solving the integral yields the following expression:
h i2
8 ðpi pwf Þ j 2Lð1þ2nÞ t hp xi i
X1 p
pðtÞ ¼ pwf þ 2 2
e cos ð1 þ 2nÞ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð3Þ
n¼0
p ð1 þ 2nÞ 2L
The drainage PV is defined as
Vp ¼ /AL; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð4Þ
and the diffusivity constant as

k
j¼ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð5Þ
/lct
Applying Eqs. 3, 4, and 5 in Eq. 1, the following expression is obtained after the proper algebraic manipulation:
hp i2
kA X 1

k
ð1 þ 2nÞ t
hp x i
i
qðtÞ ¼ ðpi pwf Þ 2e /lct 2L cos ð1 þ 2nÞ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð6Þ
lL n¼0
2L
which corresponds to the second Jacobi h function (h2),
qðtÞ ¼ qi h2 ðv; egt Þ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð7Þ
where qi is the virtual initial rate,

kA
qi ¼ ðpi pwf Þ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð8Þ
lL
v is a geometric factor accounting for the initial delay and buildup in the production rate,
p xi
v¼ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð9Þ
2L
and g is the reciprocal characteristic time,
p2 k
g¼ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð10Þ
L2 /lct

v is named a geometric factor because it is expressed only in terms of variables with dimension of length (xi and L). However, for the
same reasons previously stated for xi, an explicit physical interpretation is not attributed to v, but there are a variety of factors that cause
v=0, which will be further discussed.
As exemplified previously, the Jacobi h functions are intrinsically related to the analytical solution of certain partial-differential
equations. For this reason, these functions also have been applied in other fields of science and engineering, such as heat transfer
(Chouikha 2005), cosmology (D’Ambroise 2010), and quantum field theory (Tyurin 2002). The main concern in applying Eq. 7 in the
decline-curve analysis of large data sets is the fact that it is an infinite summation, but there are computational routines available that
are capable of computing it in a time-effective manner (Wolfram Research 1988; Igor 2007; Johansson et al. 2013).
The h2 model has been derived for liquid rates. Even though a strict physical derivation might be unfeasible for the gas case, this
model has also been validated with field data in gas wells (discussed in the later section Case Study: 992 Barnett Shale Gas Wells).
Moreover, Al-Hussainy et al. (1966) proved that the solutions to the diffusivity equation for gas and liquid have a similar format when
/lðpÞct ðpÞ
using pseudopressure function in the gas case, assuming is constant. On the other hand, the material-balance equation is
k
intrinsically related to the definition of isothermal compressibility, which is expressed in terms of average reservoir pressure instead of
a pseudopressure function. Furthermore, it is necessary to assume that Vpct is constant, as well as to compute average reservoir pressure
from the pseudopressure solution. Therefore, it seems to be impossible to pursue such derivation without making several hard assump-
tions. On the other hand, it is also important to emphasize that the behavior of the h2 ð0; egt Þ model proposed by Wattenbarger et al.
(1998) has been observed in a number of gas wells. In addition, the empirical models [the Arps (1945) hyperbolic, stretched exponen-
tial, and Duong (2011) model] are usually applied in the industry without distinction whether the fluid is oil or gas.
Subcases and Extensions of the h2 Model. Wattenbarger et al. (1998). If v ¼ 0, the solution presented in Eq. 7 is equivalent to the
one introduced by Wattenbarger et al. (1998) for tight reservoirs, and derived in Carslaw and Jaeger (1959) for linear heat-conduction
problems. It has only two parameters, is valid for transient and boundary-dominated flow, and presents a continuous decline. Easley
(2012) proposed an approximation function to h2 ð0; egt Þ that does not require evaluation of the infinite summation term.
Double-Porosity Model (Ogunyomi et al. 2016). Ogunyomi et al. (2016) proposed a rate/time relationship by coupling material bal-
ance and the analytical solution for pressure (Eq. A-30) in a double-porosity system. Their model is a time-domain approximation of
the Laplace space solution proposed by Bello (2009). It assumes a constant pressure at the fracture face because of the high contrast in
the permeability at the fracture/matrix interface, which causes pressure to reach a quick equilibrium with pwf in the fracture compart-
ment. Using this assumption and the material-balance equations presented by Ogunyomi et al. (2016), the double-porosity model can be
recast in terms of h2 functions as
qðtÞ ¼ qi;m h2 ð0; egm t Þ þ qi;f h2 ð0; egf t Þ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð11Þ
where the subscripts m and f refer to the matrix and fracture compartments, respectively.
Fig. 2 shows the double-porosity h2 approximation. Note that this model has four parameters, which can be a problem when dealing
with sparse data, such as the monthly reported production rates. If the fracture-boundary effect happens before the end of the first month
of production, the parameters related to the fracture-control volume (i.e., qi;f and gf) will be overfitting the production history and not
improving the predictions. In this case, the model requires production-rate measurements at a higher frequency. For typical values of
these parameters in unconventional reservoirs, it is expected that the transition from fracture to matrix transient flow to happen in the
order of minutes, whereas monthly reported production information are the data analyzed in this work. Therefore, the assumption of
infinitely conductive fracture is plausible here.
* θ (0, e –ηm t )
qi,f* θ2 (0, e –ηf t ) + qi,m 2
107 Fracture-
Fracture- Matrix-
transient
boundary boundary
Production Rate, q (Mcf/month)
flow
effect effect
105
Matrix-
transient
flow
1000
10
qi,f* = 50000, ηf = 100, qi,m

* = 50, η = 0.05
m
0.100 qi,f* = 20000, ηf = 200, qi,m
* = 20, η = 0.1
m
qi,f* = 5000, ηf = 300, qi,m
* = 5, η = 0.3
m
0.001
10–5 0.01 10
Time (months)
Fig. 2—The double-porosity-model approximation in terms of h2 functions. qi is in Mcf/month and g is in 1/months.
Comparison With the Arps (1945) Decline Model. The Arps (1945) decline-curve family has been widely applied in the industry to
estimate reserves. This practice also has been extended to unconventional reservoirs (Gong et al. 2014), where the hyperbolic model is
the most suitable to capture the production decline during the transient state:
1

qðtÞ ¼ qi ð1 þ bDi tÞ b; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð12Þ
where qi is the initial rate [q(0)], b is the decline exponent, and Di is the initial decline rate.

Fig. 3 shows the sensitivity of the production-decline profile in a log-log plot when varying the Arps parameters b and Di. In Fig. 3a,
notice that as b increases, the slope varies significantly less comparing early and late time. In Fig. 3b, the different values of Di present
only a slight difference in the early time, whereas in the late time the profile is similar (parallel straight lines), and then varying qi is
equivalent to varying Di in this case.
1 1
(1+b Di t )– b , Di = 0.1 (1+b Di t)– b , b = 1.82
1 1
0.500 0.500
Dimensionless Production
0.100 0.100
Rate (qD)
Rate (qD)
0.050 b = 0.001 0.050 Di = 0.1
b = 0.401 Di = 0.25
b = 0.801 Di = 0.63
0.010 b = 1.2 0.010
Di = 1.58
0.005 b = 1.6 0.005
Di = 3.98
b=2
Di = 10
0.001 0.001
1 5 10 50 100 1 5 10 50 100
Time (months) Time (months)
(a) (b)
Fig. 3—Sensitivity to (a) b and (b) Di parameters in the Arps (1945) hyperbolic model. In each plot, one of the parameters is fixed at
the median value of the best-fit solutions for the 992 Barnett gas wells presented in the section Case Study: 992 Barnett Shale Gas
Wells. Di is in 1/months and qD (t)5q(t)/qi .
Fig. 4 presents a sensitivity to the parameters g and v in the h2 model. The production profiles must be compared with the ones in
Fig. 3. Note that the h2 model captures the one-half slope of the transient-flow regime and presents a transition to the boundary-domi-
nated flow, achieving the exponential decline. In Fig. 4a, varying the reciprocal characteristic time (g) is equivalent to shifting the pro-
duction profile horizontally, whereas varying qi is equivalent to shifting it vertically. As one can see in Fig. 4b, the geometric factor (v)
is the one that defines the shape of the curve, adding flexibility to the model presented by Wattenbarger et al. (1998).
θ2 (χ, e –η t ), χ = 0.22 θ2 (χ, e –η t ), η = 0.053
–1/2 slope
Exponential –1/2 slope Exponential
1 decline decline
1
χ = 0.05
η = 0.01
Rate (qD)
Rate (qD)
χ = 0.34
0.100 η = 0.022 χ = 0.63
η = 0.048 0.050
χ = 0.92
0.010 η = 0.105 χ = 1.21
η = 0.229 χ = 1.5
η = 0.5
0.001 0.001
1 5 10 50 100 500 1,000 1 5 10 50 100 500 1,000
Time (months) Time (months)
(a) (b)
Fig. 4—Sensitivity to (a) g and (b) v parameters in the h2 model. In each plot, one of the parameters is fixed at the median value of
the best-fit solutions for the 992 Barnett gas wells presented in the section Case Study: 992 Barnett Shale Gas Wells. The one-half
slope indicates transient flow, and the exponential decline indicates boundary-dominated flow. g is in 1/months and qD (t)5q(t)/qi .
The parameter v allows the model to have an initial delay and buildup in the production rates. This feature is also present in the
Duong (2011) and the stretched exponential (Valkó 2009; Valkó and Lee 2010) models. In general, v can be interpreted as a deviation
from the assumptions of the Wattenbarger et al. (1998) model. Such a deviation can be related to the reservoir physics or field opera-
tions. An example of a physical reason is that as the reservoir starts to be depleted, pore pressure declines and the effective stress (res)
increases. If res exceeds the strength of the shale, the fractures are reactivated, propagating and causing the initial buildup in the produc-
tion rates (Duong 2011). Another plausible physical reason is that a variable skin factor caused by cleanup of drilling fluids and unload-
ing gas condensate while starting production can create this initial increase (Larsen and Kviljo 1990; Clarkson et al. 2013; Hashmi et al.
2014). Operational factors can also contribute to an increasing q(t) in the production history, such as long shut-in time during part of the
sampling period, oil-price fluctuations, restimulation, and gas/condensate-unloading operations. Therefore, the v parameter improves
the model’s flexibility, which is a desirable feature when dealing with a more-complex production history.
As proved by Lee and Sidle (2010), the Arps (1945)ðcurves family has the problem that the EUR is infinity when b 1 and no eco-
t
nomical or time constraints are imposed; i.e., limt!1 qðtÞdt ¼ 1, which is physically impossible. In such cases, the Arps (1945)
0
hyperbolic model should not be applied for long-term forecast because the observed data only exhibits transient flow or other nuances
of the production mechanism, and the model does not present a transition to boundary-dominated flow embedded in its functional form.
In contrast, Appendix B presents a general proof that the h2 model has a finite EUR and a simple equation for the special case of v ¼ 0.

History Matching. The best-fit model is obtained by solving a least-squares problem with the objective function
minz ¼ minðlogqobs logqpred ÞT C21
e ðlogqobs logqpred Þ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð13Þ
where qpred and qobs 2 <Nt 1 are vectors of the production rates predicted by the model and observed in the production history, respec-
tively; Ce 2 <Nt Nt is the covariance matrix of the measurement and modeling errors, which is discussed in more detail in the section
Heuristics: Treating the Bad Data; and Nt is the number of timesteps. Because q(t) can have different orders of magnitude in the same
production history, logqðtÞ is considered in the objective function.
Table 1 presents the box constraints for the h2 and Arps (1945) hyperbolic models, where the “practical” constraints were the ones
applied to the case study in the section Case Study: 992 Barnett Shale Gas Wells. In the h2 model, even with these box constraints, occa-
sionally a negative q(t) is computed, and adding the following linear constraint is a simple way to solve this problem (g is in 1/months):
v 12g: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .ð14Þ
θ2 Arps (1945) Hyperbolic Model

Theoretical Practical Theoretical Practical
η≥0 0.01 ≤ η ≤ 0.5 Di ≥ 0 0.1 ≤ Di ≤ 12
0 ≤ χ ≤ π/2 0.05 ≤ χ ≤ 1.5 0≤b≤2 0.001 ≤ b ≤ 2
* * *
q ≥0
i 0.05 ≤ q / qmax ≤ 2
i q ≥0
i 0.05 ≤ qi* / qmax ≤ 3
Table 1—Theoretical and practical box constraints in the h2 model and Arps (1945) hyperbolic model. g
and Di are in 1/months.
Fig. 5a shows this constraint in the v vs. g solution space, where the dots represent the best-fit h2 models for the 992 Barnett gas
wells (in the section Case Study: 992 Barnett Shale Gas Wells). Fig. 5b confirms a fair correlation between qi and qmax, which allows
to define the box constraints in terms of qi =qmax (Table 1).
χ vs. η – Solution Space

qi* vs. qmax
1.4
1.2 χ > 12 η
1×105
χ (dimensionless)
1.0 5×104
q*i (Mcf/month)
0.8
1×104
0.6 5,000
0.4
1,000
0.2 500
0.0
0.1 0.2 0.3 0.4 5,000 1×104 5×104 1×105
η (1/months) Maximum Production Rate (Mcf/month)
(a) (b)
Fig. 5—Best-fit solutions for the 992 Barnett gas wells with the h2 model: (a) distribution in the v vs. g space, with the yellow area
depicting the linear constraint; (b) relationship between qi and qmax.
Statistics: Uncertainty Analysis

When analyzing production data to forecast reserves, it is preferable to proceed with a probabilistic rather than deterministic approach
so that risk awareness is improved before decisions are taken. The Bayes’ theorem has been widely used for uncertainty assessment and
data integration in reservoir-engineering problems (Oliver et al. 2008; Gong et al. 2014). It reconciles the following elements: expert’s
judgment, embedded in the prior distribution, Ppr ðcj Þ, and the value of data acquired and model proposed, embedded in the likelihood
function, Pl ðqobs jcj Þ. Thus, the Bayes’ theorem provides the posterior distribution of the parameters:
Pl ðqobs jcj ÞPpr ðcj Þ
Ppost ðcj jqobs Þ ¼ ð ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð15Þ
Pl ðqobs jcÞPpr ðcÞdc
where cj is a vector with candidate values for each parameter of the decline model; e.g., cj ¼ ðg; v; qi ÞTj .
Fig. 6 illustrates the application of Bayes’ theorem in the estimation of the parameters of the h2 model (3D space) for Well 8 (Fig. 7). The
color bar indicates the normalized values of the probability-distribution functions (PDFs), with the hot colors being the most-likely region for
the h2 parameters. Note how the posterior is generated from the interplay between the prior distribution and the likelihood function.
Analyzing Eq. 15, even when the prior and likelihood functions are expressed in a closed form, the integral in the denominator might
be very difficult or impossible to solve. For this reason, sampling algorithms are frequently incorporated in a Bayesian framework, such
that a large-enough sample that resembles the posterior distribution is generated without solving the integral. From the percentiles of
this sample, it is possible to obtain the P10, P50, and P90 of a certain property. The sampling method used in this work is the Markov
chain Monte Carlo (MCMC) with the Metropolis algorithm. The algorithm is briefly explained here, and for more theoretical details the
reader is referred to Gong et al. (2014) and Oliver et al. (2008). Compared with Gong et al. (2014), the only modifications made were to
incorporate Ce and the linear constraint (Eq. 14) in their framework.

Joint Prior Distribution Likelihood Function Posterior Distribution

χ 0.6 χ 0.6 χ 0.6
0.4 0.4 0.4
0.2 0.2 0.2
1.0
30,000 30,000 30,000 0.8
20,000 20,000 0.6

qi*
qi*
qi*
20,000
10,000 10,000 10,000 0.4
0.2
0.05 0.05 0.05
0.10 0.10 0.10
0.15 0.15 0.15
η 0.20 η 0.20 η 0.20 0
0.25 0.25 0.25
(a) (b) (c)
Fig. 6—The Bayes’ theorem idea applied to Well 8 with the new work flow and model. Normalized PDF values are depicted by the
color scale in the three-parameter-solution space for (a) prior, (b) likelihood, and (c) posterior. g is in 1/months and qi is in Mcf/
month. [Well 8 is listed as American Petroleum Institute Well Number (API No.) 42121329920000.]
Well 1, API No: 42097341450000 Well 2, API No: 42497369130000 Well 3, API No: 42121341110000
1×105 5×104 1×105
5×104 5×104
1×104
1×104 5,000 1×104
5,000 5,000
1,000
1,000 1,000
500 500 500
100 100 100

0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
Production Rate (Mcf/month)
5×104 5×104
104
1×104 1×104 5,000
5,000 5,000
1,000 1000
1,000
500 500 500
100 100 100

0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
5×104
104 104
5,000 1×104 5,000
5,000
1,000 1,000
1,000
500 5,00
5,00
100 100 100

0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
Time (months)
Production history θ2 (bestfit) Filter line
Fig. 7—Best-fit solutions of the h2 model using the heuristic rules to filter the data.
For the likelihood function, it is assumed that the error between the proposed model and production history [i.e.,
ðlogqobs logqprop Þ] follows a normal distribution with zero mean; that is, Nð0; rbfÞ. This results in
!
1 r2prop
Pl ðqobs jcprop Þ ¼ pffiffiffiffiffiffi exp 2 ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð16Þ
2prbf rbf þ e
where rbf is the standard deviation (SD) of the residual for the best-fit model and is given by
0:5
1
rbf ¼ ðlogqobs logqbf ÞT C21
e ðlogq obs logqbf Þ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð17Þ
Nt 3

and where rprop is the SD of the residual for the proposed model, defined as
0:5
1 T 21
rprop ¼ ðlogqobs logqprop Þ Ce ðlogqobs logqprop Þ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð18Þ
Nt
and e is the inherent error of the production data, which is introduced to avoid extremely small acceptance ratios (a) and consequently
unrealistically low uncertainties. As successfully experienced by Gong et al. (2014), e ¼ 0:001 is the value used here.
Then, a Markov chain is built. A model (cj;prop ) is proposed from a proposal distribution. It has a probability of being accepted (i.e.,
being aggregated to the chain) and (1–a) of being rejected, in which case the previous model (cj;s21 ) is repeated in the chain. The accep-
tance ratio (a) is computed by comparing the posterior probabilities of cj;prop and cj;s21 .
Considering Eqs. 16, 17, and 18, as well as proposal distributions that are independent truncated normal distributions for each pa-
rameter with bounds defined in Table 1, the acceptance ratio is obtained from
2 3
! tup ts1 tlow ts1
2 2 Y U U
6 rs1 rprop Ppr ðcprop Þ rt rt 7
a ¼ min6 41; exp 7; . . . . . . . . . . . . . . . . . . . . . . . . ð19Þ
r2bf þ e Ppr ðcs21 Þ t¼g;v;q tup tprop tlow tprop 5
i U U
rt rt
where t represents each of the decline-curve parameters; tup and tlow are the upper and lower bounds (Table 1) for each parameter,
respectively; rt is the SD for each parameter, which is estimated from the best-fit solutions for the full data set; and U() is the cumula-
tive distribution function of the standard normal, N(0, 1).
The MCMC with the Metropolis algorithm applied to the h2 model for a sample with size nMCMC can be summarized as
1. Set s ¼ 1 and cs ¼ cbf .
2. s ¼ s þ 1. Draw cprop from the proposal distribution, Ntruncated ðcs21 ; rcbf Þ, until it satisfies Eq. 14.
3. Compute a from Eq. 19.
4. Draw a random number, rn, from a standard uniform distribution, U(0, 1).
5. If rn < a, then cs ¼ cprop . Otherwise, cs ¼ cs21 .
6. If s < nMCMC, then return to Step 1. Otherwise, the Markov chain is complete.
Heuristics: Treating the Bad Data

Data processing for the application of decline-curve analysis can be a tedious and subjective task, especially when dealing with large
data sets. Production data from unconventional wells can present several discontinuities, which can be caused by physical processes,
operations, or other nonstochastic factors. For example, increasing drawdown, unloading condensate or refracturing operations will
cause production to suddenly increase, and shutting in a well for a half-month for pipeline maintenance results in a lower monthly pro-
duction. This section introduces an automatic and consistent way of dealing with erratic production histories and calibrating uncertainty
of the forecasts.
For this purpose, heuristic rules are defined to assign a weight, wi, to each data point in a production history, q(ti). These weights are
then incorporated in the covariance matrix of the measurement and modeling error, Ce, in the history matching (Eq. 13) and uncertainty
analysis (Eqs. 17 and 18). The basic idea is that the value of the weight wi is a measure of the importance and confidence on q(ti) for the
forecast period. Each of the heuristic rules are presented here and are explained in the sequence in which they are implemented.
1. Start setting the vector of weights with unit elements:
winitial ¼ ½w1 ; w2 ; …; wi ; …; wNt T ¼ 1Nt 1 : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð20Þ
2. Flow rates less than a threshold value, qlim, are assigned zero weight; i.e., if q(ti) qlim, wi ¼ 0. This is necessary to eliminate
unrealistically low flow rates that do not correspond to the actual reservoir potential and deviate the models toward lower produc-
tion rates. Now, winitial is a vector of zeros and ones.
3. Decline-curve analysis is mostly concerned with propagating the last state of production; i.e., higher weights must be assigned as
time increases. Therefore, the weights can be redefined as an increasing function of winitial and ti:
w ¼ ½ fw ðw1 ; t1 Þ; …; fw ðwi ; ti Þ; …; fw ðwNt ; tNt ÞT : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð21Þ
In this paper, fw ðwi ; ti Þ is a linear function, so Eq. 21 can be written as
w ¼ aw winitial T t þ bw : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð22Þ
4. It is common to have data points that when compared with the general production trend are significantly deviated downward, but
are still higher than qlim, as shown in Fig. 7. Therefore, it is important to reduce the contribution of these data points to the fore-
cast; i.e., reduce wi. For this reason, a straight line (unconstrained exponential model) is fit to the logarithmic production history,
logqobs , yielding
logqexp ðtÞ ¼ aexp t þ bexp : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð23Þ
This line is shifted downward by introducing the multiplier ml:
logqexp;l ðtÞ ¼ aexp t þ ml bexp ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð24Þ
where 0 ml 1. This is the red dashed line shown in Fig. 7. If logqobs ðti Þ logqexp;l ðti Þ, then wi is reduced by a multiplier 0 b 1,
such that wi ¼ bwi.

Once the weights (w) have been defined, they are introduced in Ce, such that C21 e is computed as
( )
w1 wi wNt
C21
e ¼ Diag ; …; ; …; ; Ce 2 <Nt Nt : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð25Þ
½logqobs ðt1 Þ2 ½logqobs ðti Þ2 ½logqobs ðtNt Þ2
The denominator is defined as ½logqobs ðti Þ2 as a way to normalize the errors, thus dealing with relative errors in the history matching
and uncertainty analysis. For numerical stability, if qobs ðti Þ qlim , it is redefined as qobs(ti) ¼ qlim, and note wi ¼ 0 in this case.
Tuning the Heuristics. Even though the reasoning of heuristic rules has been previously explained and exemplified, it is necessary to
have a consistent procedure to define its parameters (aw, bw, ml, and b). Here, the criterion established is that such parameters must be
defined in a way that calibrates the uncertainty of the forecasts. Also, it is desirable that these parameters keep the uncertainty calibrated
as more production history is obtained and new forecasts are made.
For this purpose, the concept of hindcasts is used, where each production history is split in two periods: the first period is used for
model fitting and the second period is the blind data that are compared with the forecast projected from the model generated with the
first period data. As done by Gong et al. (2014), the total production during the second period (PDTSP, or Q2nd ) is considered as a met-
ric for the calibration. Then, for the full data set (992 Barnett gas wells), the frequency that the observed PDTSP is higher than the ones
for a predefined percentile model (e.g., P50 is the 50th percentile) is computed, which is here denoted as xðQ2nd;obs > Q2nd;percentile Þ and
is the real percentile. The models are probabilistically calibrated if xðQ2nd;obs > Q2nd;percentile Þ match the predefined percentiles. There-
fore, by comparing the forecasts and actual production history in a large data set, it is possible to check if the projected percentiles (e.g.,
the P10, P50, and P90 models) correspond to the distribution of the actual data set.
Considering the P10, P50, and P90 models, this can be framed as an optimization problem with the objective function depending on
the heuristic parameters (aw, bw, ml, and b):
X 2
minaw ;bw ;ml ;b ½xth ðQ2nd;obs > Q2nd;P10 Þ 0:1 þ ½xth ðQ2nd;obs > Q2nd;P50 Þ 0:52 þ ½xth ðQ2nd;obs > Q2nd;P90 Þ 0:92 ; . . . . . ð26Þ
th
where th represents total time used in the first period, and in the case study in the section Case Study: 992 Barnett Shale Gas Wells, it
takes the values 6, 12, 18, 24, 30, and 36 months. The following constraints were applied:
0:05 aw 2; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð27Þ
0:85 ml 1; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð28Þ
0 b 0:5; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð29Þ
and bw ¼ 0 was defined to reduce the number of parameters in the problem.

For comparison, Fig. 8 shows the need for probabilistic calibration in the base case, where the heuristic rules are not applied
(aw ¼ 0, bw ¼ 1, ml ¼ 1, and b ¼ 1). Fig. 9 shows the probabilistically calibrated case with the adjusted heuristic parameters, which is
further discussed in the following section.
Frequency of True PDTSP > Probabilistic
1.0
0.8
θ2 – best fit
Arps – best fit
0.6
PDTSP
θ2 – P10
Arps – P10
0.4 θ2 – P50
Arps – P50
θ2 – P90
0.2 Arps – P90
0.0
10 15 20 25 30 35
Production Data Used to Hindcast (months)
Fig. 8—Base case, no heuristic rules applied; i.e., aw 5 0, bw 5 1, ml 5 1, and b 5 1. The uncertainty is not calibrated.
Case Study: 992 Barnett Shale Gas Wells

For this case study, the gas-production history of 992 wells from the Barnett Shale was analyzed using the new methodology and model.
Fig. 10 shows the number and percentage of wells of each fluid type. Even though there are 66 oil wells in this data set, in this text, the
whole data set is referred as gas wells for the sake of simplicity and because only the gas-production history is being analyzed.
A full comparison with the results obtained with the Arps (1945) hyperbolic model is also included. The data are publicly available
because the producers are obligated by law to report production on a monthly basis, and were accessed in the DrillingInfo (1998–2017)
database. The Barnett Shale was chosen because it was the first unconventional play to be massively drilled and start to produce in com-
mercial scale. Therefore, the Barnett Shale is the most abundant unconventional play in terms of longer production histories, which

makes it the best candidate for the validation of the objectives of this study. An overview from the early to recent developments and
operations in the Barnett Shale can be found in Parshall (2008) and Browning et al. (2013). To evaluate the performance of the proposed
framework to horizontal multistage hydraulically fractured wells, only the ones that started to produce in 2010 or after were included in
the data set. Also, only the wells that had at least 40 months of production higher than qlim ¼ 100 Mcf=month were taken into account.
1.0
Frequency of True PDTSP > Probabilistic 0.8
θ2 – best fit
0.6 Arps – best fit
PDTSP
θ2 – P10
Arps – P10
0.4 θ2 – P50
Arps – P50
θ2 – P90
0.2 Arps – P90
0.0
10 15 20 25 30 35
Fig. 9—Case with adjusted heuristic rules for probabilistic calibration.
Fluid Classification Based on Initial Producing Gas – Liquid Ratio
108
107 Dry gas – 426 wells (42.94%)

GLRi (scf/STB)
106
105
Wet gas – 255 wells (25.71%)
104
Gas condensate – 245 wells (24.70%)
Volatile oil – 23 wells (2.32%)
1000 Black oil – 38 wells (3.83%) Indeterminate by GLRi – 5 wells (0.50%)
0 200 400 600 800

Ordered Wells
Fig. 10—Fluid classification using the initial producing-gas/liquid ratio (GLRi) for 992 wells in the Barnett Shale.
The results presented in this section are for the probabilistically calibrated models with the heuristic parameters in Table 2. As
shown in Fig. 9, these parameters provide a significantly better uncertainty estimation than in the base case (Fig. 8). The highest mis-
matches in the percentile distribution are for the P50 models from 6 to 18 months. Also, note that the best-fit models are significantly
higher than the 50% frequency, which means that in the beginning they tend to provide a more-pessimistic forecast. However, as more
data are acquired, these history-matched models tend to the 50% frequency.
θ2 Arps
aw 1.747 0.241
ml 0.954 0.966
β 0.003 0.044
Table 2—Heuristic parameters for probabilistically calibrated models.
The computational code was implemented in Mathematica (Wolfram Research 2015). For the h2 model, the prior distribution for
each parameter is obtained from the PDF that best fits the histogram of the history-matched solutions for the 992 wells among 27

candidate PDFs. The result is shown in Fig. 11; the prior PDF varies smoothly within its domain. The same procedure was initially
attempted for the Arps (1945) hyperbolic model, but the parameters b and Di have a significantly higher frequency at a narrow range
(Fig. 12), [1.8, 2] and [0.1, 0.12], respectively. Therefore, using a smooth prior was problematic for calibrating the uncertainty, and
instead the prior is defined as the sum of two PDFs, where one of them is a uniform distribution for the narrow range. The combination
that best fits the histogram was chosen, as shown in Fig. 12.
Inverse Gaussian (0.072,0.085) Gamma (2.05,0.16) Inverse Gamma (3.30,0.78)

14 3.5 3.5
12 3.0 3.0
10 2.5 2.5
PDF
PDF
PDF
8 2.0 2.0
6 1.5 1.5
4 1.0 1.0
2 0.5 0.5
0 0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.2 0.4 0.6 0.8 1.0 1.2 1.4 0.0 0.5 1.0 1.5 2.0
η (1/months) χ (dimensionless) *
qi /qmax
(a) (b) (c)
Fig. 11—Prior distributions for the parameters of the h2 model. It is assumed that the parameters are independent of each other.
0.54 Uniform (0.001,2) + (0.34 Half-Normal (0.93) + Inverse Gamma (8.11,9.89)

0.46 Uniform (1.8,2) 0.66 Uniform (0.1,0.12)
2.5 3.5
30
3.0
2.0 25 2.5
20
PDF
PDF
PDF
1.5 2.0
15 1.5
1.0
10 1.0
0.5 5 0.5
0.0 0 0.0
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0
b (dimensionless) Di (1/months) qi* /qmax
(a) (b) (c)
Fig. 12—Prior distributions for the parameters of the Arps (1945) hyperbolic model. It is assumed that the parameters are inde-
pendent of each other.
Fig. 13 compares the average PDTSP for all wells for the production history and best fit, and the P10, P50, and P90 models. As
expected, PDTSP decreases as time increases because of the second period being shortened and the natural decline of production rates.
The P50 models are very close to the production history. As in Fig. 9, this plot also confirms that the best-fit solutions provide pessimis-
tic estimates initially but gradually approach the production history, being much closer after 2 years of production. In addition, the best-
fit solutions from the Arps (1945) model are generally closer to the production history than the ones from the h2 model. This is because
the reduced flexibility of the Arps (1945) hyperbolic model causes b and Di to fall in a narrow range (Fig. 12) for unconventional reser-
voirs, which is identified with less data on the price of generating similar forecasts for most of the wells. On the other hand, the more-
flexible h2 usually will require a longer production history, but captures more features in the data, making a better distinction between
wells when forecasting.
1.4
PDTSP Averaged for All Wells (Bcf)
1.2
Production history
1.0
θ2 – best fit
0.8 Arps – best fit
θ2 – P10
0.6 Arps – P10
θ2 – P50
0.4 Arps – P50
θ2 – P90
0.2 Arps – P90
0.0
10 15 20 25 30 35
Fig. 13—Average production during the second period (PDTSP) for probabilistic and best-fit models compared with the production
data for hindcasts.

The normalized P90–P10 range is taken as a measure of the uncertainty and is shown in Fig. 14 for the calibrated and uncalibrated
models. The uncalibrated h2 model predicts a higher uncertainty than the calibrated case. In contrast, the uncalibrated Arps (1945)
hyperbolic is overconfident. In fact, both uncalibrated models show an increasing uncertainty with time, which is inconsistent, because
the new data acquired should be adding value to the identification of the representative parameters. Therefore, the calibrated h2 model
is the most consistent in the sense that it recognizes the large uncertainty in the beginning of production because of the lack of data;
uncertainty smoothly decreases and becomes lower than that of the calibrated Arps (1945) hyperbolic model when at least 18 months of
production data are available. These results show the need for tuning the heuristic rules and validating the uncertainty quantification in
the data set, as well as the benefit of using a physics-based model.
PDTSP for All Wells

Median[(P10–P90)/True] 1.4
1.3
1.2
1.1 θ2
Arps
1.0
θ2 (uncalibrated)
0.9 Arps (uncalibrated)
0.8
0.7
10 15 20 25 30 35
Fig. 14—The probabilistic calibration is necessary for reliable uncertainty assessment. Uncertainty reduces as more data are
acquired for calibrated models.
Because the objective of decline models is to estimate reserves by extrapolating the current production history, it is essential to com-
pare the responses generated from different models. Fig. 15 contrasts the probabilistic and best-fit responses of the Arps (1945) hyper-
bolic and h2 models for the EUR considering a time horizon of 40 years (EUR40), where the cumulative production from the history
was summed with the model prediction for the remaining time to complete 40 years. It is clear that the EUR estimates from the Arps
(1945) hyperbolic model were optimistic, whereas the h2 model is more conservative. In fact, comparing the lines from two different
models, the closest ones are h2-P10 and Arps-P90, which indicates the enormous discrepancy in the values generated by these models.
Reserves estimation can greatly affect the economic feasibility of a project; being too optimistic can challenge the implementation and
operations during the field development because of lack of budget, and in the worst-case scenario can cause bankruptcy of companies.
EUR (40 Years)
10
θ2 – best fit
EUR40 (Bcf)
Arps – best fit

1 θ2 – P10
Arps – P10
0.50
θ2 – P50
Arps – P50
θ2 – P90
0.10 Arps – P90
0.05
0 200 400 600 800 1,000

Ordered Wells
Fig. 15—Comparison of cumulative production during 40 years for probabilistically calibrated models.
Lee and Sidle (2010) has proved that when b 1, the Arps (1945) hyperbolic predicts infinite EUR (no time or rate constraint con-
sidered). Therefore, the optimistic results of the Arps (1945) hyperbolic model in Fig. 15 were expected. At this point, it is important to
compare these also with the Duong (2011) and stretched exponential (Valkó 2009) models, which also only have three parameters. This
comparison is presented in Fig. 16. The Duong (2011) model is the most optimistic because it is designed to capture the transient flow
in unconventional reservoirs, but it does not have a feature indicating a transition from transient to boundary-dominated flow. For this
reason, its EUR40 estimates are fairly close to those from the Arps (1945) hyperbolic model in most wells. However, there are wells
with extremely high and unrealistic EUR40 estimates from the Duong (2011) model, and these are wells that presented a persistently
increasing or steady production history. Even though this model allows the fitting of an initial buildup in the production history, it is not

capable of predicting a decline if it is not present in the data, which causes an infinite EUR estimation. Even for these anomalous pro-
duction histories, an engineering solution must be achieved, so the h2 provides more-reasonable results. The stretched exponential
decline model of Valkó (2009) agrees with the h2 model for wells with lower EUR40, but it is more optimistic in some wells, forecast-
ing a plateau or very-slow decline for production rates. Therefore, the h2 model is the most-conservative estimate, which is not because
of an empirical function, but a physical phenomenon: that the reservoir is a limited resource and eventually boundary-dominated flow
will start.
EUR (40 Years)
1,000
100
EUR40 (Bcf)
θ2 – best fit
10
SEDM – best fit
Duong – best fit
Arps – best fit
1
0.1
0 200 400 600 800 1,000

Ordered Wells
Fig. 16—Comparison of cumulative production during 40 years for best-fit solutions of the h2, stretched exponential decline
(SEDM), Duong (2011), and Arps (1945) hyperbolic models. Heuristic parameters are aw 5 1.747, ml 5 0.954, and b 5 0.003.
Fig. 17 shows a comparison between the forecast and production history when using 2 years of data in the first period. The wells are
the same ones shown in Fig. 7; they were purposely chosen because of their erratic production history, and therefore they can provide a
good understanding of how the methodology works under such circumstances. As expected, the uncertainty tends to be higher in the
presence of erratic data; e.g., the production during the first period in Wells 1, 2, 5, 7, and 9. In contrast, if a clear trend is shown in the
first period, the predicted uncertainty will be lower in the second; e.g., Wells 3, 4, 6, and 8. In some cases, the production history will
deviate significantly from the trend in the first period and neither the best-fit nor the probabilistic model are capable of providing an ap-
proximate response (e.g. Well 9); in other cases, the history-matched model provides a bad forecast, but the probabilistic models are
predictable (e.g., Wells 2 and 5). As more data are acquired, the quality of the predictions improves (Fig. 7), and the uncertainty
decreases if a trend is kept (Fig. 14). Therefore, using a probabilistic approach provides robustness to reserves estimation because the
best-fit model by itself will many times not be predictable.
Fig. 18 depicts the ability of the model to fit and predict the transient- and boundary-dominated-flow states (e.g., Wells 8, 10, 11, 12,
and 13). In addition, it shows the importance of adding the parameter v to the model because the production delay and initial buildup
can happen in some wells (e.g., Wells 14, 15, 16, and 17).
An analysis of the v parameter separately for each fluid type (Fig. 19) reveals additional causes for deviation of the behavior pre-
dicted by the analytical solution of Wattenbarger et al. (1998), where the h2 model proposed in this work is advantageous. As shown in
Fig. 19, as liquid content increases, it is observed that the central tendency (e.g., mean, median) of v increases, as well as its uncertainty.
Even though the sample sizes for the categories of oil wells (i.e., volatile oil, indeterminate, and black oil) are not statistically signifi-
cant to draw conclusions, this trend is also observed there and should be further investigated in future works. Wells with higher liquid
content are more prone to the occurrence of liquid loading. In addition, other phase-behavior aspects become important. For example,
in black-oil wells, the initial gas/liquid-producing ratio is expected to be very low. As the reservoir is depleted and pressure falls to less
than the bubblepoint, gas will come out of the solution in the reservoir. If a gas cone is established, gas will be more mobile than oil. As
a result, an initial increase in gas rates is observed. At some point, the total producing gas/liquid ratio stabilizes, and the gas rates will
start to decrease with similar characteristics to the total system.
Discussion
Because a large data set is being analyzed and the h2 model is an infinite summation, computational time could be a concern. However,
fast algorithms for the computation of the Jacobi h functions have been implemented in a number of high-level programming languages
(Wolfram Research 1988; Igor 2007; Johansson et al. 2013), and Mathematica was the one used in this work. As shown in Table 3, 992
wells can be successfully history matched in 109.3 seconds using an average desktop computer with eight cores computing in parallel.
The most time-consuming step in one analysis is to generate the Markov chains. One full analysis with six hindcasts usually takes less
than 2 hours. The probabilistic calibration is the most time-consuming procedure, because several analyses (usually 10–40) need to be
run to tune the heuristic parameters, which can take a few days. However, supercomputers are an alternative to speed up this process,
and the values of the parameters obtained here might be a helpful initial guess.
The proposed h2 model has the advantage that it is a physics-based model. However, there is one empirical assumption, which is the
modified definition of the average pressure (Eq. 2). As it was proved, discussed, and exemplified, the transient- and boundary-domi-
nated-flow states are embedded in this functional form (Fig. 4), which also always provides a finite EUR (Appendix B). These features
distinguish it from the previous empirical models. It is also important to mention that it is possible to include an additional linear con-
straint between qi and g from Eq. B-3 if there is a maximum plausible EUR established.

1×105 5×104
1×105
5×104
5×104
1×104
1×104 5,000
5,000
1×104
1,000 1,000
500 500 5,000
100 100
0 10 20 30 40 50 0 10 20 30 40 50 60 0 10 20 30 40 50
5×104 2×104
104
5,000 1×104 1×104
5,000 5,000
1,000
1,000
500
500 2,000
100 100
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 60
Well 7, API No: 42497370560000 Well 8, API No: 42121329920000I Well 9, API No: 42121338360000
104 104
5,000 104 5,000
5,000
1,000 1,000
500 500
1,000
500
100 100
0 10 20 30 40 50 0 10 20 30 40 50 60 0 10 20 30 40 50
Time (months)
Production history Best fit P10 P50 P90
Fig. 17—Prediction from history-matched and probabilistic h2 models considering the first 24 months of production and compar-
ing prediction with the actual production history.
The framework developed here for automatic decline-curve analysis aims to reduce the number of preprocessing steps. Wells are
not rejected a priori dependent on discontinuities or other features of the production history, as in Gong et al. (2014) and Fulford et al.
(2016). Instead, first, the algorithm is performed for the selected database and generates probabilistic forecasts for all the wells. Then,
the engineer judges which wells presented satisfactory results and which ones require further investigation, saving significant time in
the analysis.
For example, the probabilistic calibration implies that for 10% of the wells, the observed production during the second period will
be higher than the P10 estimate, and for the other 10% of the wells, it will be lower than the P90 estimates. Therefore, it is expected to
observe situations like in Well 9 (Fig. 17), where the P90–P10 range completely missed the production history in the second period.
There are many possibilities for post-treatment in such wells that are not in the scope of this paper and are case dependent, such as man-
ually defining a time window, acquiring bottomhole pressure or tubing head pressure data for superposition calculation, and defining a
more-suitable model using reservoir characteristics and available data.
To improve the robustness of the automatic decline-curve analysis, it is necessary to implement functional forms that reduce the sub-
jectivity involved in tasks such as selecting time windows for history matching or classifying outliers. The fact that the proposed model
is capable of presenting an increasing rate in the beginning of production reduced the need for selection of a time window in several
wells. In addition, the heuristic rules tend to be more important for those wells that would have been initially excluded from the data set
in the previous approaches.
In general, it is not recommended to use the h2 model to estimate properties such as fracture half-length, matrix permeability, or ini-
tial reservoir pressure. Instead, the model is applied solely for production forecast and to compute EUR. The reason is that the parame-
ters are a number of lumped physical quantities (fluid, rock, and completion properties).
In this case study, information from any of these properties is not available. It is possible to formulate an inverse problem to compute
some of these physical quantities if more-detailed information is available, especially if v 0 and boundary-dominated flow has been
observed. However, a large uncertainty is still expected because of the lumped parameters. Another possibility in the case of a more-com-
prehensive data set is to incorporate the information of the parameters and their uncertainty in the prior distribution of each well.
Conclusions
1. The h2 model accounts for the transition from transient to boundary-dominated flow, allows an initial delay and buildup in the pro-
duction rates, and has a finite EUR.
2. If at least 18 months of production history are available, the h2 model has a lower uncertainty than the Arps (1945) hyperbolic model.
3. The h2 model is more conservative than the Arps (1945) hyperbolic, Duong (2011), and stretched exponential models regarding
reserves estimation.

4. The heuristic rules implemented improve the predictability of the models and allow probabilistic calibration.
5. The Bayesian approach with the tuned heuristic rules can effectively estimate uncertainty in reserves, which allows assessment of
risk during the decision-making process.
Well 8, API No: 42121329920000 Well 10, API No: 42121338200000 Well1 1, API No: 42121341100000
1×105
5×104 104
104 5,000
1×104
5,000 5,000
1,000
1,000 500
1,000 500
500
100 100
1 5 10 50 100 1 5 10 50 1 5 10 50
6×104
5×104 5×104
104 4×104
5,000 2×104 3×104
1×104 2×104
1,000
5,000
500
1 5 10 50 1 5 10 50 1 5 10 50
2×104 5,000
104
1×104 5,000
5,000 2,000
1,000
2,000 1,000
500
1 5 10 50 1 5 10 50 1 5 10 50
Time (months)
Production history Best fit P10 P50 P90
Fig. 18—h2 models compared with field data showing evidence of transition to boundary-dominated flow and initial production buildup.
Dry Gas – 426 Wells Wet Gas – 255 Wells Gas Condensate – 245 Wells
5 5 5
4 4 4
PDF
PDF
PDF
3 3 3
2 2 2
1 1 1
0 0 0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
χ (dimensionless) χ (dimensionless) χ (dimensionless)
(a) (b) (c)
Volatile Oil – 23 Wells Indeterminate – 5 Wells Black Oil – 38 Wells
5 5 5
4 4 4
PDF
PDF
3
PDF
3 3
2 2 2
1 1 1
0 0 0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
χ (dimensionless) χ (dimensionless) χ (dimensionless)
(d) (e) (f)
Fig. 19—Histograms for the v parameter considering the best-fit solutions for the full gas-production history and organized by res-
ervoir-fluid type.

θ2 Arps (1945) Hyperbolic Model

History matching 992 wells 109.3 seconds 33.4 seconds
Generating Markov chains of 2,000 samples for the 992 wells 6.90 minutes 11.08 minutes
Full analysis with six hindcasts 1.38 hours 1.85 hours
Table 3—Time elapsed during the automated decline-curve analysis on an average desktop computer.
Nomenclature
1mn ¼ m n matrix of ones
A ¼ reservoir-drainage area (fracture face), L2
aexp ¼ exponential fit parameter
aw ¼ linear-weights parameter
b ¼ Arps decline exponent, dimensionless
bexp ¼ exponential fit parameter
bw ¼ linear-weights parameter
ct ¼ total compressibility (rock and fluid), LT2/M
Ce ¼ covariance matrix of the errors, NtNt
Di ¼ Arps initial decline rate, T–1
k ¼ matrix permeability, L2
L ¼ reservoir length, L
ml ¼ lower shifting parameter
nMCMC ¼ size of Markov chain
Nt ¼ number of timesteps
p ¼ pressure, M/LT2
pi ¼ initial reservoir pressure, M/LT2
pwf ¼ bottomhole flowing pressure, M/LT2
p ¼ average reservoir pressure, M/LT2
Pl ðqobs jcÞ ¼ likelihood function
Ppost ðcjqobs Þ ¼ posterior distribution
Ppr ðcÞ ¼ prior distribution
q ¼ well-flow rate, L3/T
qD ¼ dimensionless flow rate, dimensionless
qlim ¼ threshold valid production rate, L3/T
qmax ¼ maximum well-flow rate, L3/T
qi ¼ virtual initial flow rate, L3/T
Q2nd ¼ production during the second period
t ¼ time vector, Nt1
t ¼ time, T
Vp ¼ total PV, L3
w ¼ vector of weights, Nt1
wi ¼ ith element of w
xi ¼ initial position, L
a ¼ acceptance ratio
b ¼ heuristic multiplier
c ¼ vector of model parameters
e ¼ inherent error of the production data
g ¼ reciprocal characteristic time, T–1
h2 ¼ Jacobi h function No. 2
j ¼ diffusivity constant
l ¼ fluid viscosity, M/LT
r ¼ standard deviation of the residual
/ ¼ matrix porosity, dimensionless
U ¼ cumulative distribution function of N(0,1)
v ¼ geometric factor, dimensionless
x ¼ frequency of an event
Subscripts
bf ¼ best-fit model
f ¼ fracture compartment
low ¼ lower bound
m ¼ matrix compartment
obs ¼ observed data
pred ¼ predicted by model
prop ¼ proposed model
s ¼ sth element in the Markov chain
up ¼ upper bound

Acknowledgments
Authors Rafael Wanderley de Holanda and Eduardo Gildin acknowledge financial support from the Foundation CMG Research Chair
at Texas A&M University. This publication also was made possible by a National Priorities Research Program Grant, NPRP Grant No.
7-1482-1278, from the Qatar National Research Fund (a member of The Qatar Foundation). The authors also thank Thomas A. Blasin-
game and Gorgonio Fuentes-Cruz for several discussions.
References
Al-Hussainy, R., Ramey, H. Jr., and Crawford, P. 1966. The Flow of Real Gases Through Porous Media. J Pet Technol, 18 (5): 624–636. SPE-1243-A-
PA. https://doi.org/10.2118/1243-A-PA.
Arps, J. J. 1945. Analysis of Decline Curves. Trans. AIME 160 (1): 228–247. SPE-945228-G. https://doi.org/10.2118/945228-G.
Bello, R. O. 2009. Rate Transient Analysis in Shale Gas Reservoirs with Transient Linear Behavior. PhD dissertation, Texas A&M University, College
Station, Texas.
Bello, R. O. and Wattenbarger, R. A. 2008. Rate Transient Analysis in Naturally Fractured Shale Gas Reservoirs. Presented at the CIPC/SPE Gas Tech-
nology Symposium 2008 Joint Conference, Calgary, 16–19 June. SPE-114591-MS. https://doi.org/10.2118/114591-MS.
Browning, J., Ikonnikova, S., Gülen, G. et al. 2013. Barnett Shale Production Outlook. SPE Econ & Mgmt 5 (3): 89–104. SPE-165585-PA. https://
doi.org/10.2118/165585-PA.
Burnham, K. P. and Anderson, D. R. 2002. Model Selection and Multimodel Inference. New York City: Springer.
Camacho-Velázquez, R. G. 1987. Well Performance Under Solution Gas Drive. PhD dissertation, University of Tulsa, Tulsa.
Camacho-Velázquez, R. G. and Raghavan, R. 1989. Boundary-Dominated Flow in Solutions-Gas-Drive Reservoirs. SPE Res Eval & Eng 4 (4):
503–512. SPE-18562-PA. https://doi.org/10.2118/18562-PA.
Carslaw, H. S. and Jaeger, J. C. 1959. Conduction of Heat in Solids, second edition. Oxford, UK: Clarendon Press.
Castineira, D., Mondal, A., and Matringe, S. 2014. A New Approach for Fast Evaluations of Large Portfolios of Oil and Gas Fields. Presented at the SPE
Annual Technical Conference and Exhibition, Amsterdam, 27–29 October. SPE-170989-MS. https://doi.org/10.2118/170989-MS.
Chang, C.-P. and Lin, Z.-S. 1999. Stochastic Analysis of Production Decline Data for Production Prediction and Reserves Estimation. J. Pet. Sci. Eng.
23 (3–4): 149–160. https://doi.org/10.1016/S0920-4105(99)00013-3.
Chaudhary, N. L. and Lee, W. J. 2016. Detecting and Removing Outliers in Production Data to Enhance Production Forecasting. Presented at the SPE/
IAEE Hydrocarbon Economics and Evaluation Symposium, Houston, 17–18 May. SPE-179958-MS. https://doi.org/10.2118/179958-MS.
Cheng, Y., Lee, W. J., and McVay, D. A. 2008. Quantification of Uncertainty in Reserve Estimation From Decline Curve Analysis of Production Data
for Unconventional Reservoirs. J. Energy Resour. Technol. 130 (4): 043201-1–043201-6. https://doi.org/10.1115/1.3000096.
Chouikha, A. R. 2005. On Properties of Elliptic Jacobi Functions and Applications. J. Nonlin. Math. Phys. 12 (2): 162–169. https://doi.org/10.2991/
jnmp.2005.12.2.2.
Clarkson, C. R., Behmanesh, H., and Chorney, L. 2013. Production-Data and Pressure-Transient Analysis of Horseshoe Canyon Coalbed-Methane Wells,
Part II: Accounting for Dynamic Skin. J Can Pet Technol 52 (1): 41–53. SPE-148994-PA. https://doi.org/10.2118/148994-PA.
Cronquist, C. 1991. Reserves and Probabilities: Synergism or Anachronism? J Pet Technol 43 (10): 1258–1264. SPE-23586-PA. https://doi.org/10.2118/
23586-PA.
D’Ambroise, J. 2010. Applications of Elliptic and Theta Functions to Friedmann-Robertson-Lemaitre-Walker Cosmology with Cosmological Constant.
In A Window into Zeta and Modular Physics, ed. K. Kirsten and F. L. Williams, Vol. 57, 279–294. New York City: Cambridge University Press.
Drillinginfo. DI Desktop, 1998–2017, www.hpdi.com.
Duong, A. N. 2011. Rate-Decline Analysis for Fracture-Dominated Shale Reservoirs. SPE Res Eval & Eng 14 (3): 377–387. SPE-137748-PA. https://
doi.org/10.2118/137748-PA.
Easley, T. G. 2012. A Nonpressure Dependent Method for Forecasting Rate and Reserves in Linear Flowing Conventional and Unconventional Wells.
Presented at the SPE Hydrocarbon Economics and Evaluation Symposium, Calgary, 24–25 September. SPE-159391-MS. https://doi.org/10.2118/
159391-MS.
Fetkovich, M. 1980. Decline Curve Analysis Using Type Curves. J Pet Technol 32 (6): 1065–1077. SPE-4629-PA. https://doi.org/10.2118/4629-PA.
Fuentes-Cruz, G. and Valkó, P. P. 2015. Revisiting the Dual-Porosity/Dual-Permeability Modeling of Unconventional Reservoirs: The Induced-Inter-
porosity Flow Field. SPE J. 20 (1): 124–141. SPE-173895-PA. https://doi.org/10.2118/173895-PA.
Fulford, D. S. and Blasingame, T. A. 2013. Evaluation of Time-Rate Performance of Shale Wells using the Transient Hyperbolic Relation. Presented at
the SPE Unconventional Resources Conference Canada, Calgary, 5–7 November. SPE-167242-MS. https://doi.org/10.2118/167242-MS.
Fulford, D. S., Bowie, B., Berry, M. E. et al. 2016. Machine Learning as a Reliable Technology for Evaluating Time/Rate Performance of Unconven-
tional Wells. SPE Econ & Mgmt 8 (1): 23–39. SPE-174784-PA. https://doi.org/10.2118/174784-PA.
Gong, X., Gonzalez, R., McVay, D. A. et al. 2014. Bayesian Probabilistic Decline-Curve Analysis Reliably Quantifies Uncertainty in Shale-Well-Pro-
duction Forecasts. SPE J. 19 (6): 1047–1057. SPE-147588-PA. https://doi.org/10.2118/147588-PA.
Hashmi, G., Kabir, C. S., and Hasan, A. R. 2014. Interpretation of Cleanup Data in Gas-Well Testing From Derived Rates. Presented at the SPE Annual
Technical Conference and Exhibition, Amsterdam, 27–29 October. SPE-170603-MS. https://doi.org/10.2118/170603-MS.
Igor, M. 2007. Theta Function of Four Types, MathWorks, Inc., http://www.mathworks.com/matlabcentral/fileexchange/18140-theta-function-of-four-
types (accessed 13 January 2017).
Johansson, F. et al. 2013. mpmath: A Python Library for Arbitrary-Precision Floating-Point Arithmetic, Version 0.18, http://mpmath.org (accessed 18
February 2017).
Kuchuk, F., Morton, K., and Biryukov, D. 2016. Rate-Transient Analysis for Multistage Fractured Horizontal Wells in Conventional and Un-Conven-
tional Homogeneous and Naturally Fractured Reservoirs. Presented at the SPE Annual Technical Conference and Exhibition, Dubai, 26–28 Septem-
ber. SPE-181488-MS. https://doi.org/10.2118/181488-MS.
Larsen, L. and Kviljo, K. 1990. Variable-Skin and Cleanup Effects in Well-Test Data. SPE Form Eval 5 (3): 272–276. SPE-15581-PA. https://doi.org/
10.2118/15581-PA.
Lee, W. J. and Sidle, R. 2010. Gas-Reserves Estimation in Resource Plays. SPE Econ & Mgmt 2 (2): 86–91. SPE-130102-PA. https://doi.org/10.2118/
130102-PA.
Ogunyomi, B. A., Patzek, T. W., Lake, L. W. et al. 2016. History Matching and Rate Forecasting in Unconventional Oil Reservoirs With an Approximate
Analytical Solution to the Double-Porosity Model. SPE Res Eval & Eng 19 (1): 70–82. SPE-171031-PA. https://doi.org/10.2118/171031-PA.
Oliver, D. S., Reynolds, A. C., and Liu, N. 2008. Inverse Theory for Petroleum Reservoir Characterization and History Matching. New York City: Cam-
bridge University Press.
Parshall, J. 2008. Barnett Shale Showcases Tight-Gas Development. J Pet Technol 60 (9): 48–55. SPE-0908-0048-JPT. https://doi.org/10.2118/0908-
0048-JPT.

Purvis, D. C. and Kuzma, H. 2016. Evolution of Uncertainty Methods in Decline Curve Analysis. Presented at the SPE/IAEE Hydrocarbon Economics
and Evaluation Symposium, Houston, 17–18 May. SPE-179980-MS. https://doi.org/10.2118/179980-MS.
Robertson, S. 1988. Generalized Hyperbolic Equation. SPE-18731-MS.
Shahamat, M. S., Mattar, L., and Aguilera, R. 2015. A Physics-Based Method To Forecast Production From Tight and Shale Petroleum Reservoirs by
Use of Succession of Pseudosteady States. SPE Res Eval & Eng 18 (4): 508–522. SPE-167686-PA. https://doi.org/10.2118/167686-PA.
Sidle, R. E. and Lee, W. J. 2010. The Demonstration of a “Reliable Technology” for Estimating Oil and Gas Reserves. Presented at the SPE Hydrocarbon
Economics and Evaluation Symposium, Dallas, 8–9 March. SPE-129689-MS. https://doi.org/10.2118/129689-MS.
Sidle, R. E. and Lee, W. J. 2016. An Update on Demonstrating “Reliable Technology” - Where are We Now? Presented at the SPE/IAEE Hydrocarbon
Economics and Evaluation Symposium, Houston, 17–18 May. SPE-179991-MS. https://doi.org/10.2118/179991-MS.
Tyurin, A. 2002. Quantization, Classical and Quantum Field Theory and Theta Functions. Washington, DC: American Mathematical Society
Valkó, P. P. 2009. Assigning Value to Stimulation in the Barnett Shale: A Simultaneous Analysis of 7000 Plus Production Histories and Well Comple-
tion Records. Presented at the SPE Hydraulic Fracturing Technology Conference, The Woodlands, Texas, 19–21 January. SPE-119369-MS. https://
doi.org/10.2118/119369-MS.
Valkó, P. P. and Lee, W. J. 2010. A Better Way To Forecast Production From Unconventional Gas Wells. Presented at the SPE Annual Technical Con-
ference and Exhibition, Florence, Italy, 19–22 September. SPE-134231-MS. https://doi.org/10.2118/134231-MS.
Warren, J. E. and Root, P. J. 1963. The Behavior of Naturally Fractured Reservoirs. SPE J. 3 (3): 245–255. SPE-426-PA. https://doi.org/10.2118/426-
PA.
Wattenbarger, R. A., El-Banbi, A. H., Villegas, M. E. et al. 1998. Production Analysis of Linear Flow Into Fractured Tight Gas Wells. Presented at the
SPE Rocky Mountain Regional/Low-Permeability Reservoirs Symposium, Denver, 5–8 April. SPE-39931-MS. https://doi.org/10.2118/39931-MS.
Weijermars, R., Sorek, N., Sen, D. et al. 2017. Eagle Ford Shale Play Economics: US versus Mexico. J. Nat. Gas Sci. Eng. 38 (February): 345–372.
https://doi.org/10.1016/j.jngse.2016.12.009.
Wolfram Research, Inc. 1988. EllipticTheta, http://reference.wolfram.com/language/ref/EllipticTheta.html (accessed 13 January 2017).
Wolfram Research, Inc. 2015. Mathematica, Version 10.3, Champaign, IL: Wolfram Research.
Yu, W., Tan, X., Zuo, L. et al. 2016. A New Probabilistic Approach for Uncertainty Quantification in Well Performance of Shale Gas Reservoirs. SPE J.
21 (6): 2,038–2,048. SPE-183651-PA. https://doi.org/10.2118/183651-PA.
Zhao, Y.-L., Zhang, L.-H., Zhao, J.-Z. et al. 2013. “Triple porosity” Modeling of Transient Well Test and Rate Decline Analysis for Multi-Fractured
Horizontal Well in Shale Gas Reservoirs. J. Pet. Sci. Eng. 110 (October): 253–262. https://doi.org/10.1016/j.petrol.2013.09.006.
Appendix A Derivation of Pressure Solution

The diffusivity equation for a 1D homogeneous reservoir in linear flow is given by
@ 2 p /lct @p
¼ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-1Þ
@x2 k @t
Consider that the pressure in the matrix is initially in equilibrium:
pðx; 0Þ ¼ pi : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-2Þ
In addition, assume that the pressure drop in the fracture is negligible; i.e., the fracture is infinitely conductive, and pwf is held con-
stant. Thus, the following boundary condition applies:
pð0; tÞ ¼ pwf : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-3Þ
A no-flow boundary defines the other end of the drainage volume:

@pðL; tÞ
¼ 0: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-4Þ
@t
Once the partial-differential equation and the initial and boundary conditions are defined, the problem can be solved with the proper
mathematical manipulation. First, it is necessary to change variables so that the problem becomes more tractable; in this case, the pres-
sure function is redefined as
uðx; tÞ ¼ pðx; tÞ pwf : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-5Þ
Therefore, Eq. A-1 becomes

@ 2 u @u
j ¼ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-6Þ
@x2 @t
k
where j is the diffusivity constant; i.e., j ¼ . The initial and boundary conditions (Eqs. A-2, A-3, and A-4, respectively) can be
rewritten as /lc t
uðx; 0Þ ¼ pi pwf ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-7Þ
uð0; tÞ ¼ 0; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-8Þ
@uðL; tÞ
¼ 0: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-9Þ
@t
Assume that the solution of the problem can be written as the product of two functions, f(x) and g(t):
uðx; tÞ ¼ f ðxÞgðtÞ: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-10Þ

Then, Eq. A-6 can be rearranged so that each side is a function of only one independent variable (x or t), which means that each side
is equal to a constant (–k):
1 dgðtÞ 1 d2 f ðxÞ
¼ ¼ k: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-11Þ
jgðtÞ dt f ðxÞ dx2
This allows us to solve each side of Eq. A-11 independently. Starting from the spatial function, f(x), if k > 0, the solution has the
form
pffiffiffi pffiffiffi
f ðxÞ ¼ c1 cosð kxÞ þ c2 sinð kxÞ: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-12Þ
Applying Eq. A-8,

f ð0Þ ¼ 0; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-13Þ
c1 cosð0Þ ¼ 0; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-14Þ
c1 ¼ 0: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-15Þ
Applying Eq. A-9,

df ðLÞ
¼ 0; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-16Þ
dx
pffiffiffi pffiffiffi
kc2 cosð kLÞ ¼ 0; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-17Þ
where c2=0 to have a nontrivial solution. Thus, the cosine term must equal zero, which implies
p pffiffiffi
þ np ¼ kL; n 2 Z; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-18Þ
2
2
1 p
k¼ þ np : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-19Þ
L 2
Therefore, each n value provides a function:
pffiffiffi h x p i
fn ðxÞ ¼ kc2 sin þ np : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-20Þ
L 2
If k ¼ 0, the spatial function has the form
f ðxÞ ¼ c3 þ c4 x: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-21Þ
Applying the boundary conditions (Eqs. A-13 and A-16),
c3 ¼ c4 ¼ 0: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-22Þ
Thus, if k ¼ 0, only a trivial solution is obtained:
f0 ðxÞ ¼ 0: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-23Þ
Analogously, if k < 0, it can be proved that only a trivial solution is obtained.

From Eq. A-11, the solution of the first-order ordinary-differential equation for the temporal function is easily obtained:
j p 2
þ np t
gn ðtÞ ¼ c5 ejk t ¼ c5 e L2 2 : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-24Þ
Combining Eqs. A-10, A-20, and A-24, the solution in a series form is given by
X
1 j p p2 hxi
þ np t
uðx; tÞ ¼ cn e L2 2 þ np ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-25Þ
sin
n¼0
L 2
pffiffiffi c2 c5 p
where cn ¼ c2 c5 k ¼ þ np . Notice that the negative values of n were neglected in Eq. A-25, because they result in fn(x) < 0,
L 2
which when considered individually as a solution results in u(x,t) < 0 and p(x,t) < pwf, which does not correspond to the physics of
the problem.
The last step to obtain the solution is to determine cn. Applying the initial condition (Eq. A-7),
X
1 h x p i
uðx; 0Þ ¼ cn sin þ np ¼ pi pwf : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-26Þ
n¼0
L 2
Then, the coefficients cn can be determined as the Fourier sine series of the initial condition. The following equations result from the
orthogonality of the sines:
ð h x p i
2 L
cn ¼ ðpi pwf Þsin þ np dx; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-27Þ
L 0 L 2

4ðpi pwf Þ
cn ¼ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-28Þ
pð1 þ 2nÞ
The solution for Eq. A-6 is finally obtained as
X
1
4 ðpi pwf Þ 2 hp x i
ej½2Lð1þ2nÞ t sin
p
uðx; tÞ ¼ ð1 þ 2nÞ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-29Þ
n¼0
p ð1 þ 2nÞ 2L
Substituting Eq. A-29 into Eq. A-5, an analytic solution for the pressure equation in the 1D reservoir is obtained:
X
1
4 ðpi pwf Þ 2 hp x i
ej½2Lð1þ2nÞ t sin
p
pðx; tÞ ¼ pwf þ ð1 þ 2nÞ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-30Þ
n¼0
p ð1 þ 2nÞ 2L
Appendix B Finite EUR

Integrating the material-balance equation (Eq. 1) and considering the initial condition (Eq. A-2) and that p ! pwf as t !1, the EUR
when v ¼ 0 can be obtained as
ð pwf ð1
ct Vp dp ¼ qðtÞdt ¼ EUR; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðB-1Þ
pi 0
EUR ¼ ct Vp ðpi pwf Þ: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðB-2Þ
From the definitions in Eqs. 4, 8, and 10, the EUR can be expressed in terms of qi and g:
qi
EUR ¼ p2 : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðB-3Þ
g
To prove that there is a finite EUR for any value of v, it is necessary to take into account that 0 v p2 and the cosine term in Eq. 6
is bounded:
1 cos½vð1 þ 2nÞ 1: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðB-4Þ
As one can realize, for any value of n, the maximum for this cosine term is obtained when v ¼ 0. In addition, the production rates
must be positive:
qðtÞ 0: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðB-5Þ
As a result of Eqs. B-4 and B-5, when comparing models with the same vales of qi and g, the following applies:
0 qðv; tÞ qð0; tÞ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðB-6Þ
which leads to
EURðv; tÞ EURð0; tÞ: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðB-7Þ
Because EUR(0, t) is finite, EUR(v, t) is finite also. This result can be confirmed in Fig. 4b and is intuitive from the modified defini-
tion of the average pressure (Eq. 2).
Rafael Wanderley de Holanda is a PhD degree candidate and research assistant in the Harold Vance Department of Petroleum
Engineering at Texas A&M University. His research interests include reservoir dynamics and control, physics-based decline-curve
analysis, capacitance/resistance models, uncertainty quantification, and reservoir-production optimization. Holando holds a
master’s degree in petroleum engineering from Texas A&M University and a bachelor’s degree in chemical engineering from
Universidade Federal de Pernambuco, Brazil.
Eduardo Gildin is an associate professor in the Harold Vance Department of Petroleum Engineering at Texas A&M University. He
is the holder of the Ted H. Smith ’75 and Max R. Vordenbaum ’73 DVG Developmental Professorship and the Foundation CMG
Research Chair in Robust Reduced Complexity Modeling in Reservoir Engineering. Gildin’s research interests include the mathe-
matics of reservoir simulation, numerical methods for control and model reduction of large-scale systems, finite-element model-
ing, numerical analysis, and optimization with an emphasis on petroleum-engineering problems. He holds a PhD degree in
aerospace engineering from the University of Texas at Austin and master’s in mechanical and mechatronics engineering from
Universidade de São Paulo (USP), Brazil, and bachelor’s degree in mechanical engineering from Faculdade de Engenharia
Industrial (FEI), Brazil.
Peter P. Valkó is professor and holder of the R. L. Whiting Chair in the Department of Petroleum Engineering at Texas A&M Univer-
sity. Before joining Texas A&M University in 1993, he was a researcher at the Mining University, Leoben, worked for MOL (Hungar-
ian Oil Company), and was a faculty member at Eötvös Lorand University, Budapest. Valkó’s research interests include design
and evaluation of hydraulic-fracturing treatments, performance of stimulated wells, and numerical methods for Laplace inver-
sion. A native Hungarian, he holds bachelor’s and master’s equivalent degrees from Hungary, and a PhD equivalent degree
from the Institute of Catalysis, Novosibirsk, Russia.

Combining Phy, Stats and Heuristicsw

Caricato da

Informazioni sul documento

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Combining Phy, Stats and Heuristicsw

Caricato da

Copyright:

Formati disponibili

REE185589 DOI: 10.

2118/185589-PA Date: 5-January-18 Stage: Page: 1 Total Pages: 20

Combining Physics, Statistics, and

2018 SPE Reservoir Evaluation & Engineering 1

ID: jaganm Time: 14:40 I Path: S:/REE#/Vol00000/170036/Comp/APPFile/SA-REE#170036

Physics: Jacobi h Function No. 2 as a Decline-Curve Model

2 2018 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 14:40 I Path: S:/REE#/Vol00000/170036/Comp/APPFile/SA-REE#170036

Linear flow No-flow boundary

Initial condition: p (x,0) = pi

The definition of average reservoir pressure is modified to

The drainage PV is defined as

and the diffusivity constant as

which corresponds to the second Jacobi h function (h2),

qðtÞ ¼ qi h2 ðv; egt Þ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð7Þ

where qi is the virtual initial rate,

2018 SPE Reservoir Evaluation & Engineering 3

ID: jaganm Time: 14:40 I Path: S:/REE#/Vol00000/170036/Comp/APPFile/SA-REE#170036

qi,f* = 50000, ηf = 100, qi,m

4 2018 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 14:40 I Path: S:/REE#/Vol00000/170036/Comp/APPFile/SA-REE#170036

θ2 (χ, e –η t ), χ = 0.22 θ2 (χ, e –η t ), η = 0.053

2018 SPE Reservoir Evaluation & Engineering 5

ID: jaganm Time: 14:40 I Path: S:/REE#/Vol00000/170036/Comp/APPFile/SA-REE#170036

θ2 Arps (1945) Hyperbolic Model

χ vs. η – Solution Space

Statistics: Uncertainty Analysis

6 2018 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 14:40 I Path: S:/REE#/Vol00000/170036/Comp/APPFile/SA-REE#170036

Joint Prior Distribution Likelihood Function Posterior Distribution

30,000 30,000 30,000 0.8

20,000 20,000 0.6

10,000 10,000 10,000 0.4

100 100 100

100 100 100

100 100 100

Production history θ2 (bestfit) Filter line

2018 SPE Reservoir Evaluation & Engineering 7

ID: jaganm Time: 14:40 I Path: S:/REE#/Vol00000/170036/Comp/APPFile/SA-REE#170036

Heuristics: Treating the Bad Data

winitial ¼ ½w1 ; w2 ; …; wi ; …; wNt T ¼ 1Nt 1 : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð20Þ

w ¼ ½ fw ðw1 ; t1 Þ; …; fw ðwi ; ti Þ; …; fw ðwNt ; tNt ÞT : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð21Þ

In this paper, fw ðwi ; ti Þ is a linear function, so Eq. 21 can be written as

logqexp ðtÞ ¼ aexp t þ bexp : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð23Þ

This line is shifted downward by introducing the multiplier ml:

logqexp;l ðtÞ ¼ aexp t þ ml bexp ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð24Þ

8 2018 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 14:40 I Path: S:/REE#/Vol00000/170036/Comp/APPFile/SA-REE#170036

and bw ¼ 0 was defined to reduce the number of parameters in the problem.

Case Study: 992 Barnett Shale Gas Wells

2018 SPE Reservoir Evaluation & Engineering 9

ID: jaganm Time: 14:40 I Path: S:/REE#/Vol00000/170036/Comp/APPFile/SA-REE#170036

Frequency of True PDTSP > Probabilistic 0.8

Fig. 9—Case with adjusted heuristic rules for probabilistic calibration.

Fluid Classification Based on Initial Producing Gas – Liquid Ratio

107 Dry gas – 426 wells (42.94%)

0 200 400 600 800

Table 2—Heuristic parameters for probabilistically calibrated models.

10 2018 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 14:40 I Path: S:/REE#/Vol00000/170036/Comp/APPFile/SA-REE#170036

Inverse Gaussian (0.072,0.085) Gamma (2.05,0.16) Inverse Gamma (3.30,0.78)

0.54 Uniform (0.001,2) + (0.34 Half-Normal (0.93) + Inverse Gamma (8.11,9.89)

2018 SPE Reservoir Evaluation & Engineering 11

ID: jaganm Time: 14:40 I Path: S:/REE#/Vol00000/170036/Comp/APPFile/SA-REE#170036

winitial ¼ ½w1 ; w2 ; …; wi ; …; wNt T ¼ 1Nt 1 : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð20Þ

w ¼ ½ fw ðw1 ; t1 Þ; …; fw ðwi ; ti Þ; …; fw ðwNt ; tNt ÞT : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð21Þ

1 cos½vð1 þ 2nÞ 1: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðB-4Þ