Dynamic Flow Analysis - v4.

02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p1/24

4 – Production Analysis (PA)

OH – OSF - DV

4.A Introduction and a brief history

Production Analysis (PA) is seen as a natural complement of Pressure Transient Analysis when
one has pressure and rate data and wants to make some kind of analysis. This is the case with
the spread of Permanent Downhole Gauge (PDG) data, which contain candidate data for both
analysis techniques.

PA and PTA share a large technical kernel, and are often performed by the same engineers.
This was not always the case, and we will start with a short history of PA.

PA started in the 1920s on a purely empirical basis, and as a financial tool. There was no
technical background to these relations, the objective was to find the right decline function
that fit the past history and would be able to assess the US$ revenue in the future.

In the 1940s, the formulation of constant pressure exponential, hyperbolic and harmonic rate
decline was published (Arps, 1945). This was still partly empirical, but some parameters could
be quantified using specific analyses.

In the 1960s came the first series of type-curves, still assuming constant flowing pressure. The
Fetkovich type-curve combined two families of curves: one for the transient flowing period and
one for the late time boundary dominated response. Ten years later Carter extended it to the
gas case. Other type-curves were later published to address further complex configurations
including layered and fractured reservoirs. This was done in parallel to the theoretical work
done in PTA.

At this stage the methodology was somewhat equivalent to the standard procedure in PTA in
the late 1970s. The Arps plot was the counterpart of the Horner plot, and the constant
pressure type-curves were the counterpart of the well test drawdown type-curves.

Fig. 4.A.1 - Arps plot Fig. 4.A.2 - Fetkovich type-curve

Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p2/24

As we have seen the introduction of the Bourdet derivative and PCs dramatically changed PTA
in the 1980s and 1990s. This did not happen as fast in PA, where most work continued to be
done using Arps and Fetkovich methods, generally as side applications linked to the production
databases. Unlike PTA, classical methodology in PA was not phased out. In many mature
reservoirs, permanent gauges cannot be justified economically, and PA methods will stay as
they are, because there are generally no data to justify a more sophisticated approach.

However the theory had evolved in ways akin to those in PTA. Blasingame et al. introduced a
variable rates/variable pressure type-curve as a loglog plot of productivity index vs. material
balance time, complemented by the equivalent of the Bourdet derivative. An adapted version
of the loglog plot, where rate-normalized pressure replaces the productivity index, was also
published. Additional solutions accounted for various well and reservoir configurations. So the
modern tools were available in theory before the end of the 1980s, but they were only recently
implemented in commercial PA applications, such as Topaze.

Fig. 4.A.3 - Blasingame plot Fig. 4.A.4 - Loglog plot

Advancements in PA started to progress in the late 1990s and early 2000s, partly due to the
development of permanent pressure gauges. When engineers started receiving long term
continuous pressure data, the first reaction was to load this into a PTA program: ‘I have rates;
I have pressures, so I treat this as a well test’. However PTA methodology was not designed
for this type of data, and engineers would sometimes perform incorrect interpretation by
overlooking specific assumptions that are no longer valid on the time scale of a permanent
gauge survey. Material balance errors and over-simplifications using Perrine’s approach for
multiphase flow property evaluation were, and are, among the most frequently
encountered errors.

4.B The old stuff

4.B.1 Arps
Decline Curve Analysis methods, as formalized by Arps in 1945, have been for many years the
conventional technique for analysis and forecasting of well production data. Decline type-
curves are based on an empirical rate-time and associated cumulative-time equation, which
can be expressed in the general form:
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p3/24

q( t ) =
qi
Q( t ) =
qi b
Di (1 − b)
(
qi1−b − q ( t )
1−b
)
[1 + bD t ]
1
b
i

where: qi is the initial rate, Di is the decline factor, and b a parameter varying between 0 and
1, defining the decline type. Three types are usually considered: hyperbolic, exponential and
harmonic.

Exponential decline, b = 0

It can be shown that the general decline equation tends to an exponential decline when b
tends to 0:

qi − q( t )
q( t ) = qi e − Di t Q( t ) =
Di

Harmonic decline, b =1
qi qi ⎛ qi ⎞
q( t ) = Q(t ) = ln⎜ ⎟
[1 + D t ]i
Di ⎝ q( t ) ⎠

Hyperbolic decline, b ∈ ]0,1[

The expressions are those above.

Decline curve equations are applicable only after the transient part of the response has ceased,
i.e. during boundary dominated flow. A general approach consists in the determination of the
three parameters directly, by non-linear regression. The traditional usage however, is centered
on the use of some specific presentations where linearity is sought after the value of b has
been fixed. Practically, the following scales/linearity can be used:

log(q ) vs t : Linear plot if the decline is exponential, concave upward otherwise.

q vs Q : Linear plot if the decline is exponential, concave upward otherwise.

log(q ) vs Q : Linear plot if the decline is harmonic, concave downward otherwise.

Most PA software allows the scale to be set to the above and more. The regression of the
decline is non linear, i.e. it is possible to have the value of b determined from the regression
rather than assuming a value of b linked to the particular scale.
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p4/24

Fig. 4.B.1 – Arps

Once the decline parameters have been obtained, and since the analytical expression of the
rate and cumulative are known, it is possible, from a given abandonment rate, to calculate the
corresponding abandonment time, and hence the recovery at abandonment.

qa
The abandonment rate is usually defined as either q a or the ratio . The abandonment time
qi
( )
is noted t a and the recovery at abandonment N p t a . Figure 4.B.2 illustrates this extrapolation
of the Arps plot.
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p5/24

Fig. 4.B.2 - Abandonment

Exponential decline is widely used because of the simplicity of the associated graphical
methods. It leads to conservative reserves estimates. Besides, it can be demonstrated that
exponential decline is the late time behavior of a constant pressure production in a closed
reservoir, with a slightly compressible fluid assumption.

The equation governing the PSS behaviour is:

Δp = mQ + bq

With Δp = pi − p w

Q = cumulative production
q : instantaneous production rate

1
and m =
Nct

Differentiating the two terms of the equation with respect to the time:

dΔp dQ dq
=m +b
dt dt dt
Under constant production well pressure conditions

dΔp dQ dq
=0=m +b
dt dt dt
We have

dQ
=q
dt
Therefore
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p6/24

dq m dq m
= − q or = − dt
dt b q b

dq m m
∫ q
= − ∫ dt or ln q = − t + cst
b b

m m
q = exp(− t + cst ) that can be written q = qi exp(− t )
b b

There are many situations however, where the general hyperbolic decline is more adequate.
This is the case in solution gas drive reservoirs.

In our opinion, and given the power of the non-linear regression, it is better to try and
determine all three parameters, including b, systematically. Above all, it is important to stress
that decline curves have many limitations:

• The bottom-hole pressure must be fairly constant.

• The well behavior is assumed constant, e.g. no change in skin with time.
• The drainage area of the considered well is constant, i.e. the producing behavior of
surrounding wells must also be stabilized.
A refinement can be made for the case where the decline in the oil rate is caused by an
increase in the water cut, mostly in water drive reservoirs with unfavourable mobility ratio. If
one replaces the oil rate by the oil cut, fo the Arps equation can be used for wells with variable
gross production. The same plots can be made and extrapolated: log(fo) vs t, fo vs Q and
log(fo) vs Q.

4.B.2 Fetkovich
In 1980, Fetkovich introduced a type-curve combining the theoretical response of a well in a
closed reservoir, and the standard Arps decline curves. The motivation behind this work was to
come up with a loglog matching technique applicable to both the transient part of the data and
the boundary dominated flow period. By presenting both periods, the Type-Curve would avoid
incorrectly matching transient data on decline curves.

A determining preliminary step was that the exponential decline can be shown to be the long-
term solution of the constant pressure case. The Fetkovich type-curve is derived assuming a
slightly compressible fluid and constant flowing pressure. Extension to gas can be made with
the appropriate dimensionless rate expression, as described below. The original type-curve
presented by Fetkovich displayed rate only. A composite presentation including the cumulative
was later introduced to bring more confidence in the matching process and to reduce the effect
of the noise.
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p7/24

Fig. 4.B.3 – Fetkovich type curve

In Figure 4.B.3 the left region of the curves (green to blue) corresponds to the transient part
of the response. On the right hand side, are the Arps decline curves (red to yellow). Note the
legend on the left: the red Arps curve is for an exponential decline (b=0), the last yellow curve
is for a harmonic decline (b=1).

The Fetkovich type-curve displays dimensionless values qDd, QDd versus tDd as defined below.
The dimensionless variables can be expressed in terms of the Arps decline curve parameters,
or in terms of the transient response parameters. The duality is due to the composite nature of
the type-curve showing once again a merge of a theoretical closed reservoir response, and the
empirical Arps stems.

All equations below are in Oil Field Units.

Time

Decline curve dimensionless time: t Dd = Di t

0.00634 kt
Dimensionless time: tD =
φμct rw2
tD
Related by: t Dd =
1 ⎡⎛ re ⎞ ⎤⎡ ⎛ r ⎞ 1 ⎤
2

⎢⎜ ⎟ − 1⎥ ⎢ln⎜ e ⎟ − ⎥
2 ⎢⎣⎝ rw ⎠ ⎥⎦ ⎣ ⎝ rw ⎠ 2 ⎦

Rate

q( t )
Decline curve dimensionless rate: q Dd =
qi

. q( t ) μB
1412
Dimensionless flow rate, oil: q D oil =
kh( pi − p w )
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p8/24

And the equivalent expression for gas is:

50300Tq( t ) psc
p
2 pdp
q D gas =
(
Tsc kh m( pi ) − m( p w ))
; with m( p) = ∫0
μZ

⎡ ⎛ r ⎞ 1⎤
q Dd and q D are related by: q Dd = q D ⎢ln⎜ e ⎟ − ⎥
⎣ ⎝ rw ⎠ 2 ⎦

Cumulative production

Q( t )
Decline curve dimensionless cumulative: QDd =
Npi
Where Npi defines the ultimate recovery.

A match will bring values of re and kh, Di and qi. The type of decline, b is not linked to any of
the match ratios, obtained by selecting the correct type-curve. From the external boundary
distance, the reservoir pore volume can be calculated. From the Arps parameters, the future
performance can be forecast; Npi can be calculated as well as Np for any specified
abandonment rate.

Fetkovich type-curve and gas

It is worth noting that specific methods or extensions have been studied for gas production,
such as the Carter type-curve. When using the Fetkovich type curve with gas, the expression
of the rate match is modified to use the pseudo pressure m(p), in a manner analogous to what
is done in Pressure Transient Analysis. In addition, Fraim and Wattenbarger suggested the use
of a normalized time rather than time itself, with the following definition:

t
(μct )i
tn = ∫
0 ( ) ( )dt
μ p ct p

The gas diffusion equation can then be re-written in terms of pseudo pressure m(p):

∂m( p) k
= 0.0002637 ∇ 2 m( p )
∂t Φμct

When the pressure varies significantly, the term μct and therefore the diffusion term varies, If
we introduce the pseudotime:

t (μct )i
t n (t ) = ∫ I ( p wf (τ ))dτ I ( p) =
0
where
()()
μ p ct p
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p9/24

∂m( p) k
The diffusion equation becomes: = 0.0002637 ∇ 2 m( p )
∂t n Φ (μct )i

Note : the use of the normalised pseudo time requires the knowledge of the average pressure
at each step, that makes its use difficult.

They demonstrated that the real gas constant pressure solution, when plotted using
normalized time, exhibits a boundary-dominated period that matches the exponential decline.
The major drawback however, is that the normalized time expression requires an a priori
knowledge of the average pressure, hence of the reserves. The method is thus iterative in
nature. Note: the normalized time is not used; hence a gas response in depletion mode will not
necessarily follow the exponential decline.

Using Fetkovich type-curves

It is important to remember that the Fetkovich type-curve is based, for the depletion part, on
the Arps decline curves. Like the decline curves it suffers some limitations:

- It is assumed that the bottom-hole pressure is fairly constant. Fetkovich suggests that if
the pressure is smooth, and uniformly decreasing, one could use a Δp normalized rate.
- The well behavior is assumed constant, e.g. no change in skin with time.
- The drainage area of the considered well is constant, i.e. the producing behavior of
surrounding wells must also be stabilized.

4.C The right stuff

4.C.1 Blasingame plot
Previous sections have described the conventional Decline Curve Analysis methods, and their
extension using Fetkovich type-curves. We recall that the latter were obtained by combining a
theoretical model response and the empirical decline stems of Arps.

Broadly speaking, one could say that the introduction of type-curve matching techniques in
production analysis has opened the way to applying methods developed for well test
interpretation to the analysis of production data. The main limitation in the Fetkovich type-
curve is the assumption of constant flowing pressure. Blasingame and McCray noted that using
a pressure normalized flow rate when the bottom-hole pressure varies significantly did not
remedy the problem. They sought functions that would transform the variable
pressures/variable rates solution into an equivalent constant pressure or constant rate
solution. They introduced two specific time functions, tcr the constant rate time analogy, and tcp
for constant pressure. For the liquid case, the constant rate analogy time function is defined as
the ratio of the cumulative and the flow rate:

Q( t )
t cr =
q( t )
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p10/24

q( t )
When the normalized rate is plotted versus this function on a loglog scale, the
pi − p w ( t )
boundary dominated flow period follows a negative unit slope line:

The Black Oil Pseudo Steady State flow rate equation is:

qo 1
=
Δp bo , pss + mo , pss t
With:

1 Bo
mo , pss = ;
Nct Boi

μ o Bo ⎡ 1⎡ 4 1 A⎤ ⎤
bo , pss = 141.2 ⎢ ln ⎢ γ ⎥ + s⎥ ;
kh ⎣ 2 ⎣ e C A rw2 ⎦ ⎦

Np
and t =
qo

qo
When the Pseudo Steady States dominates is function of t at exponent (-1),
Δp

qo
Therefore a loglog plot of vs t will show a negative unit slope straight line
Δp

Np
Note: Periods of very low rates give artificially high values of t = then the equation
qo
qo 1 qo 1
= tends to = the points are found on the same -1 unit slope
Δp bo , pss + mo , pss t Δp mo , pss t
straight line.

Based on this result, Palacio and Blasingame introduced type-curves that could be used for
variable flowing pressure conditions. In order to improve the type-curve analysis the Bourdet
derivative was also considered. However, due to the noise inherent to the production data, the
derivative was not applied to the normalized flow itself but to its integral. More precisely, the
Palacio-Blasingame type-curve plot shows the following:

Normalized rate:

q(t )
PI ( t ) =
pi − p w (t )
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p11/24

Normalized rate integral:

q(τ )
te te
1 1
PI Int .=
te ∫0 PI (τ )dτ = t e ∫p
0 i − p w (τ )
dτ

Normalized rate integral derivative:

∂ (PI Int )
PI Int. Derivative =
∂ ln (te )

All three curves are plotted against t e on a loglog scale as illustrated in Figure 4.C.1.

Fig. 4.C.1 – Blasingame plot

The traditional method of using this presentation is in conjunction with type-curves for a
particular well model. This plot is used as a diagnostic tool, where the data and a model
response are compared. The model can be any model, analytical or numerical, single or multi-
well, etc. One can either display the ‘true’ model response, i.e. the response to the full
pressure history, or the response to a single pressure step. The single step response shows the
signature of the model in a clear, and usable form, whereas the response to the real history is
usually very erratic, because the equivalent time is jumping back and forth in time.

Nonlinear cases

q( t )
The cornerstone of the Blasingame plot is the linearity between the normalized rate
pi − p w ( t )
1
and during boundary dominated flow. This relation, valid for a slightly compressible fluid,
te
does not apply to gas unless the rate is normalized with respect to Δm p and the time is ( )
modified as follows:
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p12/24

μ gi cti t qg (t )
qg (0) ∫0 μ g ( p )ct ( p )
tegas = dt

qo 1
In the PSS oil flow rate equation = , the slope mo , pss is a function of the fluid
Δp bo , pss + mo , pss t
compressibility, which is highly pressure dependent in gas cases.

The objective is to keep the linearity and the PSS flow rate equation for gas in the same shape
as for oil,:

qg 1
= with a constant slope mg , pss
Δp p bg , pss + mg , pss t egas

We take the varying viscosity and compressibility into account by introducing the ‘pseudo’
pressure and ‘pseudo’ normalized time:

μ gi zi p p
pp =
pi ∫ pbase μg z
dp

μ gi cti t q g (t )
(t ) ∫ μ
t egas = dt
qg 0
g ( p ) ct ( p )

1
The slope is then mg , pss =
Gct

the intersect becomes:

μ gi Bgi ⎡ 1 ⎡ 4 1 A⎤ ⎤
bg , pss = 141.2 ⎢ ln ⎢ γ 2⎥
+ s⎥
kh ⎣ 2 ⎣ e C A rw ⎦ ⎦

However, this time function is not used in Topaze. The consequence is that the linearities
expected during the various flow regimes may be distorted. It is important to realize however,
that the Blasingame plot only provides a basis for comparing the data and a model response
calculated for the particular case. Handling the non linearities is the model responsibility, not
that of the plot. If the model is representative, the model and the data will be consistent. So it
does not matter whether such or such period exhibits a linearity or not. When using a type
curve this is different as the type curve embeds modelling assumptions.
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p13/24

4.C.2 Loglog plot

By replacing the time with an equivalent time, defined as the ratio of the cumulative to
the flow rate, one can transform a variable flowing pressure test into a constant rate
equivalent, at least for a liquid case. The parallel with constant rate solution can be taken
one step further if, rather than working with a pressure-normalized rate, we work with
rate-normalized pressure.

pi − pw ( t ) Q( t )
In other words for the liquid case, if we plot versus t e = on a loglog scale the
q( t ) q( t )
boundary dominated flow will exhibit a unit slope line, similar to pseudo-steady state in
Pressure Transient Analysis. Furthermore, if we take the derivative of the normalized pressure
with respect to the logarithm of t e , the transient part will exhibit a stabilization at a level
linked to the mobility.

Fig. 4.C.2 – Loglog plot

The similarity with PTA is thus complete. Yet, the noise level on the derivative is usually too
high, as illustrated in Figure 4.C.2. One workaround is to work with a normalized pressure
integral, in a manner analogous to what was done on the Palacio-Blasingame type-curves.

pi − p w (τ )
I (te ) =
1 te
Integral of normalized pressure:
te ∫
o q(τ )
dτ

∂I (te )
Bourdet derivative of the Integral of normalized pressure: I ' (te ) =
∂ ln (te )
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p14/24

Figure 4.C.3 – Loglog plot, integral of normalized pressure

Using the integral preserves the signature of the flow regimes while significantly reducing the
noise. Hence such definitions provide a diagnostic tool where most of the usual well test
methods can be used. In particular, it is clearly possible to get an estimate of the reservoir kh
from the derivative stabilization level. The kh being known, one can then get a first estimate of
the reservoir size from the unit slope late time trend. These calculations are an integral part of
the loglog plot. It is possible to either display the ‘true’ model response, i.e. the response to
the full pressure history, or the response to a single pressure step. The single step response
shows the signature of the model in a clear and usable form (Figure 4.C.2 and 4.C.3), whereas
the response to the real history is usually very erratic, because the equivalent time is jumping
back and forth in time as illustrated in Figure 4.C.4.

Fig. 4.C.4 – Loglog plot with ‘true’ model response

Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p15/24

Nonlinear cases

The linearities expected during the various flow regimes may be distorted when the diffusion
does not follow a linear equation, with gas or in multiphase cases, it is important to realize that
the loglog plot, like the Blasingame plot, only provides a basis for comparing the data and a
model response. Handling the non-linearities is the model responsibility, not that of the plot. If
the model is representative, the model and the data will be consistent. It does not really
matter whether such or such period exhibits linearity or not. When using a type-curve this is
different as the type-curve embeds modelling assumptions.

4.C.3 Material balance plot

Agarwal et al presented a Cartesian plot of dimensionless rate qD versus dimensionless
cumulative QDA.

They show that the responses corresponding to distinct reservoir sizes all exhibit a straight line
with a negative slope during boundary dominated flow, and all curves converge to the same
value on the X axis, equal to 1 2π . In other words, the following relation is established in all
cases during boundary dominated flow:

1
qD = − Q DA
2π

Fig. 4.C.5 – Agarwal et al plot

The expression of the dimensionless variables varies depending of the fluid type and a specific
treatment must be applied in each case.
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p16/24

Oil

For an oil case, the expression of the dimensionless parameters is defined below:

141.2qBμ 0.8936QB
qD = and Q DA =
kh( p i − pw) φ hAc t ( p i − p w )

All equations are in Oil Field units

The dimensionless cumulative production can be expressed in terms of the fluid in place, in
STB/D:

φhA
N=
5.615 B
0.8936Q Q
Q DA = =
5.615 Nc t ( p i − p w ) 2πNc t ( p i − p w )

So the linear relationship between dimensionless rate and cumulative becomes:

141.2qBμ 1 0.8936QB
= −
kh( p i − pw) 2π 5.615 Nc t ( p i − p w )

Using the full definition of the dimensionless variables requires an a priori estimate of PV,
basically what we are after. Therefore the method presented by Agarwal-Gardner is iterative.
q Q
However we see from the above equation that if we plot versus
pi − p w ct ( pi − p w )
boundary dominated flow will exhibit a straight line which intercept with the X axis gives
directly N.

Fig. 4.C.6 – Material balance plot

Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p17/24

Note: In the case of constant flowing pressure, it is interesting to draw a parallel between this
rate cumulative plot and the rate cumulative plot used in traditional decline curve analysis. The
traditional decline methods yield a maximum recovery rather than fluid in place. The relation
between the two methods is established by considering a recovery factor of RF= c t ( p i − p w ) .

Gas

The boundary dominated flow obeys the same equation:

1
qD = − Q DA
2π
provided that the dimensionless rate and cumulative production be defined as:

qD =
1422.T .q
and QDA =
[
4.50.T .zi .Gi . m( pi ) − m( p) ]
kh(m( p i ) − m( p w ) ) φ hA. pi .[m( pi ) − m( pw )]
All equations are in Oil Field units.

Unlike the oil case, we cannot find a simple expression of a normalized cumulative that is
independent of the fluid in place. This is because the gas in place is involved in a non-linear
fashion in the expression of the dimensionless cumulative. However by extension with the
previous method for oil we can choose to plot:

q
versus Q DA =
[
G i . m ( p i ) − m( p ) ]
m( p i ) − m ( p w ) [m( p i ) − m( p w )]

The value of ‘X’ at the intercept is:

[ ]
G i . m ( p i ) − m( p )
@ Intercept =
φ .h. A.2.π . pi PV .2π . pi
= =
PV .2π . pi. . p sc .T
=
PV
= Gi
[m( p i ) − m( p w )] 4.50.T .zi 4.50.T .zi 4.50.T .Bgi .Tsc. pi Bgi

Note: the X-axis value Q DA =

[
G i . m ( p i ) − m( p ) ] depends on the gas in place Gi value,
[m( p i ) − m( p w )]
therefore a change in the straight line coefficients changes the intersect therefore the abscissa
of the data, in other word moving the straight line will move the data points through which we
draw it, it becomes an iterative process that converges easily.
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p18/24

4.C.4 History plot

For complex cases and noisy data where no specific behavior is seen on these diagnostic plots,
the linear plot of pressure and rates vs. time becomes the main tool. There is no real
diagnostic, just an optimization process. Under these conditions, it is not realistic to expect to
estimate more than a productivity index, mobility and a drainage area. In the absence of any
other information, the simplest analytical solution, homogeneous circular reservoir, will usually
be suitable to model the well drainage area. The use of more complex models can be
acceptable if complementary information is available from other sources, and the number of
parameters allowed to change in the optimization process is reduced to a minimum.

Fig. 4.C.7 – Production history match

4.D Modern PA methodology

Modern Production Analysis is based on the use of PC based PA software products. The key for
any modern software is to combine user friendliness to a powerful technical kernel, requiring
both analytical and numerical capabilities. In terms of methodology, the central diagnostic
tools are the Blasingame and loglog plots, which are used whenever such a diagnostic is
possible. However, because of the very scattered nature of production data, the ultimate
diagnostic tool will often be the history plot, where the coherence of the model and the data, in
terms of simulated pressures, rates and cumulative productions, will be the final decision tool
for the interpretation engineer.

Once the interpretation is initialized and production data loaded, the first task will be to extract
the interval of time on which the analysis will be performed. If the pressures are not available,
only the ‘old’ tools can be used. If both rates and pressures are available, the interpretation
will be performed with the four main diagnostic tools. The interpretation engineer can select
one or several analytical and/or numerical models, set their parameters and generate these
models for comparison with the actual data. For models that are believed applicable, the
engineers can refine the model leading parameters, either manually or by using nonlinear
regression.
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p19/24

Once this is finalized, the engineer may use the model(s) to forecast future production by
defining a scenario of producing pressures. The user can then run a sensitivity analysis on a
selection of the model parameters.

The path described is the default path when all is well. In reality, for complex problems, it
becomes a trial-and-error process where the interpretation engineer may decide to go back
and forth as needed when part of the process is unsatisfactory.

4.D.1 Preparing a project and loading data

The initialization procedure is largely the same as that used in PTA. The interpretation engineer
inputs PVT data, geometric well and reservoir information.

The load option imports flat ASCII files, allows manual input, copy-paste from spreadsheets,
and increasingly input through links to databases or intermediate repositories using advanced
filtering tools.

After the load, the cumulative production is automatically calculated by integration of the
production history, and is displayed on the history plot together with the rate. Pressures is
loaded and displayed in the history plot.

Quality control is not as critical as in PTA, because wellbore effects are normally not dominant,
except when the pressure is recorded at surface. In this case, the validity of the well intake
curves used to correct the pressure to sandface during extraction can become a potential weak
point.

4.D.2 Editing data

Loaded data may be the result of careful post-processing after the data has been acquired, in
which case no or little editing may be needed. However, often the interpreter will gather data
of unequal quality from different sources. Pressures will often be acquired downhole in real
time or with a memory gauge or come from permanent gauges (PDG), while rates are mostly
measured at surface but in some cases, can also come from permanent measuring devices
downhole.

Beyond the usual cleaning of irrelevant data and the correction of load errors, the main
challenge is to end up with at least one coherent, synchronized set of rate and pressure data.
To get there the engineer may have to perform the following tasks:

• Synchronise all data acquired electronically to the same reference time.

• If rates are not loaded from a file, create the rate history by identifying the pressure breaks
and get the rate values from hard copy reports.

• Refine the production history, when the time sampling of rate measurements is too crude.

• Conversely, if the production history goes into useless detail, simplify the rate history to
reduce the CPU time required to run the models.
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p20/24

4.D.3 Extraction and diagnostics

Once the data have been synchronized and validated, the analysis itself will start. The time
range over which the extraction of the data will take place is defined and the following plots
are built by the software in addition to the history plot:

• ARPS plot
• Fetkovich type-curve plot
• Fetkovich plot
• Blasingame plot
• Loglog plot
• Normalized rate-cumulative plot

At extraction time the option to invoke a defined lift curve or flow correlation to correct the
pressure profile from the measurement depth to sandface can be chosen.

The loglog plot is used for diagnostic purposes to identify the two main flow regimes hopefully
present in production data, infinite acting radial flow (IARF) and pseudo steady state (PSS).
The pressure match is fixed to coincide with a stabilization of the derivative of the normalized
pressure integral and the time match is fixed to the unit slope line of PSS at late time.
Figure 4.D.1, the match is adjusted by click and drag of the mouse. The loglog plot is linked to
the Blasingame and the Fetkovich plot so any change in the loglog match is mirrored in the
others. In case the data is of high quality and the sampling frequency is high enough it is
sometimes possible that more than the IARF transient develop thus extending the diagnostic
possibilities to approach those of PTA and both well and reservoir models can be recognized in
the test data. This is however rare in low frequency data typically used in production analysis.

If the loaded pressure history contains any decent build-ups with high frequency pressure data
or a link to a database that allows the repopulation of this data without a filter then the
interpreter is in luck. This data can be transferred to a PTA module to determine all of the
classical parameters including the model and these can be transferred back to the PA package
and finally the modelling can begin; adjusting for parameters that can typically change over
the longer time intervals involved in production analysis (i.e. skin).

Fig. 4.D.1 – Loglog match

Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p21/24

4.D.4 Model generation

After the diagnostics candidate models, analytical or numerical, are selected and an attempt is
made to obtain a match between the models and the real data in all the relevant plots
including the history plot. To obtain a match the interpreter will run with a first estimate of the
model parameters generally obtained by generating the default, or automatic, model based on
the initial match made in the loglog plot as described in section 4.D.3. The default model is the
homogenous model with constant skin in a closed circle. At generation time a first estimate of
the constant skin is automatically made by the software.

After a comparison between the model and the data, changes can be made to the model
parameters and any known well configuration can be imposed such as knowledge of the well
being fractured, horizontal or partially penetrating. In the event that PTA was performed on
part of the pressure data the model used can be transferred to the production analysis.

Finally the objective is to vary the model parameters until there is a reasonable match
between the model and the data in all relevant plots, including the history plot.

4.D.5 Model refinement

Before using the software non linear regression routine to optimize the model parameters and
the match, the engineer should manually deal with the gross parameter errors and change the
values until the model and the data are closer. An experienced analyst with a good
understanding of the sensitivity to the various model parameters should get a fit, by changing
the parameters by hand, between the model and the data very quickly. This will increase the
chance for the regression to succeed and converge faster.

The principle of non linear regression is to use numerical optimization to refine the parameter
estimates by minimizing an error function, generally the standard deviation between the
simulated and real values at selected times. The most commonly used optimization algorithm
is Levenberg-Marquardt, but there are many variants. The engineer has the possibility to run
with some or all the leading parameters of the model and he can also fix the upper and lower
limits of the allowed parameter variation. The data points on which the error function will be
calculated may also be controlled. See Figure 4.D.2.

Fig. 4.D.2 – Final match after optimization

Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p22/24

4.D.6 Forecast
Once the model has been selected a production forecast can be easily performed by defining a
flowing (producing) pressure scenario. Figure 4.D.3 illustrates this with a constant pressure
production for 100 days.

Fig. 4.D.3 – Forecast 100 days constant pressure production

4.D.7 Sensitivity study

At the end of the nonlinear regression it is also possible to recover some confidence intervals.
They can be used to assess the sensitivity to individual parameters and any eventual
parameters cross-correlations.

Another possibility is to run and display a series of model generations corresponding

to different values of a given parameter, in order to compare theses on the history.
See Figure 4.D.4.

Fig. 4.D.4 – Sensitivity to skin

Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p23/24

4.D.8 Reporting Guidelines

A typical analysis report will be split into two components: the ‘mechanical’ part, basically the
result tables and plots generated, directly or indirectly, by the Production Analysis package,
and the ‘verbose’ part, where the engineer will report the objectives, the operations, the
analysis, his confidence in the results and forecast, and possible recommendations for
stimulation, workover and continued or future measurements and equipment to use.

Typically, professional analysis reports are generated with two possible set-ups:

• A header document, from a word processor, with some ‘copy-paste’ of plots and results
from the PA software, but with most of the ‘mechanical’ report delivered as an annex,

• An integrated document, typically from a word processor, where some plots and tables are
dynamically connected to the PA software using some OLE or COM automations. The
advantage of this solution is that it is much more flexible. Once a model template has been
defined, the reporting process will get shorter and shorter from one analysis to the next.

4.E PA versus PTA

4.E.1 Common tools
PA and PTA methods share the same assumptions in terms of the use of the diffusion equation
and limiting conditions. Most of the analytical and numerical models developed in PTA may be
used in PA with minor adjustments, such as the ability to use the pressures as the input to
simulate the rates with superposition.

Modern PA and PTA share a similar path. After loading, synchronizing and extracting data, one
first tries to run a diagnostic using specialized plots and straight lines. An analytical or
numerical model is then run, and an optimization process adjusts the parameters to minimize
the difference between the simulated model response and the observed data.

4.E.2 PSS vs. IARF

The main regime of interest in PTA is Infinite Acting Radial Flow (IARF). We look primarily for a
stabilization of the Bourdet derivative. When IARF is detected, specialized analysis will give a
value of mobility and a total equivalent skin factor. We can refine this and diagnose other well,
reservoir and boundary behaviors from various parts of the response; however the starting
point will always be IARF.

The main regime of interest in PA is Pseudo Steady State (PSS). We look primarily for a unit
slope on the loglog or the Blasingame plot. Specialized analysis will determine the size of the
well drainage area from the slope, and the intercept will be a function of three main factors:
the well productivity index, the mobility and a shape factor. More complex models could be
used, but there may not be enough information to determine the additional parameters.
However the pressure transient results may be used to determine these.
Dynamic Flow Analysis - v4.02 - © KAPPA 1988-2007 Chapter 4 – Production Analysis (PA) - p24/24

4.E.3 Diagnostic capabilities

One of the key steps in PTA is the diagnostics, where, based on the derivative behavior the
engineer decides which model could most appropriately be used to perform the analysis. This
is made possible by very clean data and constant production i.e. zero, during the build-up.

Production history may be so scattered that the responses will be dominated by transients. In
this case there is no way to identify pseudo steady state behavior. This may happen even
though the well is still producing and the pressure is declining globally.

Despite the lack of pure PSS behavior it will be possible with a model to history match the data
and obtain a reliable drainage area estimate and even sometimes discriminate mobility, skin
and shape factor. No specialized plot will show such a behavior. So the use of models and
optimization is likely to change the way PA is performed completely, even more radically than
happened with PTA.

4.E.4 Validity of the PTA hypothesis in Production Analysis

PTA can provide a clean snapshot of what the well and reservoir system is at a given time. PA
covers a much wider time range, and some of the assumptions valid during a single well test
will not be true over the well producing history. The three main differences are related to the
well productivity, the drainage area and multiphase production.

PTA models account for rate-dependent skin. It is also known that the well may be cleaning up
during the initial production phase. So the well productivity may not be constant during a well
test. However this is a reasonable assumption for a single build-up, and optimization will be
possible with a single mechanical skin model. In PA this is unlikely. Well productivity does
change over time, and no optimization process is reasonably possible over a long period of
time without considering a time-dependent skin.

In PTA, boundary effects are generally material boundaries, even though interfering wells can
produce the same effects as boundaries. In PA we consider the well drainage area. Except
when there is only one producing well, part or all of the drainage area boundaries are
immaterial, depending on the flow equilibrium between the neighboring wells. The drainage
area will change in time when new wells are produced, or even when the flow rates change
asymmetrically. To account for these changes, a multi well model, either analytical or
numerical, may be required.

In PTA we approximately handle multiphase cases using pseudopressures or considering that

saturations are constant and the flow can be modelled with an equivalent single-phase fluid as
in Perrine’s method. In PA, solutions exist that consider that individual fluids develop PSS
independently. However these solutions make a global correction for multiphase production but
they are unable to forecast the breakthroughs. There is a point where only a history match
with a numerical model can account for multiphase production.