You are on page 1of 9

SPE 84880

In-Situ Diagnosis of Formation Characteristics in Horizontal Wells


H. Cui, SPE, U. of Tulsa; Y. Dong, SPE, U. of Oklahoma; R. Kalita, SPE, S. Sinha, SPE, Y. Jalali, SPE, Schlumberger

Copyright 2003, Society of Petroleum Engineers Inc.


reservoir permeability field through a far-field scaling
This paper was prepared for presentation at the SPE International Improved Oil Recovery technique. Therefore, good estimates of the near-wellbore
Conference in Asia Pacific held in Kuala Lumpur, Malaysia, 20–21 October 2003.
permeability profile can be obtained with highly uncertain
This paper was selected for presentation by an SPE Program Committee following review of
information contained in an abstract submitted by the author(s). Contents of the paper, as
knowledge of the reservoir permeability field. This is
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to important as the technique can be applied not only to early-
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at time but also late-time data. We illustrate the application of
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
the multipoint pressure method through a series of examples
for commercial purposes without the written consent of the Society of Petroleum Engineers is and discuss its potential as an interventionless method of well
prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300
words; illustrations may not be copied. The abstract must contain conspicuous diagnosis, for estimation of both flux and permeability
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
distribution in high-angle completions.

Introduction
Abstract Horizontal wells can be diagnosed based on information
We present a method of estimating the distribution of derived from open-hole and cased-hole surveys. These
permeability along the length of a liner-completed horizontal include petrophysical logs, dynamic formation testers,
well from measurements of well flowing pressure at multiple production logging, and pressure transient testing. With the
points along the path of flow in the wellbore. We do not advent of permanent sensing technologies,1-3 and the
model the deployment mechanism of the sensors, although in development of methods of production data inversion or
principle this can be achieved with stinger completions. history-matching,4-5 a new form of cased-hole diagnosis can
be envisaged, with improved spatial and temporal coverage,
The method applies to flow under constant-rate conditions and and without the need for in-well intervention and interruption
yields estimates of permeability along well trajectory in the of production. The impact of such method on reservoir-scale
principal directions normal to well path. Therefore, estimates characterization can also be significant.
of horizontal permeability perpendicular to well orientation
and vertical permeability can be obtained for the gridblocks There are two main preconditions for the development of such
intersected by the well. Estimation accuracy, however, methodology, one concerning sensing technology, the other
improves when the geometric mean of the normal interpretation methodology. Permanent sensing technology
permeabilities is obtained, as opposed to their individual has made great progress during the last decade, with the
values. This permeability group is what governs the influx development of single-point and distributed measurements
into the wellbore. We neglect the effect of skin, but the that can be deployed with the completion (pressure, flowrate,
method can be used to invert for a piecewise well index and distributed temperature). However, these systems are
coefficient, that is proportional to the permeability group and typically developed as standalone units, and do not enjoy the
inversely proportional to skin. Therefore, the method can required degree of integration. One reason for this, besides
generate the longitudinal profile of a parametric group that the basic drive for the development of discrete measurements
represents the quality of the formation and the integrity of high metrological performance for direct diagnosis of flow
of completion. conditions, is the absence of a methodology to determine what
combination of measurements are required to resolve
The quality of the inversion is governed by the spatial density anticipated flow problems. Today, basic modeling tools
of measurements, accuracy of measurements, knowledge of required to arrive at such determination are available,6-7
wellbore hydraulics, and knowledge of the relative although an efficient methodology for such analyses has yet to
permeability characteristics of the formation. We obtain fair be developed. Thus, the sensor configuration problem, in the
estimation of wellbore-gridblock permeability distribution for absence of robust inversion techniques, can be addressed by
a five-node pressure configuration in a 2000 ft horizontal well recursive simulation of measurement scenarios. The basic
experiencing 60 psi pressure drop (10,000 STB/day), given problem, therefore, is not to determine what can be diagnosed
realistic bands of uncertainty associated with the governing with a predefined set of sensors, but to determine what sensors
parameters, including measurement drift, and errors in liner are required to resolve a predefined or anticipated set of
roughness, and relative permeability exponent. Also, the problems. One object of this study is to motivate the
inversion can be rendered insensitive to knowledge of the
2 SPE 84880

development of such methodology and associated We generate the pressure profile of the well at every time step
sensing assemblies. and invert for the wellbore-gridblock permeabilities using all
pressure data from initial production to the time at which the
We decouple the well diagnosis problem to diagnosis of flow inversion is made. (The total production rate of the well is
condition in the wellbore and diagnosis of near-wellbore imposed as a constraint.) Therefore we perform an inversion
formation characteristics. (By near-wellbore is meant the of the pressure profile corresponding to successive time steps.
wellbore gridblock scale.) This is partly to adhere to the We compare the quality of the inversion based on short-time
conventional demarcation between production logging and and long-time data (e.g., 10 days versus 100 days).
formation evaluation, and partly is a natural consequence of
the mathematical problem. However, although the wellbore A perturbation technique is used to determine the sensitivity of
diagnosis problem (determination of flux distribution as in wellbore pressure at each segment to formation permeability
production logging) can treat the formation simply as a at each gridblock intersected by the well. This is done
boundary condition, the formation evaluation problem cannot numerically using the coupled wellbore-reservoir model. The
do the same, as evaluation is based on measurements made sensitivity coefficient is computed with respect to permeability
inside the wellbore. Thus both media, porous and nonporous, along the three principal axes (x-direction, parallel to well
have to be taken into account. (Sensors that are in direct orientation, y- and z-directions normal to well cross-section).
contact with the formation, however, are emerging.8 Thus, A gradient-based optimization technique is used to minimize
evolution of this problem is to be expected.) In this study, we the mismatch between measured and computed wellbore
examine the interventionless or in-situ analogue of formation pressures, using information on sensitivity coefficients and an
evaluation. The in-situ analogue of production logging is initial guess of the wellbore permeability profile. This is done
examined in a parallel study.9 iteratively until a termination criterion is reached. Details of
the inversion technique are provided in the Appendix.
Method
We model a coupled wellbore-reservoir system, where We then examine how imperfect information about the key
wellbore hydraulics is taken into account through a multiphase parameters used in the inversion affects the quality of the
flow model that accounts for slippage between the phases, but inversion. This includes imperfect measurements, imperfect
not segregated flow patterns (drift flux model).10 This flow knowledge of wellbore hydraulics, and imperfect knowledge
model has been found to agree relatively well with of the reservoir permeability field. We also examine the
nontransient liquid flow in horizontal conduits.11 One impact of relative permeability. We show first the impact of
advantage of this model is that it allows a fully implicit each parameter individually on the quality of the inversion and
coupling of the reservoir and the wellbore.6 However, care then consider their combined effect.
must be taken to analyze with this coupled model, only flow
problems for which the wellbore flow model is applicable. Results
This precludes use of this model for transient or segregated- Figure-1 shows the wellbore pressure profile at two snapshots
flow problems. Alternatively, a more elaborate flow model after initial production, 10 and 115 days. The rate of pressure
for the well can be used,12 but this renders the modeling drop along well length is governed by the rate of increase of
exercise more complex. In this study, therefore, we adhere to fluid velocity in the liner. The pressure gradient, therefore, is
the simple drift flux model for multiphase flow in the governed by the distribution of influx into the wellbore.9 The
wellbore. This is adequate for the purposes of this study, latter is in turn governed by the permeability profile along
which serves to illustrate the methodology of in-situ well length. The decline in the pressure profile over time is
formation evaluation. caused by the decline in fluid mobility in the formation with
the onset of two-phase flow (relative permeability). Thus,
We analyze first the forward problem. We consider drawdown has to increase to supply the same rate, given the
homogeneous and heterogeneous reservoirs subject to bottom reduced mobility of fluids in the formation.
water-drive. We consider formation anisotropy and model
normal crude properties. We consider a liner completion of Figure-2 shows the sensitivity of the pressure in the heel
2000 ft length draining the reservoir at a constant liquid rate of segment of the wellbore to the permeability of gridblocks
10,000 STB/day. The basic parameters of the model are traversed by the well. Permeability along the three principal
summarized in Table-1. directions is taken into account. The sensitivity of wellbore
pressure to axial permeability (with respect to the well
We divide the wellbore into segments to compute the wellbore orientation, x direction) is much lower than that to normal
flowing pressure along the length of the well at any time step. permeabilities. This is physically sensible because the well is
We divide the well into as many segments as there are principally fed along streamlines normal to the circular cross-
gridblocks intersected by the well (ten in our example). We section of the well. Also wellbore pressure in each segment is
consider that every other well segment contains a pressure most sensitive to the permeability of its adjacent gridblock.
measurement node. Therefore we consider a sensor spacing of With increasing distance of gridblocks from the measurement
400 ft. We do not explicitly model the sensor configuration, point (along well length), sensitivity decays. This means that
but note that distributed sensing (fiber-optic temperature) has the resolution of the permeability inversion is governed by
been deployed in horizontal wells on extended tail pipe or measurement density.
sealed stinger.13
SPE 84880 3

Figure-3 shows the inversion of the permeability normal to addressed by calibration of pressure measurements when there
well path for a homogeneous anisotropic problem. Early time are interruptions to production. Upon decay of transients, all
data is used (10 days). The true horizontal and vertical nodes must read the same value, unless the well undulates
permeabilities are 200 and 100 md. The initial guess is 150, through the formation (i.e, must adjust for trajectory), or a
and 50 md. The estimation is about 185 and 110 md. regional pressure gradient exists in the reservoir.
Although the estimated permeabilities deviate from the true
permeabilities, their geometric mean closely tracks the true Figure-8 shows the inversion of the permeability group for the
mean. This is because the geometric mean governs the influx five-node configuration with ‘drift-free’ measurements, but
to the well, which governs the pressure profile. For the when knowledge of wellbore hydraulics is imperfect. This is
remainder of the examples we compare only estimates of the modelled as the uncertainty in liner roughness (underestimated
geometric mean to the true mean. Also, unless otherwise by a factor of two). Inversion deteriorates towards the heel of
stated, the results pertain to early time inversion (10 days). the well, where pressure gradient is greatest, and error in
roughness translates into error in permeability. This effect is
As described in the Appendix, inversion can also incorporate less pronounced when the magnitude of pressure drop in the
the concept of skin, such that for each wellbore gridblock the wellbore is smaller, such as in larger pipes (Figs. 9-10). But
ratio of geometric mean of normal permeabilities to a term the effect is still notable. We address this problem by making
containing the skin is estimated. This would be similar to pipe roughness also a parameter of inversion. Figure-11
estimation of the productivity index of wellbore gridblocks. shows the permeability estimation for cases of large error in
The variation of this composite group along well length is roughness. The quality of the inversion is quite satisfactory.
informative of the quality of the wellbore-reservoir coupling This is because the near-wellbore characterization problem,
and the condition of the completion. with multiple wellbore measurements, behaves like a well-
posed problem. (The total number of parameters being
Figure-4 shows the impact of measurement density on the estimated here is 21, i.e., one liner roughness and 2
quality of inversion. A good estimation is obtained when permeability values for each wellbore-gridblock. The total
every other well segment is instrumented with a pressure number of data points is 30, i.e., five pressure points per time
sensor (five-node configuration). It can therefore be suggested step, times 6 time steps for early-time inversion of 10 days.
that due to the partial sensitivity of the wellbore segment The data outnumbers the unknowns, although there might be
pressure to nonadjacent gridblocks, the spatial resolution of redundancy in the information content of the data. There is
permeability profile is greater than the spatial density of however a strong constraint for the inversion, which is the
pressure measurements. The spatial density of measurements total production rate of the well.)
required for any application, however, is determined by the
degree of variability of permeability profile in the axial or Figure-12 shows the inversion of the five-node configuration
longitudinal direction. in a uniform permeability field, but when reservoir
permeability is imperfectly known. As before, the true
Figure-5 shows the inversion of the five-node configuration horizontal and vertical permeabilities are 200 and 100 md. In
when sharp permeability variations exist along well path. This the inversion, however, initial guess of the reservoir
can occur when the well traverses a formation with bedding permeability field is either (500, 250 md) or (100, 50 md).
planes, which creates lateral variation of facies along well The initial guess of the wellbore gridblock permeabilities is as
path. Fractured intervals would produce the same effect. before (150, 50 md). The inversion nevertheless produces a
Here the estimated and true profiles deviate, although the fair estimation of the near-wellbore permeability distribution.
character of the permeability profile is captured fairly well. The error of the far-field permeability is attenuated, such that
any overestimation of the far-field results in a smaller
Figure-6 shows the inversion for a five-node configuration underestimation of the near-field (and conversely if the far-
(homogeneous anisotropic example), when the sensors field is underestimated). This is however true because the
experience uniform drift. A negative drift, i.e., when a inversion is based on early time data. This problem can be
pressure lower than true pressure is measured, causes an overcome by inverting for a far-field permeability multiplier,
underestimation of permeability. This is because negative as well as the wellbore gridblock permeability distribution,
drift exaggerates the actual drawdown, which has to be which generates the least mismatch between measured and
compensated by reduced permeability to respect the true computed pressures. This technique is described in the
production rate of the well. Likewise, positive drift Appendix and was suggested to us by Reference 14. Figure-
overestimates the permeability. 13 shows the inversion when the multiplier technique is used
(compare to Figure-12).
Figure-7 shows the inversion for when the sensors experience
nonuniform drift. This is modelled as an oscillation of ±1 psi Figure-14 shows the inversion of the five-node configuration
or ±2 psi from the true value. Therefore, this simulates in a uniform permeability field when there is imperfect
uncorrelated drift amongst the nodes. knowledge of the relative permeability function. Straight-line
relative permeability functions are used for inversion, whereas
Comparison of Figs. 6 and 7 indicates that the quality of the the true relative permeability is a concaved Corey-type
inversion is determined by the relative accuracy of the function (see Figs. 15-16). The quality of the inversion,
pressure measurements. The drift problem, therefore, must be therefore, is dependent on knowledge of relative permeability.
4 SPE 84880

One way to improve the inversion is to invert pressure profiles Kx,y,z = Permeability in x, y, z directions, md
at earlier times, when relative permeability effects are N = Number of wellbore segments (numerical discretization)
less pronounced. OWC = Oil-Water Contact
Pb = Bubble point pressure, psi
Figure-17 shows the inversion of the five-node configuration Pi = Initial reservoir pressure at specified datum, psi
with ‘drift-free’ measurements but when the combined effects Pwfo,i = Observed flowing pressure in segment i, psi
of error in liner roughness, far-field permeability, and relative Pwfc,i = Computed flowing pressure in segment i, psi
permeability function are taken into account. The true Rsb = Solution gas-oil ratio at bubble point pressure, scf/STB
permeability profile is moderately heterogeneous with Sg = Gas saturation
variation of normal permeabilities along the length of the well. Sgc = Critical gas saturation
The inversion result is a fair characterization of the true Sw = Water saturation
permeability profile. This is partly because the influencing Swc = Connate water saturation
parameters can have opposite effects on the δkj = Perturbation of jth wellbore-gridblock
permeability estimation. ∇kf(kl) = Gradient vector expressing dependency of f on
permeability array at iteration l
Conclusions ε = Convergence tolerance
φ = Porosity
Modeling and inversion techniques illustrate that it is feasible µob = Oil viscosity at bubble point pressure, cp
to estimate the permeability profile of horizontal wells given ρo = Oil density, lbm/ft3
measurements of well flowing pressure at multiple points
along the path of flow. Such measurements contain References
information about the permeability of the formation normal to
well path. We obtain reasonable results in estimation of the 1. Baker, A., Gaskell, J., Jeffrey, J., Thomas, A.,
geometric mean of permeabilties normal to the transverse Veneruso, A., and Unneland, T.: “Permanent
cross-section of the well. The quality of the inversion is Monitoring Looking at Lifetime Reservoir
governed by certain measurement, wellbore, and reservoir Dynamics,” Oilfield Review, 7, no. 4, (Winter 1995)
characteristics. This dependency, however, is not 32-46.
fundamentally different than that pertaining to classical 2. Unneland, T., Manin, Y., and Kuchuck, F.:
methods of well diagnosis or reservoir characterization, e.g., “Permanent Gauge Pressure and Rate Measurements
pressure transient testing or production data inversion (history- for Reservoir Description and Well Monitoring: Field
matching). Therefore, there are theoretically no barriers to the Cases,” paper SPE 38658 presented at the 1997 SPE
implementation of this technique. A parallel study9 illustrates Annual Technical Conference and Exhibition held in
that measurements examined in this study also reveal the San Antonio, Texas, 5-8 October.
distribution of fluid influx along the completion, and under 3. Laurence, O.S., Brown, G.A.: “Using Real-Time
certain conditions can discriminate the influx of distinct fluid Fibre Optic Distributed Temperature Data for
phases. Therefore, a sound basis exists for the assessment of Optimizing Reservoir Performance,” paper SPE
the well diagnosis problem with completion-conveyed sensors. 65478 presented at the International Conference on
This can have an important impact on technology Horizontal Well Technology, Calgary,
development programs (integration of permanent sensing 6-8 November 2000.
systems), well diagnosis and remediation practices, and 4. Li, R., Reynolds, A.C., Oliver, D.S.: “History
reservoir-scale characterization. Further investigation of these Matching of Three Phase Flow Production Data,”
problems is required. paper SPE 66351 presented at the 2001 Reservoir
Simulation Symposium, Houston, 11-14 February.
Acknowledgements 5. Zhang, F.: “Automatic History Matching of
Production Data for Large Scale Problems,” Ph.D.
We thank Schlumberger Well Completions and Productivity dissertation, University of Tulsa, 2003.
for sponsoring this study and permission to publish the results. 6. Holmes, J.A., Barkve, T., Lund, Ø: “Application of a
We thank Professor Albert C. Reynolds for helpful Multisegment Well Model to Simulate Flow in
suggestions concerning the inversion methodology. Advanced Wells,” paper SPE 50645 presented at the
1998 European Petroleum Conference, Hague,
Nomenclature 20-22 October.
7. Stone, T.W., Bennett, J., Law, D.H.S., and Holmes,
Bob = Oil formation volume factor at bubble point, RB/STB J.A.: “Thermal Simulation with Multisegment
f = Objective function (based on data mismatch) Wells,” paper SPE 66373 presented at the 2001 SPE
H(kl) = Hessian matrix of f Reservoir Simulation Symposium held in Houston,
ID = Internal Diameter, inch Texas, 11-14 February.
kj =Permeability of jth wellbore-gridblock, md 8. Powers, J.: “Incorporating DTS into Sand Control
Krg = Gas relative permeability Completions,” case study presented at the 2003
Kro = Oil relative permeability
Krw = Water relative permeability
SPE 84880 5

Intelligent Wells Europe Conference, Stavanger, pressure. In this work, the wellbore pressures are calculated
29 –30 April. with the multisegment facility of Eclipse over any prescribed
9. Bui, T.D., Vicencio, O.A., Sinha, S., Kalita, R., time duration.6 N is the number of wellbore segments in which
Jalali, Y.: “In-Situ Diagnosis of Inflow behavior in flowing pressure can be measured.
Horizontal Wells,” paper SPE 84873 to be presented
at the 2003 SPE International Improved Oil Recovery To minimize the objective function, a non-linear optimization
Conference in Asia Pacific, Kuala Lumpur, algorithm is required. We use the Levenberg-Marquardt
20-21 October. method. Using k as the vector of model parameters, the
10. Zuber, N., Findlay, J.: “Average Volumetric iterative equation is given by:
Concentration in Two-Phase Systems,” Trans. AIME,
J. Heat Transfer, November 1965, v. 87, 453-468. ⎡ H ⎛ k l ⎞ + µ l + 1I ⎤δk l + 1 = −∇ f ⎛ k l ⎞
11. Banu, A.: “Drift-Flux Models for Multiphase Flow in ⎢⎣ ⎜⎝ ⎟⎠ ⎥⎦ k ⎜⎝ ⎟⎠
Wells,” M.S. Thesis, Stanford University,
September 2002.
12. Petalas, N., and Aziz, K.: “A Mechanistic Model for
Multiphase Flow in Pipes,” J. of Canadian Petroleum where δk is the parameter incremental vector, ∇kf(kl) is the
Technology (2000), 39, No. 6, 43. gradient vector, H(kl) is the Hessian matrix, I is a unit
13. Al-Aisimi, M. et al: “Advances in Well and diagonal matrix, µ is the damping factor and l is the
Reservoir Surveillance,” Oilfield Review, 14, no 4, iteration number:
(Winter 2002/2003) 25.
14. A. C. Reynolds Jr.: personal communication. ⎡ δ K1 ⎤
15. Eclipse Technical Description 2003A, pp 957. ⎢δK ⎥
⎢ 2⎥
δ k = ⎢δ K 3 ⎥
⎢ ⎥
Appendix ⎢ M ⎥
⎢δK ⎥
Inversion Parameters: The wellbore pressure response is ⎣ n⎦
computed in this study by a numerical simulator that accounts
for the coupling of the reservoir and the wellbore through a ⎡ ∂f l ⎤
connection factor.15 The connection factor for a horizontal ⎢ ⎥
well geometry, where the well is aligned along the x direction ⎢ ∂k l ⎥
⎢ 1⎥
is given by: ⎢ ∂f l ⎥
⎢ ⎥
d x (k y k z )1 / 2 ⎢ ∂k l ⎥
⎢ 2⎥
CF = ∇ k f ⎛⎜ k l ⎞⎟ = ⎢ l ⎥
ln(ro / rw ) + S ⎝ ⎠ ∂f
⎢ ⎥
⎢ ∂k l ⎥
⎢ M ⎥ 3
where CF is the connection factor, dx is the length of the
gridblock along the x direction, ky and kz are the ⎢ l ⎥
⎢ ∂f ⎥
permeabilities normal to well cross-section, S is the skin, rw is ⎢ l ⎥
the wellbore radius, and ro the radius at which the pressure ⎢⎣ ∂k n ⎥⎦
corresponds to the ‘reservoir’ pressure. (ro itself is a function
of the gridblock dimensions and permeabilities in the y and z
directions.) In this study, we neglect skin and individually ⎡ ∂2 f ∂2 f ∂2 f ⎤
invert for ky and kz. We then report the results as the ⎢ K ⎥
⎢ ∂k1k1 ∂k1k 2 ∂k1k n ⎥
geometric mean of these permeabilities, i.e., (ky*kz)1/2. It ⎢ 2 ⎥
would however have been possible to invert for the entire ⎢ ∂ f ∂2 f ∂2 f ⎥
⎛ l ⎞
H ⎜ k ⎟ = ⎢ ∂k k K
‘connection factor’ which is, for all practical purposes, the
⎝ ⎠ ⎢ 2 1 ∂k 2 k 2 ∂k 2 k n ⎥
productivity index of the wellbore-gridblock (times viscosity). ⎥
⎢ M M M M ⎥
⎢ ∂2 f ∂2 f ∂2 f ⎥
Inversion Methodology: We define the objective function in ⎢ K ⎥
the form: ⎢⎣ ∂k n k1 ∂k n k 2 ∂k n k n ⎥

N 2
f = ∑ ⎛⎜ P ⎞ The elements of the gradient vector are computed as follows:
−P
wfo, i wfc, i ⎟⎠
i = 1⎝ ∂P
∂f N ⎞ wfc, i
= −2 ∑ ⎛⎜ P −P
where Pwfo,i, denotes the observed wellbore pressure of the ith ∂k wfo, i wfc, i ⎟⎠ ∂k
segment, and Pwfc,i is the corresponding calculated wellbore j i = 1⎝ j
6 SPE 84880

Above contains the derivative of wellbore pressure to Table-1 Basic Parameters of the Wellbore-
gridblock permeability. This is the sensitivity coefficient and Reservoir Model
is calculated by the perturbation method. The sensitivity of Model Dimensions 4000- × 3000- × 360-ft
wellbore pressure at the ith segment to the jth wellbore Grids (uniform) 20 × 15 × 9
gridblock permeability is given by: Kx, Ky, Kz (base case) 200, 200, 100 md
φ 0.15
Swc 0.2
Sgc 0
⎛ k + δk ⎞ − p ⎛k ⎞
∂p p ⎜
wfc, i ⎝ j ⎟
j⎠ wfc, i ⎜⎝ j ⎟⎠
Kro(Sw=Swc), Krw(Sw=1) 0.8, 1.0
wfc, i Krg (Sg=Sgc), Krg. (Sg=1-Swc) 0.8, 0.9
= ρo 35 lbm/ft
3

∂k δk µob 1 cp
j j
Bob 1.2 RB/STB
where δkj is the perturbation of kj. In the perturbation method, Rsb 1.13 scf/STB
the magnitude of the perturbation is critical for the accuracy of Pb 2800 psi
the sensitivity. This can be determined by numerical Pi (at OWC) 4000 psi
experimentation. Well Placement (heel to toe grids) (6,8,6) to (15, 8, 6)
Well Standoff (to OWC) 140 ft
Well Length 2000 ft
Finally, the elements of the Hessian are calculated as follows: Well ID 0.25 ft (base case), 0.45 ft
⎛ ∂2P ⎞ Well Roughness 0.001ft

∂2 f N ⎜ ∂Pwfc,i ∂Pwfc, i ⎛ ⎞ wfc,i ⎟


Well Rate 10,000 STB/Day
=2 ∑ ⎜ −⎜P −P ⎟
∂k k i =1⎜ ∂k ∂k ⎝ wfo, i wfc, i ⎟⎠ ∂k k ⎟
kl ⎜ k l kl ⎟
⎝ ⎠ 4020

Wellbore Pressure (Psi)


4000

The convergence criterion for the Levenberg-Marquardt 3980


method is given by: 3960
3940
f ⎛⎜ k l + 1 ⎞⎟ − f ⎛⎜ k l ⎞⎟ ≤ ε 3920
⎝ ⎠ ⎝ ⎠ 0 1 2 3 4 5 6 7 8 9 10 11
Segment Number
where ε is a small number representing the convergence
tolerance. Ref. 4 includes more information about the 10 Days 115 Days
inversion methodology.
Fig. 1 -- Wellbore pressure profile at two different times (.25 ft ID).
Segment-1 is the heel, segment-10 the toe.
Minimizing the Far-Field Effect: If k is the vector of
parameters to be estimated, we define k as k = [m k α ]T where
1
mk is the column vector representing the permeabilities (ky and
kz) of the gridblocks penetrated by the well, and α is a
Sensitivity values

0.1
multiplier for the rest of the permeability field. Thus, α is one
more parameter to be estimated. Estimation of α along with 0.01

mk allows us to adjust the far-field values along with mk. 0.001


Otherwise mk needs to accommodate for the disparity between
assumed and true states of the reservoir permeability field. 0.0001

Introduction of α is critical when late-time pressure data is 0.00001


used for the inversion. A similar approach can be used for 0 2 4 6 8 10 12
inclusion of liner roughness as an inversion parameter, i.e., Segm ent num ber
adding one more parameter to the vector of unknowns.
Sensitivity t o Kx Sensitivit y t o Ky Sensit ivity to Kz

SI Metrix Conversion Factor


bbl × 1.589 874 E – 01 = m3 Fig. 2 – Sensitivity of pressure at heel segment (no. 1) to
°F (°F – 32)/1.8 = °C permeability of wellbore gridblocks
ft × 3.048* E – 01 = m
in × 2.54* E – 00 = cm
lbm × 4.535924 E – 01 = kg
psi × 6.894 757 E + 00 = kPa
*Conversion factor is exact.
SPE 84880 7

210 160

SQRT(Ky*Kz) (md)
150
180
Permeability (md)

140

150
130

120 120
0 1 2 3 4 5 6 7 8 9 10 11
Segment Number
90
0 1 2 3 4 5 6 7 8 9 10 11 True Values 0 Psi Drift
Segment Number +2 Psi Drift -2 Psi Drift

True Kz Estimated Kz Fig. 6 – Estimation of permeability profile with uniform drift (5-
node configuration)
True Ky Estimated Ky
True SQRT(Ky*Kz) Estimated SQRT(Ky*Kz)

Fig. 3 – Estimation of permeabilities normal to well path and their 250

SQRT(Ky*Kz) (md)
geometric mean 200

180 150
SQRT(Ky*Kz) (md)

100
150 50

0
120 0 1 2 3 4 5 6 7 8 9 10 11
Segment Number
90
0 1 2 3 4 5 6 7 8 9 10 11 True Value No Drift
Segment Number Drift (+/- 1 Psi) Drift (+/- 2 Psi)

Fig. 7 – Estimation of permeability profile with nonuniform drift (5-


True Values 2 Nodes Near the Heel node configuration)
5 Nodes Near the Heel 5 Nodes Along the Well

Fig. 4 – Effect of measurement density on permeability estimation


250
SQRT(Ky*Kz) (md)

200
700
150
600
SQRT(Ky*Kz) (md)

500 100
400 50
300 0 1 2 3 4 5 6 7 8 9 10 11
200 Segment Number
100
True Values
0
0 1 2 3 4 5 6 7 8 9 10 11 Inversion with True Roughness
Roughness Underestimated by x2
Segment Number
Fig. 8 – Estimation of permeability profile with imperfect
True Values 5 Nodes Along the Well knowledge of wellbore hydraulics, 5-nodes, (0.25 ft liner ID)

Fig. 5 – Estimation of permeability profile when lateral


permeability variations are pronounced (5-node configuration)
8 SPE 84880

180 400
SQRT(Ky*Kz) (md)

350

SQRT(Ky*Kz) (md)
160
300
140 250
120 200
150
100
100
80 50
0 1 2 3 4 5 6 7 8 9 10 11 0
Segment Number 0 1 2 3 4 5 6 7 8 9 10 11
Segment Number
True Values
Inversion with True Roughness True Values Far Field Underestimated
Roughness Underestimated by x2 Far Field Overestimated

Fig. 9 – Estimation of permeability profile with imperfect Fig. 12 – Estimation of permeability profile with imperfect
knowledge of wellbore hydraulics, 5-nodes, (0.45 ft liner ID) knowledge of reservoir permeability field (without
far-field multiplier).

3990 146
Wellbore Pressure (Psi)

145

SQRT(Ky*Kz) (md)
3988
144
3986 143
3984 142
3982 141
140
3980
139
3978 138
3976 0 1 2 3 4 5 6 7 8 9 10 11
0 1 2 3 4 5 6 7 8 9 10 11 Segment Number
Segment Number
True Values Far Field Underestimated
10 Days 115 Days Far Field Overestimated

Fig. 10 – Wellbore pressure at two different times (0.45 ft liner ID) Fig. 13 – Estimation of permeability profile with imperfect
knowledge of reservoir permeability field (with far-field multiplier).

160
SQRT(Ky*Kz) (md)

160
SQRT(Ky*Kz) (md)

150
150

140 140

130 130
0 1 2 3 4 5 6 7 8 9 10 11
Segment Number 120
0 1 2 3 4 5 6 7 8 9 10 11
True Values Underestimated by x2 Segment Number
Overestimated by x2 Underestimated by x5
Overestimated by x5 True Values Short-time Estimation
Long-time Estimation
Fig. 11 – Estimation of permeability profile with imperfect
knowledge of wellbore hydraulics. (0.25 ft liner ID; roughness Fig. 14 – Estimation of permeability profile with imperfect
used as an inversion parameter; curves correspond to different knowledge of relative permeability function. (Inversion
values of initial guess for pipe roughness.) corresponds to production times of 10 days and 100 days.)
SPE 84880 9

1.2
Relative Permeability

1
0.8
0.6
0.4
0.2
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
Sw

True Krw True Kro


Straight Line Krw Straight Line Kro

Fig. 15 – True and assumed (straight-line) relative permeability


curves (oil-water)

1
Relative Permeability

0.8

0.6

0.4

0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Sg

True Krg True Kro


Straight Line Krg Straight Line Kro

Fig. 16 – True and assumed (straight-line) relative permeability


curves (oil-gas)

160
SQRT(Ky*Kz) (md)

140

120

100
0 1 2 3 4 5 6 7 8 9 10 11
Segment Number

True Values Short-time Estimation


Long-time Estimation

Fig. 17 – Estimation of permeability profile with combined effects


of uncertain wellbore hydraulics, reservoir permeability field, and
relative permeability function. Inversion corresponds to
production times of 10 days and 100 days. (In the inversion true
roughness is underestimated by a factor of two; true permeability
field has mean of 200 and 100 md, for horizontal and vertical
permeability, and variance of 40 md; initial guess assumes a
variance of zero; finally inversion uses a straight-line relative
permeability function instead of the true functions, Figs. 15-16.)

You might also like