Escolar Documentos
Profissional Documentos
Cultura Documentos
M.J. O’Sullivan1, A.E. Croucher1, J.A. Newson1, E.B. Anderson2, A. Hochwimmer3, S. Yamazawa3 and S. Lin4.
e-mail: m.osullivan@auckland.ac.nz
[ ]
numerically generated model responses, the
following major ingredients are essential: E in = (1 − φ i )ρ r i c r i Tin + φ i ρ lni Slni ulni − ρ vni Svn i (4)
The first of these requires the development of fast ρ vni density of steam
numerical solvers (see ‘Fast Numerical Solvers’) and Slni water saturation (volume fraction)
the second requires a robust and fast fitting technique
(see ‘Model Fitting’). The development of the user S nv i steam saturation
interface and related database management is ρr i rock density
described in the ‘Graphical Interface’ section.
cr i rock specific heat
FAST NUMERICAL SOLVERS Tin temperature
For a single layer uniform porous medium model, the ulni specific energy of water
flow near a well is radially symmetric and a 1D radial
grid can be used. For such a grid the discrete mass u vn i specific energy of steam
and energy balance equations can be written as
[
Vi M in +1 − M in + ] These formulae apply for general two-phase
[
Vi E in +1 − E in + ] calculation (see UK Committee on the Properties of
Steam, 1970).
∆t n +1
[FE n +1
i+ 12 2 2 2
]
A i + 1 − FE in−+11 A i − 1 = 0
(2)
The total flow is given as the sum for the individual
It is an important feature of finite difference formulae phases:
(1) and (2) that all the flux terms are evaluated at FM in++11 = FMLni++11 + FMVin++11 (5)
time tn+1. Such schemes are called “fully implicit”. 2 2 2
FM n +1 mass flux the relative permeabilities (see below) and ν lni , ν vni
i + 12
are the kinematic viscosities of each phase.
FE in++11 energy flux
2
time step between times tn and tn+1 Similarly, the total energy flux is given by
∆t n +1
FE in++11 = hlni + 1 FMLni + 1 +
2 2 2
The mass and energy contents are given in terms of
the other quantities by the formulae Tn
i +1 − Tin (8)
hvni + 1 FMVin+ 1 + K
[ ] ∆ri + 1
2 2
M in = φ i ρ lni Slni − ρ vni Svn i (3) 2
Here hlni + 1 is the enthalpy of water and hvni + 1 is S nv +i 1 ). The system (13) is solved iteratively using the
2 2
The last term in (8) is the energy flux resulting from The k+1st estimate of the solution Xk+1 is calculated
conduction. The quantity ∆ri + 1 is the distance using
[ ]
2
At the first time step the initial data provides values The process is initiated by using the solutions at time
for p i0 … and Ti0 (or S 0v i ) for i = 1, 2, … , M. The tn as the first estimate of the solution at time tn+1. If
boundary conditions must also be specified. It is the scheme does not converge the time step is
usually assumed that the model is sufficiently large decreased (see Pruess, 1991).
so that it is “infinite acting” and no flow occurs at the
end of the model. In TOUGH2 and in the simplified simulator
developed here the Jacobian derivative entries are
Thus calculated numerically using difference
approximations.
FM nM++1 1 = 0 (9)
2
Equation (14) represents a large system of sparse
linear equations. In TOUGH2 there are several
FE nM++1 1 = 0 (10) methods available for solving them. Some of the
2
latest methods available are the conjugate gradient
At the well base or sand face end of the model the methods such as GMRES and BICGSTAB (Bullivant
total flow is specified: thus et al., 1991, Moridis and Pruess, 1997).
The MINC system (Pruess and Narasimhan 1982) is The first of these steps may be carried out manually,
a numerical generalisation of the double porosity idea using knowledge of the reservoir and modelling
introduced by Warren and Root (1963). The idea experience, or (at least in simpler cases) some degree
of automation may be introduced by applying ‘model the layout of wells in the test, the positions of
recognition’ techniques (Horne 1994). measuring instruments, the field data and the model
specifications.
Tuning of the model parameters may be cast as a
non-linear optimisation problem, in which the Input and output of data is also handled by the
discrepancy between the data and model response is graphical interface. Data are stored in three
viewed as a function of several variables (the standardised databases:
unknown model parameters) to be minimised.
Commonly, this discrepancy is measured using a • ‘Field’ database- contains information
‘weighted least-squares error’ ε defined by: concerning the field in which the test is carried
out, including the locations of the wells and their
d i − mi (p )
nd 2
∑
deviations
ε (p ) = (17)
•
i =1 σi
Observation data table- contains the pumping
and observation data
where p is the vector of model parameters, nd is the • AWTAS test database- contains information on
number of data points, di is the ith data point, mi is the the configuration of action and observation wells
modelled approximation to di and σi is the estimated in the test, details of the models used, and the
measurement error in di. test results.
This approach has the advantage of flexibility, in that Once an analysis has been set up and set running, the
it is applicable to any data set (e.g. mixtures of progress of the model fitting process and current
pressure and temperature measurements). More model run are displayed graphically. Final fitted
importantly, any type of model can be used- be it model results are displayed alongside the observation
analytical or numerical. As numerical models are a data, together with the fitted model parameter values.
necessary ingredient in the AWTAS software, this is
the parameter tuning approach adopted. EXAMPLE
Perhaps the most demanding of the various tests so
An efficient and robust nonlinear optimisation
far conducted using AWTAS is a drawdown-buildup
algorithm is required to minimise the function ε(p), test in a hot water reservoir, starting from an
since this function can become time-consuming to undisturbed pressure close to the boiling point. Such
evaluate and very complex when numerical models a test involves very non-uniform processes near the
are used. AWTAS employs the so-called ‘Full wellbore. During the drawdown phase of the test, a
Newton’ method of Dennis et al. (1981) (better boiling zone develops that expands outwards as time
known as the ‘NL2SOL’ algorithm) which is a increases. After the well is shut in, the liquid
slightly more advanced version of the well-known saturation increases in the boiling zone as the
Levenberg-Marquardt method. Both of these pressure near the well rebounds. A ring of condensed
methods are specifically designed for least-squares water forms around the wellbore and slowly spreads
minimisation problems and offer advantages over outwards. Eventually the whole of the boiling zone,
other methods in both efficiency and robustness. The which was established during drawdown, re-
algorithm requires the Hessian matrix of the function condenses and liquid conditions are again present
ε(p) to be calculated, which is carried out using finite throughout the reservoir.
difference approximations.
The results of numerical simulations (using the solver
When more complex numerical models are used, it is described above in ‘Fast Numerical Solvers’) of two
possible that the function ε(p) could develop more drawdown-buildup tests are shown in Figures 1-5.
than one local minimum, in which case it may These results represent pressure and temperature
become necessary to find the absolute minimum results at the wellbore. The parameter values used
using a global searching method (such as the are given in Table 3.
‘simulated annealing’ algorithm- see Kirkpatrick et
al. (1983)). Unfortunately such methods generally Test I is for a reservoir with an undisturbed pressure
require far more function evaluations than do high enough to prevent any boiling from occurring
conventional local optimisation methods (Goffe et al. during drawdown. The Horner plot of pressure
1994), so their use is to be avoided if at all possible. during buildup gives the expected straight line
(Figure 2). However test II has an undisturbed
GRAPHICAL INTERFACE pressure low enough to cause boiling (and
consequently the processes described above) to
Setting up a well-test analysis using the AWTAS
occur. The two distinct sections of the Horner plot
software is carried out via an easy-to-use graphical
(see Figure 5) correspond to two-phase and liquid-
interface. This enables the user to view graphically
only reservoir behaviour respectively.
Parameter Value 36
35.5
Initial pressure 50 bar (test I), 36 bar (test II)
35
Initial temperature 240 °C
Pressure (bar)
34.5
Pumping rate 50 kg.s-1 34
Layer thickness b 100 m 33.5
Permeability 10-13 m2 33
46
45
simulated results from test II. It was found that
44 AWTAS was able to recover the reservoir parameters
43 (permeability, porosity and fractional dimension)
42
when started from some other initial parameter
41
40 values. Table 4 shows the initial and fitted parameter
0 10 20 30 40 50 60 70 80 values. The fitted parameter values are close to those
Time (hr)
used to generate the original data set (see Table 3),
Figure 1: Pressure vs. time (test I) with the discrepancy being attributable to the added
50
noise. The artificial test data and the model results
for the fitted parameter values determined by
49.5 AWTAS are shown in Figure 6 and Figure 7.
Pressure (bar)
49
Parameter Initial value Fitted value
48.5 Permeability 8×10-14 m2 9.83×10-14 m2
Porosity 0.15 0.108
48
Fractional 2.0 1.801
47.5 dimension
50 20 10 5 2 1
Horner time
Table 4: Initial and fitted parameter values
Figure 2: Horner plot (test I)
36 37
36
35
35
Pressure (bar)
Pressure (bar)
34 34
33 33
32
32
31
31 30
30 29
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
Time (hr) Time (hr)
Figure 3: Pressure vs. time (test II) Figure 6: Pressure data and fitted curve (test II)
240 241
240
239
Temperature (deg C)
Temperature (deg C)
239
238
238
237
237
236
236
235 235
234 234
233 233
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
Time (hr) Time (hr)
Figure 4: Temperature vs. time (test II) Figure 7: Temperature data and fitted curve (test
II)
A number of tests of AWTAS on field data from Finsterle, S. (1997), “iTOUGH2 user’s guide”,
various geothermal fields have been carried out, with Version 3.1, Report LBNL-40400, Lawrence
mixed success. The main difficulties that have arisen Berkeley Laboratory, Berkeley, CA.
are related to the quality of the data and the types of
tests. Most reasonable quality data sets are for multi- Finsterle, S., Pruess, K., Bullivant, D. P. and
rate injection-fall off tests. Past analyses of these O'Sullivan, M. J. (1997), “Application of inverse
tests have used only the part of the record that mostly modeling to geothermal reservoir simulation”, Proc.
closely fits the traditional constant rate injection-fall 22nd Workshop on Geothermal Reservoir
off test. The analysts have used graphical methods Engineering, Stanford University, Stanford,
such as a Horner plot. California, 8 p.
With AWTAS the whole complex injection record Goffe W.L., Ferrier G.D. and Rogers J. (1994),
and corresponding pressure response can be Journal of Econometrics, 60, 65-.
incorporated. However the non-isothermal nature of
the test, often with an injection temperature very Grant, M. A. and Sorey, M. L. (1979), “The
different from the reservoir temperature, makes compressibility and hydraulic diffusivity of a water-
analysis based on a simple reservoir model pointless. steam flow”, Water Resources Research 15(3), 684-
It is apparent even from a visual examination of the 686.
data that the temperature dependence of porosity and
permeability is an important factor. A model for Horne, R. N. (1994), “Advances in Computer-Aided
representing temperature dependence of rock Well-Test Interpretation”, JPT (July 1994), 599-606.
properties suggested by Nakao and Ishido (1998) has
been included in AWTAS and some success has been Horne, R.N. (1995), Modern Well Test Analysis, 2nd
achieved with it in matching field data. Further work ed., Petroway Inc., Palo Alto, CA.
on the topic is proceeding.
Kirkpatrick S., Gelatt C.D. and Vecchi M.P. (1983),
ACKNOWLEDGEMENTS Science, 220, 671-680.
NEDO (New Energy and Industrial Technology Moridis, G., and Pruess, K. (1997), “T2SOLV: An
Development Organization of Japan) provided enhanced package of solvers for the TOUGH2 family
funding for this work as part of the New Sunshine of codes”, Proc. 22nd Workshop on Geothermal
Programme. Reservoir Engineering, Stanford University,
Stanford, California, 295-301.
REFERENCES
Black, J. H., Barker, J.A. and Noy, D.J. (1986), Nakao S. and Ishido T. (1998), “Pressure-transient
“Crosshole investigations- the method, theory and behaviour during cold water injection into
analysis of crosshole sinusoidal pressure tests in geothermal wells”, Geothermics, 27(4), 401-413.
fissured rock”, Stripa Project IR 86-03, SKB,
Stockholm. Noy, D. J., Barker, J. A., Black, J.H. and Holmes,
D.C. (1988), “Crosshole investigations-
Bullivant, D. P., O'Sullivan, M. J. and Zyvoloski, G. Implementation and Fractional Dimension
A. (1991), “Enhancements of the MULKOM Interpretation of Sinusoidal Tests”, Stripa Project IR
geothermal simulator”, Proc. 13th New Zealand 88-01, SKB, Stockholm.
Geothermal Workshop, University of Auckland, 175-
182. O’ Sullivan, M. J., Bullivant, D. P., Follows, S. E.
and Mannington, W.I. (1998), “Modelling of the
Dennis J.E., Gay D.M. and Welsch R.E. (1981), Wairakei- Tauhara Geothermal System”,
ACM Transactions on Mathematical Software, 7, Proceedings TOUGH Workshop ’98, Berkeley,
348-368. California, 1-6.