Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
T.J.Woodhead, E.J.M. Willemse and R.M. Mills: Shell International: The Hague.
The field contains light oil in an initially saturated oil column 3. Static Modelling
and with an overlying primary gas cap. The primary The main producing interval consists of limestones and
production mechanism is gas-oil gravity drainage (GOGD) in dolomites with some isolated sand layers and a few anhydrite
which gas liberated from matrix oil fills the fractures allowing and shale layers. The shales are more common in the lower
oil to drain into the fractures via gravity forces and then to be part of the formation and anhydrite is found as beds, nodules
produced. and secondary infilling, mainly in the upper part of the
reservoir.
The field production was initially increased up to a plateau and
this plateau period was maintained for a period of 10 years. Gamma ray, neutron and resistivity curves were used in
The production rate was then significantly reduced due to the combination to define the shale layers while the neutron, sonic
rapidly thinning fracture oil column. The fracture oil column and resistivity data were used to define the layers of dense
then expanded allowing the production rate to be increased. rock in the reservoir. Porosity was calculated from sonic logs,
Production since then has slowly declined and there has been a calibrated with core data. Hydrocarbon saturations were
gradual further thinning of the fracture oil column (see Figure calculated from capillary pressure data measured on core
1). samples. These measurements were predominately performed
on air-mercury representing a strongly non-wetting system.
A secondary recovery scheme for based on gas injection is Typically, carbonate reservoirs exhibit mixed or oil wet
now being considered. A detailed evaluation of the feasibility characteristics and hence the laboratory capillary pressure and
of this scheme was the primary objective of the current study. relative permeability data is not likely to be representative of
A base-case development plan has been constructed in which fluid flow in the reservoir. Wettability therefore, has to be
gas is injected at various crestal locations to repressurise the taken as a significant uncertainty parameter in the simulations.
reservoir back to the initial saturation pressure.
Matrix permeability was calculated directly from the porosity-
The physical mechanism being employed in this recovery permeability relationship derived from the core data. This
process is the enhancement of the GOGD process due to the relationship showed a high degree of scatter in the data.
effects of a reduction in the gas-oil interfacial tension (IFT) at Matrix permeability was therefore treated as a major
increased reservoir pressures. This process is enhanced by the uncertainty in the simulations.
re-solution of liberated solution gas (oil swelling) at higher
saturation pressures. A full 3D geological model was constructed of the field
consisting of the major reservoir intervals identified on logs.
The physical processes involved in the production mechanism The model building process was accelerated by selectively
critically depend on the interaction between many different picking the tops of the flow units in the key wells, and
reservoir and fluid parameters. Furthermore the available data interpreting the structure between them based on a kriging
set was relatively sparse This emphasises the underlying algorithm. This kriging procedure was also constrained by the
problem of making predictions based on reservoir models for top structure map. A spatial inter-well porosity distribution
which a large degree of uncertainty exists, both in the was calculated by interpolating between well values. Figure 2
understanding of the reservoir geology and in the data used for shows a view of the geological model. This model was further
describing the fluid flow processes. It was recognised that any used to calculate an expectation GRV and its associated range
predictive reservoir study would have to take full account of due to structural uncertainty.
the impact of static and dynamic uncertainties on model
forecasts. Following analysis of the full data set, the following 4. Fracture Modelling
emerged as the key sub-surface uncertainties High well productivities as well as substantial mud losses
during drilling indicate the existence of an extensive field-
Table 1 : Key Sub-Surface Uncertainties wide network of conductive fractures. A conceptual model for
1 matrix permeability levels and distribution the fracture network was constructed, based on outcrop
2 fracture permeability levels and distribution observations and in which the presence of axial and cross-
3 aquifer influx axial fractures as well as stylolites is indicated. The fracture
4 reservoir volumes in place (STOIIP) model is illustrated in Figure 3.
5 matrix porosity and capillary pressures
6 fracture porosity The fracture system is a key factor influencing gas-oil-gravity-
7 continuity of matrix blocks and matrix block heights drainage. In order to simulate GOGD dynamically, the
8 fluid PVT properties and compositional grading following geological fracture network parameters must be
9 hydrocarbon saturations and relative permeabilities investigated:
• block height
• fracture permeability
The multiple scenario modelling strategy was therefore based • fracture porosity (as a fraction of the gross rock volume)
on designing and constructing a set of history-matched sub- • fracture surface area per matrix volume.
surface models to represent the above uncertainties.
SPE 59453 Integrated Modelling and Uncertanty Management of a Gas Injection Scheme in a Highly Fractured Reservoir 3
Based on worldwide experience, a range of fracture porosity Several alternative geological hypotheses have been tested to
was used between 0.01% and 0.3%, with a base case of 0.1%. investigate the aerial fracture distribution. Both bending and
Fracture permeability was derived by first establishing flexural slip have been proposed as important deformation
permeability thickness from a study of steady state flow mechanisms. In addition to these mechanisms, it is possible
analysis in 70 wells. An absolute high value in the range of the that fracture development has been influenced by movement
average effective permeability was computed by assuming the over the underlying thrust. The areal strain distribution due to
thickness to be equal to the flowing interval, yielding an these various mechanisms was calculated by applying several
average effective permeability of 9D. A low value was alternative algorithms to the top structure map.
obtained by assuming the thickness to be equal the entire
formation, yielding an average effective permeability of 0.7D. The best qualitative correlations between computed strain and
A range of 1-5D was used in the simulations, with a base case well inflow data were obtained from the curvature analysis
of 1D. and the vertical shear reconstruction. Both models were
translated into fracture flow properties by subdividing
The block height represents the distance between barriers curvature or vertical shear magnitude into four classes. Within
across which there is no capillary continuity. Capillary each class, the cumulative frequency of permeability thickness
boundaries may be formed by shales and bed-parallel was established. These frequency distributions were then used
stylolites. Well logs were used to measure the distance to convert the curvature and vertical shear maps into
between shales (both true vertical as well as corrected for layer permeability thickness maps
dip). Assuming negligible capillary hold-up, the proper block
dimension is the arithmetic average of these values, which is The permeability thickness maps were upscaled to the same
just over 70ft after correction for layer dip. This would serve grid size and incorporated into the 3D geological model.
as the most optimistic case. For comparison, the median value Finally, the permeability thickness maps were converted into
would suggest a block height of about 27ft. A minimum block fracture porosity and intrinsic fracture permeability maps
height, based on the harmonic mean, would be in the order of using the standard relationships between permeability
10ft. thickness, effective permeability, intrinsic fracture
Anhydrite and dense carbonate layers also occur in the permeability, fracture aperture and fracture spacing. These
formation, and may also form capillary barriers depending on maps were also upscaled for use in the simulations.
relative time of diagenesis and fracturing. If the dense
carbonates formed after fracturing, the average block height is 5. Dynamic Modelling and Uncertainty Analysis
reduced significantly to 35ft, with a median of about 10ft. An uncertainty analysis based solely on a large full field
Smaller structures with less lateral extent such as thin shales simulation model is prohibitively time consuming and
and stylolites may also affect capillary continuity. However, resource intensive. A combination of simple and more detailed
this effect is likely to be more localised. Nonetheless, the thin models was therefore constructed, the simpler models being
shales and stylolites are unlikely to be identified on wireline used for sensitivity and risk analysis. The models constructed
logs and add uncertainty to the above estimates. The results of were as follows:
the geological analyses and the values used for the 1D • a one-dimensional stack model of the entire reservoir, in
simulation are summarised in Table 2. which the GOGD flow mechanism was taken into account
by detailed modelling of the vertical flow (including
Table 2: Calculated vertical shale and dense block-to-block interaction and capillary continuity)
rock separation between matrix blocks
Shales Shales Dynamic Reservoir
& Simulation • a two-dimensional stack reservoir model in the field was
Dense
divided into areal reservoir sectors, each of which was
Rock
High (arithmetic 73 35 70
individually represented by a vertical flow stack. These
mean) sectors were allowed to partially communicate with each
Medium 27 10 30 other in the two areal directions.
Low (harmonic 11 4 10
mean) • a three dimensional model in which the flow within two
particular reservoir sectors was modelled in detail using a
fine scale three dimensional grid
For volumetric purposes, and to allow meaningful 2D or 3D
dynamic reservoir simulations, stratigraphic and areal
variations in fracture properties were also studied and Taking advantage of its relatively small size and simplicity,
compared against well performance. The fracture spacing was the one-dimensional model was used extensively to perform a
established to be of the same order of magnitude as the layer full uncertainty analysis and to derive reservoir forecasts for
thickness which the impact of various uncertainties was explicitly
evaluated. The 2D model was used to confirm the results of
4 T.J.WOODHEAD, E.J.M.WILLEMSE & R.M.MILLS SPE 59453
the 1D model and to evaluate cross flows between sectors and The strength of this aquifer was taken to be a further history
their impact. The 2D model, in addition, allowed future matching parameter. In all cases the aquifer strength used in
offtake rates from each of the sectors to be individually the history matches fell in the range 0.5-3, where the aquifer
optimised. Finally, the 3D model was used to look in detail at strength is a parameter representing the aquifer volume as
the fluid displacement mechanism within the sectors. The multiple of the reservoir gross rock volume.
impact, for example, of areal or along-dip flow in addition to
GOGD was evaluated. Different well configurations and Compositional grading has been observed in the field oil
geometries were also evaluated using the 3D model. column. For the purposes of the current modelling exercise,
the impact of this variation on fluid GOR’s, viscosities and
5.1 One Dimensional Reservoir Model densities was considered to be minor relative to the
Figure 4 illustrates the one dimensional reservoir model, uncertainties associated with understanding fluid flow in the
showing a saturation plot for initial conditions as represented matrix-fracture system. An important parameter for
the reservoir simulator in fracture mode. The presence of determining GOGD rates in a fractured reservoir however, is
fractures is indicated by a column of 'fracture' grid blocks on the interfacial tension (IFT) between the gas and oil phases.
both sides of the matrix blocks. The impact of uncertainty was This parameter is strongly dependent both on reservoir
evaluated by performing multiple history matches and pressure and fluid compositions. It was essential therefore, to
forecasts using this model. estimate the likely variation in gas-oil IFT within the reservoir
due to pressure and composition variations with depth. An
Of the uncertainty parameters listed in Table 1, volumes, analysis was carried out to quantify this variation. From this
porosities and permeabilities can be explicitly represented in analysis, a likely mid-column value of IFT was determined to
the simulator as individual parameters. Capillary continuity be 6dyn/cm with a range of 2dyn/cm < IFT < 10dyn/cm. This
was modelled by choosing grid block layer thicknesses range of values was carried forward to the uncertainty
corresponding to a given matrix block height i.e. the grid analysis.
block thickness represented the assumed vertical distance
between shale discontinuities in the reservoir. In the 1D Another parameter having strong impact on capillary pressures
model, the reservoir volume distribution was represented by is matrix porosity. From the petrophysical analysis, capillary
taking equidistant horizontal slices through the 3D static pressures relationships were derived giving capillary pressures
model and assigning the gross rock volumes and average as a function of porosity. These relationships were then used
porosities of each of these slices to the appropriate layer in the to generate both water-oil and gas-oil capillary pressures for
simulation model. Variations in STOIIP were then obtained by use in the simulations. For a particular simulation, the porosity
using volume multipliers in the simulator. was first assigned a value from which the corresponding
capillary curves were selected. A reference value of IFT was
Volumetric uncertainty was represented in the model using a then chosen to convert air-mercury capillary pressures into the
range of (0.8-1.15)*expectation STOIIP based on the likely equivalent water-oil and gas-oil curves. For water-oil, a value
variation in GRV due to areal uncertainty in reservoir extent. was chosen from the literature. For gas-oil however, this factor
These variations were taken to be independent of porosity was given a value in the range 2dyn/cm < IFT ref < 10dyn/cm
variations, which were taken into account using a separate for each of the history matches performed, where IFT ref is the
uncertainty parameter. reference value of IFT at initial reservoir pressure. The impact
of the uncertainty in gas-oil interfacial tension is thereby
The petrophysical analysis showed that considerable modelled. The dependence of IFT on pressure was then
variability existed in the porosity-permeability data derived modelled by invoking the simulator option to allow the gas-oil
from matrix core analysis. For the 1D model history matching capillary pressures to re-scaled at each reservoir pressure
exercise therefore, the matrix permeability was taken to be a according to a table specifying IFT v pressure.
history matching parameter, i.e. allowed to vary for the
purposes of obtaining a history match. In all cases, the matrix From SCAL data derived from experiment the porosity
permeability used in the history matches fell in the range dependence of drainage relative permeability curves is
0.2mDarcy < Kmatrix < 10mDarcy uncertain. More significant in the field simulations however, is
the rock wetness and its likely impact on the imbibition of
From the fracture analysis, a range of matrix block height was water via the fracture system into the matrix as the FWOC
estimated. The impact of matrix block height (i.e. capillary rises. This process will have a significant impact on the oil
continuity) on the history match was modelled by history saturations remaining in the matrix oil column after it has been
matching the two extreme cases of matrix block height (10 ft swept by rising water levels. Using an imbibition curve for
and 70 ft) in addition to the mean (30ft). which oil wetness was assumed, a parameter α was introduced
with which the imbibition curve was re-scaled to give a family
Aquifer support in the field is thought to be small, although of curves representing oil wetness. In this representation a
there has been an observed historic rise in the fracture water- value α=0 represents water wet, whereas α=1 represents oil
oil contact. An aquifer was therefore represented in the wet. An intermediate value represents a mixed wetness. Given
simulation model analytically to represent this water influx. the fact that the underlying geology in the field is carbonate, it
SPE 59453 Integrated Modelling and Uncertanty Management of a Gas Injection Scheme in a Highly Fractured Reservoir 5
is likely that the field will exhibit predominantly oil-wet permeability and aquifer strength were systematically varied
behaviour. A base case value of α=.75 was therefore chosen until the simulated fracture fluid contact levels were in
for the simulations. agreement with the historically observed values.
Matrix-fracture relative permeabilities were chosen to allow The impact of the various uncertainties was then established
block to block interaction (BTBI) i.e. re-imbibition of oil into by varying one of the uncertainty parameters in the table
matrix blocks after its has drained into fractures (see Ref. 2). above and repeating the history matching exercise. The key
This process can have a significant retarding effect on the assumption in this approach is that each of the uncertainty
GOGD process. parameters can be assessed independent of the other
parameters. In total, 13 different history matches were
Finally, the reservoir fracture analysis showed that fracture obtained i.e. representing the base case, together with a high
porosity (as fraction of bulk rock volume) could be assumed to and low variation in each of the six uncertainty parameters.
be in the range .0001-.003 with an expectation value of 0.001.
Figure 5 shows the history match for the base case, as
Of the uncertainty parameters identified in Table 1, the first indicated by the historical FGOC and FWOC contact levels,
three (matrix permeability, fracture permeability and aquifer together with the history matches obtained for each of the
influx) were treated as adjustable variables in the history model variations. The history match in each case is seen to be
exercise. From the discussion above, the uncertainty arising of very good quality. The fact that a history match could be
from the remaining six was concluded to be fully represented obtained with such a wide range of matching parameters
by the following: - emphasises the significant remaining uncertainty in the
predictive capability of the current dynamic model. The
• Volumes in Place (STOIIP) impact of these uncertainties was therefore represented in the
• Matrix Porosity model forecasts by performing forecast runs for all thirteen of
• Gas/Oil IFT the history-matched cases.
• Matrix Block Height
• Oil Wetness of Water-Oil Relative permeabilities 5.2 One Dimensional Model Forecasts.
Using the history matches described in the previous section,
• Fracture Volume
forecasts could then be performed by running the simulations
with assumed offtake levels.
Table 3 shows these uncertainty parameters together with their
associated uncertainty ranges as represented in the simulator.
As a preliminary forecast, the base case model was run
without any gas injection to represent the no further activity
(NFA) case. The results of this run are shown in Figure 6, in
Table 3: Ranges of Uncertainty Parameters
which oil offtake rate, cumulative oil production and contact
Parameter Low Median High
levels are illustrated respectively. From the current cumulative
STOIIP as multiple of 0.77 1 1.15 oil level the recovery factor is predicted to increase by 15% by
expectation value the year 2030.
Porosity 7% 9% 11%
IFT 10dyn/cm 6dyn/cm 2dyn/cm The thirteen history-matched models were then run until for
Matrix Block Height 10ft 30ft 70ft the length of the forecast period. Gas injection volumes were
Oil Wetness (α) 0. .75 1.0 set to limit injection volumes after the reservoir has returned to
Fracture Volume .0001 .001 .003 initial saturation pressure. After commencement of gas
injection offtake levels were assumed to gradually increase.
A base case simulation model was established by taking the
median value for each of the above parameters. The model For each simulation, the incremental offtake rate was
was then history matched by varying the matrix and/or optimised by successively increasing the increment and re-
fracture permeabilities and the strength of aquifer influx into running the forecast until the oil column reduced to below the
the reservoir. a cut-off at some stage during the forecast period. In addition
a maximum offtake rate was imposed on the production. This
Historical production was simulated by representing reservoir resulted in an optimal forecast for each of the thirteen models.
production with a single well in the model for which the
productivity was adjusted to ensure that the well could be The base case results are included in Figure 6, where the
constrained using historical oil production levels. Hence the incremental benefit of gas injection is clearly illustrated.
historical fluid offtake volumes were automatically honoured Figure 6 also shows the corresponding results for the
in the model. remaining twelve history matched models. For each forecast,
three plots are presented, oil production rate, cumulative oil
The average historical fracture contact levels FWOC and production and fluid contact levels as a function of time
FGOC were used as history matching criteria. The matrix throughout the forecast period.
6 T.J.WOODHEAD, E.J.M.WILLEMSE & R.M.MILLS SPE 59453
Table 4 summarises the results. independence, the probabilities can be formally combined by
multiplication
Table 4: Summary of Forecast Runs Performed
Forecast Incremental Recovery as
%fraction of STOIIP
NFA - p[RF(t)] = Π ( p[RFPi(t)] )
Base Case 5.7 i=1,6
Low STOIIP 2.3
High STOIIP 8.5 where
Low Porosity 2.5
High Porosity 7.9 RF(t) = Π ( RFPi(t) )
i=1,6
Low IFT 3.5
High IFT 8.4 Here p[RFPi(t)] is the probability of the recovery factor for
Low Block Ht 7.2 parameter i at time t during the forecast period and RF(t) is the
High Block Ht 6.2 combined recovery factor at time t, resulting from a particular
Low Oil Wetness 4.6 value of each of the six parameter variables.
High Oil Wetness 6.7
Low Frac. Vol. 3.0 The recovery factors RFPi can be taken directly from the
High Frac. Vol. 9.6 simulation results at each timestep where the combined
* compared with NFA forecast with base case parameter probabilities are required. This allows a complete range of
values recovery profiles to be generated using the formula above.
Since the number of uncertainty parameters is, in this case, six
The large number of history matches possible within the and each has three discrete (low, median and high) outcomes,
uncertainty ranges of the various key reservoir and fluid the total number of permutations using the above formulae is
parameters illustrated the need to take uncertainty into account 36 = 729. From these permutations, a cumulative probability
in any prediction made by the reservoir model. Table 4 distribution can be generated for the recovery at 2030. Figure
illustrates the large range in incremental recovery predicted by 8 shows this curve.
the different model variations. These forecasts were made
using models in which only one parameter had been varied Reading-off the P85, P50 and P15 percentiles of this
from its base case value. Clearly many more possible distribution and then comparing with individual statistical
interpretations are possible where more than one parameter is recoveries allows the construction of the P85, P50 and P15
allowed to vary at the same time. The six parameters used in recovery profiles. These are shown in Figure 9.
the uncertainty assessment, listed in Table 3, can be assumed
to be independent parameters. This allows a statistical analysis 5.3 Two Dimensional Reservoir Model
to be made of the simulation results. Figure 10 illustrates the two-dimensional reservoir model
showing a saturation plot for initial conditions. Here again the
The full field forecasts were generated using a set of presence of fractures is indicated by a column of 'fracture' grid
parameters, which were taken to be fully representative of the blocks on either side of the matrix blocks. The model consists
complete range of uncertainties associated with the field data. of eight areal grid block stacks, each stack representing a
Furthermore, these parameters were selected in such a way particular sector of the reservoir in which distinct contact
that they could be assumed to be mutually independent. This behaviour has been observed. A history match was performed
allows a rigorous statistical analysis to be made of the forecast with the two dimensional model in which the parameters listed
results. in Table 3 were assigned their base case values.
As a first step in this analysis, it was necessary to establish Each sector was then history matched by varying the matrix
estimates of the probability associated with each of the various and/or fracture permeability values and aquifer strength
parameter values. The results of these discussions are assigned to that sector. In addition, the history matching
illustrated in Figure 7. The probabilities are expressed as procedure made use of a further set of parameters representing
triangular distributions from which the probabilities of the the partial communication between sectors. These parameters
low, median and high parameter values have been extracted as were multipliers on the transmissibility values that would
discrete values. The distribution for porosity has been taken result if the sectors were in complete communication. This
directly from the porosity distribution of the 3D static model. procedure resulted in a history match for each sector as
For the remaining parameters, STOIIIP, IFT, block height, oil illustrated in Figure 11. Here again, the history match in each
wettability and fracture volume, probabilities were estimated sector is seen to be very good. This confirms the applicability
by consideration of all the available analysis and by discussion of the base case parameters values used to the individual
with the team members of the relevant disciplines for each sector level and therefore strengthens the conclusions made
parameter concerned. Because of the assumption of from the one dimensional model.
SPE 59453 Integrated Modelling and Uncertanty Management of a Gas Injection Scheme in a Highly Fractured Reservoir 7
Figure 12 shows the full field forecast from the 2D model on a viewed at intervals during the historical production period.
sector by sector basis. In each sector the offtake rate has been The movement of the fracture GOC is clearly illustrated, the
successively increased until the local fracture oil column width swept oil region eventually reaching the bottom of the
reaches a critical value. For late production, the offtake rate in formation in the crestal region. The influence of preferential
each sector is reduced to maintain the fracture oil column production from certain wells is also shown by the uneven
width. The figure illustrates the different offtakes from the descent of the FGOC in the transverse cross section plots. The
various sectors due to their varying volumes initially in place average fracture contact levels however, are consistent with
and varying reservoir and fracture properties. The total field the historical levels for the two sectors hence showing that the
production at 2030 is however, almost identical with the 1D impact of areal flow on the overall sector production
forecast. This confirms the validity of and increases the behaviour is relatively minor. This further confirms the
confidence in the conclusions made from the simpler model. validity of the one- and two-dimensional model results.