You are on page 1of 16

PETROLEUM SOCIETY PAPER 2006-129

CANADIAN INSTITUTE OF MINING, METALLURGY & PETROLEUM

Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
History Matching Standards; Quality Control
and Risk Analysis for Simulation
R. O. BAKER, S. CHUGH, C. MCBURNEY, R. MCKISHNIE
Epic Consulting Services Ltd.

This paper is to be presented at the Petroleum Society’s 7th Canadian International Petroleum Conference (57th Annual Technical
Meeting), Calgary, Alberta, Canada, June 13 – 15, 2006. Discussion of this paper is invited and may be presented at the meeting if
filed in writing with the technical program chairman prior to the conclusion of the meeting. This paper and any discussion filed will
be considered for publication in Petroleum Society journals. Publication rights are reserved. This is a pre-print and subject to
correction.

Abstract the confidence level of a reserves determination and decrease


the range of uncertainty. The reverse is also true some
Despite forty years of simulation, the petroleum industry simulations with poor data are only applicable in a semi-
has been reluctant to discuss history match quality standards. quantitative sense. This paper has important implications for
This lack of standard is often justified by statements such as using simulation as a reserves determination tool. It is
“the reservoir is unique” and the data type and quality vary important that simulation studies be graded in terms of history
from field to field. While those statements are true to a degree, match quality, number of variables matched, percentage of
they leave the non-simulation personnel confused as to whether wells matched, data quality, what parameters were adjusted
a model is good or bad. This paper discusses quality control and by how much so that assignments of proved P90, probable
and history matching from experience of over 200 field P50, and possible reserves can be estimated.
simulation studies.

History match parameter tolerance depends upon: Introduction


q Drive mechanisms and related key parameters
q Reservoir architecture Although new technology and techniques hold great
q Study objective promise, workflow and quality control of data are still central
q Data quality issues in insuring reliable simulation forecasts as well.

There is often a big trade off between the quality of history In order to evaluate success and failure, as well as the need
match, size of model and time requirements. This paper for simulation, we must clearly understand why
discusses key parameters and how they apply to field simulation simulation/reservoir forecasts are successful and why they fail.
cases. There are three general categories that control a simulation
study:
It must be noted that flow simulation tools are aimed at the 1. Timing versus accuracy
most likely or P50 case. Higher quality studies with matching
multiple parameters and good data can significantly increase

1
2. Initial diagnostics
3. Coupling history match parameters versus history match Simulation studies directly integrate geological parameters
objectives with engineering data, such as production signatures and
pressure data, but that integration of data requires time and an
The following is a more detailed list of why sometimes understanding of reservoir mechanisms. In comparison,
reservoir forecasts have sometimes been unsuccessful: techniques such as conventional decline analysis are much
Timing quicker but there is not necessarily a direct link with geological

Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
1. The simulation study is taking too long relative to the or physical parameters. As a result, simulation studies will
business objective/decision require more time. The input of geological data and processing
2. Paralysis by analysis; reservoir engineers being of information has greatly improved in recent years1. This
paralyzed by the analysis of data. additional effort in integration in simulation however does come
3. The engineer assuming every variable in a simulation with benefits. Studies on revisions of reserves by a major oil
model is critical. company have shown that analytical/decline type forecasts
4. Confusing complexity with accuracy. Assuming bigger show larger revision in recovery factors and production profiles
and more complex simulation is always better or more than fields with simulation studies. In other words, with more
accurate. data and information incorporated via simulation the accuracy
Characterization of the prediction improved. This is especially true during early
5. Failure to adequately represent reservoir heterogeneity. development of fields.
6. Improper initial well diagnostic and reservoir diagnostic
and failure to recognize reservoir mechanisms.
7. Failure to calibrate the geological model's overall Initial Quality Control and Reservoir
average permeability.
Modelling Characterization Are Necessary to Pre-
8. Failure of coupling history match parameters to physical Condition the Model
parameters that strongly control simulation forecast
results.
9. The lack of understanding of how history match Reservoir Permeability Heterogeneity
parameters are controlled by physical parameters, such Understanding how reservoir properties vary within a
as relative permeability, permeability distribution and reservoir is critical to correctly predicting future performance
porosity and translating those parameters to predict for some reservoirs. This is especially true in the fluid injection
future performance. processes. Reservoir characterization and scale up are the
10. Mechanisms and operational limitations should be primary obstacles to improving simulation forecasts. We
recognized up front. should be clear on our definition of reservoir heterogeneity.
11. Selecting a small simulation study area is not trivial. Reservoir heterogeneity is the condition of physical and textural
Outer boundary effects significantly control recovery variation that actually interferes with the flow of fluids.
factor within an area Reservoir heterogeneity often is used as convenient “catch-all”
12. An assumption that increased computer/simulation speed phrase of uncertain meaning. The confidence level of a
directly translates to simulation time/turn around. simulation relies heavily on the accuracy and uncertainty in the
13. Poor understanding of how gridding effects affect input geological model. Permeability homogenization is very
simulation results. common to see in the initial mapping/geological model and is a
result of averaging in the vertical scale due to log reading,
Most of the above factors are easily controlled by better averaging in an areal scale by contour mapping, averaging due
quality control, as well as better reservoir characterization, to porosity-permeability transforms and limited data sampling.
resulting in less error in simulation forecasts and better business
coherence. "Error bars" are ±10 to 40%, as estimated by History matching of the dynamic data can narrow the error
Saleri1, on field wide simulation studies and greater error bars bars significantly. Our experience and the experience of others
are estimated for individual wells. In our opinion the quality of is that matching watercut in waterflood/water drive cases or
history matching and the number of parameters matched often GOR behavior in gas cap or gas injection case usually
control the range of uncertainty of an answer. introduces more permeability heterogeneity into the model and
improves performance forecasts. Thus history matching
Tables 1 to 3 give some guidelines to standards for GOR/watercut profiles is an important step and often results in
simulation. Table 1 gives typical history match parameters and local changes in permeability/porosity distribution. For
some acceptable ranges. Table 2 shows typical adjustment example, history matching of water breakthrough time for
parameters and some guidelines for their ranges. Table 3 gives individual wells, very often results in a reexamination of
key history match parameters sorted by drive mechanism. The reservoir hot streaks to explain early breakthrough. Similarly
purpose of this paper is not set absolute standards but rather to lack of water breakthrough in waterflooding, high GORs / low
stimulate discussion of guidelines. pressures in waterfloods may result in a search for lack of
continuity in some sands2. Issues such as reservoir
Length of Time of Simulation Study compartmentalization and permeability "hot streaks" often
Note that point one to four (re: study time) can directly totally dominate the forecast. Matching overall field behavior
conflict with points five to eight (including reservoir but not individual wells is a common mistake and throws away
heterogeneity, the need for areal large models) respectively. the ability of the model to suggest permeability heterogeneity.
The key to a successful execution of a simulation study,
limiting study time and manpower requirements is to focus on Figure 1 shows a comparison of geological model, an old
key parameters only, as shown in Table 1. An understanding of averaged simulation model and a more current model. The old
the reservoir mechanisms is critical, to identify those key model had great difficulty achieving a history match of oil,
parameters. water and gas production rates despite numerous history match

2
attempts. Yet a refined simulation model achieved a marked distribution are adjusted to history match watercut. This type
improvement in the first few runs. Figure 1 shows with less on history match doesn't tell us anything about the average
vertical averaging the permeability highs and lows were permeability. If average permeability in a drainage area is not
preserved which allowed a better history match. matched, it is not possible to accurately forecast infill wells.
This focus on matching watercut trends and neglecting well
Permeability heterogeneity does not always dominate drawdown is very common in the hundreds of field waterflood
reservoir performance. Therefore, when determining the level simulation studies we have seen.

Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
of detail the reservoir description needs in a simulation model,
drive mechanism and reservoir architecture must also be From the history matching process, it is also critical to match
carefully considered. More complex geological models are not inflow performance. Inflow performance will often be most
always the answer and can substantially slow down the progress sensitive to the average permeability in a well's drainage radius
without adding additional value. Some drive mechanisms such whereas, GOR/watercut production profile will be most
as expansion drive or solution gas drive are relatively sensitive to permeability distribution as well as relative
insensitive to permeability heterogeneity. Other drive permeability. Current research and practical field experience
mechanisms such as gas injection (horizontal gas floods) and show that constraining the well inflow models (average
waterflooding viscous oil projects are extremely sensitive to permeability in a drainage area) makes simulation models more
permeability heterogeneity. unique in their solutions. "Using watercut data alone does not
result in a very unique solution even in a homogenous
Gravity can mitigate permeability effects. For example, permeability field with permeability heterogeneity the problem
many high permeability reefs in Canada have high structural becomes less defined. However, using flowing pressure
relief and very efficient water drive or gas cap drive resulted constraints the history match runs substantially so that a more
because of gravity stabilization. These systems were simulated unique solution is possible ". (D. GUPTA)
with 1D flow models or simple material balance methods were
used2. Forecasts of these systems were very satisfactory Often companies do not record an entire flowing bottomhole
without detailed permeability description. This was a result of pressure history for all wells in a field. However, most North
excellent gravity stabilization of interface movements. Failure American fields or mature onshore fields are in the pumped off
to recognize drive mechanisms and doing "brain simulation" condition late in the life of the field. This condition allows for a
first can result in overkill when describing the permeability field "snap shot" history match of bottomhole flowing pressure.
for some systems.
A commonly taken short cut, in the interest of a quick look
Calibration of Average Permeability approach is to not match inflow and thus average permeability
but regulate withdrawal in the forecast by limiting wells to
In our experience, average effective permeability is the most
historic fluid and injection rates. This is an acceptable method
likely parameter to be incorrect in simulation studies. Yet
as long as fluid and injection rates are calibrated by
permeability is very often a controlling parameter in production
performance. When wells are not calibrated by performance
forecasts because of its effect on injectivity, productivity, and
often artificial peak oil or gas rates occur in the forecast.
aquifer influx, coning and induced hydraulic fracturing. It is
important to understand how initial permeability estimates were
Having said that we need to examine inflow performance
derived in the model, (i.e., were they derived from build up test,
and pressure, it is important that in coarse grid models (grid size
core air test or productivity testing). What makes permeability
>50m) that there are some “acceptable or reasonable” error bars
a difficult parameter to consider is that it is a tensor property
on producing pressures. Small scale effects such as gas break
and scale dependent. In-situ stress, shale baffling, fracturing
out or turbulence in gas wells will limit well productivity in the
and relative permeability effects need to be considered.
field very near wellbore but be transparent to coarse gridded
Because of the various factors above, simulation models often
models. Again it depends upon the objectives of the model, but
need to calibrate permeability. Pressure transient analysis
generally fine tuning producing well pressures with finely
and/or simple productivity index calculations should set the
gridded models may add very little accuracy to the forecast but
initial guess on reservoir permeability.
considerably lengthen the time line.
We often see (~80% of the time) where the sum of the
A common problem in some fields (as shown in Figure 2) is
capacity (kh) from core is not equal to larger scale data such as
that we see historical peak oil or fluid rates which require large
AOF, pressure buildups or productivity index. Figure 3 shows
pressure drawdown and high levels of permeability. But at later
the results from thousands of fields comparing (kh)core versus
stages wells are not in the pumped off condition (Pwf<500-1000
(kh)AOF tests. (not all data is shown) The plot shows averaged
KPa). In these cases, there is often a trade off between
(kair h) vs. pressure AOF tests. Clearly cored the air
matching peak oil rates and cumulative oil values vs. inflow
permeability for stress, initial water saturations and Klinkenberg
performance. To reconcile this problem we can calculate with
effects would shift the points to the right but not totally reduce
IPR analysis what the maximum operating bottomhole pressure
the scatter in the plot. Accounting for baffling in the reservoir
should be.
and convoluted flow paths again will shift the core calculated
permeability downwards where as fracturing will increase
Reservoir and Well Diagnostics; the Key to Good
permeability. The key point here is that we must examine large
Reservoir Characterization
and small scale permeability and reconcile them. Just looking at
core data alone may be naïve but not examining the core leaves One of the most difficult situations occurs when a field is
important insight on the table. new and has limited well data as well as limited dynamic data,
as noted by Saleri (1993). Whenever possible, an extensive
Often, in the interests of saving time and costs, the reservoir and well diagnosis should be done first. In the cases
simulation objectives are only to match watercut in water drive where we have dynamic data (production/pressure data), the
or waterflood cases. In physical terms this means that some data is extremely helpful in building good simulation models
combination of relative permeability and permeability and selecting key variables. Response to fluid injection, peak

3
oil rates, decline rate behavior, pressure vs. withdrawals profiles simulation forecasts. However, further grid refinement and
and GOR/watercut profiles should be carefully analyzed for adding complicated process is also problematic. There are two
parameters such as hot permeability streaks or lack of problems associated with a never-ending search for more grids
continuity. Rapid watercut or gas-oil ratio break throughs can and more features. Gringarten5 describes the problem for
indicate hot streaks in fluid injection studies3. We often see pressure transient analysis but it is fitting for simulation as well.
cases where reservoir/well diagnostics have been completely The first is that we can get caught up in simulation paralysis and
bypassed to “get on with the simulation” and the history the second is that the ability to resolve petrophysical properties

Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
matching exercise. An example of the importance of reservoir for each cell becomes poorer and poorer with smaller and
diagnostics is in a naturally fractured field where an anisotropic smaller grids. This has several consequences 1) a study is never
ratio (kx/ky) was determined first, then input in the simulation complete because there is always a new method or better
resulting in a 70% history match of water cut in the first run. refinement that has not been implemented, 2) there is never
4
Reservoir simulation will yield a result that is very dependent consensus on the results and 3) the general opinion is that the
upon the initial geological model and initial guess of reservoir simulation was useless because of points one and two4.
mechanisms or in other words the initial interpretation. Well
and reservoir diagnostics are key to getting good initial guesses. Grid resolution is dictated by 1) the size of geological
heterogeneity 2) areal model grid resolution; needed for
Well behavior should be analyzed to de-convolve operational evaluating infill wells 3) vertical grid resolution; need for
vs. reservoir effects. Reservoir simulation is strong at picking modeling gravity and capillary forces in a vertical sense 4)
out reservoir fluid flow phenomena that control well behavior, numerical dispersion. The grid resolution requirements from
but often operational effects can strongly control well behavior points two, three and four are largely met in today’s models or
as well. High water cut could be due reservoir effects such as at least we are very close to that level. The grid resolution
coning or channeling but it could also be due to poor cement needed to model gigascopic or megascopic (large scale)
bond. Out of nine possible causes of rising watercut, only three geological details is largely already incorporated in developed
are reservoir fluid flow controlled. Failure to recognize fields with < 40 acre spacing. The weakness of today’s model
operational or near wellbore effects vs. reservoir-controlled is the modeling of features that are smaller than interwell
variables can waste valuable time and results in poor study distance < 400 m but still large enough where averaging is not
focus and in long time-lines. appropriate. Yet some of these features are not second order
effects. However, populating the geological models with these
Diagnostics also help to build pre-conditioned geological and features and the prior identification of these features remain the
reservoir flow models. In the diagnosis of well and reservoir main bottle neck.
behavior, focus is in the two general areas of
compartmentalization and hot streaks. Typically, initial
geological modeling sometimes may overestimate reservoir Simulation Objectives, History Match
continuity. Therefore, reservoir and well diagnostics are critical
in low continuity / permeability situations. Simulation is a Variables and Physical Parameters
quantification of one’s own bias; therefore, production and In developing a forecast it is important that the engineer
pressure data analysis are absolutely critical to establishing a understands which parameters control future production and
good initial guess in the model if the data exists. Geostatistical focus on narrowing the range of these parameters. Table 1
models built with static data only tend to give very wide ranges shows major drive mechanisms controlling parameters. For
of production profiles but dynamic data such as production and example, a common objective of most simulation studies is to
reservoir pressure data often narrow the range down identify future infill drilling locations and optimize field
significantly. This helps identify reservoir mechanism, the development. In order for a successful infill well to be
initial geological model and operational aspects. evaluated, two risk factors have be to be evaluated. The first is
geological/petrophysical risk; i.e. are there favorable conditions
Calibration of Volumes in Place of permeability, porosity and top of structure to develop a good
Material balance calculations should be done before model well? The second risk is reservoir depletion; in other words, is
construction. Material balance calculations have an advantage the oil or gas saturation too low, or is reservoir pressure
over simulation studies in that the calculations use what is depleted so low that the infill well is uneconomic? Because of
relatively well-known, like production volumes and pressure, these two risk factors, simulation studies need to match well
and calculate what is unknown, like original oil in place. drawdown pressure to come up with reasonable average
Simulation studies on the other hand use volumetric/geological permeability and match GOR/watercut trends to address
analysis as the known and then calculate pressures and saturation risk. Matching reservoir pressure is also necessary to
production volumes. Therefore, material balance calculations address depletion risk.
can give considerable insight to what the volumes in place
should be. Similarly material balance calculations can be used As pointed out by Saleri 1993, many models are not that
to gain a sense of the magnitude of aquifer influx. sensitive to OOIP6,7. This especially true in relatively immature
fields. It is our experience however that in fields that are
mature (>20-30 years of production), that movable OOIP often
Confusing Complexity with Accuracy controls late stage oil rate decline. Figure 4 illustrates the case
of a forecast in 1996 based on an Sorw of 20% established from
Many times the layman assumes the most complex multiple lab and field tests. A consistently high prediction led
simulation model and largest number of grid blocks are the to the revision of Sorw to 30% in 2000.
most accurate. Field models now have 100 to 1000 times more
grid blocks than typical models 20 years ago. This decrease in A critical component of history matching is that we need to
grid block size has allowed much better flood front resolution examine not just the XY plots but also the bubble maps of the
and allowed better reservoir description in simulation models. spatial or areal variation between the model and the field as
There is no doubt in our mind that this has resulted in better these maps identify areas of uncertainty in the model. Figure 5

4
shows an areal saturation map (using green for oil and blue for not be mapped and the initial geological description will result
water) on which the error between actual and simulated in unrealistically high well capacity (kh). With history
performance is plotted as bubbles. The plot quickly illustrates matching, long term well deliverability will be substantially
areas of high and low confidence in the saturations. lowered

Therefore, if infill drilling is a key objective of simulation, Matching multiple parameters such as reservoir pressure,
it is important that the history match have clear targets for all watercut and flowing bottomhole pressures is the key to

Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
three parts in the history match, these are: insuring a "good history match" by narrowing down the range
1. Material balance / matching average and local reservoir of uncertainty. In the interest of speed and cost, history match
pressure trends (pressure/depletion risk). target parameters are usually limited as shown in Table 1. We
2. Transport/saturation; history matching of watercut/GOR believe this is a big mistake because it allows a large
trends, more specifically match water breakthrough uncertainty in the model to be carried forward into forecasts. In
trends as well as watercuts and water rates (depletion other words, it is relatively easy to history match one parameter,
risk). such as watercut in which many solutions/models could
3. Inflow performance - matching average permeability / successfully match the trends. However, each of those models
(drawdown pressures (geological/petrophysical risk). may give very different production forecasts. Whereas when
multiple parameters, such as water cuts, average reservoir
pressure and bottom-hole flowing pressure, are all matched the
Increasing Confidence in the Model: The resulting models are more unique and the resulting forecasts
have a narrower range.
Importance of History Matching
Failure to Couple History Match Variables with
Importance of History Matching Exercise Physical Parameters that Control Forecasts

Building the "correct" reservoir model from static data alone Using the previous infill well example to adequately
is a very formidable task, because log and core data only sample forecast future performance, we need to understand:
approximately one ten billionth of the reservoir. Therefore, a Reservoir pressure vs. time/withdrawal profile
reservoir model built from geological (static) data alone will Average permeability within a drainage area
very often not match the observed field production and pressure How oil/gas saturations will change with
data. Due to a of lack of resolution of porosity/permeability time/depletion
fields, even with 3D seismic data, the static model requires Bottomhole flowing pressures vs.
history matching to tune the spatial area of porosity and time/withdrawal
permeability. Similarly, it has been demonstrated that
construction of the model using only dynamic data often results In multiphase oil and gas reservoirs, an understanding how
in a poor model. Good geological reservoir characterization saturations and pressure change with depletion, permeability
and geostatistical approaches definitely improve the speed of distribution and relative permeability is needed. Therefore,
the history match but they do not replace the history match history matching those variables is key to understanding future
process. It is critical that the reservoir model be history performance.
matched or calibrated to reproduce past history because during
the history match, more spatial heterogeneity is often introduced In our opinion the key task for flow simulation is to directly
to match production profiles. integrate the geological (static) data with the dynamic
(production, pressure) data.
Matching the dynamic data can probably yield more than one
model that can fit the history match trend. Therefore, although Importance of History Matching Spatial Trends
history matching the dynamic data can not guarantee that the A fair question to ask after the history matching is done, is
correct model is found, one thing that can be said is that lack of what confidence level are we at and where spatially are we most
history matching guarantees that the model is not correct. confident? Generally parameters "near the well" have the
However, realize that although history matching is sometimes a highest confidence level because these variables control well
slow and time consuming step it also is a necessary one. performance the most.

A key question to ask during the history match is whether However with time and depletion our confidence in far field
the modifications can be supported by physical phenomenon – parameters grows. For example in a tight gas field the initial
for instance, could the pore volume multiplier of 30% be production data (<1 year) is very sensitive to completion and
reasonable in a tight carbonate or is it applied in a homogeneous near well bore variables; with time and depletion the connected
high permeability sandstone with little areal variation to justify gas in place number and well interference then control the
the modification – and does the modification result in a net to production data. This is analogous to the radius of investigation
gross which exceeds one? formula for pressure transient tests, with time and depletion the
far field parameter becomes more and more defined.
Importance of History Matching Multiple Parameters
Two examples of where history matching is critical are the Unfortunately/fortunately, for fluid injection or gas
case of waterflood and infill drilling have already been cap/water drive cases, rather than radius of investigation the
illustrated. Another example of history match tuning is in gas region of investigation is dictated by streamlines. The shortest
wells in reservoirs with low permeability. Often in these travel time streamlines between injectors and producer will
reservoirs, core permeability data is much higher than pressure affect the early production profile whereas medium travel time
transient (buildup test) data. The differences in permeability are streamlines will affect late time response. Therefore in
due to internal (shale) baffles within the zones, which result in immature waterfloods, we usually understand what is between
tortuous flow paths. If the shales are small, very often they will

5
injectors and producers but insensitive to edge oil or even is probably due to both the complexity of the issue and because
OOIP, especially if voidage has been balanced. the requirements/standards of the study vary depending on the
objectives, desired accuracy and time/budget of the particular
It is critical that we check not only time trends but also reservoir.
spatial history match trends by looking at bubble maps of
differences between field data and model output as shown in Consider for example the case of an onshore field where an
figure 5.. It is hard to say that we have a good history match if economic well might require a recovery of 50,000 barrels to pay

Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
large regions are not matched. out (neglecting time effects). In the early stages when an
average well might recover 250,000 barrels, the error acceptable
Outer boundary effects in a forecast might be quite high – a recovery estimate of
250,000 barrels/well could have an error of 200,000 barrels/well
Various references cover how to handle pool or edge
without impacting the project’s economics. In later years when
boundaries8,9. Often simulation studies may have large edge
the average well recovers only 75,000 barrels, the acceptable
areas which have no well control in this area. Engineers would
margin of error is only 25,000 barrels. This illustrates that
do well to consult those references on how to handle edge areas.
standards may change over time. Often the risk tolerance
Having said that, in some cases where there is a high level of
tightens as the field matures. Yet, for offshore fields where the
pressure depletion or high sweep efficiency, the simulation can
capital expenditure is highest at the outset, the accuracy
considerably give insight to edge hydrocarbons. In other cases
required (and hence the standards) is the highest early in the life
the model, where there is small pressure depletion and low
of the field.
volumetric sweep, the model may be relatively insensitive to
hydrocarbon volumes
The above discussion also raises an important point – that
technical purity notwithstanding, any discussion of standards
History Matching Infill Production
should consider the economics of the particular problem.
There are three general components of a history match: Simply put, if the objective of the study is to generate a
Historical matching rates and cumulative volumes forecast, then a forecast accurate to within ±20% of the desired
History match recent data project economics may be required. This in turn requires one to
History matching infill well results consider what accuracy for pressure (which impacts
Often there may be tradeoffs between those components. hydrocarbon in place and rate), saturation or any other variable
Matching historical rates and cumulative oil, gas and water is is required to generate a forecast with the required precision.
critical to honoring volume and material balance.
However, none but the largest studies can justify the
Often the simulation engineer puts a lot of effort in the first additional expense of the above approach. In the interests of
component because one of the main functions of the model is stimulating discussion on standards and based on our practical
volume balance and material balance. If you ignore cumulative experience in many studies, we propose the following simple
trends, the volume and material balance calculations are measures as possible standards which may be used to determine
meaningless. On the other hand, the decline reserves method of if a suitable history matched model has been developed:
examining fields often puts a lot of effort on the latest history
match of oil and gas rates in the last few years of production in Material Balance Risk
oil and gas fields respectively. In our experience, often in
1. Field and pod pressures within ±10% to address the
mature field, the moveable oil and its permeability control late
pressure / material balance uncertainty
stage oil rates. So matching these latest trends are important as
2. Bottomhole flowing pressure within ± 20% - this is
well. However, matching your rates to a decline trend alone can
important for ultimate recovery projections but subject to
be a mistake if there are changing drive mechanisms or
greater uncertainty due to changing wellbore skin in the
changing conditions in the reservoir. So we either match these
field, lack of historical data. This criteria also requires
trends or explain why theses trends may change.
consideration of whether reservoir or wellbore
permeability is limiting factor.
Finally, as a check on history match quality, we recommend
that we check infill results. Often large producers are well
Saturation Risk
matched in most models but infill wells with low cumulative
production are often neglected. This is very problematic 3. Match field withdrawals for oil, water and gas rates and
because often the objective of the study is to test for infill well cumulative volumes within ±10%
feasibility. Yet matching the infill wells often weaves insight 4. 80% of the wells (when you have more than 20 wells), or
into “dead spot” areas and edges of the pool. Figure 6 shows a wells comprising of 80% of the production, be matched
case where we have history matched a large segment of a pool. for oil, water and gas rates and cumulative volumes
The corresponding check on how well the 15 infill wells within ±20% on a well level to address the saturation risk
matched showed that 80% of the infills were correctly predicted on a spatial plane.
and indicated uneconomic wells which could have been 5. Alternatively, match 80 % of the infills rates/cumulative
avoided, whereas actual infill drilling without the benefit of to within 10-20%
simulation resulted in only a 50% success rate. 6. Final rates of oil, water & gas match within 10%

The above criteria can be used provided the correct drive


Proposed Standards mechanisms have been identified and reasonable adjustments
were made during the history match,
Our experience in observing hundreds of simulations is that
there is a lack of standards as to what constitutes an acceptable It is very important to understand that when it is stated that
simulation model. There are no hard and fast rules which are an acceptable field match is within 10% of cumulative volume,
universally agreed upon and can be found in the literature. This it is assumed that the trend has been matched. In other words,

6
its not just the cumulative fluid volume produced that matter, Yes, the modeling will be only approximate and , yes,
but how you get there. A final cumulative volume can be there will always remain a significant empirical
obtained in a number of ways, so the simulation model must component to the process. But there has been a
replicate the proper physics of how the fluid was produced from demonstrably huge economic benefit from the
the reservoir. Figure 7 provides an example of an unacceptable enhanced understanding of fracture growth achieved in
history match of total field gas production that went unnoticed the last ten years11.
originally as the cumulative history match target was deemed

Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
acceptable. The model produces at too low of a GOR and then We would again add that flow simulation will never be
in late stage production, the pressure dropped enough for perfect but the exercise of tying physical parameters to
solution gas breakout and a huge amount of gas was produced, production profile has tremendous value.
resulting in a cumulative total reflective of actual historical
values. However, in actuality the GOR was consistently higher
and the late stage gas production was not nearly as significant Conclusions / Lessons Learned
resulting in a more gradual ‘path’ to the current cumulative gas
total. The gas production in the model was later matched A very real problem in the petroleum industry is an
through a change in solution gas ratio and gas relative unwillingness to set a target for history match quality. So often
permeability, as both were considered history match variable it is inexperienced engineers that define what is acceptable
due to lack of any lab measurements. history match quality. Therefore, not setting standards, in our
opinion, is a major mistake. Project objectives, data quality and
data quantity should dictate an acceptable history match,
Golden Rules for Simulation Engineers because these parameters vary from field to field and study to
10 study; some engineers are reluctant to set standards.
Aziz in 1989 presented ten golden rules. To repeat
those very valid rules; 1. Establish objectives and clearly demonstrate how history
match variables correspond to objectives.
1. Understand Your Problem and Define Your Objectives. 2. Well and reservoir diagnostics should be done before
2. Keep It Simple. model construction. Diagnostics should identify
3. Understand Interaction Between Different Parts. reservoir vs. operational effects on production signature.
4. Don’t Assume Bigger Is Always Better. Identification of volumes in place, hot streaks and poor
5. Know Your Limitations and Trust Your Judgment. reservoir continuity are also a key objectives.
6. Be Reasonable in Your Expectations. 3. History matching is a classical mathematical problem
7. Question Data Adjustments for History Matching. where the closer the initial guess is to the correct answer;
8. Don’t Smooth Extremes. the faster the convergence is to the correct answer.
9. Pay Attention to the Measurement and Use Scales. Therefore, a history match algorithm or simulation
10. Don’t Skimp on Necessary Laboratory Work. engineer’s skill will not compensate for a poor initial
reservoir description. Getting the large geological
We would include two additions. (1) to judge the quality of a features down first and preconditioning the geological
simulation you can not just examine the history match model with dynamic data is critical.
parameters, you must also examine the history match 4. Calibrate model average permeability by history-
adjustment parameters, are they “reasonable”. (2) We cannot matching drawdown.
merely look at a field wide plot of history match parameters 5. Realize that complexity doesn't always translate to
vs. time, we need to look at spatial trends and individual wells accuracy .
to judge quality. 6. History match individual well data to check for local
variation in reservoir permeability and porosity. We
often need additional data from analog fields, outcrops,
seismic data and/or geostatistics. This type of data is
Accuracy of Forecast critical to conditioning a complex geological model.
When there is no such data the model may appear to be
Analogous to flow simulation, hydraulic fracturing
more complex but we have introduced more uncertainty.
simulation has had similar difficulties as stated below:
7. History-match multiple variables to reduce the non-
uniqueness of the model.
Today’s fracture models are not reliable predictors of
8. There are three general components of a history match:
hydraulic fracture growth. They are not likely to ever
Historical matching rates and cumulative volumes
match the performance of today’s electronic circuit
History match recent data
simulators or even flight simulators. Fracture models
History matching infill well results
will always face the challenge of predicting (modeling)
Often there may be tradeoffs between those
a complex process that is taking place thousands of
components. Matching historical trends is critical to
feet below the earth’s surface with only poor
honoring volume and material balance.
characterization of the rock/reservoir/geology on the
relevant scale (10’s to 100’s of feet) for hydraulic
fracture growth. Relevant reservoir heterogeneity is
too large of a scale to detect with cores and logs but Acknowledgement
too small of a scale to be mapped as part of field The authors would like to thank all the people at Epic who
characterization. helped provide the key examples from which these lessons were
learned from.
Do these inherent difficulties mean that fracture
modeling is a futile endeavor? No……

7
NOMENCLATURE
Sorw = residual oil saturation to water displacement
(fraction)
AOF = absolute flow potential
GOR = gas to oil ratio (m³/m³)
Pwf = bottomhole flowing pressure

Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
REFERENCES

1. Saleri, N.G.: “Reservoir Performance Forecasting:


Limits, Field Examples and Acceleration by Parallel
Planning,” SPE May 1992, pp. 652-657.
2. Baker, Richard: “Reservoir Management for
Waterfloods – Part II,” JCPT 13-17, 1998.
3. Baker, Richard: “Reservoir Management for
Waterfloods – Part II,” JCPT 13-17, 1998.
4. Elsayed, S. A., Baker, Richard, Churcher, P. L., and
Edmunds, A. C.: “Multidisciplinary Reservoir
Characterization and Simulation Study of the Weyburn
Unit,” JPT October 1993.
5. Gringarten, Alain C.: “Evolution of Reservoir
Management Techniques: From Independent Methods to
an Integrated Methodology. Impart on Petroleum
Engineering Curriculum, Graduate Teaching and
Competitive Advantage of Oil Companies,” SPE39713,
1998.
6. Saleri, N.G., and Toronyi, R.M.: “Engineering Control
in Reservoir Simulation: Part 1,” SPE 18305, 1998.
7. Saleri, Nansen G.: “Reengineering Simulation: A
Bottom-Line Approach to Managing Complexity and
Complexification,” SPE 36696, October 1996.
8. “Determination of Oil and Gas Reserves, Petroleum
Society Monograph Number 1,” by Petroleum Society -
Canadian Institute of Mining, Metallurgy & Petroleum,
1994.
9. Harrell, D. Ronald, Hodgin, and John E., Wagenhofer,
Thomas: “Oil and Gas Reserves Estimates: Recurring
Mistakes and Errors,” SPE 91069, September 2004.
10. Aziz, Khalid, SPE Stanford U., JPT, Industry &
Profession (Dialog), pp. 1157, November 1989.
11. Wright, C. A., Weijers, L., Davis, E. J., and Mayerhofer,
M.: “Understanding Hydraulic Fracture Growth: Tricky
but Not Hopeless,” SPE 56724, October 1999.
12. “Guidelines for the Evaluation of Petroleum
Reserves and Resources,” Society of Petroleum
Engineers Inc., 2001, Ch. 5, pp. 44.

8
Table 1 – History Parameter

PARAMETER COMMENT METHOD

Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
Generally accept runs within 10%
Cumulative volumes of the final cumulative runs for oil Map view, bubble maps, cumulative vs.
Oil, gas, water (Np) and gas (Np) assuming gas date
was well measured

History match peak and decline


Oil, water, gas rates XY plots, bubble maps vs. date
period try to match trends

Final decline rate (last 2 years)


should be matched in simulator
(check using well’s pressure
Decline Rates Semi log, XY plots
control to see if they match decline
rate if wells are at low bottom hole
flowing pressure.

History match pressures with 20%


of measured pressure, but beware
Static Reservoir Map view, bubble plots, XY plots
of not fully built up pressure in
some wells.

9
Table 2 – History Match Adjustment Parameter12

HISTORY MATCH
ADJUSTMENT RANGE OF ACCEPTABLE VALUES QUALITY CONTROL
PARAMETER

Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
o For non fractured matrix rock pools, make sure
final history match values data are within log(k) vs.
porosity lot relationship envelope.
- Porosity vs. log
o If final values of adjustment parameters are less
permeability plot
Permeability than core values, see if baffles and tortuous flow
- Map of permeability
paths could be the reason.
original vs. final
o For naturally fractured reservoir, because of
fractures, use pressure transient data or
productivity index test as “ground truth” values.
Generally driven by:
o Variation of geological flow units (modeling small - Porosity +/- of the
Gridding geological features require more grid blocks). measured value
o Infill drilling objective. - Visual Inspection
o Numerical dispersion.
o For non thermal injection with Mobility ratio < 10
3-4 grid blocks between wells.
o For non thermal injection Mobility ratio >10 8-10
Grid Resolution grid blocks between wells. - Visual Inspection
o For dispersed drive mechanisms Solution Gas
Drive (SGD) and volumetric gas expansion 2 grid
block between wells.
Driven by:
o Ability and resolution contacts of GWC, GOC,
WOC and their movement.
Vertical Resolution o Generally 1-2m thickness preferred. - Visual Inspection
o Number of Vertical Grid >10 grid block per
geological layer for gravity over/under ride flow
geometry situations or coning problems.
Gross Rock Source: 3D Seismic, 2D
Range: +/- 30%
Volume(GRV) Seismic

Net-to-Gross Range: +/- 20% Source: Well Logs

Porosity Range: +/- 15% of the measurement, +/- 10% Source: Logs, Cores

Hydrocarbon Saturation Range: +/- 20% Source: Well Logs

Dip Range: +/- 10%, +/- 30% Source: Dipmeter, Seismic

Formation Volume
Range: +/- 5% Source: PVT Test
Factor (Bo)

12
“Guidelines for the Evaluation of Petroleum Reserves and Resources,” Society of Petroleum Engineers Inc., 2001, Ch. 5, pp. 44.

10
Table 3 - Major Drive Mechanisms / Parameters and Impact on Model

Fluid Type Drive Mechanisms Key Adjustment, Parameters History Match Parameters

Oil Compressible drive OOIP, rock, water and oil Pressure vs. (time withdrawals)
(reservoir pressure > compressibility, permeability Spatial pressure distribution

Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
bubble point) Oil rate decline characteristics
Flowing bottom hole pressure
Gas Volumetric gas OGIP, gas compressibility vs. Pressure vs. (time withdrawals)
expansion pressure, permeability Spatial pressure distribution
Oil rate decline characteristics
Flowing bottom hole pressure
Oil Solution gas drive OOIP, critical gas saturation (Sgc), Pressure vs. (time withdrawals)
vertical permeability and top of Spatial pressure distribution
structure, gas relative Oil rate decline characteristics
permeability near Sgc Gas-oil ratio vs. time, cumulative gas
vs. time
Flowing bottom hole pressure
Tight Gas Volumetric gas OGIP, pay cut offs, permeability, Spatial pressure distribution
expansion reservoir continuity, hydraulic Gas rate/tubinghead pressures
facture extents, hydraulic fracture Decline characteristics
permeability Flowing bottom hole pressure
Vertical Movable oil saturation, vertical Pressure vs. (time/withdrawals)
water drive flow barriers, permeability, eater- Spatial pressure distribution
oil density difference Oil rate decline characteristics
Watercut vs. (time /withdrawals) for
individual wells
Flowing bottom hole pressure
Coning Water oil or gas oil density Watercut/GOR vs. Time
difference, stand off distance, Pressure vs. time trends.
permeability, permeability Flowing bottom hole pressure
distribution Spatial pressure distribution
Waterflood/ Pressure Maintenance OOIP, Sorw, aquifer strength, flow Flowing bottom hole pressure
Miscible with displacement barriers, compartmentalization, Spatial pressure distribution
Flood /miscibility Sorm, miscibility, pressure Watercut/GOR vs. Time
permeability, permeability Pressure vs. time trends.
distribution

11
Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
Figure 1 - Comparison of Permeability Fields in Geological Model, Old Simulation Model and Current Refined Model; Less
vertically averaged 2006 model preserves the high permeability regions and thus history matches the data better

Figure 2 - Modifying Permeability to Match Peak Rates; In order to match historical peak oil rates permeability has to be
higher, but this leaves well with higher back pressure during late stage time

12
Permeability Derived From Core Analysis vs. Permeability
Derived From AOF Tests
Permeability Derived From Core 10000

Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
1000

100
(mD)

10

0.1
0.1 1 10 100 1000 10000
Permeability Derived From AOF
(mD)
Figure 3 - Comparison of (kh)core to (kh)AOF

1800

1600

1400

1200

1000
Actual Oil Rate
800 1996 Forecast
2000 Forecast
600 1996

400

200 2000

0
1980 1985 1990 1995 2000 2005 2010 2015

Figure 4 - Late Stage Decline Used to Tune Residual Oil Saturation

13
Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
Figure 5 - Saturation Risk Identified Via Bubble Plot of Error in Cumulative Volumes on Saturation

14
1200 6 .0 E + 0 6

1000 5 .0 E + 0 6

Cumulative Oil (sm³)


800 4 .0 E + 0 6
Oil Rate (sm³/d)

Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
600 3 .0 E + 0 6

400 2 .0 E + 0 6

200 1 .0 E + 0 6

0 0 .0 E + 0 0
1954 1958 1962 1966 1970 1974 1978 1982 1986 1990 1994 1998 2002 2006

H is t O il R a t e S im O il R a te H is t C u m O il S im C um O il

E n e r p lu s W e lls W a t e r M a t c h

100 8 .0 E + 0 6

90 7 .2 E + 0 6

80 6 .4 E + 0 6

Cumulative Water (sm³)


70 5 .6 E + 0 6
Watercut (%)

60 4 .8 E + 0 6

50 4 .0 E + 0 6

40 3 .2 E + 0 6

30 2 .4 E + 0 6

20 1 .6 E + 0 6

10 8 .0 E + 0 5

0 0 .0 E + 0 0
1954 1958 1962 1966 1970 1974 1978 1982 1986 1990 1994 1998 2002 2006

H is t O il R a te S im O il R a te H is t C u m O il S im C um O il

Figure 6A - Overall Field Match

120 1 .8 E + 0 5

100 1 .5 E + 0 5

Cumulative Oil (sm³)


80 1 .2 E + 0 5
Oil Rate (sm³/d)

60 9 .0 E + 0 4

40 6 .0 E + 0 4

20 3 .0 E + 0 4

0 0 .0 E + 0 0
1954 1958 1962 1966 1970 1974 1978 1982 1986 1990 1994 1998 2002 2006

H is t O il R a te S im O il R a t e H is t C u m O il S im C u m O i l

E n e r p lu s In f ill W e lls W a te r M a t c h

100 1 .0 E + 0 6

90 9 .0 E + 0 5

80 8 .0 E + 0 5
Cumulative Water (sm³)

70 7 .0 E + 0 5
Watercut (%)

60 6 .0 E + 0 5

50 5 .0 E + 0 5

40 4 .0 E + 0 5

30 3 .0 E + 0 5

20 2 .0 E + 0 5

10 1 .0 E + 0 5

0 0 .0 E + 0 0
1954 1958 1962 1966 1970 1974 1978 1982 1986 1990 1994 1998 2002 2006

H is t O il R a te S im O il R a t e H is t C u m O il S im C u m O i l

Figure 6B - Performance of 15 infills 1


15
Downloaded from http://onepetro.org/PETSOCCIPC/proceedings-pdf/06CIPC/All-06CIPC/PETSOC-2006-129/1829166/petsoc-2006-129.pdf by Middle East Technical University Ankara user on 15 March 2021
Figure 7 - Rate Trends to Match (As Well as Cumulative) In History Match

16

You might also like