Vous êtes sur la page 1sur 7

Uncertainties in Reservoir Simulation

Published on April 4, 2016

1. Introduction
Reservoir simulation models evolve from static models; which themselves are the
results of the assessments of discipline models[1]. The discipline models are studies
including geophysical, geological, petrophysical, and different data evaluations
and studies in the reservoir engineering. The combination of the results of such sets
of studies and assessments are organized in the context of the reservoir Static
Models. The recurrent and well data are added to the static model, to form
Reservoir Simulation Models (RSM). The RSM may be run in a dynamic state. The
complexity of the resultant model grows during its evolution from initial
geophysical model to a 3D geological static model and make another leap ahead
when it is evolved into RSM. During the preparation of such data set, the objective
is to have each part of the data-set to be the Representative of the prevailing
phenomena. Each piece of such complicated puzzle bears uncertainties in various
aspects.
Considering a model as a discrete data set, represented by grid cell properties, one
must take into account two additional sets of approximations and limitations

affecting uncertainties of the model. The first set refers to the way such
representative values are calculated or evaluated, and the other issue is related to
the limitations of software, hardware, time, and budget; actuating new constraints
on the model.
Considering the reservoir simulation as inverse problems; theoretically, reservoir
models may be matched by different combinations of parameters to the same
observed data such as well BHP, water cut and the like. On the other hand, there is
only one combination of the data that has initiated the observed data. To getting
closer to the actual model, the main challenge is to understand the nature and
sources of uncertainties and to have an appropriate methodology for handling such
uncertainties.
Oversimplifying such complicated puzzle to finding a mathematical combination of some
History-Matching parameters to match the observed data could guarantee that the
predictions may not come true.

2. Elements of Reservoir Simulation


Uncertainties
Reservoir Simulation Model is the only tool for predicting reservoirs future
performance. Irrespective of the uncertainties of reservoir modeling, there is no
better way than using such models for predicting or estimating the reservoir
production potentials and future performance. To understand the elements of
uncertainties in reservoir simulation, one may take a tour walking through the
preparation of such models.
Reservoir models are the result of multidisciplinary studies, all associated with
different degrees of uncertainty. To conclude an RSM, five independent upstream
disciplines: Geophysics, Geology, Petrophysics, Reservoir and Production
Engineering, are incorporated; while each discipline model consists of several
independent and uncertain data sources[2].
Discipline models when built, each needs to undergo quality control and
consistency checks to make sure they are sound and consistent before joining in the
geological model. When such data joined in the geological model, additional
consistency check and quality control are necessary to make sure that the discipline
data is still consistent and in agreement with the other discipline information in the
model. In this context, appropriate geological modeling workflows need to be
established and utilized to generate a 3D static geological model by combining
structure, property, saturation and fault data. While putting discipline models
together one by one, the consistency and coherency issues out surfaces; showing the
necessity for quality control of the generated compound model at each stage of
development.
By running geological modeling workflows, the completed geological model is
again needs undergoing quality monitoring and validation to make sure that the
combination of different discipline data is well-organized in the model, and the

model is sound, consistent and meaningful. Only the confirmed static geological
model may be utilized as the base for building dynamic reservoir simulation model.
The dynamic reservoir simulation model is established by utilization of static 3D
geological model and incorporating well and time-dependent data from reservoir
and production engineering disciplines. The dynamic model must undergo quality
control and validation; which, is performed in two distinct stages. At the first stage
the model is quality controlled and validated in the static mode, and at the second
stage, it is validated in dynamic mode. The first stage is called Model
Initialization and the second is called Model Validation or so-called history
matching process.
The major quality control and validation of the reservoir simulation model are
accomplished through History Matching Process. History matching is an
important part of the reservoir study process taking at least 30% of the whole
project time in most cases.
Unlike quality controls and validation processes in the static mode, the model in
history matching mode is interactive. The dynamic model is capable of calculating
some dependent data arrays provided that, the pertinent information is provided.
If production data is given, the pressure data, water cut and gas oil ratio and some
other data may be calculated. By utilizing this characteristic of the dynamic model,
the model is validated. Problems arise when a various combination of history
matching parameters may confirm the same set of observation data or the model is
insensitive to a reservoir parameter that is supposed to be influential.
As a conclusion to this end, reservoir simulation model inherited its uncertainty
from various parts of geosciences and reservoir engineering. Strictly speaking;
every data source in the simulation model is associated with inherent uncertainty.

3. The Roots of Reservoir Simulation


Uncertainties
Non-uniqueness is inherent to reservoir engineering and other upstream data
sources, which, is the root of uncertainty in upstream disciplines and simulation
models. In fact, each piece of information owns its uncertainty rendering the whole
model uncertain. Therefore understanding the model uncertainty requires
understanding the nature of each data group. The uncertainty of reservoir data
goes back to the fact that each piece of data in reservoir simulation is integrated
with other data. For example, consider, rock permeability; it is uncertain because;
it is controlled by various reservoir properties such as rock composition, rock types,
saturations, rock compaction, layering, sedimentation, fracturing, and more,
which all of them are associated with their inherent uncertainty.
The result of reservoir measurements leads one to discover that the outcomes of
measurement for each data type may share one or more of the following features
and characteristics:

1. Inexact

2. Indirect
3. Uncertain
4. Superimposed
5. Partially cover the phenomenon
6. Single Source
7. Upscaled
8. History Dependent
9. Correlated and Propagated
10.

Prevailing phenomena observed

11.

Practical Limitations

Presenting a brief explanation for each of the above features or characteristics is helpful to
understand better why no data in reservoir engineering is definite. However, among reservoir
data, the pressure, and temperature, and rate measurements could be assumed as the most
accurate.

Inexact
The reservoir measurements are inexact because the measured results may not be reproduced
if the measurement is repeated. For example, if the same set of logging tool is sent down for
the second time to the same well even in no-time after the first run, the same result cannot be
recorded. However, the two results are probably close to each other, but not the same.

Indirect
The measured object is not seen, and therefore, the measurement is indirect. For example to
measure water saturation, electrical resistivity, and other log measurements are performed.
Then, water saturation is concluded based on such indirect measurements.

Uncertain
The result of interpretation of the measured values to estimate reservoir properties renders the
calculated result uncertain. Because firstly, the measured value is inexact, and secondly
calculated result is approximate. For example, when log data is used to derive porosity, the
measured log data is inexact and indirect and also the formulae used to estimate porosity is
approximate equivalent for the actual rock in the reservoir pay.

Superimposed
Most, if not all, of the measured values, is the product of many factors affecting the value of
property intended to be measured. For example, when measuring resistivity for estimation of
water saturation, the value of the measured resistivity is controlled by many uncertain factors
in addition to the amount of water in the pore space. Alternatively, in well testing, a specific
feature is the result of many reservoir parameters superimposed creating the diagnosed
feature.

Partial coverage
The acquired data does not cover the whole reservoir. If the property is measured in wells, it
normally covers a small part of the reservoir. If measured by waves such as seismic, it does
not cover the whole reservoir with the same accuracy. Even in the same well, the
measurement accuracy is not the same, for the entire interval under investigation. Core data
may be acquired from some wells and in some intervals.

Single Source
There is one source of measurement for each data to measure. Therefore, there is almost no
chance to measure one property by different sources for verification of the result.

Upscaled
The measured data is upscaled to a certain packet in nearly any reservoir measurement. For
example in log measurements, the minimum resolution is 0.5 ft. In core analysis, the
minimum is the plug size. When data applies to the model, it must be scaled-up and averaged.
Even the measured values cannot be used as they measured; because the system is discrete
and can only be defined in the context of the grid cells.

History Dependent
The majority of data such as saturations, pressure, transmissibility, productivity indices, rates
are history dependent. When once measured, shortly after it changes.

Correlated and Propagated


Almost all reservoir data are propagated, correlated, generalized, and estimated. The reason is
there is not measured data available at every arbitrary point in the reservoir, and therefore at
any point, the data from the measured points must be propagated, generalized and estimated.

Prevailing Phenomena Observed


In reservoir measurements and data acquisition, the dominant phenomenon or characteristic is
detectable and is measured. For example, when the resistivity of an interval is measured, only
the easiest path or route is pertinent to the electrical current and, therefore, is detected by the

measuring tool. Alternatively, in well measurements only contributing parts of the well at the
time of measurement could be measured.

Practical Limitations
There is limitation all around, in time, cost, accuracy, and depth of investigation,
mathematical formulae to define natural processes, computation facilities, and more. Such
limitations apply to any data acquisition and interpretation process.
The reservoir data with the above characteristics, reveals that no data can precisely be
determined. The only thing can be done, is to try to get closer to the dominant feature of the
phenomena at a given time.

4. Reservoir Engineering as an Inverse


Problem
Reservoir engineering is an inverse problem. On the other hand, the reservoir data, of any
kind, is uncertain. In such problem; the simulation model is used to adjust reservoir data
during, quality control, consistency checks and history matching processes. In fact, the
reservoir response is used to estimate the reservoir data. Handling such a subtle issue requires
professional experience and understanding the dominant reservoir characteristics.

5. Handling Reservoir Simulation


Uncertainties
With the uncertainty in all data to conduct a reservoir simulation, the salient question is: how
to handle reservoir simulation uncertainties to do a reliable study? This issue is a fundamental
and complicated question in reservoir simulation. However in this short note, a concise and
strategic explanation is provided.
To have a reliable reservoir simulation model, one needs to walk with the evolution of the
reservoir modeling process; this means, the first, should come first. The static reservoir
simulation and modeling is started by geophysical interpretations, to come up with
preliminary reservoir structure model. The geophysical model is associated with relatively
broad uncertainty avenue. A significant part of this uncertainty may be resolved by
application of geological definitions, techniques, and information. The extent of revisions in
the model by geological information should not cross the lines of consistency in geophysics.
Utilizing petrophysical interpretation results in the model provides the next opportunity to
make the model more precise and accurate, but, again, this may not cost the coherency of the
geological and geophysical interpretations. In the same token, the results of the reservoir
engineering studies should be utilized in the model, to make the model more precise by
working inside the uncertainty avenues, without crossing extreme lines of consistencies
established earlier.
As such; the main part of static reservoir model is received from geoscience, and some added
from reservoir engineering studies. In fact, geophysics initiates the geoscience studies,
completed by geological studies, tuned by petrophysical evaluations, and get live by

incorporation of reservoir and production engineering data. At each of these steps, complete
consistency in the reservoir model is focused. The reservoir model, to be nearer to the core
disciplines, Geophysics Geology Petrophysics Reservoir Engineering Production
Engineering, is relatively more diverse with a wider avenue to be uncertain. Therefore, at any
developed stage, one must work in the uncertainty avenue of the previous stages to make the
reservoir model more accurate and targeted. The accuracy, however, is a relative term
depending on the extents of the available data. More precise and abundant data let having a
more accurate model.
During history matching process, the well performance cannot be initiated by adjusting the
characteristics of the well, or limited region around the well. In fact, the well performance is
the resultant of the reservoir properties, feeding and supporting the well to perform as it
performs. Controlling the well may affect the well performance, but only within a short time.
Trying to match the well performance by controlling the properties of the well or limited
region around the well is the biggest mistake committed by amateurs in reservoir simulation.
If matched in the short time; it deviates sooner or later in a longer period.
Reservoir rock and fluid properties which are supporting the well performance, appear in
combinations in reservoir simulation governing equations; however, changing the parameters
to match the performance is not arbitrary. One should not alter a matching parameter unless
could be justified in the context of the discipline in which it is determined.
Being excited about packages making the, so called, Automatic History-Matching is
meaningless; unless it is used in the context of a professional framework, only to speed up the
task.
Field History matching can be demonstrated when wells are matched to their performance
under certain matching criteria. However, for the history matching process one must work on
a global scale ( field or region scale) to reach local (well scale) objectives. Nevertheless, one
may use well-related, or near well region parameters, as a fine-tuning tool only. However, this
should be the last stage of the matching the history of the wells.
The last but not the least, the recorded well performance history, well rates, observation data,
well pressures, productivity indices, and the like, should not be assumed as Bibles; such data
must be critically quality controlled, verified and thoroughly checked for consistency before
embarking on history matching process.

Footnotes
[1] Discipline models are the result of the studies conducted in each main geoscience and
reservoir engineering disciplines; geophysics, geology, petrophysics, and reservoir
engineering.
[2] Data Sources are the collection of data analyzed in the context of a subtask as part of a
discipline model; e.g. CCAL, SCAL, PVT, etc. are data sources which are part of a reservoir
engineering discipline.

Vous aimerez peut-être aussi