SPE 101808

Enhancing Field Management in Siberia by Quantifying Production Uncertainties

D.E. Tipping, SPE, and M.N. Deschenya, TNK-BP, and F. Deimbacher, SPE, and D. Kovyazin, Schlumberger

Copyright 2006, Society of Petroleum Engineers

methodology gives us confidence in identifying areas that
This paper was prepared for presentation at the 2006 SPE Russian Oil and Gas Technical need additional injection support.
Conference and Exhibition held in Moscow, Russia, 3–6 October 2006.

This paper was selected for presentation by an SPE Program Committee following review of
information contained in an abstract submitted by the author(s). Contents of the paper, as Field Characteristics
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
correction by the author(s). The material, as presented, does not necessarily reflect any Geology
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at The field produces from a Lower Cretaceous deepwater
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper marine environment where deposition of sandstone and
for commercial purposes without the written consent of the Society of Petroleum Engineers is
prohibited. Permission to reproduce in print is restricted to an abstract of not more than claystones from the northwest to the southeast has created
300 words; illustrations may not be copied. The abstract must contain conspicuous sand lenses forming the productive Achimov formations.
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836 U.S.A., fax 01-972-952-9435. (Figure 1) There are 8 oil bearing beds within the Achimov
formation with 13 different oil-water-contacts (OWC’s).
Abstract Formation quality is fair with a porosity of about 16 to 20%
An analysis of production data from a Lower Cretaceous and horizontal permeabilities of about 2 to 40 m D randomly
western Siberia oilfield operated by TNK-BP suggested that distributed across the stratigraphic section. As indicated by the
the final recovery factor would not exceed 15% with one of porosity- permeability trend from core measurements one
the waterflood development plans under consideration. Such a lithofacies dominates. (Figure 2). Correlation of the wireline
plan clearly left scope for improvement. However, since there logs and results from exploration well tests suggest that
were limited surveillance data to infer current depletion from communication from one bed to another is limited except for
the 8 oil bearing beds, considerable uncertainty existed in the Achimov 1-1 though Achimov 1-4 formations. However,
locating the remaining reserves. This prompted an since dynamic data are limited this is subject to interpretation.
examination of the reservoir and well characteristics to
quantity these uncertainties in the reservoir potential and, Current field development plan
hence, ensure the future strategies to increase recovery are Material balance suggests that the final recovery under the
viable. To determine the uncertainty in predictions, it was current operating conditions may not exceed 15 % of initial-
necessary to create multiple predictive models all matched to oil-in-place (IOIP). One of the major constraints to the final
observed field data. This matching process is extremely time recovery is the water cut. It has hit 60% and its increase is
consuming when undertaken manually. Our experience shows accelerating.
that one model requires about 9 man-months. As a result, we
tested a procedure to accelerate the matching using a program Water injection is a key element of the development plan.
with an automated history matching algorithm linked to the Production is commingled: a typical production well is
dynamic simulator. Although implementation of this method is perforated in 3 or 4 of the Achimov beds. From an injector,
still ongoing, it has demonstrated the following: water is also injected into several beds. Injection rates have
been maintained to assure a voidage replacement ratio (VRR)
• Uncertainties are an inherent part of reservoir of about 1. Hence, average reservoir pressure has been
predictions. For field management, it is important to sustained above bubble point pressure (116 bar).
quantify them so they can be managed.
• Major increases in work productivity are achievable. Most of the wells have been hydraulically fractured.
Within a period of a week, after determining the Fracturing increases oil rates initially by about 2 to 3 times
uncertainties to be addressed, we obtained 7 different with a comparable jump in water cuts.
history-matched models each yielding a unique
production profile prediction. Most of the wells contain electrical submersible pumps. The
current well strategy consists of operating the pumps until they
Based on the quantification of the uncertainties, we are fail at which time they are replaced.
changing our field management practices. For example,
additional surveillance is planned to monitor the depletion in Organisational Context
each bed. A further benefit is that this uncertainty evaluation Three entities within TNK-BP are involved in developing
exploitation strategies for the field: the operational unit (BU),
2 SPE 101808

which is responsible for management of the asset, the production rates of 2 to 3 times with corresponding increases
Technical Research Centre (TNNC), which provides reservoir in water cut.
characterisation and modelling expertise, and the Reservoir
Management Department (RMD) based in Moscow, which Organisational
offers advice on best-practice technologies for evaluating The BU recognising the signs of impending risks to sustained
reservoir potential and field management. With the field water production, mobilised resources to improve field management.
cut rising rapidly plus signs of an irreversible downward trend The principal challenge was effectively deploying technical
in production, The BU sought advice from RMD on measures expertise in a concerted manner so that the right field
for sustaining production and increasing reserves. management decisions could be made. The primary difficulty
was identifying when specific technical expertise was needed
In 2005 the BU embarked on an infill drilling programme and then subsequently securing those skills from TNNC.
designed to provide additional injection support to the central
area of the field. A three-dimensional simulation model helped Within this context, RMD initiated a project designed to
justify these additional injectors. However, actual oil rates did maximise the potential of the field with the following aims:
not meet expectations. These results prompted the RMD to
undertake a review of the validity of the model as a predictive • Introduce evaluation techniques appropriate to the
tool. Discussions with subsurface technical specialists at reservoir conditions that could yield results in a
TNNC who had participated in building revealed: timely manner.
• Effectively engage technical expertise from the three
• TNNC had directed considerable effort to integrating entities for optimising field management and
petrophysical, geological and reservoir engineering maximising the value of the asset.
data to building a predictive simulation model that
gave an excellent match available data, primarily, Evaluation of Remaining Potential
field production. However, TNNC recognised that, Build geological static model
alternative matches were possible by changing The productive beds are thin ranging in thickness from 5 to 15
alternative parameters. Given the time-intensive m. Since oil has not displaced all the movable water, the
nature of history matching, the implications could not characteristics of the transition zone have a major influence on
be assessed. the initial-oil-in-place (IOIP) and well flow rate behaviour.
• Although the BU had responsibility for field Characterisation of the transition zone has involved
management, quality assurance of contractor activity, identification of lithofacies types and then obtaining capillary
such as casing cementation or wireline log pressure measurements on core samples so that these facies
acquisition, is always a challenge. For example, poor types are adequately described. We prefer measurements on
cementation in a number of wells has led to water core since the porous units in the productive beds are often 10
flow in the casing-wellbore annulus. Additionally, to 20 cm thick – substantially less than the bed definition
certain logs, such as the pulsed neutron log, were not resolution of wireline logs.
conclusive, and hence, it was not possible to diagnose
fluid saturations in selected flooded zones. Conventional approach for predicting production profiles
• Discussions revealed that, because TNNC staff was The conventional approach within TNK-BP consists of
engaged on multiple projects, they could not provide building multi-million cell numerical simulation models
ongoing operational advice to the BU. This meant designed to reproduce reservoir behaviour. This involves
quality assurance of field management practices manually adjusting reservoir parameters until the simulated
could be reactive. response matches the observed data. The method has the
Technical following major disadvantages:
Since neither bed pressures nor individual flow rates from
each bed have been monitored, subsurface specialists do not • It is exceedingly time-consuming. Our experience
have direct measurements to determine the remaining reserves indicates that an acceptable match takes about takes
from each bed. This has major significance in determining the about 9 man-months for a model of this size.
optimal reservoir management plan. Additional challenges to • It is only one representation of the reservoir. Other
quantifying reservoir depletion are: equally likely models are possible with different
• Identifying the source of water from the producer values for the reservoir description parameters.
wells. Significant amounts of water arrive at many
producer wells through the casing annulus.
• Quantifying the water arriving at the producers from
adjoining layers via hydraulic fractures. Uncertainty methodology
• Characterising the transition zone given that wireline
Upon embarking on the model building, we recognised that
logs do not have sufficient resolution.
the reservoir parameters in the model can be adjusted in a
multitude of ways so that the simulated reservoir performance
Since the Year 2000, most of the wells have been
matches field data. Every model gives a different prediction.
hydraulically fractured. This has yielded initial increases in oil
Hence, our efforts were directed towards identifying the
SPE 101808 3

distribution of possible outcomes and not on obtaining one field operations staff to identify those phenomena expected to
history matched solution. This requires creating a lead to uncertainties in predictions. These comprise:
representative selection of history matched models and then
comparing the production profiles from each of these models. • The fluxes of fluid – outflows (production) and
Evidently, it is not practical to undertake such an exercise inflows (injection) at each bed. The principal factor
manually. It would take years. As a result, we implemented an influencing these fluxes is the horizontal permeability
alternative approach using a program with an automated of each bed since flow rates are proportional to
history match routine. permeability. Hence, the uncertainties in bed
permeability have a major impact in predicting the
The procedure consists of determining the parameters in the remaining reserves in each bed. We dealt with this
model that require adjusting and then instructing the program uncertainty by allowing the permeability to vary by
to select values of these parameters and then compare the 0.2 to 5 times the base case geological model. This
results with the actual data. reflects the variation observed from core data.
• Given the difficulty in mapping shale discontinuities,
Automatic History-Matching Technology uncertainties are expected in vertical permeability. To
A global optimisation algorithm, which searches the whole reflect this, we introduced a vertical permeability
parameter space (refs), is used for history-matching. The multiplier varying from 0.02 to 0.25 times the
engineer defines the type and regional extent of uncertain horizontal permeability.
parameters as well as the parameter ranges. Different • Since the oil columns are thin, the depths of the oil
realisations of a geological model can also be used as input. water contacts have a major influence on the initial
The algorithm searches for solutions within the defined oil-in-place (IOIP) and water cut. OWCs have been
parameter space. During the optimisation a series of identified from wireline log responses and well test
simulation models with different input values is generated and results. The maximum error in these contacts is
sent to the queuing system of a Linux cluster. The queuing expected to be 7m which was input as the allowable
system distributes the serial or parallel simulation runs to variation.
available processors on the cluster. Evolutionary algorithms • Porosity and permeability measurements from core
are employed to create new populations of simulation models suggest that one facies type dominates in the
which should provide a better match to historic data than reservoir. (Figure 2). Hence, we believe that one set
previous generations (“survival of the fittest”). Fifty to one of relative permeability tables characterises flow
hundred uncertain parameters are typically used during behaviour throughout the beds. Since, there were
automatic history-matching resulting in between a few insufficient relative permeability measurements on
hundred and several thousand simulation runs. A modern core, we allowed the values to vary with likely limits
simulation cluster can finish the simulations within a few expected for the Lower Cretaceous in western
days! Siberia, namely initial water saturations (Soi) and
residual oil saturations (Sor) of 0.2 to 0.45 with
The optimiser minimises an objective function. While it is Corey exponents of 2.4 for oil and 1.5 for water.
possible to minimise field and well/group errors at the same • Produced water originating from overlying water
time, it is more convenient to reduce global field errors first, bearing beds or an underlying aquifer caused by
and then concentrate on well or regional matches. Field ineffective cementing of the casing. This was a
history-matching will usually give very good match for most problem in a number of wells.
wells; only wells with inconsistent production data or other • Water originating from water bearing beds adjoining
modelling problems will remain unmatched. Those can be the perforated intervals arriving at the producer wells
investigated in a second round of iterations (potentially with a via hydraulic fractures. Since virtually all wells have
different set of uncertain parameters keeping constant some of been hydraulically fractured, this volume of water is
the parameters that were used in the global iteration). significant. We modelled this by introducing
It must be emphasised that this optimisation procedure is fully additional perforations 10 m on either side of the
automatic and, after the initial model setup, does not require actual perforated intervals.
any user interaction! Also, the optimiser usually finds multiple
parameter combinations (i.e. history matches) that give the The following features were not considered in the uncertainty
same quality match! evaluation since their influence is expected to be minor:

The latter property was very important in the underlying • The structural features of the reservoir are readily
project so we could understand the various possible depletion defined from the 3D seismic and high well density.
scenarios in the beds. Hence, we have confidence in the bed architecture.
• The fluid properties are known with sufficient
accuracy given that the oil compressibility is
Identifying phenomena influencing uncertainties relatively low and that the reservoir pressure stays
We initiated a number of meetings with reservoir above bubble point.
characterisation specialists, well interventions personnel and
4 SPE 101808

It is essential, however, that the adjustments to the parameters • Modern, cluster-based optimisation technologies
are realistic. This requires an understanding of the geology and offer massive reductions in turn-around times for
the dynamics of in-situ flow. In the case of field under review history-matching and field development planning.
based on the reservoir and well conditions we believe that the Pre-requisites for the use of those technologies are
following parameters will have the greatest influence on consistent and quality-controlled input data, as well
initial-oil-in-place (IOIP), and final recovery: as “best practices” that guide engineers through the
optimal application of these new technologies. If the
• The oil-water-contacts (OWCs) input data are not good, the available automation
• The transition zone thickness cannot be used to its best effect.
• The initial oil saturation (Soi) and the residual oil • Uncertainty analysis demonstrates the value of data.
saturation (Sor) after waterflood. For example, bed pressures can have a major impact
in reducing the uncertainties in reservoir predictions.
Immediate implications for field management • Surveillance can help reduce uncertainties
Field surveillance significantly in fields with multiple productive beds.
We selected 7 matched cases each intended to represent a The modelling to date indicates that when bed
distinct solution. The high quality of the matches is evident in pressures are not known, large uncertainties can be
Figures 3 and 4. expected. These results have helped justify obtaining
RFTs in key wells within the field.
Figure 5 illustrates the recovery from the Achimov 2-1-1 bed
for the 7 history-matched cases. As can be seen the recovery Ongoing work
factor on the 1 July 2024 varies from 5.6 to 13.6 % of IOIP. Using seismic attribute analysis a series of geological
This is a significant variation. How can this be cut? realizations was created which addresses the uncertainty in
Surveillance can help narrow this uncertainty considerably. net-pay (i.e. baffle size and distribution) in areas of the model
Figure 6 reveals the average pressure of the bed for these where there is little well control. Approximately ten of those
cases. As can be seen, the differences are large. If bed realisations will be, after a screening process, input into the
pressures are recorded with an RFT device, then the spread of automatic history-matching algorithm, and used to further
matched solutions can be reduced. investigate uncertainties.
Optimisation of Injection Another, extremely promising technology has been applied to
The results form the representative set of history matched are the problem of waterflood optimisation. Using streamline
used to evaluate the injection support in each bed. To simulation (ref) and a heuristic optimisation algorithm, the
illustrate, Figures 7 and 8 compare the predicted recovery total field oil production, if certain reservoir conditions are
factors for the Achimov 2-1-1 and Achimov 2-2 beds, met, could potentially be increased by 20% after 10 years.
respectively. In both cases the green curve gives the base case, Water production would also be reduced. The optimisation
determined from a manual history match. The base case algorithm is based on the calculation of the efficiency of
suggests a final recovery factor of 14% for the Achimov 2-1-1 injection water flowing along streamline bundles with
after injection of 0.8 pore volumes of water while the incremental oil production. Inputs to the streamline simulator
Achimov 2-2 has a final recovery factor of 12% after the are the history-matched models plus well locations of (new)
injection of 1 pore volume. Apparently, the recovery in the producers and injectors. The preliminary results are very
Achimov 2-1-1 is better. However, the base case of the encouraging; work is ongoing.
Achimov 2-1-1 is at the more optimistic end of the predicted
recoveries. Most of the cases compute a recovery less than Acknowledgments
10%. This contrasts with the Achimov 2-2 where most of the The authors thank TNK-BP management for permission to
cases support a recovery exceeding 15%. This distinction publish this work. The authors are also grateful to Scandpower
suggests that priority for optimising injection should be Technologies for providing the automatic-history matching
directed to the Achimov 2-1-1. software, and Hewlett-Packard Moscow for giving access to
the Linux clusters in the Hi-Tech Centre.
Key Lessons
• One history-matched simulation model cannot be References
expected to give accurate predictions of future Williams, G.J.J., Mansfield, M., MacDonald, D.G., and Bush,
production. Values for reservoir parameters are never M.D.: “Top-Down Reservoir Modelling”, paper SPE 89974
known with certainty, hence multiple solutions are presented at the 2004 Annual Technical Conference &
possible. Field management decisions should account Exhibition, Houston, TX, Sept. 26-29.
for uncertainties.
• Before embarking on building a predictive reservoir Schulze-Riegert, R.W., Axmann, J.K., Haase, O., Rian, D.T.,
model, the engineer must identify the mechanisms and You, Y.-L.: “Evolutionary Algorithms applied to History-
that have a significant influence on field behaviour so Matching of Complex Reservoirs”, SPE Reservoir Evaluation
that the model represents actual performance. & Engineering (Apr. 2002) 5 (2).
• Automated-history matching techniques can give
major boosts to work productivity.
SPE 101808 5

MEPO Technical Manual, Scandpower Petroleum Figure 3: Best matched cases of field oil production
Technology, www.scandpowerpt.com
Field oil production - Simulated cases
R. Baker: Reservoir Management for Waterfloods, JCPT,
April 1997.
2500

FRONTSIM User Manual, Schlumberger Information 2000

Oil rate (sm /day)

Observed data
Solutions, 2006. Case 1

3
1500 Case 2
Case 3

1000 Case 4
Case 5
Case 7
500

Figure 1: Schematic of depositional environment – Field 0

01-Jan-92

01-Jan-94

02-Jan-96

02-Jan-98

03-Jan-00

03-Jan-02

04-Jan-04

04-Jan-06
described is deep marine (Region 12)

Figure 4: Best matched cases of field water cut

Field water cut - simulated cases

1.00

0.90
water cut (sm /sm )

0.80 Observed data

3

0.70 Case 1
3

Case 2
0.60
Case 3
0.50
Case 4
0.40
Case 5
0.30 Case 6
0.20 Case 7

0.10

0.00
01-Jan-92

01-Jan-94

02-Jan-96

02-Jan-98

03-Jan-00

03-Jan-02

04-Jan-04

04-Jan-06
Figure 2: Core porosity and permeability data suggest one
lithofacies in productive beds

Achim ov core m easurem ents

1000
Figure 5: Achimov 2-1-1 Recovery factor predictions
permeability, mD

100

Wel l 1
Achimov 2-1-1 Recovery factor predictions
10 Wel l 2
Wel l 3
30
Wel l 4

1 Wel l 5
Base case
% initial-oil-in-place

Case 1
20
Case 2
0.1
0 5 10 15 20 25 Case 3
Case 4
porosity, % 10
Case 5
Case 6

0
Jan-92

Jan-96

Jan-00

Jan-04

Jan-08

Jan-12

Jan-16

Jan-20
6 SPE 101808

Figure 6: Achimov 2-1-1 Simulation bed pressures

Achimov 2-1-1 Simulation bed pressure

350
Bed pressure (bar)

Base case
300
Case 1
250
Case 2
200
Case 3
150
Case 4
100 Case 5
50 Case 6

0 Case 7
Jan-92

Jan-96

Jan-00

Jan-04

Jan-08

Jan-12

Jan-16

Jan-20

Figure 7: Achimov 2-1-1 Recovery predictions with base

case

Achimov 2-1-1

30
Recovery factor, %

Base case
20 Case 1
Case 2
10 Case 3
Case 4
Case 5
0
0 0.2 0.4 0.6 0.8 1 Case 6

Injected water, pore volumes

Figure 8: Achimov 2-2 Recovery predictions with base case

Achimov 2-2 Recovery predictions

30 Base case
Recovery factor, %

Case 1
20 Case 2
Case 3
10
Case 4
Case 5
0
0 0.2 0.4 0.6 0.8 1 Case 6

Injected water, pore volumes