Professional Documents
Culture Documents
Scale-Squeeze-Treatment Design
in the Norne Field
O. Vazquez, SPE, and E. Mackay, SPE, Heriot-Watt University, and T. Tjomsland, SPE,
O. Nygård, and E. Storås, Statoil
Summary vide information about the reservoir dynamics and can be injected
When squeezing scale inhibitors (SIs) into oil-production wells, the at any time during injection (Cubillos et al. 2006). Tracers have
inhibitor should usually be uniformly placed in the open intervals been used to identify flow paths between injectors and producers,
to optimize squeeze lifetime. In wells with varying reservoir quality which can be used to identify water-breakthrough times and the
and/or significant crossflow, however, uniform placement is diffi- estimation of interwell oil saturations (Datta-Gupta et al. 1995).
cult to obtain. Flow diverters are frequently used to improve the There are a number of studies in the literature in which natural or
chemical placement.In many cases, it is of great interest to evaluate artificial tracers were used for the characterization of reservoir
the squeeze performance and assess the actual placement and back continuity and to improve reservoir-simulation models (Huseby
production of inhibitor to gather well information and thereby opti- et al. 2005, 2010; Valestrand et al. 2010). Radioactive tracers
mize future squeeze designs. This can be particularly interesting in have been used to assess the placement of stimulation treatments
subsea wells in which other types of data collection, such as pro- by use of gamma ray logging procedures (McLaughlin 1996).
duction logging, are not feasible because of high intervention costs The purpose of the work presented in this paper was to design
and high operational risk. This study suggests the use of tracers dur- a tracer program to identify whether or not the layer flow-rate dis-
ing squeeze treatments to evaluate the placement as an alternative tribution extracted from the history-matched model was accurate
to running production-logging tools (PLTs). enough to assist with the squeeze-design process, and thereby
The main purpose of this paper is to demonstrate the applicabil- avoid the need to run a PLT. A number of different tracer-pro-
ity of tracers [in this particular study, the injection of a potassium gram scenarios were simulated to determine layer flow-rate distri-
chloride (KCl) slug in a producer well in the Norne field] to evalu- butions at different production rates. All the scenarios had the
ate the layer flow-rate profile along the completion interval, which same target injection rate of 1130 L/min and 2000 m3 of produced
depends on the pressure and geological properties of each layer. fluid within 12 hours. However, given these constraints, the pro-
The study consists of verifying the layer flow-rate profile predicted duction rates could be varied in line with several multiple-rate
by a history-matched reservoir model. On the basis of this layer separator tests (Tjomsland et al. 2012). The resulting layer flow-
flow-rate profile, a tracer-injection program is designed, which rate profiles were included in a near-wellbore model to determine
includes two production stages at different rates. Finally, on the ba- the best tracer-program design in terms of a distinctive tracer-
sis of the reservoir-model predictions, it is identified that each layer return concentration profile. Finally, a number of simulations
is at different pressures, which leads to a distinctive return profile. were performed to optimize future squeeze treatments with the
To evaluate the match between the observed data and the simula- verified layer flow-rate distribution.
tion data, the layer flow-rate profile from the reservoir model was
used to populate a specialized near-wellbore model for scale- KCl-Tracer-Program Design
squeeze treatments. The match between the observed data and the
It was decided that KCl brine would be used as a tracer, and po-
simulated data was good. However, the near-wellbore model, in par-
tassium would be monitored during the early production stages.
ticular the layer flow-rate profile, was fine-tuned further. Finally, the
Potassium was chosen because there is a large contrast between
fine-tuned near-wellbore model was used to optimize future treat-
its concentration in the formation brine and in the injection brine.
ments more accurately with the fine-tuned layer flow-rate profile.
50 m3 of KCl tracer brine at 10% concentration would be pumped
into the reservoir section of the well at 1130 L/min. This would
be followed by a period during which the well would be shut in to
Introduction allow the KCl pill to crossflow between the layers in the event a
The Norne field is located in the southern part of the Norwegian significant pressure differential existed between the layers. The
Sea, approximately 85 km north of the Heidrun field and 200 km well would be brought back onto production, the rate being
offshore Norway. The water depth in the area is 380 m (1,250 ft). increased in a stepwise manner, and water samples would be col-
The field area is approximately 93 km. The reservoir is formed lected and analyzed until 2000 m3 of fluid was produced. A num-
by Jurassic sandstones, the Norne structure is relatively flat at a ber of scenarios involving different production rates were
depth of approximately 2525 m below mean sea level, and oil is simulated to study the effect of the production rate on the antici-
produced from the Ille and Tofte formations. The reservoir prop- pated tracer-return concentrations. Finally, the layer flow-rate dis-
erties are generally good, with porosity and permeability values tributions for the different scenarios would be implemented in a
typically in the range of 25 to 32% and 200 to 2,000 md, respec- specialized near-wellbore model to identify the most appropriate
tively. The main recovery strategy is seawater flooding; some gas tracer-program design.
injection was used for the first 10 years of production, but only
seawater injection has been used since 2007.
Tracers have been used in the oil industry for the last 50 years Scenarios Simulated in Reservoir Simulator
for reservoir characterization (Du and Guan 2005). Tracers pro- Different scenarios were simulated by use of a history-matched
reservoir model to determine the layer flow-rate diversion during
the treatment, to identify the likely SI placement, and to identify
Copyright V
C 2014 Society of Petroleum Engineers
the flow distribution at different production rates. All the scenar-
This paper (SPE 164114) was accepted for presentation at the SPE International ios had the same target of producing 2000 m3 of fluid in 12 hours.
Symposium on Oilfield Chemistry, The Woodlands, Texas, USA, 8–10 April 2013, and
revised for publication. Original manuscript received for review 28 February 2013. Revised
Descriptions of the production scenarios considered are as
manuscript received for review 29 November 2013. Paper peer approved 6 December 2013. follows:
800
600
3
Injection 1.13 m /min
400
Shut-in 30 minutes
200 3
600 m /d
3
1200 m /d
0
3
6000 m /d
–200
–400
–600
0 100 200 300 400 500 600 700 800
Completion, m
Fig. 1—Layer flow-rate profiles for injection, shut-in, and production stages. (2ve refers to negative injection and +ve refers to pos-
itive production.)
Stage Type Tracer Concentration (ppm) Flow Rate Time Volume (m3)
Concentration, ppm
Concentration, ppm
10,000 10,000
(3) 600 m3/d for 0.315 days
1,000 600 m3/d for 0.185 days 1,000 (4b) 1200 m3/d for 0.3 days
(overlies 1) 6000 m3/d for 0.2 days
100 100
(4) 1200 m3/d for 0.208 days
10 6000 m3/d for 0.292 days 10 (4c) 1200 m3/d for 0.35 days
6000 m3/d for 0.15 days
(5) 6000 m3/d for 0.292 days
1 1200 m3/d for 0.208 days 1
.1 .1
.1 1 10 100 1000 10 000 0 10 20 30
Cumulative Water Volume, 100 m3 Cumulative Water Volume, 100 m3
Fig. 2—Simulated tracer-return concentration profiles. Fig. 3—Simulated tracer-return concentration profiles—varia-
tions around Scenario 4.
scenarios, a second peak appears as a result of the breakthrough
of different layers, which is particularly evident in Scenario 2. m3 of diesel to displace the KCl slug into the formation with the
Thus, Scenarios 2 and 4 resulted in the most distinguishable displacement volume consisting of tubing volume to perforations
return profiles, with a clear change of slope occurring when the pro- (56.7 m3) and perforated-interval volume (8.9 m3). The analysis
duction rate was changed. However, in Scenario 2, the volume of of the field data was divided into interpretation of the potassium-
fluid produced at higher rate is 17 times greater than at the low ion (Kþ)-return concentration profile and modeling of the field
rate, whereas for Scenario 4, it is only 7 times greater. The ratio of data, as discussed in the following two subsections.
produced volume is an important factor to take into account for the
analysis of the effluent, and it seemed more reasonable to make this
ratio as small as possible to avoid dilution in the wellbore and to Interpretation of K1-Return Concentration Profile. Produced-
reduce the risk of failing to analyze returns during production of water samples were taken at regular intervals and analyzed for
þ
the smaller fraction. Therefore, Scenario 4 was recommended. Sce- eight ions, including SO2 4 . K was used as the reference to
nario 4 was further optimized to make the features more clear, by determine the tracer concentration because the concentration
increasing the production time at lower rate and decreasing it at contrast with the formation water is more favorable than for the
higher rate, keeping the overall 12-hour target. Fig. 3 shows a com- chloride (Cl) ion. Because there is some background Kþ concen-
parison of the following scenarios on the basis of Scenario 4: tration in the produced water, it was necessary to remove the
Scenario 4a: 1200 m3/d for 0.208 days (5 hours); 6000 m3/d background level to obtain the tracer profile. As shown in Fig. 4,
for 0.292 days (7 hours) there is clear change of slope after 0.4 days, as predicted by the
Scenario 4b: 1200 m3/d for 0.3 days (7 hours, 12 minutes); modeling.
6000 m3/d for 0.2 days (4 hours, 48 minutes) The background Kþ concentration was calculated as the mini-
Scenario 4c: 1200 m3/d for 0.35 days (8 hours, 24 minutes); mum concentration of all samples, which was 445 mg/L. Norne
6000 m3/d for 0.15 days (3 hours, 36 minutes) formation-brine Kþ concentration is 553 mg/L, but bearing in
On the basis of the simulation results, the best option is to pro- mind that the produced-water seawater cut is greater than 80%,
duce at 1200 m3/d for as long as possible, followed by 6000 m3/d, the produced-brine composition will be closer to normal North
with the rationale being to produce as much as possible at the Sea seawater (NSSW) composition (NSSW Kþ concentration is
lower rate to ensure enough fluid is available for sampling at the approximately 460 mg/L); therefore, assuming 445 mg/L as the
lower rate. There is a clear change in slope when the flow rate is Kþ background concentration seems a reasonable assumption.
increased from 1200 to 6000 m3/d, which should be clearly dis-
cernible from the tracer-return concentration profile.
Modeling Field Data. The next step was to study the change of
slope predicted in the simulations, which occurred when the flow
Field-Data Interpretation rate was increased. Comparing the Kþ profile to the total produc-
The final tracer-program design consisted of the injection of 50 tion flow rate (Fig. 5), it seems that the change of slope and the
m3 of KCl at 10% concentration at 1200 L/min, followed by 67.6 change of flow rate are not synchronized; it appears that the
10 000
1000
Return Concentration, mg/L
100 +
K Field Return
10 +
K minus background
minimum
0.1
0 0.1 0.2 0.3 0.4 0.5
Cumulative Time, days
1 2000
0.1 0
0 0.1 0.2 0.3 0.4
0 0.5
Cumulative Time, days
Fig. 5—K1 concentration (minus background) and total production flow rate vs. time.
change of slope is delayed. The observed delay is a result of the sibility is that some of the tracer pill could have been placed in
volume of fluid in the pipe tubing, which is 184.34 m3 [tubing layers at low pressure that did not flow when the well was brought
volume at perforations (8.9 m3) plus tubing volume to perfora- back onto production, or perhaps these layers produced only oil
tions (56.7 m3) plus volume of flowline and riser (118.74 m3)]. (i.e., water breakthrough had not yet occurred in these layers).
The production flow rates in the simulation were set as the av- Another alternative is that water breakthrough had occurred, but
erage of the water-flow rate at the separator for each production that gravitational effects while the well was shut in would have
stage. The first stage had an average water rate of 1797 m3/d for caused the water front to recede, such that during the initial period
0.26 days, and the second stage had an average rate of 5973 m3/d of production, no water would have flowed from these layers, and
for 0.15 days. These data were determined from the Kþ return the tracer would have been left in the reservoir until a slightly
profile in which the 0.26-day interval corresponds to the first slope later period. In the next section, a stochastic method will be
in the tracer profile and the 0.15-day interval corresponds to the applied to further tune the layer flow-rate distribution in an
second. These adjustments are considered to capture the details of attempt to obtain a quantitative match.
the tracer program adequately. The simulation profile provides a Other possible explanations that should be considered include
qualitative match in which the change of slope is matched, but it cation exchange, which is known to occur in clay-rich systems.
is a poor match quantitatively (Fig. 6). However, it is unlikely that cation-exchange capacity (CEC) will
The poor quantitative match might be attributed to different explain the loss of 2000 kg of Kþ, equivalent to 5.1107 meq. If
factors. The most obvious factor arises from uncertainty in the ac- it did, CEC would be 1020 meq/L (5.1107 meq/50103 L of
curacy of the layer flow-rate distribution extracted from the his- pore volume). It could be attributed to the presence of clays, in
tory-matched reservoir model. Second, not all of the tracer pill is particular to kaolinite with CEC of 150 meq/kg (equivalent to
back produced. The total amount of Kþ injected was 50 m3 of 186–930 meq/L, assuming bulk density of 1.86 g/cm3 and poros-
KCl at 10% concentration, which gives a total mass of 2900 kg; ity of 30%). However, the Tofte formation quartz content is more
however, only 988 kg was recovered after 12 hours of production, than 90% (Dalland et al. 1988), and consider that a reasonable
which is 34% of the total injected mass. Although this could be CEC for a sandy aquifer is approximately 100 meq/L (Appelo and
partially because of uncertainties in the water-flow-rate measure- Postma 2006).
ments, these would have to be adjusted by a factor of 3 to give a Other geochemical reactions involving Kþ that could account
closer match, which was ruled out as too high an adjustment. for this mass loss are not commonly observed in oilfield systems,
Rather than an error in measurements, it is possible that the and so it seems likely that the observation is a result of unac-
tracer pill was not produced back in full, in which case the most counted flow behavior. Behind-pipe flow can lead to unexpected
plausible suggestion is that the layer flow-rate distribution extrac- flow behavior, but, here, there is no other evidence of this occur-
ted from the reservoir model should be further adjusted. One pos- ring, and, because there is evidence of layers at different pressures
100 000
10 000
Return Concentration, mg/L
+
1000 K minus
background
minimum
100
10 Near-wellbore
simulation
1
0.1
0 0.1 0.2 0.3 0.4 0.5
10 000
10 All Stages
Only Prod
1 Lay12_13Close
0.1
0.01
0 0.1 0.2 0.3 0.4 0.5
Time, days
(the reason this project was undertaken in the first place), it seems 3. The fitness value of the particle and its best position (pbest)
probable that this, at least, is part of the explanation. are compared for each individual particle. If the current value is
better than the pbest value, then the pbest value and its position are
Further Tuning of Layer Flow-Rate Distribution replaced by the current fitness value and position.
In this section, further tuning of the layer flow-rate distribution is 4. Global best position and pbest value are updated.
used in an attempt to find a good match between the observed data 5. Velocities and positions of all particles are updated.
and the simulation data. The concept here is to use the model to 6. Go to Step 2 until a criterion for stopping is met (maximum
define various flow scenarios to identify whether any of these can number of iterations or good fitness value).
account for the observed behavior. Evidently, caution must be used PSO was applied to a number of scenarios in which the layer
because the solution may be nonunique. The layer flow-rate profile flow-rate distribution for a number of stages was varied, with all
from the reservoir model was used as the initial guess to populate the others retaining the same values as those extracted from the reser-
specialized near-wellbore model, and then flow rates were altered to voir model. The scenarios under study are as follows: all the stages
identify whether a better match could be achieved (Vazquez et al. are varied (All Stages); only production stages (Only Prod); and,
2006). To find a good match, a particular optimization method was finally, injection and production stages (shut-in stages not varied)
used, known as a particle-swarm-optimization (PSO) algorithm. except where Layers 12 and 13 are closed (Lay12_13Close)—cor-
PSO is an optimization technique proposed by Kennedy and responding to these layers not producing any water. Fig. 7 shows
Eberhart (1995) that is considered to be a powerful tool in com- the PSO results in which PSO suggested a slight improvement, but
parison with other population-based evolutionary algorithms still not a quantitative match for the All Stages case. Similar results
(Mohamed et al. 2010a). PSO is a swarm-intelligence approach, were obtained for the Only Prod case. However, in the last case in
which finds an optimal solution by performing operations with which Layers 12 and 13 are set to not produce any water, the match
given data. It is defined as a stochastic-optimization technique, was significantly improved. The suggested flow distribution per
which has been inspired by the social behavior of fish schooling stage and per layer can be found in Fig. 8. From the results, it
or birds flocking (Onwubolu and Babu 2004; Vazquez et al. seems that Layers 12 and 13 should be at lower pressure than pre-
2013a). Successful application of PSO can be found in various dicted by the reservoir model, which is consistent with the fine-
fields, applications, and research areas such as fuzzy computing, tuned solution (Lay12_13Close) because it suggests that higher
chaos theory, engineering applications, and even in psychology. injection should occur in these layers.
The basic PSO workflow is summarized in the following steps,
which have been previously described by Hajizadeh et al. (2011),
Mohamed et al. (2010a, 2010b), and Vazquez et al. (2013b): Isotherm History Matching
1. The swarm is initialized by assigning a random position in SI-squeeze treatments are the most widespread method to combat
the parameter space to each particle with random velocity. downhole scaling. To predict SI-squeeze treatments accurately for
2. Fitness position for each particle is evaluated. further optimization, it is necessary to simulate the SI retention in
0.6
Stage 1 Stage 2 Stage 3 Stage 4 Stage 5
0.5
Flow-Rate Percentage
0.4
Reservoir Model
0.3
All Stages
0.2 Only Prod
Lay12_13Close
0.1
0
S1_L10
S1_L11
S1_L12
S1_L13
S1_L14
S2_L10
S2_L11
S2_L12
S2_L13
S2_L14
S3_L10
S3_L11
S3_L12
S3_L13
S3_L14
S4_L10
S4_L11
S4_L12
S4_L13
S4_L14
S5_L10
S5_L11
S5_L12
S5_L13
S5_L14
S1_L1
S1_L2
S1_L3
S1_L4
S1_L5
S1_L6
S1_L7
S1_L8
S1_L9
S2_L1
S2_L2
S2_L3
S2_L4
S2_L5
S2_L6
S2_L7
S2_L8
S2_L9
S3_L1
S3_L2
S3_L3
S3_L4
S3_L5
S3_L6
S3_L7
S3_L8
S3_L9
S4_L1
S4_L2
S4_L3
S4_L4
S4_L5
S4_L6
S4_L7
S4_L8
S4_L9
S5_L1
S5_L2
S5_L3
S5_L4
S5_L5
S5_L6
S5_L7
S5_L8
S5_L9
Fig. 8—Layer flow-rate distribution suggested by PSO per stage and per layer.
10,000
SI-Return Concentration
1,000
Treat 1
100
Tracer isotherm
0.1
10 100 1000 10 000
the formation, which may be described by a pseudoadsorption iso- halved (Table 5). Doubling the overflush volume will effectively
therm (Vazquez et al. 2013a). mean injecting almost double the volume of water in the treat-
The well under study was treated twice before with the same ment, which may not be feasible if the well has lifting constraints
SI package, and the return data were used to derive two pseudoi- or if there are concerns about water damage.
sotherms, one assuming the original reservoir-model flow rates,
and the other by use of the fine-tuned distribution flow rates, and Conclusions
where an automatic isotherm-derivation technique proposed pre-
The applicability of a tracer program to evaluate the layer flow-
viously (Vazquez et al. 2013a) was implemented.
rate distribution as an alternative to running a PLT has been dem-
The pseudoisotherms derived with Treatment 1 (Fig. 9) provided
onstrated. Initially, the layer flow-rate distribution was extracted
matches for Treatment 2 (Fig. 10), with the match assuming the fine-
from a history-matched reservoir model that was used to design
tuned layer flow-rate distribution being much more accurate than the
the most appropriate tracer program. Then, the simulation data
one assuming the original reservoir-simulation layer flow rates.
and the observed data were compared, and only a qualitative
Fig. 11 describes the pseudoisotherms derived; the case, from which
match was obtained between the data. It was observed that only
the layer flow-rate distribution was extracted from the reservoir
30% of the injected tracer was recovered, and this did not con-
model, suggested adsorption levels in the hundreds of mg/g, which
form to the predictions made with the flow rates from the reser-
is unrealistically high. Therefore, the fine-tuned layer flow-rate-pro-
voir-simulation model.
file isotherm should be used for any optimization calculations.
An attempt to achieve a quantitative match was made by
adjusting the layer flow-rate distribution. In particular, a PSO
Optimization for the Next Squeeze Treatment algorithm was applied during which the layer flow-rate distribu-
In this section, we describe a series of calculations that were per- tion was varied per stage and per layer. The results suggested that
formed on the basis of the pseudoisotherm and fine-tuned layer SI might be placed into Layers 12 and 13, interpreted as low-pres-
flow-rate-distribution profile to optimize a subsequent squeeze sure layers, which did not then produce any water during the early
treatment. Table 3 provides a description of the generic squeeze- production stages when tracer returns were being measured. As a
treatment design. Two aspects were considered: overflush volume consequence, injected tracer remained trapped in these layers for
(Fig. 12) and SI concentration (Fig. 13). Table 4 shows the vol- at least as long as the tracer returns were being measured. Assum-
ume of produced water that is protected for three different ing this fine-tuned layer flow-rate distribution on the basis of the
minimum–inhibitor-concentration (MIC) values. MIC is the PSO algorithm, SI returns were also matched and an isotherm was
threshold concentration at or above which scale will be inhibited. derived from one prior treatment and was used in the prediction
As shown, the extra volume of water protected is very similar of a subsequent treatment.
whether the SI concentration is doubled or the overflush volume Finally, assuming this fine-tuned layer flow-rate distribution, a
is doubled, and a similar decrease is observed when either is number of calculations were performed to optimize a future-
100,000
10,000
SI-Return Concentration
1,000
Treat 2
100
Tracer isotherm
0
10 100 1000 10 000
100
Reservior model
Adsorption, mg/g
k2756 n0.49
10
Tracker k2948
n0.2
0.1
0.01 1 100 10,000 1,000,000
SI Concentration, ppm
Fig. 11—Adsorption pseudoisotherm derived assuming reservoir-simulation flow-rate profiles and assuming flow-rate profiles
from tracer-matching exercise.
100,000
10,000
SI-Return Concentration, ppm
1000
Base case
100
Double OF
10 Half OF
0.1
10 100 1000
Cumulative Water Produced, 100 m3
Fig. 12—Optimization: SI-return concentration profiles for various overflush (OF) volumes.
10,000
Base case
100
Double SI
10 Half SI
0.1
10 100 1000
Cumulative Water Produced, 100 m3
extracted from the existing full-field reservoir-simulation model, Computational Intelligence, Springer Berlin Heidelberg. http://
and by conducting some simple laboratory experiments (e.g., to dx.doi.org/10.1007/978-3-642-21705-0_8.
identify partitioning coefficients). Huseby, O., Chatzichristos, C., Sagen, J. et al. 2005. Use of natural geo-
chemical tracers to improve reservoir simulation models. J. Pet. Sci.
Acknowledgments Eng. 48 (3–4): 241–253. http://dx.doi.org/10.1016/j.petrol.2005.06.002.
Statoil is thanked for funding this work, for the care and attention Huseby, O., Valestrand, R., Naevdal, G., and Sagen, J. 2010. Natural and
put into gathering accurate data to improve the accuracy of the Conventional Tracers for Improving Reservoir Models Using the
modeling activity, and for permission to publish this paper. EnKF Approach. SPE J. 15 (4): 1047–1061. SPE-121190-PA. http://
dx.doi.org/10.2118/121190-PA.
Kennedy, J. and Eberhart, R. 1995. Particle swarm optimisation. In Pro-
References ceedings of the IEEE International Conference on Neural Networks,
Appelo, C.A.J. and Postma, D. 2006. Geochemistry, Groundwater, and Vol. 4, 1942–1948. Piscataway, New Jersey: IEEE Service Center.
Pollution, second edition (second printing). Rotterdam, The Nether- McLaughlin, J.S. 1996. Radioactive Tracers: Review of Principle Factors
lands: A.A. Balkema. in Design and Application. Presented at the Permian Basin Oil and
Cubillos, H., Torgersen, H., Chatzichristos, C., and Lamela, M. 2006. Best Gas Recovery Conference, Midland, texas, USA, 27–29 March. SPE-
Practice and Case Study of Interwell Tracer Program Designs. Pre- 35233-MS. http://dx.doi.org/10.2118/35233-MS.
sented at the First International Oil Conference and Exhibition, Can- Mohamed, L., Christie, M., and Demyanov, V. 2010a. Comparison of Sto-
cun, Mexico, 31 August–2 September. SPE-103891-MS. http:// chastic Sampling Algorithms for Uncertainty Quantification. SPE J.
dx.doi.org/10.2118/103891-MS. 15 (1): 31–38. SPE-119139-PA. http://dx.doi.org/10.2118/119139-PA.
Dalland, A., Worsley, D., and Ofstad, K. 1988. A lithostrategraphic Mohamed, L., Christie, M., and Demyanov, V. 2010b. Reservoir Model
scheme for the Mesozoic and Cenozoic succession offshore mid- and History Matching With Particle Swarms: Variants Study. Presented at
northern Norway. NPD Bulletin No. 4, Oljedirektoratet (Norwegian the SPE Oil and Gas India Conference and Exhibition, Mumbai, India,
Petroleum Directorate), Stavanger, Norway (January 1988), http:// 20–22 January. SPE-129152-MS. http://dx.doi.org/10.2118/129152-
npd.no/Global/Norsk/3%20-%20Publikasjoner/NPD%20Bulletin/NPD_ MS.
BulletinNo4.pdf. Onwubolu, G.C. and Babu, B.V. 2004. New Optimization Techniques in
Datta-Gupta, A., Lake, L.W., and Pope, G.A. 1995. Characterizing hetero- Engineering. Berlin: Studies in Fuzziness and Soft Computing,
geneous permeable media with spatial statistics and tracer data using Springer-Verlag.
sequential simulated annealing. Math. Geol. 27 (6): 763–787. http:// Tjomsland, T., Sandøy, B., Fadnes, F.H., and McCartney, R. 2012. Appli-
dx.doi.org/10.1007/bf02273537. cation of Multirate Well Tests to Scale Management: Part 2—Interpre-
Du, Y. and Guan, L. 2005. Interwell Tracer Test: Lessons Learned from tation of MRTs With Known Produced-Water Origin. SPE Prod &
Past Field Studies. Presented at the SPE Asia Pacific Oil and Gas Con- Oper 27 (3): 327–336. SPE-156519-PA. http://dx.doi.org/10.2118/
ference and Exhibition, Jakarta, 5–7 April. SPE-93140-MS. http:// 156519-PA.
dx.doi.org/10.2118/93140-MS. Valestrand, R., Sagen, J., Naevdal, G., and Huseby, O. 2010. The Effect of
Hajizadeh, Y., Demyanov, V., Mohamed, L. et al. 2011. Comparison of Including Tracer Data in the Ensemble-Kalman-Filter Approach. SPE
Evolutionary and Swarm Intelligence Methods for History Matching J. 15 (2): 454–470. SPE-113440-PA. http://dx.doi.org/10.2118/
and Uncertainty Quantification in Petroleum Reservoir Models. In 113440-PA.
Intelligent Computational Optimization in Engineering, M. Köppen, Vazquez, O., Corne, D., Mackay, E.J., and Jordan, M.M. 2013a. Auto-
G. Schaefer, and A. Abraham, Vol. 366, Chap. 8, 209–240. Studies in matic Isotherm Derivation From Field Data for Oilfield Scale-Inhibitor