Journal of Natural Gas Science and Engineering 22 (2015) 313e320

Contents lists available at ScienceDirect

Journal of Natural Gas Science and Engineering

journal homepage: www.elsevier.com/locate/jngse

Use of data reconciliation: A strategy for improving the accuracy in gas

flow measurements
Elcio Cruz de Oliveira a, b, *, Maurício Nogueira Frota b, Gisele de Oliveira Barreto a, b
a
Petrobras Transporte S.A., Project Management, Rio de Janeiro, 20091-060 RJ, Brazil
b
Posgraduate Programme in Metrology, Metrology for Quality and Innovation, Pontifical Catholic University of Rio de Janeiro, 22453-900 RJ, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: The economic impact that results from the reliability of measurements associated with natural gas (flow
Received 10 October 2014 rate and fluid properties) and caveats related to custody transfer contracts demands vigilant control of
Received in revised form the net balance in the delivery systems. The methodology for data reconciliation has proved to be an
10 December 2014
effective tool to reduce uncertainties associated with measurements used in the calculations of the net
Accepted 11 December 2014
Available online 18 December 2014
balance in distribution networks such as gas pipelines. The intrinsic nature of the calculation algorithm,
founded on the redundancy of measurements, qualifies the technique for increasing confidence in the
measurement, thus reducing the individual uncertainty associated with each physical magnitude capable
Keywords:
Data reconciliation
of affecting the measurement.
Unaccounted for gas This Brazilian gas pipeline study discusses the adequacy of the data reconciliation technique. The
Improving the accuracy proposed technique proves to be very effective as it generates lower uncertainties than those obtained by
Gas net balancing traditional techniques: the level of 1% associated with the accountability of the unaccounted for gas was
Gas flow measurements reduced to less than 0.3% when the data reconciliation methodology was used.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction operating and handling techniques, imposing challenges on their

Although oil had been avoided by ancient civilizationsd a heavy
waste that exhibits strong and viscous odord, early records confirm
that natural gas was discovered in Iran between 6000 and 2000 BC
and that the Persians had already used it to record the “eternal fire”
symbolizing the adoration of their profound religious convictions.
Since 900 BC, natural gas has been extracted in China with the help
of bamboo poles introduced into wells 1000 m deep. However, it
was only in the year 211 BC that the raw material was used to
accelerate the drying of salt stones.
Motivating multimillionaire contracts for their commercializa-
tion, international technical cooperation and even wars between
nations that dispute their exploitation, oil and natural gas are
perceived by society in general as strategic energy inputs. Their
commercialization, whether for lubrication or energy fuels, re-
quires rigorous quality controls that impose challenges to ensure
transport with safety and environmental care. Distinct in their
composition, their transport and use require special care, specific
* Corresponding author. Petrobras Transporte S.A., Project Management, Rio de
Janeiro, 20091-060 RJ, Brazil.
E-mail address: elciooliveira@petrobras.com.br (E.C.d. Oliveira).
exploratory phases (Amui, 2010).
More specifically, as far as natural gas is concerned, diverse
factors contribute to elect it as an attractive alternative less
polluting energy: the growth in demand for energy; opportunities
created by the competitive international energy market, and the
high costs associated with the production and transport of oil. Its
use has also been disregarded by the consciousness that oil con-
sumption to generate electricity uncontestably constitutes a rele-
vant portion of the consumer's ecological footprint.
Experience has shown that the control of measurements
dsensitive to different measuring techniquesd is not a simple
task. The use of inadequate measuring methods, either for non-
compliance with critical length established in measurement
norms, incorrect installation and lack of calibration of the
equipment are among factors that severely may compromise the
quality of measurement. The reliability of measurement results
depends on strict requirements related to the installation of the
measurement system and characteristics of the flow to be
measured (level of turbulence, velocity profiles upstream of the
meter, variations in fluid properties). Inaccuracies associated with
measurements certainly compromise the confidence in the audit
processes associated with the transfer of custody, particularly in

http://dx.doi.org/10.1016/j.jngse.2014.12.008
1875-5100/© 2014 Elsevier B.V. All rights reserved.
314 E.C.d. Oliveira et al. / Journal of Natural Gas Science and Engineering 22 (2015) 313e320
the case of volume flow measurement in gas pipelines. From the
economic point of view, inaccuracies in measurements can reach
intolerable amounts as easily estimated when one considers that
enormous volumes of gas are transported (about 100 million
cubic meters of natural gas are transported daily in Brazilian
pipelines). The accuracy of measurement is dependent on the
measuring technique, the capacity of the person responsible for
the supervision, the calibration of the meter and the frequency of
the measuring process. When operated correctly, ultrasonic and
orifice plate meters are able to provide acceptable levels of
accuracy.
The growing trend for the use of cleaner fossil fuels is the result
of the global consciousness that any alternative utilized in the
generation of electricity, one way or another, has a detrimental
impact on the environment. From this perspective, natural gas
should replace other fossil fuels in the energy matrix as its com-
bustion generates lower levels of pollutant gas emissions and res-
idues. However, the rational use of natural gas depends on an
efficient process to make its safe and economic distribution viable.
Despite the advances in gas pipelines (increasingly complex and
operated in compliance with modern legislation founded on the
concepts of transport logistics and appropriate regulation), the
adequate bookkeeping for the so-called unaccounted for gas re-
quires precise measurements whose accuracy and control of asso-
ciated uncertainties still remain as metrological challenges to be
overcome (Arpino, 2014). Orifice plate, turbine or ultrasound
technology flowmeters continue to be the techniques most
employed for measurements of gas flow, accepted by regulators as
an appropriate alternative for custody transfer. The inherent
properties of natural gas (e.g. low density and high volatility of its
components) facilitate leaks and losses of volume that could result
in serious economic impacts (on the order of billions of dollars) for
the investors. This is the reason why the control of gas transported
(conservation of mass) is so critical. Oliveira and Aguiar (2009) and
Bagajewicz (2003) showed that the utilization of the technique
called Data Reconciliation (DR) dthe use of measurement redun-
dancy to reduce the uncertainty associated with measurementsd
made an important contribution to the metrological control of
unaccounted for gas in gas pipelines. Here, redundancy is the
replication of critical components of a system with the intention of
increasing its reliability.
Whenever the laws of conservation are violated, random or
systematic errors are introduced to compromise the process of
measurement (Narasimhan and Jordache, 2000; Ozyurt € and Pike,
2004). In this context, the technique of data reconciliation dan
approach based on the statistical tool that considers the restrictions
imposed on the processd does in fact improve confidence when
measurements are analyzed on a global basis. Data reconciliation
incorporates redundant information to compensate and eliminate
random errors, thereby reducing uncertainties associated with
measurement. Considering the heteroscedastic behavior associated
with measurements to be reconciled, this approach deals with
multivariate nonparametric locally weighted least squares regres-
sion (Alhaj-Dibo et al., 2008; Mene ndez et al., 1998). The data
reconciliation technique enables several redundant measurements
to reconcile experimental data in one single value, yielding the so-
called reconciled data. Oliveira and Aguiar (2009) have shown in
previous work that controlled redundancy leads to lower levels of
uncertainty when compared to what it would have been obtained
in the absence of redundant data. Data reconciliation is based on
the conjecture that if a gross error resulting from an apparent
measurement bias is present in some measurement or if a note-
worthy process leak was not accounted for in the model con-
straints; reconciled data may be very inaccurate.
Implementation of the data reconciliation technique requires
that:

The reference model be defined based on the mathematical
premise that it is perfect, with no systematic errors. In other
words, random errors are usually distributed and independent
(Mansour and Ellis, 2008). An “objective function Fob ” is utilized
to evaluate the resulting difference between the data produced
by the model and the experimental data;

The formulated “objective function Fob ” be optimized by making
use of a multivariate distribution that models this routine,
whose results are expressed based on weightings attributed to
the experimental measurements of dispersion. In this work, the
use of uncertainties associated with the measurements to sup-
port the results is proposed. In this stage it is necessary to
minimize or maximize the probability of meeting experimental
measures;

The parameters be evaluated. Ramamurthi et al. (1993) sug-
gested the evaluation by using the maximum likelihood esti-
mation (MLE).
The data reconciliation approach is then applied to calculate the
reconciled values and their associated uncertainty bands, therefore
allowing detection and exclusion of the related gross errors.
Whenever the uncertainty band falls within an unacceptable range,
the experimental data will no longer be considered as the final
results but rather used to calculate the reconciliated values,
ensuring that there will be no partial overlap of their uncertainty
bands.
The objective of this work is to apply the approach of data
reconciliation to approve the analysis of the net balance in gas
pipelines. Brazilian gas pipelines utilize the tolerance criteria of 1%
as the social indicator to evaluate the quantity of unaccounted for
gas. On making use of the data reconciliation approach, it is ex-
pected that with this indicator incorporated the index could ach-
ieve the international standard of 0.3%.
2. Methodology
Measuring loops and inherent inaccuracies associated with
computer data processing and its storage usually contribute to the
relatively high uncertainties and unacceptable tolerances associ-
ated with measurement control in pipelines. Because all product
transactions associated with gas-pipeline network depend on flow-
rate measurements, control of the associated uncertainties is
absolutely critical. Assessment of the overall performance of a
pipeline system requires control and monitoring of the energy and
mass balances and a rigorous evaluation of potential leakages. Be-
sides the economic impact associated with leaks, their control
prevents possible environmental damage and increases the avail-
ability and maintainability of the pipeline network. Generally
speaking, the operation of a gas-pipeline network is evaluated and
controlled on the basis of the operating data (e.g. volume flow rate;
internal pressure and gas temperature) allowing that pressure drop
inside the pipeline and compressibility of the natural gas be
calculated.
2.1. Data reconciliation: the concept
Data reconciliation and gross error detection is thoroughly dis-
cussed by Narasimhan and Jordache (2000). According to the au-
thors, both are achieved by exploiting the redundancy property of
measurements. The technique improves the accuracy of process
data by adjusting the measured values to satisfy process constraints
while the amount of adjustment made to the measurements is
minimized whenever the random errors in the measurements are
 E.C.d. Oliveira et al. / Journal of Natural Gas Science and Engineering 22 (2015) 313e320 315

expected to be small.
Data reconciliation is a technique for validating the measure-
ε ¼ bZ e  Z c c (3)
ments by extracting the precise and reliable information from the
gross measurements that can reflect the intrinsic state of industrial Where ε is the error between the experimental data and the
processes. The process utilizes information and mathematical reconciled values.
methods to produce a single consistent set of data capable of rep- Substituting (3) in (1):
resenting the most probable operation of the process. The tech-
nique is strongly based on the concept of redundancy to correct the  
minimum possible results of measurements aiming to satisfy re- 1 1 T 1
PðXÞ ¼
1=2 exp  2ðεÞ V ðεÞ (4)
strictions inherent to the process. An intelligent reconciliation of
ð2pÞN detV
the values converted requires that the data dusually originating
from the physical measurements that incorporate associated
errorsd are statistically validated thereby guaranteeing that the
process of reconciliation is capable of eliminating gross errors to PðεÞ ¼ PðZ e  Z c Þ (5)
produce useful and reliable data.
Gross error detection and data reconciliation convey embodied  
1 1 e c T 1 e c
techniques complementing each other. When applied together, PðXÞ ¼
1=2 exp  2ðZ  Z Þ V ðZ  Z Þ (6)
identify and eliminate gross errors, therefore improving accuracy of ð2pÞN detV
measured data. Data reconciliation and gross error detection, when
applied together, contributes for reducing error by exploiting the PðZ e Þ needs to be maximized:
redundancy property that is inherent to measurements.
The reconciliation of collected operational data minimizes the
overall uncertainty associated with the closing up of the mass and maxbPðZ e  Z c Þc ¼ Fob
2
energy balances. Satisfactory use of redundant data minimizes the 
quantity of measured variables required to obtain a solution; i.e. 6 1
¼ max4
1=2 exp
optimize the use of measured variables associated with the ð2pÞN detV
respective available variances and co-variances. The methodology 3
allows measured variables derived from redundant measured 
1 e 7
 ðZ  Z c Þ V 1 ðZ e  Z c Þ 5
T
values to be consistently used to estimate true values. From a sta- (7)
2
tistical point of view, the data reconciliation technique increases
the accuracy of single measurements; treat the data by means of a
redundant network of information that takes into account the
actual state of the pipeline system, therefore improving measure- 2 3
 
ment tolerances that may lead to substantial economic benefit. 6 1 1 7
 ðZ e  Z c Þ V 1 ðZ e  Z c Þ 5
T
Fob ¼ max4In
1=2
N 2
ð2pÞ detV
(8)
2.2. The statistical modeling for data reconciliation
 
1
Fob ¼ max  ðZ e  Z c Þ V 1 ðZ e  Z c Þ
T
The normal multivariate distribution denotes the multidimen- (9)
sional model commonly used to describe variations in experi- 2
mental data, whose probability density function P(X) may be All constants in Equations (8) and (9) are considered to be
expressed by the diagonal matrix of co-variance V: negligible.
 
1 1 T 1 j k
PðXÞ ¼
1=2 exp  ðX  mÞ V ðX  mÞ (1)
Fob ¼ max  ðZ e  Z c Þ V 1 ðZ e  Z c Þ
T
2 (10)
ð2pÞN detV
Equations (10) and (11) can be considered similar because to
Based on the following parameters: minimize Fob is the same that maximize the probability to find
3
2 2 3 2 3 experimental measures (Narasimhan and Jordache, 2000).
X1 m1 s21 s212 / s21N
6 X2 7 6 m2 7 6 s2 s22 / s22N 7 j k
X¼6
4« 5
7 m¼6
4« 5
7 V ¼6
4 21
7
« 5 Fob ¼ min ðZ e  Z c Þ V 1 ðZ e  Z c Þ
T
« « 1 (11)
XN mN s2N1 s2N2 / s2N
Rewriting equation (11) for the condition: mc1 ¼ mc2 ¼ / ¼ mcN ;
s2irepresent the variances of diagonal matrix V. that is, the reconciliation of N results of the same property m can
The probability density function reflects a region of confidence then be expressed as:
for N points on the curve when:
2 3 2 3 2 3
me1 mc1 me1  mc1
PðXÞ ¼ constant ¼ ðX  mÞ V T 1
ðX  mÞ (2) 6m 7e 6m 7c 6m  m 7
e c
Ze ¼ 6
4« 5
2 7 Zc ¼ 6
4« 5
2 7 Ze  Zc ¼ 6
4«
2 1 7
5
Only random errors are considered in this process. That is, the
meN mcN mN  mc1
e
experimental data ðX or Z e Þ and the reconciled data ðm or Z c Þ
must essentially result from the uncertainties in the experimental For non-correlated quantities, Equation (11) assumes the
data: mathematical form:
316 E.C.d. Oliveira et al. / Journal of Natural Gas Science and Engineering 22 (2015) 313e320

reconciliation concept and the reason why line packs are always
22 3T 2 2 31 2 e 33
me1  mc1 s1 0 / 0 m1  mc1 positive.
66 m  m 7 6 0
e c
0 7 6 e c 77
Fob ¼ min6 6 2 2 7 6 s22 / 7 6 m2  m2 77 Experimental uncertainties used: GTR and GTE (1.5%); GCOM
44 « 5 4 « « 1 « 5 4« 55 and DEMP (3.0%).
mN  mcN
e
0 0 / s2N meN  mcN Taking the ideal situation expressed by GPOE ¼ 0 as a basis, the
unaccounted for gas (GNC) may be calculated by (16) and (17):
(12)

2 GNC ¼ GTR  GTE  GCOM  DEMP (16)

6 e 
Fob ¼ min6 c e c e c
4 m1  m1 m2  m2 /mN  mN
GNC
GNCð%Þ ¼  100 (17)
2  2 32 33 GTR
1 s1 0 / 0 me1 mc1
6 0  76 me 77
1 s22 / 0  mc2
6
4 « 54 «
76 2 77
55 (13)
« 1 «
 2 e c
0 0 / 1 sN m N  mN

Accordingly, the objective function assumes the form:

"
2 e
2 e
2 #
me1  mc1 m2  mc2 mN  mcN
Fob ¼ min þ þ/þ 4. Results and discussion
s21 s22 s2N
(14) Table 1 describes volumes of natural gas transported in the
pipeline system during a seven-day period.
As redundancy exists in the measurements, the process model
constraints are considered by Equation (18):
3. Experimental analysis

The UrucueManaus Brazilian gas pipeline located in the GTRc ¼ 0:96  GTRc þ 0:01  CGOM c þ 0:03  DEMP c (18)
northern region of Brazil started its operation in 2009. Based on the The constraints may be described by the expressions:
experimental data collected along a seven-day period (Feb 1, to Feb
7, 2013) and taking into account that the pipeline is 662 km long
(383 km of 20 inches diameter pipe and 279 km of 18 inches GTEc ¼ 0:96  GTRc (19)
diameter pipe) and that the average volume flow rate is 4.26
million m3 per day, the data reconciliation analysis can be devel-
GCOMc ¼ 0:01  GTRc (20)
oped as follows.
Fig. 1 depicts the simplified mathematical model proposed:
The data collected were treated by means of the data reconcil- DЕМ P c ¼ 0:03  GTRc (21)
iation technique, in compliance with the balance equation (15):
General equation (14), when expressed in terms of GTRc , as-
GTR ¼ GTE þ GCOM þ GPOE þ DEMP (15) sumes the following forms:
In this expression: "
ðGTRe  GTRc Þ ðGTEe  GTEc Þ ðGCOM e  GCOM c Þ

GTR: Total gas received, measured by an orifice plate meter; Fob ¼ þ þ
s2GTRe s2GTEe s2GCOMe

GTE: Total gas delivered, measured by a four-path ultrasonic #
flow meter; DEMP e  DEMP c
þ

GCOM: Fuel gas, measured by micro turbine flow meters; s2DEMP e

GPOE: Total gas lost in the operation, estimated by equations of
(22)
state from differential pressure, pressure, temperature, diam-
eter of the pipeline and gas composition;

DEMP: Variation in line packing, assessed by comparing ma-
nometers readings.

Natural gas can be temporarily stored in the pipeline system

itself, through a process called line packing (line pack). Generally, Table 1
gas volumes can vary due to temperature and pressure changes on Volumes of natural gas.
two consecutive days. However, as discussed above, experimental GTRe (m3) GTEe (m3) GCOMe (m3) DEMPe (m3) GNC (%)
data are adapted to facilitate readers to understand the data
Day 1 2674969 2560900 25145 80249 0.32
Day 2 2677264 2570173 40000 80318 ¡0.49
Day 3 3922694 3765786 39227 112888 0.12
Day 4 3917279 3760588 19333 117518 0.51
Day 5 2755777 2645546 37001 81999 ¡0.32
Day 6 2711803 2603330 38111 81354 ¡0.41
Day 7 2676892 2569816 26212 70035 0.40

The experimental values of the unaccounted for gas that appear in bold correspond,
Fig. 1. Simplified architecture of a gas pipeline. in absolute terms, to a level that exceeds 0.3%.
 E.C.d. Oliveira et al. / Journal of Natural Gas Science and Engineering 22 (2015) 313e320 317

2 2 2 2 2 3
ðGTRe Þ  2  GTRe  GTRc þ ðGTRc Þ ðGTEe Þ  2  GTEe  GTEc þ ðGTEc Þ
6 2
þ 2
þ7
6 s e s e 7
6 GT R GT E 7
6 7
6 ðGCOMe Þ2  2  GCOMe  GCOMc þ ðGCOM c Þ2 7
6 þ 7
Fob ¼6 7 (23)
6 s 2 7
6 GCOM
e 7
6 7
6 2 2 7
4 ðDEMP Þ  2  DEMP  DEMP þ ðDEMP Þ
e e c c 5
s2 e
DEM P

2 ðGTRe Þ2  2  GTRe  GTRc þ ðGTRc Þ2 3

þ
6 s2 e 7
6 GT R 7
6 7
6 7
6 ðGTE Þ  2  GTE  0:96  GTR þ 0:96  ðGTR Þ
e 2 e c 2 c 2 7
6 þ 7
6 7
6 s2 e 7
6 GT E 7
Fob ¼ 6 7
6 2 2 7
6 ðGCOM e Þ  2  GCOMe  0:01  GTRc þ 0:012 ðGTRc Þ 7
6 þ7
6 s2 7
6 GCOM
e 7
6 7
6 7
4 ðDEMP e Þ  2  DEMP e  0:03  GTRc þ 0:032  ðGTRc Þ 5
2 2

s2 e
DEM P
(24)

2 2  GTRe þ 2  GTRc 1:92  GTEe þ 1:8432  GTRc 3

þ þ
6 s2 e s2 e 7
vFob 6 GT R GT E 7
¼6
6
7 (25)
vGTR c c7
4 0:02  GCOM e þ 0:0002  GTRc e
0:06  DEMP þ 0:0018  GTR 5
þ
s2 e s2 e
GCOM DEM P

0 12
2
B C
2 B s2GTRe C
UGTR c ¼ B  UGTRe C
@ 2 þ 1:8432 þ 0:0002 þ 0:0018 A
s2GTRe s2GTEe s2GCOMe s2DEMPe

0 12
1:92
B C
B s2GTEe C
þB  UGTEe C
@ 2 þ 1:8432 þ 0:0002 þ 0:0018 A
s2GTRe s2GTEe s2GCOMe s2DEMPe

0 12
0;02
B C
B s2GCOMe C
þB  UGCOMe C
@ 2 þ 1:8432 þ 0:0002 þ 0:0018 A
s2GTRe s2GTEe s2GCOMe s2DEMPe

2GTRe þ 1:92GTE
e
þ 0:02GCOM þ 0:06DEMP
e e
0 12
c s2GTRe s2 s2 s2
GTR ¼ GTEe GCOMe DEMP e
(26) 0;06
2 þ 1:8432 þ 0:0002 þ 0:0018 B C
s2GTRe s2GTEe s2GCOMe s2DEMPe B s2DEMPe C
þB  UDEMP eC (27)
@ 2 þ 1:8432 þ 0:0002 þ 0:0018 A
By applying the applicable concepts described in the Guide to s2GTRe 2
sGTEe 2
sGCOMe
2
sDEMPe
the expression of uncertainty in measurement (GUM, 1995), one
obtain:
318 E.C.d. Oliveira et al. / Journal of Natural Gas Science and Engineering 22 (2015) 313e320

Similarly, according as GTRc , GTEc , GCOMc , DEMP c and their

associated uncertainties, the following two expressions can be
written:

2:084GTRe þ 2GTE
e
þ 0:02GCOM þ 0:06DEMP
e e

c s2GTRe s2 s2 s2
GTE ¼ 2:17
GTEe GCOM e DEMPe
(28)
s2GTRe
þ s22 þ 0:00024
s2
þ 0:0019
s2 Fig. 2. GTR: experimental values and reconciled values.
GTEe GCOM e DEMP e

0 12
2:084
B C
2 B s2GTRe C
UGTE e ¼ B 2:17  U GTReC
@ 2 þ 22 þ 0:00024 þ 0:0019 A
s s GTRe
s2 s2 GTEe GCOMe DEMP e

0 12
2
B C
B s2GTEe C Fig. 3. GTE: experimental values and reconciled values.
þ B 2:17  UGTE eC
@ 2 þ 22 þ 0:00024 þ 0:0019 A
s s s2
GTRe
s2 GTEe GCOMe DEMP e

0 12
0:02
B C
B s2GCOMe C
þ B 2:17  UGCOM eC
@ 2 þ 22 þ 0:00024 þ 0:0019 A
s s s2
GTRe
s2 GTEe GCOMe DEMP e

0 12
0:06
B C
B C
s2DEMPe
þ B 2:17  UDEMP e C (29)
@ 2 þ 22 þ 0:00024 þ 0:0019 A
s s s2
GTRe
s 2
GTEe GCOMe DEMP e

Fig. 4. GCOM: experimental values and reconciled values.

200GTRe þ 192GTE þ 2GCOM

e
þ 6DEMP
e e

s2GTRe s2 s2 s2
GCOMc ¼ 2000
GTEe GCOMe DEMP e
(30)
s2GTRe
þ 18432
s2
þ s2 2 þ s218
GTEe GCOM e DEMP e

Table 2
Reconciled volumes of natural gas and respective associated uncertainties.

GTRc UGTRc GTEc UGTEc GCOM c UGCOMc DEMP c UDEMPc

Day 1 2655989 25201 2547996 24177 26542 252 79625 756

Day 2 2741998 26149 2630383 25085 27400 261 82199 784 Fig. 5. DEMP experimental values and reconciled values.
Day 3 3908172 37079 3749262 35571 39055 371 117164 1112
Day 4 3297510 32479 3163932 31164 32958 325 98873 974
Day 5 2810158 26742 2695794 25654 28081 267 84244 802
Day 6 2772377 26405 2659537 25331 27704 264 83111 792
Day 7 2629583 24969 2522707 23955 26278 250 78835 749
 E.C.d. Oliveira et al. / Journal of Natural Gas Science and Engineering 22 (2015) 313e320 319

Table 3 Table 4
Relative uncertainties, reconciled and minimized. New calculations for GNC based on reconciled values.

UGTRc UGTEc UGCOMc UDEMPc GTR (m3) GTE (m3) GCOM (m3) DEMP (m3) GNC (%)

0.95 0.95 0.95 0.95 Day 1 2674969 2560900 26542 80249 0.27
0.95 0.95 0.95 0.95 Day 2 2677264 2570173 27400 80318 ¡0.02
0.95 0.95 0.95 0.95 Day 3 3922694 3765786 39227 112888 0.12
0.98 0.98 0.98 0.98 Day 4 3297510 3163932 32958 98873 0.05
0.95 0.95 0.95 0.95 Day 5 2755777 2645546 28081 81999 0.01
0.95 0.95 0.95 0.95 Day 6 2711803 2603330 27704 81354 ¡0.02
0.95 0.95 0.95 0.95 Day 7 2676892 2569816 26212 78835 0.08

while those that result from operational measurement were

0 12
reduced by a factor of three.
200
B C As may be seen, there is no overlap, even partially, between the
2 B s2GTRe C
UGCOM c ¼ B  UGTRe C experimental values and the reconciled values (highlighted in black
@20000 þ 18432 þ 2 þ s218 A
2
sGTRe2
sGTEe s2GCOMe DEMP e circle), as demonstrated in Figs. 2e5. This behavior generates
doubts in the experimental results; gross errors are suggested,
0 12
beyond the expected random errors.
192
B C After substituting those values considered suspect with those
B s2 e C
þ B20000 18432 GTE  UGTEe C reconciled (shown in bold), the unaccounted for gas may then be
@ 2 þ 2 þ 2 þ s218 A
sGTRe sGTEe s2GCOMe DEMPe recalculated (data in Table 4). As can be observed, the application of
the data reconciliation approach was capable of bringing the values
0 12
of GNC within the criteria proposed as a minimum limit of 0.3%,
2
B C which confirms that the objectives proposed in making use of the
B s2 C
þ B20000 18432GCOM 2  UGCOMe C
e

@ 2 þ 2 þ 2 18 A data reconciliation technique to increase the confidence in mea-

sGTRe sGTEe sGCOMe
þ s2
DEMPe surements were fully achieved.
0 12
6 5. Conclusions
B C
B s2 e C
þ B20000 18432DEMP 2  UDEMP e C (31)
@ 2 þ 2 þ 2 þ s218 A As shown, detection of errors in a multivariable system is not an
sGTRe sGTEe sGCOMe DEMPe
easy task even when the source of the abnormal values is known. If
a gross error that may result from an extreme observation is not
evidently defensible, discard of unacceptable values must always
66:667GTRe þ 64GTE
e
þ 0:6667GCOM þ 2DEMP
e e

s2GTRe s2 s2 s2 be based on statistical analysis. The data reconciliation approach

GDEMP c ¼ 2217:78
GTEe GCOMe DEMP e
(32) used depicts the process to identify and classify similar clusters of
s2GTRe
þ 2048
s2
þ s0:222
2 þ s2 2
GTEe GCOM e DEMPe inputs and outputs as a strategy to eliminate gross errors inherently
imbedded in the results. The analysis undertaken considered a real
0 12 set of seven-day real data collected at the Urucu-Manaus pipeline
66:667
B C to illustrate that the reconciliation data approach was robust
2 B s2GTRe C
UDEMP c ¼ B  UGTRe C enough to identify and exclude gross errors that eventually has a
@2217:78 þ 2048 þ 0:222 þ s2 2 A
2
sGTRe 2
sGTEe s2GCOMe DEMP e
detrimental effect in measurement systems. The study also
confirmed that the data reconciliation results do not depend on
0 12 data only, but also on the process architecture.
64
B C Contrasting to the common perception, this paper highlights
B s2GTEe C
þ B2217:78 2048  UGTEe C that measurement redundancy should not be perceived as an
@ 2 þ 2 þ 0:222 þ s2 2 A
sGTRe sGTEe s2GCOMe DEMP e
additional cost, but rather as an attractive strategy for improving
system accuracy. Redundant data, even of lower accuracy, may
0 12 satisfactorily be used to enhance the overall accuracy by generating
0:6667
B C reconciled values, therefore confirming that the correct use of the
B s2GCOMe C
þ B2217:78 2048  UGCOMe C reconciliation methodology may conveniently replace the need for
@ 2 þ 2 þ 0:222 þ s2 2 A
sGTRe sGTEe s2GCOMe DEMP e
additional experimental work.
The application of the data reconciliation approach proved to be
0 12 a useful and powerful tool to evaluate the unaccounted for gas,
2
B C therefore capable of explaining deviations associated with net
B s2DEMPe C
þ B2217:78 2048 0:222  UDEMP e C (33) balances in a natural gas pipeline. Within best international prac-
@ 2 þ s2 þ s2 þ s2 2 A
s GTRe GTEe GCOMe DEMP e
tices, the confidence level associated with gas pipeline metrology
confirm that 0.3% seems to be a realistic criteria to express the
The data in Table 2 and the data in Figs. 2 to 5 describe the unaccounted for gas in natural gas pipelines.
volumes reconciled and the respective uncertainties associated
with these volumes (all expressed in m3). References
Based on the results of Table 3 and on the data from the figures
that follow, it is observed that the reconciled uncertainties are Alhaj-Dibo, M., et al., 2008. Data reconciliation: a robust approach using a
contaminated distribution. Control Eng. Pract. 16, 159e170.
minimized. Note that the uncertainties derived from the custody Amui, S., 2010. Oil and Gas Executives to: Exploration Areas, Drilling and Comple-
transfer measurements were reduced from by more than 50%, tion of Wells and Production of Hydrocarbons. Intercie^ncia, Rio de Janeiro.
320 E.C.d. Oliveira et al. / Journal of Natural Gas Science and Engineering 22 (2015) 313e320
Arpino, F., Dell’Isola, M., Ficco, G., Vigo, P., 2014. Unaccounted for gas in natural gas
transmission networks: prediction model and analysis of the solutions. J. Nat.
Gas. Sci. Eng. 17, 58e70.
Bagajewicz, M.J., Cabrera, A.E., 2003. Data reconciliation in gas pipeline systems.
Ind. Eng. Chem. Res. 42 (22), 5596e5606.
Guide to the Expression of Uncertainty in Measurement (GUM), 1995. ISO.
Mansour, M., Ellis, J.E., 2008. Methodology of on-line optimization applied to a
chemical reactor. Appl. Math. Model 32, 170e184.
Mene ndez, A., Biscarri, F., Gomez, A., 1998. Balance equations estimation with bad
measurements detection in a water supply net. Flow. Meas. Instrum. 9,
193e198.
Narasimhan, S., Jordache, C., 2000. Data Reconciliation and Gross Error Detection:
an Intelligent Use of Process Data. Gulf Publishing Company, Houston.
Oliveira, E.C., Aguiar, P.F., 2009. Data reconciliation in the natural gas industry:
analytical applications. Energy Fuel 23, 3658e3664.
€
Ozyurt, D.B., Pike, R.W., 2004. Theory and practice of simultaneous data reconcili-
ation and gross error detection for chemical processes. Comput. Chem. Eng. 28,
381e402.
Ramamurthi, Y., et al., 1993. Control-relevant dynamic data reconciliation and
parameter estimation. Comput. Chem. Eng. 17, 41e59.