Electric Power Systems Research 97 (2013) 9399

Contents lists available at SciVerse ScienceDirect

Electric Power Systems Research

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

Probabilistic methodology for Technical and Non-Technical Losses estimation in

distribution system
Edison A.C. Aranha Neto , Jorge Coelho
LabPlan-UFSC, Electrical Systems Planning Research Laboratory, Federal University of Santa Catarina, Campus Universitario, CP 476, CEP 88040-900 Florianopolis, SC, Brazil

a r t i c l e

i n f o

Article history:
Received 27 September 2012
Received in revised form
12 December 2012
Accepted 14 December 2012
Available online 16 January 2013
Keywords:
Probabilistic Energy Balance
Technical Losses
Non-Technical Losses
Distribution systems
Measurement data
Case study

a b s t r a c t
This work contemplates the development and application of a probabilistic methodology for the Technical
and Non-Technical Losses estimation in a feeder in the presence of load variations of a large distribution
system. Due to such variations in load, the feeder losses are also characterized as random variables
which are related to statistical moments, such as mean and variance. Comparing the Measured Energy
consumption in the feeder with the Billed Energy by the utility plus the Technical Losses, the energy
balance is made and the Non-Technical Losses are estimated. Along a feeder, meters are installed allowing
its division into sub-networks, therefore, the Non-Technical Losses of each circuit are estimated with
greater precision. This methodology is useful in countries that Smart Grids are far from reality and the
resources are scarce. A real system was used as a case study for the Probabilistic Energy Balance developed.

1. Introduction
The electrical losses represent a signicant share in the cost
matrix of distribution systems and, therefore, have always had
major highlight in planning studies. For utilities, due to the management model which emphasizes the productivity and protability,
a reduction in the costs can be achieved by improving the system
performance. In this sense, the decrease of Technical Losses of the
system, as well as improving the efciency of system operation
are desired goals for utilities and are required by the Regulatory Agency. Therefore, Technical Losses should be reduced to its
optimal level, i.e. the level at which no additional investment is
economically justied.
Thus, this work aims to develop a new methodology for the
probabilistic estimation of Technical1 and Non-Technical2 Losses
in a feeder in the presence of load variations of a large distribution
system. The proposed model determines the sensitivity of voltage
and Technical Losses variations in each bus without the need of a

Corresponding author. Tel.: +55 48 3721 9731; fax: +55 48 3721 7538.
E-mail address: earanha@gmail.com (E.A.C. Aranha Neto).
1
Technical Losses occur naturally and consist mainly of power dissipation in electricity system components such as transmission and distribution lines, transformers,
and measurement systems.
2
Non-Technical Losses are caused by actions external to the power system and
consist primarily of electricity theft, non-payment by customers, and errors in
accounting and record-keeping.
0378-7796/$ see front matter 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.epsr.2012.12.008

2012 Elsevier B.V. All rights reserved.

new power ow solution. After determining the actual amount of

Technical Losses from periodic measurements, the Non-Technical
Losses can be segregated of the total system losses. Since both
the demand and the Technical Losses are being modeled as random variables, the separation of the Non-Technical Losses results
in a relation of probability density functions involving convolution
operations between these functions.
In Section 2, a brief state-of-the-art review on Losses Calculation
is presented. Section 3 details the developed probabilistic methodology that is applied, in Section 4, in the study case of a real system.
Lastly, Section 5 presents some conclusions.
2. Losses Calculation
Regarding the methods for determining losses in transmission
and distribution systems, there is a wide variation in the processes
adopted. In transmission systems, losses are estimated mainly by
power ow studies or energetic balance. In distribution systems,
the vast majority of utilities use procedures such as network management, power ow, statistical processes, and geometric models,
among others [1].
According to [1,2], the choice between a more developed and
a simplied methodology depends both on the available data and
the objective. The more elaborate methods (network management
and load ow, for example) present results that should be closer
to reality, and can even be used for individual and localized analysis, but they require an extensive database and constantly updated

94

E.A.C. Aranha Neto, J. Coelho / Electric Power Systems Research 97 (2013) 9399

records. The simplied methods (statistical processes, geometric

model, etc.) require a reduced volume of data and provide the
estimated loss quickly; however tend to produce satisfactory
results only when applied in a general form.
Several simplied models of loads and respective modes of loss
calculation by distributed load and geometric areas modeling (rectangular, triangular, and trapezoidal) with constant load density
are developed in [3]. Such techniques are useful for rapid calculations, obtaining estimated values. The methodology used by several
Brazilian utilities consists in calculating the Technical Losses by
segment in the distribution system within a policy of periodical
calculations (monthly) with the use of a software, as stated in [4].
In any system, making the difference between Total Loss and
Technical Loss (calculated/estimated) returns an estimation of the
Non-Technical Losses [1,2,4].
3. Probabilistic methodology
In this work, the development of a new analytical model of the
Technical Losses variation versus load change is proposed. This
model estimates adequately the sensitivity of the voltages and their
respective angles to the variation of the load including the electrical
losses, and also includes a new analytical formulation sensitivity to
losses with the variation of load. Thus, through the variation of the
loads, the variation in the Technical Losses can also be estimated,
which is used in the Probabilistic Energy Balance for calculating the
Non-Technical Losses.
3.1. Installation of meters
Once dened the feeder to be studied, strategic points for the
installation of meters are chosen. In a general way, being M the
number of meters installed, the network can be divided into M subnetworks. Each sub-network has as a start point one of the meters,
being the meter downstream, modeled as a load.

new sensitivity matrix with the inclusion of the loss (named MSit).
Therefore, the next step is to establish the desired accuracy for the
update of the losses of MSit. The accuracy inuences considerably
in the computational time that the algorithm takes to be executed.
The iterative process consists of n backward iterations, through
the end of the feeder to the beginning, in order to update Paci /Pj
and Qaci /Qj , which in the sensitivity matrix without losses were
binary matrices equal to the Reachability Matrix [9], and Qaci /Pj
and Paci /Qj that previously were null matrices. The accumulative
active (Pac) and reactive (Qac) powers are given by:
Paci = Pi + Paci+1 + LPaci+1 and Qaci = Qi + Qaci+1 + LQaci+1 (1)
Differentiating these expressions in relation to the active power
(P) results in:
(Pi + Paci+1 + LPaci+1 )
Paci+1
LPaci+1
Pi
Paci
=
=
+
+
Pj
Pj
Pj
Pj
Pj
(Qi + Qaci+1 + LQaci+1 )
Qaci+1
LQaci+1
Qaci
Qi
=
=
+
+
Pj
Pj
Pj
Pj
Pj
(3)
where the active (LPac) and reactive (LQac) losses are given by:
LPaci =

Ri (Paci2 + Qaci2 )

and LQaci =

Vi2

Vi (Paci2 + Qaci2 )

LQaci+1
= Xi
Pj

 
Paci

Vi3

Vi (Paci2 + Qaci2 )

3.3. Sensitivity matrix of voltages in relation to the load variation

The objective is to calculate the voltage variation sensitivity in
each distribution network bus (node) when there is a variation in
its demand, without requiring a new power ow solution.
The analysis of the voltage sensitivity in relation to the load
variation is based on the DistFlow method and was developed by
[8]. In order to improve the accuracy of results and reduce the
computational time, some modications and optimizations were
introduced, such as the use of graph theory [9] and inclusion of
losses in the calculation.
The voltage sensitivity calculation with the load variation without losses is used as an initial step for the iterative calculation of a

Xi (Paci2 + Qaci2 )
Vi2

(4)

  

so:
LPaci+1
Paci
Qaci
2
= Ri
Paci
+ Qaci
Pj
Pj
Pj
V3

3.2. Network reduction and sorting routine

In order to apply the methodology in larger distribution networks, algorithms to reduce the number of electrical points (nodes)
were developed. It consists of grouping the points where there is no
load, which serves only to denote the spatial layout of the feeder
or consist in some equipment with no inuence in this analysis,
such as switches or fuses. The lengths of the grouped sections are
combined to maintain the real values of resistance and reactance
of the total section. After the reduction, a reordering of the points is
made in order to optimize the performance of back/forward sweep
methods as the DistFlow method [57] which was used in this
work.
With the use of these network reduction and sorting routines a
high gain in computational time and processing is obtained without
altering the results.

(2)

Vi
Pj

(5)

Paci
Qaci
+ Qaci
Pj
Pj
Vi
Pj

(6)

Similarly, the accumulated active and reactive powers derivatives in relation to the reactive power (Q) can be calculated. The
process continues until there is convergence of the voltage derivatives (Vi /Pj and Vi /Qj ) to the accuracy set.
3.4. Loss sensitivity in relation to the load variation
In order to determine a relationship that denotes the sensitivity
of the total Technical Loss in the feeder to the load variation, the
total Technical Loss, LS, is given by:
LS = LP + j LQ =

  E E 2 
2
i
zi  i1
=
yi ||Ei1 Ei ||


z

(7)

where zi is the impedance between the buses i 1 and i, and yi

is the admittance between the buses i 1 and i. Separating the
above expression in their real and imaginary parts, since yi = gi + j bi
results in:
LP =

gi ||Ei1 Ei ||

and LQ =

bi ||Ei1 Ei ||

(8)

where gi is the conductance between the buses i 1 and i, and bi is

the susceptance between the buses i 1 and i. Now, notice that the
square of the voltage drop modulus in the lines is given by:
2
+ Vi2 2Vi1 Vi cos i
||Ei1 Ei ||2 = Vi1

(9)

Differentiating the above expression in relation to Pm and Qm :

E.A.C. Aranha Neto, J. Coelho / Electric Power Systems Research 97 (2013) 9399

||Ei1 Ei ||2
V
Vi
= 2Vi1 i1 + 2Vi
2
Pm
Pm
Pm


Vi

i
Pm

cos i Vi1 Vi sen i

V
||Ei1 Ei ||2
Vi
= 2Vi1 i1 +2Vi
2
Qm
Qm
Qm

Vi1
Vi
+ Vi1
Pm
Pm

(10)

(11)

Therefore, once the voltages and angles sensitivities are

obtained, the Technical Losses sensitivity in relation to the variation
of the load is calculated as follows.

 ||E E ||2
 ||E E ||2
LP
LP
i1
i
i1
i
=
gi
and
=
gi
Pm
Pm
Qm
Qm
(12)

 ||E E ||2
 ||E E ||2
LQ
LQ
i1
i
i1
i
=
gi
and
=
gi
(13)
Pm
Qm
Pm
Qm
It can be observed that the losses variation in the above equations refers to the total Technical Losses in the feeder. The rst
vector element refers to the total losses sensitivity in relation to
variation in the rst load, the second element refers to the total
losses sensitivity in relation to the variation in the second load and
so on.
3.5. Technical Losses variation versus load variation model
Given the power ow calculation performed for a given average
load prole, the variation of the active Technical Losses in relation
to the change in the load may be estimated by:
LP

  LP
m

LP
Pm +
Qm
Pm
Qm

(14)

Similarly, the active Technical Loss variance can be estimated

by:
2

2
= E{(LP) } E
LP




LP
LP
Pm +
Qm
Pm
Qm

2

(15)

Furthermore, assuming a constant power factor FPm (during the

period of analysis) on each node m of the network results in:

2
E
LP





m =

LP
LP
Pm + m
Pm
Pm
Pm

2

2
1 FPm

(16)

2
FPm

Thus, for a single increment Pm P, m = 1, . . . , n:

2
E
LP




  LP
m

Pm

LP
LP
+ m
Pm
Pm

LP
+ m
Pm

 2
(P)

 2
2

E{(P) }

  LP
m

V
Vi
Vi i1 + Vi1
Qm
Qm
i
Qm


2
LP



cos i Vi1 Vi sen i

(17)

Pm

LP
+ m
Pm

95

 2

P2

P2


  LP
m

Pm

2
(1 + m )
(18)

The above expression relates the active Technical Loss variance

of the feeder in relation to the variance of the loads. Analogous
relations for reactive Technical Losses can be obtained.
3.6. Probabilistic Energy Balance
The measured values in the feeder are compared with the billed
values by the utility. The comparison of these values characterizes the so-called Energy Balance, resulting in total losses. As both
variables have mean and standard deviation, the result will also
have mean and standard deviation, characterizing a Probabilistic
Balance.
From the determination of the mean () and standard deviation
() of Technical Losses in the presence of uncertainties (variations)
in the loads, the Non-Technical Losses can also be estimated from
the difference between the Total and the Technical Losses.
3.6.1. Transformation of Daily Measured Power and Technical
Loss in energy
From measurements performed in the feeder, the demand factor (DF) and power factor (PF) in each of the measured intervals can
be calculated. By multiplying these values by the installed power
of each bus of the feeder (in this case, a transformer), a matrix
with n rows (number of buses) and m columns (number of daily
measurements) for each day of analysis is obtained.
Instead of performing a power ow for each of the m daily points,
only one power ow with average values is performed, and then,
the methodology based on the sensitivity matrices which determines the value of the Technical Losses standard deviation without
the need to perform new power ow analyses is applied. From the
power data (hourly average) and Technical Losses (in power) the
conversion to energy is made. The Daily Measured Energy is the
sum of the 24 hourly averages values of the power.
It is known that the variance of a variable multiplied by a constant is equal to the multiplication of the square of the constant
by the variable variance. Thus, the Technical Loss in energy is the
average Technical Loss multiplied by 24 h; the same is true for the
standard deviation.
3.6.2. Daily Billed Energy and the daily seasonality representation
At rst, it should be noted that there are always different dates
of consumption readings, which can result in signicant differences
when considering a period (e.g. 1 month) for analysis. To minimize
these uncertainties, from the monthly consumption data and the
reading date performed for each consumer, a daily average is made,
obtaining, for each consumer, a vector with daily average.
The daily average Billed Energy is calculated according to the
following steps for each transformer: (1) First, the day on which
the reading was performed for the consumer is identied. (2) Then,
the measurement period is calculated (day of reading minus billed
days the reading day is not accounted for that month, only for
the following month). (3) The amount of energy consumed per day
(total billed in the month divided by the billed days) is calculated.
(4) The lowest and highest day of the total period of the analysis is
found. (5) A spreadsheet where the lines represent the days and the
columns represent the consumers is assembled. (6) Finally, there
is a column with the sum of all columns representing the value
consumed per day for the transformer. The process is made for all
transformers of the feeder.

96

E.A.C. Aranha Neto, J. Coelho / Electric Power Systems Research 97 (2013) 9399

Fig. 1. (a) Curves L and S do not cross; (b) Curves L and S do cross, forming an intersection area.

Established the daily average consumption of each consumer, a

matrix is built for each transformer grouping the consumers connected to the same distribution transformer and obtaining thereby
a daily amount of energy for each transformer for the entire period.
Knowing where in the power grid the transformer is connected,
the total amount of consumption for each sub-network can be
obtained. Using the daily average consumption, a more accurate
comparison with the measurement data can be made, thereby mitigating the problem of differences between the reading dates of
consumers.
To better represent the seasonality of the week days, a transformation is done in the billing curve to represent the load curve
variability of the feeder, maintaining the same average. This transformation is done by means of a standardization in relation to the
Measured Energy data.
3.6.3. Daily Total Losses
With the measured values of power, transformed into energy,
together with data from Billed Energy, the Total Losses of the feeder
can be calculated. This calculation is done daily for the period under
analysis, which is usually of one month.
3.6.4. Probability of Non-Technical Losses occurrence
Knowing that the Non-Technical Losses can be obtained by the
balance between Total and Technical Losses and that both values
have mean and standard deviation that have been properly estimated, and assuming that these random variables follow a normal
distribution, the subtraction of these two normal curves can be
performed by means of the convolution process of these curves
[10].
By doing an analogy with the model of convolution of the
loadstrength interference [1012], there are two possible cases,
as shown in Fig. 1a and b. In the graphs, L (load) is the Technical
Losses and S (strength) the Total Losses.
In the rst situation, shown in Fig. 1a, the two curves have no
common area, indicating that there are big differences in the energy
loss balance and that it is highly likely to occur Non-Technical
Losses. In the second situation, shown in Fig. 1b, the two curves
intersect, forming an intersection area. The larger the area of intersection, the more similar the two curves are and the smaller the
difference in energy balance is, indicating that it is unlikely to occur
Non-Technical Losses.
By means of the convolution expressions of the two curves, the
area of intersection between them can be calculated, as:


Pf =

The probability of occurrence of Non-Technical Loss (PS ) is given

by:
PS = (PCom ) = 1 Pf

where (z) is the cumulative distribution function of a standard

normal distribution and is the reliability index or safety index
given by:

(z) =

1
S L
2
eu /2 du and = 
2
S2 + L2

4. Results
A real feeder from southern Brazil [13] was used to test and
validate the developed probabilistic methodology. This feeder has
mixed load characteristics: commercial, residential, small industries and a small rural region. The feeder serves mainly the city
downtown and is approximately 10 km long. A total of 4129 consumers is connected with an installed load of about 11 MVA. This

(19)

where FS is the cumulative distribution function of S (Total Losses)

and fL is the probability density function of L (Technical Losses).

(21)

3.6.5. Estimation of Non-Technical Loss for the feeder and for the
sub-networks
Through the Probabilistic Balance of the Losses, subtracting the
Technical Losses from the Total Losses, the Non-Technical Losses for
the feeder are obtained. When more than one meter is installed on
the feeder, it can be separated into sub-networks. Thus, each subnetwork will operate like an independent feeder. By applying the
developed methodology in each of these sub-networks, the probability of occurrence and Non-Technical Losses for each of them are
estimated.

FS (l)fL (l)dl

(20)

Fig. 2. Feeder and location of the meters.

E.A.C. Aranha Neto, J. Coelho / Electric Power Systems Research 97 (2013) 9399

97

Fig. 3. Measured Power (in W) on the SS (a) and in the intermediate point (b).

Fig. 4. Daily Measured Energy (in kWh) on the SS (a) and in the intermediate point (b).

feeder was initially composed of 605 buses and after the reduction
routine, it remained with 368 buses.
In this feeder, two meters were installed for a period of three
months: one in the substation (SS) and the other at an intermediate
point of the feeder, as shown in Fig. 2.
4.1. Feeder
Measurements were made from 10 to 10 min per day, obtaining
a total of 144 daily measurement points. Thus, a matrix with 368
lines, representing the number of system buses, and 144 columns
for each day of analysis is obtained. The results of one month are
presented.
In Fig. 3, the Measured Powers, discretized in hours, in the two
meters can be viewed. The month of analysis has a total of 720 h
(30 days).
The conversion from power (hourly average) to energy is made,
as shown in Fig. 4.
In Fig. 5, the Daily Billed Energy for the feeder can be visualized.
Through the execution of the power ow, with the daily average
for the month of analysis, the average value of the Technical Loss
is estimated and by using the developed analytical methodology,
the standard deviation of these losses is calculated. By making the
difference between the Measured Energy and Billed Energy, the values of Total Loss are evaluated. Non-Technical Loss is the difference

Fig. 5. Daily Billed Energy (in kWh) of the feeder.

Table 1
Average and standard deviation of feeders variables.

Daily Measured Energy

Daily Billed Energy
Daily Total Losses
Daily Technical Losses
Daily Non-Technical Losses

Average (kWh)

St. dev. (kWh)

52,923.39
46,050.24
6873.15
917.90
5955.26

7636.48
6644.73
991.75
395.27
1067.61

98

E.A.C. Aranha Neto, J. Coelho / Electric Power Systems Research 97 (2013) 9399
Table 2
Average and standard deviation of Sub-network IIs variables.

Daily Measured Energy

Daily Billed Energy
Daily Total Losses
Daily Technical Losses
Daily Non-Technical Losses

Average (kWh)

St. dev. (kWh)

44,809.07
38,808.68
6000.39
209.16
5791.23

6264.00
5425.19
838.81
92.92
843.94

Fig. 6. Total Loss and Technical Loss curves of the feeder.

between Total and Technical Loss. In Table 1 the values calculated

for the feeder can be seen.
In Fig. 6, the Total and Technical Loss curves can be observed.
Visually, the curves do not cross, indicating a high probability of
Non-Technical Losses.
The value of the reliability index () calculated was 5.5781
resulting in a value of the cumulative probability function of a
standard normal distribution, (), of 0.99999, meaning that the
probability of Non-Technical Losses is practically 100%.
4.2. Division of the feeder into sub-networks
As two meters were installed in the feeder, it could be divided
into two sub-networks, where the section from the SS to the central
meter was called Sub-network I and the section from the central
meter to the end of the feeder was called Sub-network II. To facilitate the calculation and the understanding of the methodology, the
results of Sub-network II are rst presented.
4.2.1. Sub-network II
Its measurement data were shown in Fig. 3, and the conversion
to Measured Energy in Fig. 4. With the information of which clients
are connected to the transformers that belong to Sub-network II,
the Daily Billed Energy for that section of the feeder can be calculated. In Fig. 7, the values of Daily Billed Energy in Sub-network II
are shown.

Fig. 8. Total Loss and Technical Loss curves of Sub-network II.

Making use of the power ow tool with the modied network

to include only the section downstream of the central meter along
with their measurement data, the average Technical Loss is evaluated, as shown in Table 2.
In the graph of Fig. 8, the curves of Total and Technical Loss
can be viewed. Again, it appears that the curves do not intersect,
strongly suggesting the presence of Non-Technical Losses.
The calculated values for and (), were 6.8621 and 0.99999,
respectively. An almost 100% probability of occurrence of NonTechnical Losses is also veried.
4.2.2. Sub-network I
Its Measured Energy and Energy Billed data are the difference
between the values considered for the feeder as a whole and Subnetwork II. These values are represented in Fig. 9.
For calculating the Technical Loss, the power ow tool was used
considering only the network section of the substation until the
central meter, which was modeled as a load using the values of its
own measurement, as shown in Table 3.
In Fig. 10, the curves of Total and Technical Losses for Subnetwork I can be observed. Unlike the previous situations, the
curves intersect and have a large common area, indicating a lower
probability of occurrence of Non-Technical Losses.
The Sub-network I has a of 0.5316 and a () of 0.70250. As
noted in the graphics, there is a lower probability of occurrence of
Non-Technical Losses, being of 70.25%. However, this probability
Table 3
Average and standard deviation of Sub-network Is variables.

Fig. 7. Daily Billed Energy (in kWh) of Sub-network II.

Daily Measured Energy

Daily Billed Energy
Daily Total Losses
Daily Technical Losses
Daily Non-Technical Losses

Average (kWh)

St. dev. (kWh)

8114.33
7241.56
872.77
696.93
175.83

1406.80
1248.67
158.27
290.43
330.75

E.A.C. Aranha Neto, J. Coelho / Electric Power Systems Research 97 (2013) 9399

99

Fig. 9. Daily Measured Energy (a) and Daily Billed Energy (b) (in kWh) of Sub-network I.

variations due to the uncertainties inherent in the process, in large

distribution systems.
Dividing the network into sub-networks showed that, as
expected, a higher number of meters in the feeder can point out,
with greater precision, the places where the majority of NonTechnical Losses are. Thus, from a more supervised measurement
system (for example, by placing 4+ meters) the Non-Technical
Losses and its probable location can be evaluated, without making
exhaustive measurements across the feeder.
This methodology is useful in countries that Smart Grids are far
from reality and the resources are scarce, so inspection teams for
fraud combat can work with greater efciency.
References

Fig. 10. Total Loss and Technical Loss curves of Sub-network I.

already indicates the need for an anti-fraud enforcement on this

section.
4.3. Considerations about the analyzed feeder
From the obtained results, a strong presence of Non-Technical
Losses in the feeder is veried. Comparing the values of
Non-Technical Losses of Sub-network II with the values obtained
for the feeder, it appears that almost all of the Non-Technical Losses
are in the section downstream of the central meter until the end of
the feeder.
5. Conclusions
Knowing that the electrical losses represent a signicant share
in the cost matrix of distribution systems, it is necessary for companies to know their share of Technical and Non-Technical Losses
and verify exactly where and why they occur.
Thus, this paper presented the development of a probabilistic
methodology for the determination and segregation of Technical
and Non-Technical Losses in distribution systems. One of the goals
was to determine the Technical Losses in the presence of load

[1] M.E. Oliveira, A. Padilha-Feltrin, S.A. Oliveira, D.B. Camargo, F.J. Candian, M.A.
Pereira, Calculation methodology for technical losses in distribution systems,
in: CIDEL2006 Congreso Internacional de Distribucin Elctrica, Buenos Aires,
Argentina, 2006 (in Portuguese).
[2] Brazilian Association of Electricity Utilities (ABRADEE), Method for the determination, analyzes and optimization of technical losses in distribution systems,
Technical Document ABRADEE-19.34, Rio de Janeiro, 1996 (in Portuguese).
[3] W.H. Kersting, Distribution System Modeling and Analysis The Electric Power
Engineering Series, 2nd ed., CRC Press, Boca Raton, 2007, 421 p.
[4] A. Mffe, C.C.B. Oliveira, Technical loss calculation by distribution system segment with corrections from measurements, in: CIRED 2009 20th International
Conference and Exhibition on Electricity Distribution, Prague, Czech Republic,
2009.
[5] R.P. Broadwater, A. Chandrasekaran, C.T. Huddleston, A.H. Khan, Power ow
analysis of unbalanced multiphase radial distribution systems, Electric Power
Systems Research 14 (1) (1988) 2333.
[6] M.E. Baran, F.F. Wu, Network reconguration in distribution systems for loss
reduction and load balancing, IEEE Transactions on Power Delivery 4 (2) (1989)
14011407.
[7] R.G. Cespedes, New method for the analysis of distribution networks, IEEE
Transactions on Power Delivery 5 (1) (1990) 391396.
[8] A.B.K. Sambaqui, Voltage prole improvement methodology in distribution
systems, Thesis (Doctors Degree), Federal University of Santa Catarina, Florianpolis, 2005, 248 p. (in Portuguese).
[9] N. Christodes, Graph Theory: An Algorithmic Approach, Academic Press, London, 1975.
[10] G.J. Anders, Probability Concepts in Electric Power Systems, Wiley, New York,
1990, 682 p.
[11] M.L. Shooman, Probabilistic Reliability: An Engineering Approach, McGrawHill, New York, 1968, 524 p.
[12] D.C.S. Cordovs, Load-stress interference applied in umbilical cables, Dissertation (Masters Degree), State University of So Paulo, So Paulo, 2008 (in
Portuguese).
[13] E.A.C. Aranha Neto, D. Issicaba, J. Coelho, A.L. Bettiol, A. Carniato, R. Marcelino,
S.M. Barcelos, Technical and non-technical losses calculation in distribution
systems, in: XIII ERIAC Encuentro Regional IberoAmericano del CIGR, Puerto
Iguazu, Argentina, 2009 (in Portuguese).