You are on page 1of 9

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO.

1, FEBRUARY 2007

179

Variance-Based Energy Loss Computation


in Low Voltage Distribution Networks
Obrad M. Mikic

AbstractThis paper describes a method for calculating energy


losses in any system element assuming that the transmitted energy
across it is known. The method identifies two components of losses:
load variance and load mean. The load variance component can be
neglected for large loads, but it is significant for small loads, which
are dominant in low voltage networks. Needed parameters for loss
calculation are determined by two load duration curves (LDCs) per
load type. The formulae for loss calculation in transformers and
distribution lines of various phase systems are given, including the
formulae for energy losses and average current in the neutral conductor. A new algorithm for voltage calculation at network nodes is
proposed and shown to provide high accuracy of calculated losses.
The method enables loss calculation for compounded (nonhomogeneous) loads and gives the exact relationship between traditional
loss factor and the load factor.
Index TermsDistribution networks, energy losses, load measurement, load variance, neutral conductor current.

I. INTRODUCTION

HERE are three main reasons for developing reliable


energy loss calculation methods in a distribution system:
1) verifying measured losses and their allocation, particularly
after the introduction of a free market for electrical energy,
2) fast evaluation of various actions in distribution system
corrections,
management (reconfigurations, power factor
etc.), and 3) distribution systems development planning. Existing approaches [1][5] provide correct methods for large
loads (primary feeders and larger) but not for small loads and
low voltage networks [6], in which the losses are greatest. With
the exception of load flow programs (LFPs), whose application
is limited by the lack of load data, there is no method for
calculating losses and currents in neutral conductors.
The most accurate approximate method is described in [1]
and [2]. The load squared integral, as the basis for loss calculation, is determined in [2] by means of second statistical moments
of the representative daily load curves (DLCs) of all component
loads. It is assumed that all moments are constants, independent
of the load size. This approach is significantly improved in [1],
by applying a new method of voltage calculation. Voltages at
network nodes are computed for average loads by a load flow
program. According to test examples, use of these voltages, instead of constants, significantly increases the accuracy of calculated losses.
Manuscript received June 14, 2006; revised August 16, 2006. The idea for this
paper appeared during the authors work on distribution systems planning (see
IEEE TRANSACTIONS ON POWER SYSTEMS, Vol. 1, No. 1, Feb. 1986, pp. 3440).
Paper no. TPWRS-00360-2006.
The author is retired from the Institute Mihailo Pupin, Belgrade, Serbia
(e-mail: obradmikic@gmail.com).
Digital Object Identifier 10.1109/TPWRS.2006.889143

The percent loading method [3] indirectly differentiates


DLCs, by introducing the load profile (LP) correction coefficients. The loading and actual voltages of the primary lines are
estimated using a random sampling technique. The loads of
secondary lines are based either on average loads of line transformers or on average customer loads. This method enables fast
evaluation of losses in the whole distribution system. However,
it involves two key simplifications: 1) load profile correction
factor is constant for all conductors of the same voltage level,
and 2) all line section loadings are replaced by the average
loading, equal for all lines of the same type. It is not shown
how they influence the methods accuracy.
The most frequently used method is the loss factor
method, comprehensively described in [4]. Unfortunately, this
method is very unreliable, as it is based on vague or highly stochastic variables like peak loads, number of customers, and
factor. This statement is proven below, and similar conclusions
can be found in [1] and [3].
as a function of the
Although there are some graphs of
number of customers [4], the loss factor is mainly determined
using the following formula of Buller and Woodrow [7]:
(1)
where

is the load factor, and is recommended to be


.
method, I analyzed
To illustrate the difficulties with the
how loss factors are determined in the most frequently cited
EPRI reference for distribution system losses [4]. Among loss
factors for 56 transformers, (cluster 1, Appendix E), 14 have
, or
, and in
impossible values: in 13 cases,
, or
. Only 13 values of are within
one case,
the recommended limits. In [7], it was shown that has a stable
value of around 0.08 but for large systems and very large loads
(up to 15 000 MW). This is, however, irrelevant for a distribution
network, particularly for lower voltage levels.
calculation
The method described in [5] enables
without factor
but also uses peak loads, with the same
difficulties.
Very important research on neutral conductor currents and
losses in low voltage lines is presented in [8]. By detailed modeling, applying LFP, and using two-minute load records for 900
households, authors proved: 1) the high value of this current (up
to 70% of the phase currents) at a quite balanced connection
of consumers, 2) significant energy losses in the neutral conductor, justifying its cross-section of equal size as that for the
phase conductors, 3) nonlinear change of neutral currents with
load change, and 4) the vital importance of the actual load data
collection for correct presentation of load characteristics.

0885-8950/$25.00 2007 IEEE

180

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO. 1, FEBRUARY 2007

The method proposed in this paper is based on the fact that


load curve shapes for small and large loads are significantly
different, even when loads are homogeneous. The differences
are very pronounced in the range of up to 100 kW of the
single-phase peak load. Therefore, calculation of actual load
curve squared integral, as the basis for energy loss calculation,
cannot be accurate if based on one representative DLC or
corresponding second statistical moment (described in [2]).
Splitting this moment into two components, one of which is
constant and dependent on load expected value, while the other
is variable and dependent on load variance, enables accurate
calculation of the load curve squared integral for any load
size. Three different methods are suggested in this paper for
calculating these components, producing the same results if
the loads are homogeneous. Using all of them can serve to
test the homogeneity level of loads. The input data for these
calculations are two load curves per homogeneous load area.
The formula for calculating the load curve squared integral
is a second-order parabola with two constant coefficients, with
the energy transmitted through the network element as a variable. The only condition for this relation is the load (or load
mix) homogeneity. In the case of nonhomogeneous loads, it is
necessary to calculate additional parametersmixed statistical
moments.
Identification of the load variance influence on losses leads
to three additional results presented in this paper. Two of them
are formulae for energy losses and for average current in the
neutral conductor at symmetrically connected loads, due to the
stochastic nature of loads. The third result is the formula for
exact calculation of traditional loss factor. It is shown that the
loss factor depends not only on load factor but also on peak load,
which explains difficulties in defining constant .
In addition, a new algorithm for voltage calculation at network nodes and loss calculation is proposed. The algorithm is
based on the load duration curve (LDC) approximation by a
step function with the same sum of load squared and asymmetry
of phase loads, sufficient to produce corresponding currents in
the neutral conductor. The error in computing total losses in
four-wire, three-phase systems by this algorithm is less than
0.5%, for losses in the neutral conductor less than 2%, and for
the current in this conductor less than 1%. The test results of
this and two other approximate algorithms are compared with
the accurate ones, computed by LFP.
Finally, this paper emphasizes the problem of LDC recording.
Present load recorders are actually integrators, which record average loads within a recording interval, and are not recorders of
instantaneous loads. Even one-minute interval integrator may
hide significant changes of small loads, as was shown in [9]. The
losses in secondary lines, neutral conductors, and small distribution transformers calculated with the present 1-h or 15-min
load recordings may underestimate the true values by 30% or
more. Precise evaluation of load-dependent energy losses is not
possible without instantaneous load (kW) recordings.
II. PROPOSED METHOD
A. Determination of the Load Squared Integral
The proposed method relies on two assumptions. The first is
, or its components, can be classified into
that each load
some of the known load profiles. The load profile represents a
set of homogeneous loads, and I assume that the whole load area

can be divided into homogenous subareas. The second assumption is that all component loads for any given are uncorrelated
stochastic variables. In general, the assumption on stationary
linear independence
is correct, although it is argued in [10] that high stationary correlation
exists between the load profiles, particularly within the same
profile. To my knowledge, the only evident stationary dependence exists among symmetrical three-phase loads. Some synchronized loads (street lighting, automatically controlled loads,
etc.) do not increase loads covariance if they are randomly distributed. If they are concentrated, they can be treated as specific load profiles among which some do not have variance at
all (street lighting).
Although statistical properties of loads are a frequent research
subject, there is a lot of confusion on their independence and
statistical distribution. A load is a variable that depends on two
argumentsspace and timeand can be analyzed along each
one. Load analysis at fixed space (load area) is quite simple and
reduces to time series analysis, which cannot give answers on
the load stationary dependence and statistical distribution. On
the contrary, load analysis at a fixed time is very complicated,
as it is very difficult to create a large number of coincident comparable homogenous loads, necessary for statistical inference.
Such effort was made in [11] with the conclusion that peak loads
are independent. In order to get comparable loads, some authors
[12] normalize the loads dividing them by their peaks or averages, which drastically changes the load variance and the probability density function. Having in mind that load data represent
their hourly (rarely 15-min) averages, which also drastically reduces load variance, some of the existing arguments on load statistical properties are not correct.
For accurate evaluation of load-dependent energy losses in a
distribution system element (transformer, the section of a secondary line, etc.) in period , it is necessary to calculate the
integral
(2)
,
, and
are the voltage, active, and reactive
where
component of the load flow respectively, measured at the same
end of the elementinput or output. Here, I assume that the
voltages are constant in time (but different at nodes, as suggested
in [1]). With this simplification, (2) reduces to the calculation of
load squared integral. Here, I will demonstrate how to do it for
the active component; the calculation is the same for the reactive
one.
squared is
The expected value of

where
is the mean and
the variance of
. For
and
, resulting from and consumers
two loads
, respectively, and
, we can write

: VARIANCE-BASED ENERGY LOSS COMPUTATION


MIKIC

181

(3)
Components of each load need to be mutually independent but
and
(some componot necessarily resulting loads
may be a part
nents may be a part of both loads, or whole
).
of
Ratio can be determined through the corresponding energies. It is easy to prove that two homogeneous loads with equal
means at any time have equal expected energies in the same
period. For long periods, expected and actual energies can be
equated. Also, energies are available and the most reliable data
on loads.
and
denote the corresponding energies in period ,
If
i.e.

Fig. 1. Load squared integral as a function of energy. Triangles represent the


values of I for 11 residential loads measured during one week at 15-min intervals and crosses represent the sums of pairs of these loads.

(4)

and
than

, then it follows that


. For simplicity, let us define

is

times greater

and

TABLE I
), ORIGINAL (I ) AND FITTING VALUES (I ), THE
AVERAGE LOAD (P
FITTING ERROR IN %, AND AVERAGE LOSS FACTOR (L
) FOR 11 LOADS

(5)

and (3) is now


(6)
Solving (6) for

and

, we get
(7)

]. Fig. 1 shows the values of for 11 residential loads and


sums of pairs of these loads fitted by

According to (6), the equation for calculating any load


squared integral is
(8)
where
and
terms of energy, (8) is

is the corresponding energy. In

or

(9)

or
(10)
where
,
, and
.
and are constants for given homogeneous loads or LP.
and
The proof follows directly from (5) and (9). If
change times, so will
, and (9) remains unchanged.
Formula (9) is a second-order parabola with constant coeffiand
cients, which can be determined from two LDCs [

For this type of loads, we have


and
kW. Table I gives the average values for 11
loads and original and fitting data of . The difference between
the original and fitting data is less than 1.35%.
The term in parenthesis of (10) is a dimensionless value close
to 1, which some authors [3] call the LP correction coefficient,
or second statistical moment in per unit of average load [2].
This value, however, has the same meaning as the traditional
loss factor, which refers to the average instead of the peak load,
and it is reasonable to call it average loss factor. Let us denote
, and we will have
it by

(11)
where represents
for very large loads, as the second
(we denote it
addend in (11) can be neglected for large

182

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO. 1, FEBRUARY 2007

by

due to load variance. The expected value of the integral of that


, which means
current squared is
are equal to the correthat losses in neutral conductor
sponding component of losses in all three phase conductors

). That is an alternative way to determine


if LDC is available for a large homogeneous load. The
for 11 loads are given in Table I.
values of
It is important to note the relatively small value of for large
loads, compared to . To avoid errors in calculating , the two
and
should not
LDCs should be for low loads, i.e.,
have peaks greater than a few hundred and a few tenths of kW,
respectively. For higher accuracy of both coefficients and ,
the ratio of loads should be, according to the authors experience, greater than 3, since the position of the curve defined by
two points is not stable if the points are too close. Better accuracy can be achieved using more LDCs and applying weighting
coefficients at fitting in order to have approximately equal percent deviations of all points from the fitting curve. Otherwise,
should be much
the percent deviations for small values of
greater than for large ones, which especially influence the value
of . The best weighting is
i.e.
making all

(12)

(16)
Inserting three-phase variables in (16) and adding it to (15),
we get

(17)
If we compare expressions in parenthesis in (14), (15), and
(17), we see that average loss factor depends on the phase system
to which the loads are connected. The general equation for it is

close to one.
(18)

B. Variable Power Factor


If the power factor varies over a wide range, so that DLCs for
and
active and reactive loads notably differ, the coefficients
for reactive component should be also computed in the same
is constant
way as for the active one. That is not necessary if
). In that
or varies within a small range (say,
case, without a large error, it can be considered that coefficient
is the same for both active and reactive components, i.e.,
, and
, where average

, 3, or 6, for
, 3, or 3 0 (for single-phase,
where
three-phase, or three-phase system with neutral conductor, respectively). If (9) is divided by the square of the yearly peak
of the element and by
, the result is
load
the traditional loss factor

(19)
and
Average
or
is required for
data on reactive energy do not exist.

(13)
calculation, because

C. Calculation of Energy Losses


1) Line Losses: According to (10), the formula for losses
in a single phase line section is

(14)
where

is line resistance
,
is the power factor,
is the th average load (kW) in , is the actual
phase voltage (kV) at the same point on section where
is identified, and is length of the th line section (km).
To get the loss formula for three-phase lines with neutral conby line to
ductors, we replace 2 by 3 in (14), phase voltage
, and average phase load by
line voltage
three-phase average load

Neutral Losses

(15)

In the Appendix, it is shown that a current exists in the neutral


conductor, even for symmetrical allocation of connected loads,

where
.
As we see, formulae (19) and (1) are quite different, and there
is no constant parameter (like ) that can be used together with
load factor to determine the loss factor.
Equation (19) is in contradiction with existing practice of assigning a unique loss factor to each load. The difference between
loss factors for single-phase lines and three-phase lines with
neutral conductor is very large for small loads. Unfortunately,
load records available to the author are all for large loads, with
a peak of 172 kW for the smallest among them. To illustrate the
difference between the loss factors, ten simulated homogeneous
load curves for a year period in 15-min intervals were generated,
and
with peaks ranging from 18 to 1075 kW and
kW. Simulation data highly conform to (9), with deviations less than 0.2%. A MATLAB load simulation program
with explanations is given in the Appendix. Using the simulated
data, above three values for average and traditional loss factors
are given in Table II.
The difference is caused by the component of losses, which
depends on variance and can be neglected for large loads, but
for loads below a few hundred kW of the yearly peaks, as are
loads of distribution transformers and secondary lines, it can be
shows that this
significant. The sixth column in Table II
part of losses for loads below 150 kW may range from 9% up to
). Obviously,
100% or more (calculated by
the use of any method of loss calculation in low voltage distribution network in which load variance is neglected may reduce
calculated energy losses to half of the true value.

: VARIANCE-BASED ENERGY LOSS COMPUTATION


MIKIC

183

TABLE II
AVERAGE (P
) AND PEAK LOADS (P ), AVERAGE LOSS FACTORS (L
;L
VARIANCE (LB ), AND TRADITIONAL LOSS FACTORS (LS F ; LSF

2) Transformer Losses: The formula for load-dependent enin period is


ergy losses in three-phase transformer

;L
; LSF

), PERCENTAGE INCREASE IN LOSSES DUE TO LOAD


) FOR SIMULATED HOMOGENEOUS LOADS

Fortunately, this moment does not depend on the load size.


and
for compounded load
are
The constants

(20)
is the load loss at rated load (kW),
is the rated
where
is the actual line-to-line voltage (kV), and
voltage (kV),
is the transformer rated load (kVA).
For single phase transformer, the formula is quite similar
(21)
is not used in above equations, they should be transIf
formed as (17) into

(24)
Formulae for
and
of reactive load are quite similar, replacing by in (24). The power factors for different
LP usually differ, and loss computation for compounded loads
are used (except for
is much simpler if formulae without
calculation).
Equation (23) can be used to define a third way for calculating coefficient . If compounded load consists of only two
and , their sum
is also
independent homogenous loads
homogeneous, and we can write (omitting )

(22)

(25)

If the load of a network element is composed of two or more


different load profiles, to determine (9) for compounded
load, it is necessary to have, in addition to coefficients and
for each component load,
of second mixed mobetween all component loads. That follows from
ments
the square of the compounded load

This is true only if


. Therefore, the coefficient
is equal to the second mixed moment in p.u. of average homogeneous loads. If these values differ, two loads are not homocan be used for homogeneity
geneous, and the value of
for 11 loads from Table I
testing. The average value of
(for
is 1.0902, which is slightly different from
0.65%). For simulated data, this difference is negligible (
,
). Note that loads
and
, if dependent, cannot be used for this testing.
Now we have all the needed parameters for loss calculation,
whose accuracy depends on methods for voltage calculation.
Test results are given in Section IV.

D. Case of Nonhomogeneous Loads

III. CURRENT IN THE NEUTRAL CONDUCTOR


Corresponding approximation of integral

over period

is

(23)
where

In the Appendix, it is shown that the expected value of current


squared in neutral conductor of balanced three-phase system is
equal to the sum of the phase current variances. The current
alone represents the deviation of its expected value, which is
and in
zero, and is very variable both in absolute value
in some period
phase angle. However, its average value
can be determined as average absolute deviation, since we
know the average value of its variance in the same period. The

184

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO. 1, FEBRUARY 2007

relation between mean absolute deviation


and variance for
statistical set with nearly symmetrical distribution and sufficient
number of elements is

Considering that
is a vector sum of three-phase currents,
,
is specific, and the above
sequence
relation does not apply to it. It can be shown that
Fig. 2. Test radial network.

(26)

The coefficient
is determined for a sufficiently
large load of constant expected value in (
and
). For variable loads represented by LDC,
is higher and depends on , , and
. If
,
depends only on , with good approximation by
coefficient
(27)
, approximate
is 1.305 and accurate 1.308,
For
with difference less then 0.23%.
for loads with
is more
Determination of
complicated and is not considered here.
If (32) is divided by , the result is average value of in

and

or

(28)
The term in the last parenthesis depends on and is a constant for given , with dimension of the current. Let us denote
it by , and we will have

(29)
where
and
.
Formula (29) gives minimal average current in a neutral conductor in period , since the balanced connection of loads on
all phases has been assumed, and is entirely the result of the
stochastic nature of loads. If expected values of phase loads are
unbalanced, this current greatly increases, which is not considered here. For simulated homogeneous loads used in this paper,
, and let
,
, then
, and
. For
, we
will have
.

IV. NEW METHODS FOR VOLTAGE CALCULATION


AT NETWORK NODES
The method suggested in [1] calculates average voltages in
period at network nodes, resulting in much higher accuracy
of loss computation than using nominal voltage. However, for
small loads, this method has two shortcomings: equal absolute
.
phase voltages and simple approximation of DLC only by
If phase voltages are measured at the same intervals as loads
and computed averages for maximal (of the three), middle, and
minimal voltages for all intervals, significant differences among
,
,
them should be observed, e.g.,
, caused by continuing phase imbalance.
and
These differences are proportional to the current in the neutral
conductor. Since this current can be computed by (28) or (29)
at each node, it is possible to make an imbalance of phase loads
at nodes, corresponding to computed currents. One simple imbalance is
,
, and
, where
(node index omitted).
There are more ways for phase load determination depending
or
on LDC approximation: by horizontal line (
), by step function, with equal or unequal
duration of steps, with various definitions of step heights, etc.
With phase loads defined in this manner and applying LFP for
four-wire three-phase radial network, it is possible to compute
the total losses and the losses and average currents in a neutral
conductor with high accuracy in one or two (if step function is
used) LFP convergences.
The results of three approximate algorithms for radial network shown on Fig. 2, compared with the accurate one, are
given in Table III for six test cases. Algorithm denoted as NEW
is the one described above, with unbalanced phase loads and
LDC approximated by step function, with different durations of
steps and two convergences of LFP. Two other approximate algorithms are based on [1]. The one denoted by T&R (initials of
the authors) is a modification of the original denoted by T&R
ORG, where the variable second statistical moments calculated
by (18) are used instead of fixed ones. Accurate total losses, and
losses and average currents in a neutral conductor, are computed
inputs of phase loads are simuby LFP, for which
lated, in 8760 intervals (three phases, nine nodes, and three load
are used as inputs to
levels). The averages of these inputs
the approximate algorithms.
All algorithms are of a constant load type, since the input data
are based on the energy already consumed. The ratio of load
levels is 1: 5: 50, and two voltage drop maximums are about 5%
and 2.5%. The loads are of two profiles: , , , , and

: VARIANCE-BASED ENERGY LOSS COMPUTATION


MIKIC

TABLE III
RESULTS OF THREE APPROXIMATE ALGORITHMS COMPARED
WITH ACCURATE (LFP) ONE

185

TABLE IV
LOADS AND LINE RESISTANCES OF LEVEL 5

Fig. 3. Same load as a function of time recorded in 15-min (thin line) and 1-h
(bold line) intervals.

have
,
,
, and ,
, , and
have
,
,
. Second moments are
,
,
near 100 kW, the mixed moment is
computed for DLCs of
. Also,
, average loads of level 5 and
line resistances
in
, at load level 5 and voltage
drop of 5% are given in Table IV (reactance neglected); quoted
voltage drops are obtained changing line resistances by factors:
5, 2.5; 1, 0.5; 0.1, 0.05 for load levels 1, 5, and 50, respectively.
The error in computing total losses by algorithm NEW is less
than 0.4% in all test cases, and the error for losses and current in neutral conductor is less than 2.25% and 1.35%, respectively. For large loads (level 50), all algorithms have small errors, but for small loads (level 1), computed total losses by algorithm T&D ORG underestimate the true ones by nearly 30%.
Test results show two equally significant contributions of the
proposed method to loss calculation accuracy: through accurate
calculation of both, load squared integral (compare T&R and
T&R ORG) and voltages at network nodes (compare NEW and
T&R).

V. PROBLEM OF LOAD RECORDING


For load-dependent loss calculation, the availability of true
instantaneous load values is of key importance. However, loads
are recorded as 1-h (rarely 15-min) averages, hiding all load
changes in the recording interval, as was shown in [9]. This type
of recording significantly underestimates load variance, whose
almost disappears. Calculated losses in secondary
measure
lines, where the load coefficient of variation is the highest, can
underestimate the true losses by 20% to 30% using 1-h recordings. Even 1 min of integration is too long for small loads. Here,
we show the effect of applying 1-h average records instead of
the 15-min one. In Fig. 3, it is visible how DLC based on hourly
intervals cuts all peaks and troughs of the 15-min one. The difference of integrals of two DLC squared is 7.438%, with 1-h integral as a base. If coefficient for 11 residential loads had been
,
calculated by means of 1-h records, it should be
or only 49% of the value for 15-min records. For accurate loss
calculation, it is necessary to have instantaneous kW and kvar
load records. The length of recording interval is not critical if
the recording period is sufficiently long, as DLCs have similar

186

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 22, NO. 1, FEBRUARY 2007

All these values are functions of time. The relation between


and
is
phase load variance

From (5) and (10), we know that

and
Fig. 4. Components of the current in neutral conductor.

(32)

shapes and repeat many times. According to simulations, 1-h interval is not too long for 1-yr recording period.
B. MATLAB Simulation Program
VI. CONCLUSION
In addition to improved accuracy of total loss calculation,
the presented method introduces several new results. They are:
1) current and losses in the neutral conductor, 2) exact value
of traditional loss factor, and 3) load homogeneity testing. Besides, the proposed algorithm is very fast and can be used for
real-time network analyses. Input data on loads are just energies consumed. However, application of the method is conditioned on accurate determination of parameters and for all
component loads and corresponding mixed statistical moments.

% Probability vector
;
% Frequency distribution vector

;
% Number of electrical appliances with average installed
capacity Pa (kW)

APPENDIX

A. Current in the Neutral Conductor

As illustrated in Fig. 4, the orthogonal components of the


current in neutral conductor of three-phase system are

;
,

for

;
,

for
;

and

% Load curve-matrix

(30)
where , , and
Expected value of

are the absolute values of phase currents.


is

end;
% Peak load;
;
% Energy during

intervals;

where
,
, 1, 2, 3, and
is the variance of the
neutral, and
of the phase currents.
If the phase currents are mutually independent and their
means and variances are equal, we have

% Sum of load squared;


;
% All load curves;
;

(31)
As we can see,
is entirely a variance, since
balanced three-phase system.

in
end;

: VARIANCE-BASED ENERGY LOSS COMPUTATION


MIKIC

This simple load simulation with only one type of appliances


enables generation of perfectly homogenous loads of any size,
and vectors and , it
simply by changing . By changing
is possible to simulate any LP and any real load curve, with the
,
, and LDC (sorted
) but not
same , ,
for any , which is not important for model use in above testing.
ACKNOWLEDGMENT
The author would like to thank I. Smon of the Laboratory
for Power Systems at the University of Ljubljana who made
available load data recordings used in this paper.
REFERENCES
[1] R. Taleski and D. Rajicic, Energy summation method for energy loss
computation in radial distribution networks, IEEE Trans. Power Syst.,
vol. 11, no. 2, pp. 11041111, May 1996.
[2] A. L. Shenkman, Energy loss computation by using statistical techniques, IEEE Trans. Power Del., vol. 5, no. 1, pp. 254258, Jan. 1990.
[3] D. L. Flaten, Distribution system losses calculated by percent
loading, IEEE Trans. Power Syst., vol. 3, no. 3, pp. 12631269, Aug.
1988.
[4] EPRI EL-3261, Improved Methods for Distribution Loss Evaluation
vol. 1, Analytic and Evaluative Techniques, Nov. 1983.
[5] R. D. Schultz, G. L. Hickey, and A. D. Brunhart, Determination of
subsystem loss factors for loss allocation with limited data, IEEE
Trans. Power App. Syst., vol. PAS-101, no. 5, pp. 10531058, May
1982.

187

[6] J. A. K. Douglas and N. J. L. Randles, Report on Distribution System


Losses The Office of the Regulator-General, Victoria, Australia, Doc.
no. 10025 D008 (Australian Report), Jun. 2000, p. 28.
[7] M. W. Gustafson and J. S. Baylor, The equivalent hours loss factor
revisited, IEEE Trans. Power Syst., vol. 3, no. 4, pp. 15021508, Nov.
1988.
[8] N. Ozay, N. Guven, and A. Aksu, Accurate determination of neutral
conductor current distribution along low voltage feeders, in Proc. 7th
Mediterranean Electro-Technical Conf., 1994, pp. 948951.
[9] T. H. Ortmeyer and S. Krishnamurthi, Short term residential load data:
measurement and analyses, in Proc. 10th Int. Conf. Harmonics Quality
Power, 2002, vol. 2, pp. 430434.
[10] K. Livik, N. Feilberg, and J. A. Foosnaes, Estimation of annual coincident peak demand and load curves based on statistical analysis and
typical load data, in Proc. 12th Int. Conf. Electricity Distribution, May
1719, 1993, pp. 6.20.16.20.6.
[11] J. Nazarko, R. P. Broadwater, and N. I. Tawalbeh, Identification of
statistical properties of diversity and conversion factors from load research data, in Proc. 9th Mediterranean Electrotechnical Conf., May
1820, 1998, vol. 1, pp. 217220.
[12] A. Seppala, Statistical distribution of customer load profiles, in Proc.
Int. Conf. EMPD, Nov. 2123, 1995, vol. 2, pp. 696701.
, near Belgrade,
Obrad M. Mikic was born in Sopic
Serbia, in 1931. He received the B.Sc. degree in electrical engineering from the University of Belgrade.
Until 1964, he worked on automation in the Nuclear Research Institute in Vinca and then in the Operations Research Department of the M. Pupin Institute, Belgrade, until he retired in 1996 as the Head
of Science and Technology Policy Research Centre.

You might also like