Sie sind auf Seite 1von 7

694

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 18, NO. 3, JULY 2003

Distribution System Performance Evaluation


Accounting for Data Uncertainty
Jovan Nahman and Dragoslav Peric

AbstractThe effects of uncertain input data on the performance evaluation of a distribution system are analyzed. A criterion is introduced for assessing the grade of uncertainty of the results obtained in the calculation of maximum loads, voltage drops,
energy losses, and characteristic reliability indices of a network if
some input parameters are only guesses based on limited experience, measurements, and/or statistical data. Reasonable outputs
bounds are determined based upon the shape of the function measuring the uncertainty. High uncertainty of a result obtained indicates that a re-examination of relevant uncertain input data would
be recommendable for a more precise quantification. The method
proposed is applied to a real life example for illustration.
Index TermsDistribution systems, fuzzy mathematics, operating performances, uncertainty.

I. INTRODUCTION

OR a proper planning of construction and exploitation of


distribution networks, various network performances have
to be analyzed. The most important among them are the maximum currents to be carried by the distribution feeders and associated voltage drops, annual energy losses, and the reliability of
supplying consumers demands. Unfortunately, many of the inputs forming the basis for these studies are often assessed with
some uncertainty, for many reasons. The annual load duration
diagrams of consumers load demands can be only roughly predicted. The same is the case with the failure transition rates of
network elements and associated renewal duration which are
important for reliability evaluation [1]. As these input data substantially affect the results of network analysis, it is important to
have some idea how uncertain are the results obtained if some
of the inputs can be only roughly assessed. An adequate tool for
incorporating the uncertainties in distribution network studies
and for assessing the grade of acceptability of the results obtained can provide the fuzzy algebra. Some applications in reliability analysis of power systems using fuzzy arithmetic and
fuzzy logic have been suggested in the past [2][4]. Fuzzy technique was also successfully used in consumer demand prediction [5], [6]. This paper proposes a method for quantifying the
uncertainties of the input and output data in a distribution system
performance analysis. The method models the input and output
quantities in distribution network analysis as fuzzy variables.
The uncertainty grade for a fuzzy variable is measured by the
interval encompassing its most credible values. This interval is
obtained by weighting the possible intervals of the values of the
Manuscript received November 22, 1999.
J. Nahman is with the University of Belgrade, Belgrade 11000, Yugoslavia.
D. Peric is with the School of Electrical Engineering, Belgrade 11000, Yugoslavia.
Digital Object Identifier 10.1109/TPWRD.2003.813868

Fig. 1.

Characteristic function of a FN.

variable by the uncertainty levels characterizing these intervals.


Such a quantification of the uncertainties makes it possible to
judge the credibility of the results of network analysis based
upon uncertain inputs. If the uncertainty grade of a result obtained is too high, some of the most uncertain and most effective
inputs should be reconsidered for a more precise quantification.
II. MATHEMATICAL MODEL
A. Engineering Interpretation of Fuzzy Algebra
Consider a variable which values are not known with certainty. This variable may be modeled as a normalized unimodal
fuzzy number (FN) as depicted in Fig. 1 [7]. FN models of
guessed quantities are further on denoted by capital letters.
( cut) is introduced that may be
Parameter
interpreted as the level of uncertainty of the guess made at .
To each , an interval of possible values is attached with
and upper bound
. For increasing , these
lower bound
bounds become closer to one another tending to a single value
.
as approaches to 1. This value is the kernel of , denoted
is the highest of all as the presumption
The uncertainty of
that has exactly a specified value must be taken with the least
confidence. If is modeled by a triangular FN, then this FN is
.
completely defined by the triple
, may be determined
The uncertainty grade of , for
as

0885-8977/03$17.00 2003 IEEE

(1)

NAHMAN AND PERIC: DISTRIBUTION SYSTEM PERFORMANCE EVALUATION ACCOUNTING FOR DATA UNCERTAINTY

695

As can be seen, the integral in (1) gives the interval of variable


values obtained by weighting all possible intervals of these
values by associated uncertainty levels. This interval may be
considered as a most reasonable prediction of variable values
concerning its uncertainty. Interval bounds are given in relative
terms with the kernel of being the base value.
may
The variable lower bound relative declination of
be determined as
(2)
The corresponding declination of

upper bound equals


(3)

It is clear that
(4)
On the basis of the former definitions, the following bounds
for variable values may be taken as most reasonable for engineering decisions made under uncertainty
(5)

B. Calculation Flow
of inputs ,
. If
Presume that is a function
inputs are modeled as FNs, to encompass their uncertainty, then
is also a FN which may be formally expressed as
(6)
has to be conTo define , its characteristic function
structed. In order to determine this function, a series of values
is generated from the whole interval (0,1). For each , the lower
and upper bounds of are determined as

Fig. 2. Sample distribution network feeder.

III. NETWORK MODEL


A. Maximum Feeder Branch Currents

(7)
for
(8)
and
for all define
.
Bounds
equals the minimum of
As can be observed from (7),
if values are within intervals (8). These intervals
function
for for
are determined by the lower and upper bonds of
equals the maximum
which the calculation is performed.
obtained for the same intervals of values.
of function
and
is trivial if
is a monoThe calculation of
tonic increasing or decreasing function with regard to all arguments being within the intervals in (8). In the first case,
is obtained from
for
and
for
,
. In the latter case,
should be inserted to
and
to obtain
. As will be shown hereobtain
after, all distribution network quantities of interest are monotonic increasing functions of their arguments which makes the
analysis easy and straightforward.

Consider a branching feeder of a radial distribution network


presented in Fig. 2. It is supposed that the feeder consists of
branches characterized by their length and the consumer demand load at their receiving end. The branches can be separated
from each other by opening the disconnectors at their sending
ends that are not marked for simplicity.
To determine the maximum feeder branch currents and maximum voltage drops along the feeder, the peak rms magnitudes
of load demand currents are used as input variables. They are
presumed to be fuzzy variables quantified by triangular fuzzy
numbers. The following relationships correlate these currents
with currents flowing through feeder branches [8], [9] (here and
further on fuzzy variables are designated by capital letters)

(9)
,
are by 1 column vectors of real and
where
,
are
imaginary parts of branch currents while
corresponding column vectors of load demand peak currents.

696

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 18, NO. 3, JULY 2003

The elements of the

by

Boolean matrix

are

if branch supplies branch


otherwise.

(10)

.
It is implied that
Equations (9) and (10) simply state that the current flowing
through a feeder branch is equal to the sum of the load demand
currents supplied by this branch.
for the feeder in Fig. 2
For illustration, the first 5 rows of
are (see equation at the bottom of the page).
From (9), it follows that the maximum branch currents also
are fuzzy quantities. The rms of the current in branch equals
(11)
row vector built of the row

Branch k load duration diagram.

elements

magnitudes are dominantly determined by


values
affect the voltage phase shifts that are
while drops
very small usually. Therefore

Let us denote by , the diagonal by matrix of the lengths


of feeder branches. Then, the voltage drops at load points are

(18)

where
.
of

is a 1 by

Fig. 3.

B. Maximum Voltage Drops

(12)

which means, according to (16), that voltage drops are practically monotonic increasing functions of load point demand currents. This facilitates the calculation of their bounds for various
uncertainty levels .

(13)

C. Load Demand Duration Diagram and Energy Losses

where

It is assumed that the feeder branches have the same impedance


per unit length . From (12), it is clear that the elements of by
are fuzzy quantities as they depend
1 column vector
.
on
Bearing in mind that
(14)
with and being the real and imaginary parts of , from (12)
and (14), it follows:
(15)
where

The load duration diagram can be represented by several real


current load levels of given duration, spread in descending order.
The load levels are treated as uncertain data and quantified by
corresponding triangular FNs. In this application, three load
levels are presumed to be sufficiently representative: maximum,
medium, and minimum load level (Fig. 3). The load level duration and the power factor are presumed to be the same for all
consumer loads. Consumer loads have been considered to be of
the same type and synchronous which means that all consumers
have the characteristic load levels approximately at the same
time. However, the load levels and their uncertainty grades for
various consumers generally differ from consumer to consumer.
Based upon the assumptions made, the annual energy losses
equal

(16)
The rms values of voltage drops at feeder load points are

and

with
and

(17)
being elements of vectors
, respectively. As known, the voltage

(19)
It is clear from (19), that the energy losses are monotonic increasing functions of load demand current maximum, medium,
and minimum values represented as fuzzy variables. It makes it
bounds for various .
possible to easily determine

NAHMAN AND PERIC: DISTRIBUTION SYSTEM PERFORMANCE EVALUATION ACCOUNTING FOR DATA UNCERTAINTY

697

TABLE I
LOAD DEMAND CURRENTS

D. Feeder Reliability

TABLE II
BRANCH LENGTHS

If there is no back feed available, as presumed, each failure


of a branch, say , interrupts the supply to all branches and
associated loads fed by this branch during its repair. Hence, the
expected energy not supplied due to branch failures is
(20)
It is assumed that the feeder failure rate per unit length , sustained interruption duration , and the average real power deare fuzzy variables as all of these quanmands of customers
tities cannot be predicted with certainty.
Here, with reference to Fig. 3

TABLE III
NUMBER OF CUSTOMERS

(21)
being the rated network voltage and
year.
with
The total expected energy not supplied annually equals
(22)
The second term in (22) encompasses the energy not supplied
is the uncertain duration of this
during the fault location.
activity.
The system average interruption frequency index is defined
as the ratio of the total number of customer interruptions and
the total number of customers served
(23)

where is the number of customers served at load point .


The system average interruption duration index is defined as
the ratio of the sum of customer interruption durations and the
total number of customers

(24)

being by 1 column vector of numbers of customers


with
served at feeder load points.
,
, and
are functions of fuzzy variAs
ables, these indices are fuzzy variables too. We can observe
is a monotony increasing
from (20), (21), and (22) that
function of consumer currents, failure transition rate per unit
and
are profeeder length, and repair duration.

portional to the failure transition rate per unit feeder length.


is also a linearly increasing function of fault repair and
location duration. Thus, lower and upper bounds for all of these
three performance indices can be easily obtained for each uncertainty level from the lower and upper bounds of the associated
aforementioned input data for the same .
IV. APPLICATION EXAMPLE
A. Feeder Data
A real life example of a 10-kV overhead feeder is considered,
that is displayed in Fig. 2. The main feeder data are given in
Tables IIV. The data are acquired from experience and limited
measurement. Triangular fuzzy numbers quantifying the input
.
is
quantities are given in the general form
the guess used in the conventional analysis.
B. Calculation Procedure
The calculation procedure comprises the following steps.
and
for
1) Using (10) and (13), determine matrices
the feeder under consideration.

698

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 18, NO. 3, JULY 2003

Fig. 4. Characteristic functions of calculated performance indices.


TABLE IV
OTHER SAMPLE DATA

2) Determine lower bounds of all fuzzy input data for


from their membership functions.

3) By inserting the values from the previous step into (11)


and (18), determine the lower bounds of branch peak
currents and voltage drops for
.
4) Using (9) and data from step 2, determine the lower
,
, and
bounds of branch currents
for
and insert these values in (19) to calculate the
.
lower bound of annual energy losses for
,
, and
5) By inserting the lower bounds of
for
in (21), the lower bound of elements
for
are obtained.
of power demands

NAHMAN AND PERIC: DISTRIBUTION SYSTEM PERFORMANCE EVALUATION ACCOUNTING FOR DATA UNCERTAINTY

TABLE V
CALCULATION RESULTS

quantities is comparatively low with UG being less than 30%.


However, the annual energy losses, which depend on the load
duration diagrams of consumers, characterized by three load
levels for each load point, are more considerably affected by
the fuzziness of input data. The uncertainty of the
is
high if all input data are fuzzy. This particularly holds for the
, indicating that
might be much
upper bound of
greater than its kernel value which would be taken as relevant
in the conventional approach. For illustration, from (5) and the
corresponding data in Table V, we obtain the following bounds
if all inputs are uncertain
of
MWh

6) By inserting
values from the previous step and lower
in (20) and (22), lower
bounds of , , and for
is obtained for
.
bound of
, and for
7) By inserting lower bounds of ,
in (23) and (24), the corresponding lower bounds of
and
are calculated.
8) Determine upper bounds of all fuzzy input data for
from their membership functions.
9) Using the data from step 8 repeat steps 37 with upper
bounds to determine the upper bounds of all considered
.
performance indices for
, and repeat steps 29
10) Increase for, say
for this increased value of to determine lower and
upper bounds of all considered performance indices for
the new value of .
11) Repeat step 10 by gradually increasing values until
is reached. In such a way, the data for the
value
construction of membership functions of all considered
performance indices are obtained.
12) Using (1) and (5) and the data on membership functions obtained in previous steps, determine the uncertainty grades and reasonable bounds of considered performance indices.
C. Calculation Results
Table V quotes the results of the analysis. The effects of uncertain input data upon the uncertainty of the quantities characterizing the distribution system performances are examined. The
first column lists the analyzed outputs. The next three columns
give the calculated kernels (these coincide with the values obtainable using the conventional, crisp approach), and the associated certainty weighted percentage bounds of declination of
kernels. The last column indicates which input quantity is taken
to be uncertain and modeled as the corresponding fuzzy number
according to Tables IIV. Characteristic functions calculated for
the analyzed outputs are presented in Fig. 4.
The maximum load current that is flowing through branch 1
and the maximum voltage drop occurring at node 43 depend on
the load peak currents only. The uncertainty of both of these

699

MWh

(25)

is closely related to this of the failure


The fuzziness of
transition rate and may be taken as comparatively low. Howis more pronounced as it is addiever, the fuzziness of
tionally affected by fault location and repair times that we usually cannot precisely predict. It is worth noting that upper bound
are high in absolute terms and much greater
declinations
for the majority of the analyzed outputs. This means
than
that, in reality, the crisp valuesbased analysis might underestimate the examined performance indices and lead to too optimistic conclusions. The uncertainty of the calculated outputs is
tolerable if only one among the inputs is fuzzy to some extent.
value
The fuzziness of two or more inputs produces a high
that is greater than the sum of the declinations generated by the
inputs individually.
V. CONCLUSIONS
A method for assessing and incorporating the effects of uncertain data upon the performance analysis of distribution systems
is proposed. Criteria for determining the uncertainty of both the
input and output quantities are suggested as well as a reasonable way to determine the bounds of calculated performance indices reflecting the uncertainties of input data. The application
of the suggested approach to a real life example has shown that
is the performance index most affected by the unthe
certainties of input data and should not be taken as a relevant
index for any decision making unless the associated inputs are
well known. Second most input data sensitive parameters are
and annual energy losses but to a considerably lower
degree. Maximum feeder currents and voltage drops and the
index are moderately fuzzy.
REFERENCES
[1] R. E. Brown and J. R. Ochoa, Distribution system reliability: Default
data and model validation, IEEE Trans. Power Syst., vol. 13, pp.
704709, May 1998.
[2] V. Miranda, Fuzzy reliability analysis of power systems, in Proc. 12th
Power Syst. Comput. Conf., Dresden, Aug. 1923, 1996, pp. 558566.
[3] J. Backes, H. J. Koglin, and L. Klein, A flexible tool for planning transmission and distribution networks with special regard to uncertain reliability criteria, in Proc. 12th Power Syst. Comput. Conf., Dresden, Aug.
1923, 1996, pp. 567573.
[4] J. Nahman, Fuzzy logic based network reliability evaluation, Microelectron. Reliab., vol. 37, no. 8, pp. 11611164, 1997.
[5] M. Chow, J. Zhu, and H. Tram, Application of fuzzy multi-objective
decision making in spatial load forecasting, IEEE Trans. Power Syst.,
vol. 13, pp. 11851190, Aug. 1998.

700

[6] Sprinivasan, C. S. Chang, and A. C. Liew, Demand forecasting using


fuzzy neural computation, with special emphasis on weekend and public
holiday forecasting, IEEE Trans. Power Syst., vol. 10, pp. 18971903,
Nov. 1995.
[7] G. J. Klir and B. Yuan, Fuzzy Sets and Fuzzy Logic. Englewood Cliffs,
NJ: Prentice-Hall, 1995.
[8] M. Papadopoulos, N. D. Hatziargyriou, and M. E. Papadakis, Graphic
aided interactive analysis of radial distribution networks, IEEE Trans.
Power Delivery, vol. 2, pp. 12971302, Oct. 1987.
[9] J. Burke, Power Distribution Engineering. New York: Marcel Dekker,
1994, p. 96.

Jovan Nahman was born in Belgrade, Yugoslavia. He received the Dipl. Eng.
Grade and the TechD degree in electric power engineering from the Faculty of
Electrical Engineering at the University in Belgrade, Yugoslavia, in 1960 and
1969, respectively.
Currently, he is a Professor with the Power System Department at the University of Belgrade, where he has been since 1960.

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 18, NO. 3, JULY 2003

Dragoslav Peric was born in Raca, Yugoslavia, in 1958. He received the


Dipl.Eng., M.Sc., and PF.D. degrees in power engineering from the Faculty
of Electrical Engineering at the University in Belgrade, Yugoslavia, in 1983,
1989, and 1997, respectively.
Currently, he is a Professor with the School of Electrical Engineering, Belgrade, Yugoslavia. His main fields of interest are distribution systems operation
and planning and computer applications.