IEEE Power Engineering Society

Transmission and Distribution

Committee
3152 IEEE Transactions on Power Apparatus and Systems, Vol. PAS-104, No. 11, November 1985
DISRlBUrIOKIA VARENTION OF DISTRIBUTIt(N
SYSTEM RELIABILITY IDICES
R. Billinton, Fellow IEE
Power Systems Research Group
University of Saskatchewan
Saskatoon, Saskatchewan
Canada
ABSTRACT
Reliability analyses are mainly concerned with the
mean values of reliability measures, however, the
distributional variation about the mean is becoming
increasingly important as a supplemental measure.
This paper presents the results of a series of
simulations studying the distributional variation
associated with indices for simple distribution
systems. Distributions for the Load Point Failure
Rate, Load Point Outage Duration, Load Point Annual
Interruption Time, SAIDI, SAIFI, and CAIDI are
presented and discussed.
IT1DUCTION
Conventiona1 reliability analyses are normally
only concerned with the expected or average value of
the particular measure of reliability. Little
consideration has been given in the past to the
variation of that measure about its mean. For
example, when the frequency of failures at a load
point is predicted, only the average value of that
quantity is typically calculated. The probability
that the load point will suffer a specified number of
failures in a year is not normally calculated.
Similarly, the expected values of the duration indices
are determined but the probabilities of various
durations are not calculated. The mnean values are
extremely useful and are the primary indices of load
point adequacy. There is, however, an increaesd
awareness of the need for information related to the
variation of the reliability measures around their
means [1,2,3,4,5,6] This information can prove
. more statistical data on load point interruptions will
useful in studies involving:
1) the probability of the interr'uptions being longer
than the Critical Service Loss Duration Time or
some othher time of interest [4,7,8,9]. This
information is especially useful in the design of
distribution systems for industrial customers with
critical processes or conercial customers with
very non-linear costing functions.
2) the probability of a certain number of failures
occurring in a particular year [2,3,10,11,12]
(perhaps determine annoyance factor).
to
3) the comparison of performance indices of different
years or different systems to determine the
probability of their having a different average
value. Such a comparison would assist plaanners to
judge whether differences in indices indicate real
changes in performance or are due to statistical
variation. (eg. the SAIDI, SAIFI, and CAIDI of a
system could be compared on a year-to-year basil
for significant variation or could be compared
85 Wi 229-0 A paper recommended and approved
by the IEEE Transmission and Distribution Committee
of the IEEE Powqer Engineering Society for presenta-
tion at the IEEE/PES 1985 Winter Meeting, New York,
New York, February 3 - 8, 1985. Manuscript submit-
ted August 23, 1984; made available for printing
December 12, 1984.
0018-9510/85/1100-3152$01.00O1985
E. Wojczynski
Manitoba Hydro
Winnipeg, Manitoba
Canada
with a utility's criteria of acceptable
performance) [6].
4) the validation of reliability models and the
applicability of reliability data [4,5,10,11,12],
as outage data is often insufficient and
inadequate. The index probability distributions
can be useful in estimating the errors resulting
from inaccurate data. Confidence intervals could
be comuted. This type of study is even more
applicable in generation reliability studies.
5) the determination of customer costs of
interruptions using non-linear cost functions.
Probability distributions provide a practical
vehicle to describe the variation of reliability
measures about their means. The approach taken in
this study to determine these distributions was to
perform probabilistic (Monte Carlo) simulations of
typical radial distribution systems. Monte Carlo
simulation is a widely applied technique and as it has
been described fully elsewhere [13 ,14, 151, an
explanation is not provided here.
An alternative approach is to analyse actual
interruption data for distributional information. A
number of such studies have been published
[4,6,10,11,121. Mlost of these are studies of outages
of components such as lines rather than studies of
load point interruptions. Load point index
distributions are dependent not only on combinations
oE component outages but also on system configurations
and restoration activities. With the increasing
mphasis that utilities are placing on data
collection, it is possible that in the near future
be available. Not only can simulation studies provide
useful information before comprehensive historical
data is available but they can provide information
that would not otherwise be possible to obtain. For
example, simulations can provide information
concerning the effects of very specific system
configurations. Furthermore, simulations can deal
with predictive reliability indices not just
performance indices. Their advantages become obvious
when making use of distributional studies of
performance data such as the one by Koval and Erbland
[6]. The paper provides quite useful information and
an illustration of the distributional tendencies.
Such as aproach however is not a good means to study
the effects of specific characteristics such as manual
sectionalizing capabilities. The reliability of
specific designs for a line can not be analyzed by
performance data studies but could be anal]yzed using
predictive reliability indices and simulations.
Studies exanining the di stributions associated
with the basic reliability indices indicate that the
load point failure rate is approximately Poisson
;istributed [2,3]. The failure rate probabilities can
be readily obtained using the mean load point failure
rate, since the Poisson distribution is a
single-parameter function. This information can be
calculated directly using the Poisson equation [2,31
or from a set of published graphs [16].
It has been noted that if the restoration times
can be assumed to be exponentially distributed, the
load point outage duration is approximately gamma
Jistributed and the desired probability information
caa be readily calcuilated [2,3]. There are, however,
IEEE
 3153
many distribution systems for which the ganma
distribution does not adequately describe the load
point outage duration. These systems are ones in
which some of the restoration t.imes may be better
described by non-exponential distributions, eg.
log-normal repair or manual sectionalizing times.
Studies carried out at the Univrersity of Saskatchewan
indicate that when the restoration times are assumed
to be non-exponential then the interruption duration Load
can not generally be represented by a gana ', 2miles Point
distribution [3,16,17]. In this case, obtaining the ~-~- ---- --- B
desired probability information is more difficult.
Distributions for the Annual Load Point Outage Time,
SAIDI, SAIFI, and CAIDI indices can also not readily I mile
be represented by common distributions. This paper
presents example distributions resulting from studies
of a sample radial system.
DISTRIBUTION SYSTE RELIABILITY ASSESSMENT Load
A C
Reliability assessment in distribution systems is
concerned with system performance at the customer end, F'igure 1. Simple Radial Configuration.
ie. at the load points. The basic indices normally
used to predict the reliability of a distribution utilizing the following basic equations:
system are: Average Load Point Failure Rate, Average
Load Point Outage Duration, and Average Annual Load A EXi failures/yr.
Point Outage Time. These indices are used to predict EX 1
future system performance. Utilities also calculate EA. r.
service performance indices to describe statistically r
s
=- i
-- hours/failure
the past performance of the system. The performance
indices - System Average Interruption Duration Index
(SAIDI), System Average Interruption Frequency Index
(SAIFI), Customer Average Interruption Duration Index Us = As .rs hours/year
(CAIDI), and others [3,18] - can also be calculated This procedure [18] is shown in Table I.
directly from the three basic predictive indices. An
example calculation of reliability and performance Sunmmarizing the results.
indices is presented for the example radial system
shown in Figure 1. Table II
In this base case configuration, all switches are Case 1 Indices
rormally closed and the customer load points A, B, C
are supplied from the primary main by fused laterals. A B C
The feeder breaker and the substation supply bus are
ass umed to be fully reliable. The individual
component data are as follows: X - failures/year 1.35 1.1 0.885
r - hours/failure 1.15 1.86 2.41
Primary main 0.10 failure/circuit mile/yr U - hours/year 1.55 2.05 2.05
3.0 hrs. average repair time
Primary lateral 0.25 failures/circuit/mile/yr There are many configurations particularly in
1.0 hrs. average repair time rural locations which have a topology similar to that
shown in Figure 1. The results shown in Table II can
Manual sectionalizing time for any switching action = be used to obtain the standard performance indices.
0.5 hrs. Assume that there are 250, 100 ahd 50 customers
The simplest approach is to perform a failure res,pectively at load points A, B and C, giving a total
modes and effect analysis [18] in a table form A:-Jt 400 customers in the system.

Table .I
Case 1 (Base Case) Calculatio'ns
LoDad Point A Load Point 1B ___
Load Point C
Component X r Ar X r Ar A r Ar

f/yr_ hrs hrsZy fZyK_ hrs _hrszyE f/yK hrs hrs/yr

Primary Main
2 m section 0.2 3.0 0.6 0.2 3.0 0.6 0.2 3.0 0.6
3 m section 0.3 0.5 0.15 0.3 3.0 0.9 0.3 3.0 0.9
1 m section 0.1 0.5 0.05 0.1 0.5 0.05 0.1 3.0 0.3

Primary Lateral
3 m section 0.75 1.0 0.75 - - -

2 m section - 0.5 1.0 0.5

1 m section -__ 0.25 1.0 0.25
1.35. 1.15 1.55 1.1 1.86 2.05 0.85 2.41 2.05
3154
Annual Customer Interruptions -
(250) (1.35) + (100) (1.1) + (50) (0.85) = 490]
System Average Interruption Frequency Index
SAIFI
total number of customer interruptions
total number of customers served
SAIFI = 490 = 1.23
400
Customer Interruption Duration =
(250) (1.55) + (200) (2.05)
System Average Interruption Duration Index = SAIDI
sum of customer interruption durations
total number of customers served
SAIDI = 695 = 1.74
400
Customer Average Interruption Duration Index
CAIDI
sum of customer interruption durations
total number of customer interruptions
CAIDI = 695 = 1.42
490
Similar calculations are presented in Reference 16
for the following operating assumptions:
1) Base Case (As in Figure 1)
2) Base Case with alternate supply available
3) Base Case with alternate supply conditional
load transfer probability of .5
4) Base Case with solidly connected laterals
5) Base Case with probability of successful
lateral fault clearing = 0.9.
These case number designations are used later in this
paper to illustrate the distributional effects.
PROBAABILISTIC SIMULATION PROGRAM
A program was developed at the University of
Saskatchewan to simulate the performance of any
N-section radial distribution system with loads
connected to laterals or directly to the primary mains
[17]. Any combination of exponential, normal,
log-normal, and gamma distributions can be used to
sinulate the failure, repair, manual sectionalizing,
alternate supply and fuse times. Costs of each
interruption can be calculated from 1 minute, 20
minute, 1 hour, 4 hour and 8 hour cost data. The
program outputs for each load point: the mean,
standard deviation, and distribution histogram of the
annual interruption time, interruption duration,
annual intorruption frequency, and annual interruption
cost. It provides similar outputs for the entire
system, in terms of SAIFI, SAIDI, CAIDI, cost per
interruption, and annual interruption cost. Studies
were performed on the 6 section example system of
Figure 1 and on a larger 18 section system.
Distribution of Load Point Failure Rate
The studies indicate that the load point failure
rate is reasonably described by the Poisson
distribution with a Chi-squared level of significance
= .1 [3,16, 17]. This result is in agreement with
theoretical considerations and a previous study
Patton [2]. Only one parameter is required to
describe the Poisson distribution, ie. the expected
annual failure rate. Since this value is the index
normally calculated, the distributional information
can be obtained from the Poisson distribution equation
= with minimal extra effort.
Figure 2 depicts the distributions associated with
the failure rates of load points A, B and C for Case
1. The distributions are noticeably different for the
three load points. At load point A, years with one
failure occur most frequently while at load point C,
years with no failures occur most frequently.
Concurrent with an increase in the average failure
rate, the shape of the distribution varies
significantly and the individual failure rate
probabilities increase in a non-linear fashion.
+ (50) (2.05) = 695
Load Point Outage Duration Distributions
Patton [21 noted that if the repair and other
restoration times can be assumed to be exponentially
distributed, the load point outage duration can be
approximated by a gamna distribution [2]. This is
confirmed by the studies described in this paper.
Figure 3 shows the simulation results for the
outage durations of load points A, B and C for Cases
= 1, 2 and 4 when the restoration times are assumed to
be exponentially distributed. These distributions can
also be reasonably described by the gamma distribution
(Chi-square level of significance - .05). As can be
seen, the general shape of the distributions does not
vary. Although the gamna distribution can take on
many different shapes, when it is the result of
combinations of exponential distributions, the shape
is always of this general form. The ganma
distribution becomes more or less spread out, or more
peaked, depending on the average outage durations and
the average restoration durations. If it can be
assumed that the restoration times are exponentially
distributed, the resulting outage duration
distribution can be assumed to be of the forms in
Figure 3 and the outage duration probabilities can be
readily calculated from the gamrna distribution.
Patton describes a relatively simple approach for
calculating the ganna distributed duration
probabilities using only the average outage durations
and failure frequencies of the load points and
contributing sections [2].
In many practical systems, the restoration times
cannot be assumed to be exponentially distributed. It
is often unrealistic to assume that the probability of
a repair or restoration increases as the duration
approaches zero. Restoration times may be better
described by non-exponential distr ibutions, eg.
log-normal repair times. The studies carried out
indicate that when the restoration times are assumed
to be non-exponential, the load point outage duration
cannot generally be represented by a ganma
distribution. The remainder of this section will
discuss the resulting distributions and how they vary
with certain factors.
It should be emphasized that the average values of
the load point outage duration indices are not
affected by what the underlying distributions are. A
set of averages such as those calculated for the
example system can have any set of distributions
associated with it.
The durations associated with repairs and other
restoration activities may often be weJll described by
1og-normal or other similarily skewed distributions.
Figure 4 shows the simulation results for the outage
durations when repair times are assumed to be
log-normally distributed with a standard deviation
equal to one third of the mean. These distributions
of the load point outage durations have a radically
b., d,,! £erent shape than those a^ssumng expnential
r>- ;toration times (Figure 3) . In Case 1, the form of
 3155

combination of those at A and C.

The backfeeding in Case 2 reduces the number of
failures with three hour repairs at load point C. The
CAS E second mode in the distribution is eliminated. The
3 mode resulting from the lateral repairs is now more
FAILURES/YEAR
pronounced for both load points B and C. When the
AVERAGE
* 1.345

laterals are connected solidly to the mains as in Case

SIMUJLATED RESULT
4, the first mode is even more pronounced for all
three load points. The predominant cause of outages
are failures on the primary laterals resulting in one
hour average repairs.
A visual inspection of the distributions in Figure
4 indicates that the distributions are so different
from those of Figure 3 that attempts to predict the
duration probabilities using the gamma distribution in
these cases of non-exponential restoration times could
O 1 3 4 S 6 7 lead to large errors. Similar results have been
2A RRESUS/YEAR obtained from simulations that assumed restoration
activities are garma and normal distributed. This
.4 indication is verified by goodness of fit testing
;
CASE 1
~~~~~~LOADPOINT B
(level of significance = .01) and by example
calculations [3].
The study results indicate that no known
distributions can universally describe the outage
duration distributions. Studies of outage duration
distributions for entire regions and service areas
confirm the reasonableness of this conclusion and the
distributions depicted in Figure 4 [4,6].
~o~~ ~ ~ ~ ~ ~ ~ As already discussed, varying the means of the
restoration times can significantly affect the shape
of the outage duration distributions. If the
underlying distributions are assumed to be similar to
the lognormal or gamma distributions, the effect is
greater. When the different components all have
restoration times with averages near to each other and
close to the origin, the distribution tends to
FAI LURES/YEAR resemble the exponential. As the differences between
the averages increases, the resemblance decreases.
The distribution may even be multimodal.
The distributions vary with the associated
standard deviations as well as with the component
means. Figure 5 shows outage durations for the sample
.4 4 _ 7 ~~~~~~~LOAD
POINT C
system with the assumptions that the standard
deviations for the repair times are equal to m (MMTR),
m/2, m/3, and m/6. When the standard deviations are
relatively large, the contributing distributions
overlap with the result that the outage duration
distribution is almost a monotonically decreasing
function. As the standard deviations decrease, the
contributing distributions become apparent and the
distribution definitely multimodal.
The sizes of the primary main or lateral sections
L L _ affect the mean, standard deviation, and possibly the
type of the ccrponent restoration distribution. The
load point outage duration distributions are then
indirectly affected by these factors in the ways
discussed above. It is possible that the number of
sections in the system affects the individual
O l ~2
~ 3 4 5 b 7
component restoration distributions. A more likely
FAILURES/REAR possibility is that the larger number of contributing
distributions may tend to overlap more and to obscure
Figure 2 Load Point Failure Rate Distributions - modal tendencies that might be apparent in smaller
Exponentially Distributed Restoration systems. This tendency is most pronounced if the
Activities. restoration activities of the different system
sections are dissimilar.
the distribution of load point A appears to be Simulations were performed for an 18 section
decreasing with duration except for a peak a couple of system similar to the example system but with 9 load
bars wide. This peak is attributable to the large points. Figure 6 plots the outage duration
number of repairs of 1 hour average duration which are probabilities for the first, middle, and last load
made on the first primary lateral. Due to it being points of the 18 section system. Visual and
further down the line, load point C has a larger statistical comparisons (Chi-Square level of
numer of repairs of three hour average duration. significance = .05) indicate that the outage duration
These repairs to the mains in combination with repairs distributions resulting from the assumption of
to the third lateral result in a bimdal distribution lognormal distributed repair times are significantly
with peaks of durations just less than one and three different from those resulting from an assumption of
hours
long. Load point B has a distribution that is a exponential distributed repair time. This confirms
3156

.351 .35 1- distributed restoration

times resulted in annual
.30 .30
interruption time
distributions that contain a
CASE 1 CASE 2 CASE 4
LOAD POINT A
sharp peak for the interval
LOAD POINT A
.25
LOAD POINT A
AVERAGE-1.1709 HRS. AVERAGE-1. 1707 HRS. .25 AVERAGE-0.9411 HRS. indicating the number of
years with no failures. The
distributions are steadily
.20
r.20 decreasing ones with long
-3 tails. The distributions
.: .15 go C . 15 are not of the exponential
*s
C;
I=
or gamma form.
The simulations that
.10 .10 assumed lognormal
distributions for the repair
.054 .05
times and exponential
distributions for the other
times also resulted in
0 distributions with a sharp
U 1 2
OUTACE DURATION IN
3 4
XJhOURS
5 1 2 3 4
OUTAGE DURATION IN HOURS peak for the no failure
.30 .30 interval. Inspection of the
annual interruption time
distributions of all the
.25 .25 other Case 1, six section
CASE 1 CASE 2 CASE 4
LOAD POINT B LOAD POINT B LOAD POINT B system simulations reveals
.20
AVERAGE-1.8739 HRS. AVERAGE-1.5363 HRS.
.20
AVERAGE1.L4416 HRS. that the number of years
with zero hours of
J interruption is independent
;; 15
.2
0 < .15 of the form and standard
0*
: co
deviation of the restoration
10
10
time distributions. This
.
independence occurs because
the number of years with
.05 zero interruption time is
dominated by the failure
ll
rate distributions which
1 4 5
0
determine the number of
years in which no failures
.0 2 3

OUTAGE DURATION IN HOllRS

30 .30 1 occur.
CAS E 1 CASE 2 CASE 4 The lognormal
LOAD POINT C LOAD PONTM C LOAD POINT C
AVERAGE-1.6350 HRS.
simulations did not result
AVERAGE=2.4151 HRS. AVERAGE=1.2597 HRS. .251
in steadily decreasing
distributions but in
20 distributions with multiple
I,-
modes. The multiple
-0 modality is more prcninent
:;,
:
2: Is5
co
when the restoration times
0 are assumed to have small
10 .10 standard deviations. With
standard deviations equal to
the means, the zero failure
.05 peak and a second mode
31) before 1.2 hours are
observable. With standard
1 2 3 4 S 0 1 2 3 4 5 1 2 3 4 5 deviations equal to 1/6
OlJTAGE DlURATION IN IIOURS OllTAGE DURPUl I!l IN IIOIIRS OUlTAI(T IIUJRATION IN HOURS means, the zero failure peak
and modes about 1.0, 2.0,
Figure 3 Load Point Outage Duration Distributions - 3.0, and 6.0 hours are evident. The peaks of certain
Exponentially Distributed Restoration duration (ie 2 and 6 hours) are related to years in
Activities. which more than one interruption occurs and the
restoration times are multiples of the dominent modes.
the conclusion drawn for the smaller 6 section system
that the outage duration distributions cannot in Distribution of SAIDI, SAIFI, CAIDI
general be described by the gania distribution
ccmputed fran the carponent average repair times. System performance indices and load point
reliability indices are not normally carpared or
Distribution of Load Point Annual Interruption Time simultaneously calculated. Performance and
reliability indices are in reality related by the fact
The annual interruption time distribution is that both are based on a cmmon set of data. The
dependent on both the failure rate and outage duration distributions of both indices are dependent on the
distributions. Because of this, it is even more distributions of the cmponent failure rates and the
difficult to describe the interruption time restoration activities. The SAIFI and SAIDI average
distributions by known functions. Figure 7 depicts indices are independent of the underlying
distributions resulting from simulations of the six distributions but the CAIDI average index is
section example system. distributionally dependent (17).
The simulations that assumed exponentially Figure 8 depicts distributions resulting fran
 3157

.30 .30
likely that there would be
at least oneinterruption.
CASE 1
LOAD POINT A
CASE 2
LOAD POINT A
CASE 4
LOAD POINT A The SAIFI distributions
.2S AVERAGE.1.1668 HRS. .25 AVERAGE.1.1560 HAS. AVERAGE.0.9252 HRS. are identical for the
,_I
exponential, log-normal S.D.
.20 .20 = m, and S.D. = m/6
simulations because SAIFI is
only dependent on the
is -.15 component failure time
4 distributions which do not
.10 .10 .I
vary with the simulations
runs and on the number of
customers served at each
.05 .OS load point. The irregular
variation of the
distribution shape is
related to the small number
0

0 1 2 3 4 5 I 3 4 S
of load points and the
2

OUTAGE DURATION IN iOURS OU1 AGE DURATION IN HOURS OUTAGE DURATION IN HOURS
.30 .30 resulting discrete weighting
CASE 1 CASE 2
LOAD POINT B
CASE 4
LOAD POINT B
by the number of customers
LOAD POINT B
AVERAGEal.5084 HRS. AVERAGE-1.4010 HRS. factor. As for the SAIDI
.25 AVERAGE-1.8772 HRS. .25$ .25
distributionss, in large
systems the probability of
.201 .20 SAIFI equalling zero
diminishes with the
distribution less resembling
.%.is
15
-. is
an exponential one and more
40 go one with a mode about the
.10 .1 average.
The CAIDI distributions

tl
are non-linearily related to
.05.
both the failure and
restoration times. This
0 0
1f
results in a somewhat
1 2 3 4 S D 1 2 3 4
OIuTAGE DURATION IN HOURS
similar modal distribution
OUTAGE DURATION IN HOURS
for the three simulations.
.30 The exponential and
CASE 1 CASE 2 CASE 4
LOAD POINT C
lognormal S.D. = m
LOAD POINT C
AVERAGE-2.4096 HRS. .2S
LOAD POINT C
AVERAGE.1.2455 HRS. AVERAGE=1.5825 HRS. simulation distributions are
more similar than the
lognormal S.D. = mV6
.20 .2( simulation distribution.
This and other comparisons
.- indicate that the standard
deviations of the underlying
.15
an
co co
CL.
1'
distributions can affect the
.10 .1 shape of the final index
distribution as much or more
.0s5
than the actual form chosen
.05
for the underlying
distributions. For large
0
0
systems, the CAIDI
0 1 2 3 4
OUTAGE DURATION IN HOURS
5 0 1 2 3 4
OUTAGE DURATION IN HOURS
distribution also tends to
OUTAGE DURATION IN HOURS "tighten up"' around the
,u'en .
;.ure 4 Load Point Outage Duration Distributions -
Pepair Times Lognormally Distributed (SD CXTLUSIC
1/3 Mean) Other Times Exponentially
Distributed. The means of the reliability and system
performance indices are and will continue to be the
simulations of the 6 section example systen. The main indice characteristic but the indice
SAIDI distribution is dependent on the distributions will increasingly be utilized as
distributions of the restoration times. The number of supplementary measures. The Load Point Failure Rate
customers at each load point and the average failure distribution can be assumed to be Poisson distributed
rates are weighting factors that are independent of with the distributional information easily obtainable
the associated distributions. The SAIDI distributions from the Poisson equation. The Load Point Outage
of Figure 8 are similar to the Annual Interruption Duration distribution can be assumed to be Gamma
Time distributions of Figure 7. This is because SAIDI distributed when it is possible to assume that
is a linear combination of annual interruption times. restoration activities are exponentially distributed.
In a large system, the resemblance tends to decrease Otherwise the duration distributions must be obtained
because of the averaging effect of the large number of by simulations or other means. The Load Point Annual
load points which are aggregated. In this small Interruption Time, CAIDI, SAIFI, and SAIDI
system, the number of years with a SAIDI equal to zero distributions cannot in general be described by any
is relatively high (ie. P [SAIDI 0] is high). This = known distribution and must be found by simulations or
might be expected in a small or moderate size system. other means.
In the case of a large system or region it is -very Distributional information obtained from
3158

.30 LOAD POINT A LOAD POINT A .30 LOAD POINT A

S.D. (r)= S.D. (r) =mf/2 S.D. (r=1)=m/2 .30 LOAD POINT A
AVERAGE=1 .1743 AVERAGE=1. 1698 (=3)/3
S ,D r.- S. D. (r)=m/6
S.D. (ra)= .6183 S.D.(rS)=1 .1166 AVERACE=1 .1682 AVRRAGE=1.1657
.25 .25 S.D. (ra)=1 .0236 S.D. (r8)=.8954
.25

.20. .20 1
.20

q .15 .15
N .15

,,.
.10 .10 K .10

4 .0.5o
.05 I'
.°S
0
1 Z 3' 4 -3 1 2 3 4 5' 0
OUTAGE DURATION IN HOURS OUTAGE DURATION IN HOURS 1 2 3 4' 5
OUTAGE DLRATION IN HOURS

LOAD POINT C .25 LOAD POINT C

S.D. (r)=m/2 LOAD POINT C
S.D.(r)=a S .D.(ral )=A/2 LOAD POINT C
AVEIAGE=2.3979 A VEAGE=2.4098 S.D.(r)=v/6
S .D .(r-3) =013
S.D. (r8)=2.6414 AVERAGE=2.4141
S.D. (r8)=1 .5637 AYERAGE=2.4118
.20 .20 S.D.(r )=1.2618 .20
S.D.(r,)=1.0054

f"; .15
i .15
H

I
1
X .10

.05
.05

0 -rF

OUTAGE DURTION IN HOURS 1 2 3 4 5

OUTAGE DURATION IN HOURS

Figure 5 Load Point Outage Duration Distributions As
A Function of Ccmponent Standard Deviations
- Repair Times Lognormally Distributed, Case
1.
simnulation studies such as these shown in this paper
complement actual data on index variation collected
for similar load points over a number of years.
Together they provide additional valuable information
on system and load point reliability.
1. Koval,, D.O., and Billinton, R., "Statistical and
Analytical Evaluation of the Duration and Cost of
Consumers Interruptions",, IEEE/PES Paper No. A79
057-1, Winter Meeting, February 1979.
2. Patton, A.D., "Probability Distribution of
Transmission and Distribution Reliability
Performance Indices", 1979 Reliability Conference
for the Electric Power Industry, pp. 120-123,
1979.
3. Billinton, R., Wojczynski,, E., and Rodych, V.,
"Probability Distributions Associated
Distribution System Reliability Indices",
Reliability Conference for the Electric Power
Meeting, 1980.
4. Koval, D.O., and Erbland, M.J., "Statistical
Analysis of the Duration of Outages", CEA
Transactions, presented at Fall Meeting, Winipeg,
1980.
5. Sahinoglu, M., Longnecker, M., Ringer, L., Singh,
C., Ayoub, A., "Probability Distribution Functions
for Generation Reliability Indices - Analytical
Approach", IEEE TPAS-102, No. 6, June 1983, pp.
1486-1493.
6. Koval, D., Erbland, M., "Distribution Reliability
Assessment Using a Proprietory Statistical
Package", CEA Distribution System Planning,
Conference #1, October 1983.
7. IEEE Cnittee Report,"Report on Reliability
Survey of Industrial Plants, Part II: Cost of
Power Outages, Plant Restart Time, Ctitical
Service Loss Duration Time, and Type of Loads Lost
Versus Time of Power Outages", IEEE Transactionis
on Industry Applications, Vol. IA-10, No. 2, pp.
236-241, March/April, 1974.
8. Power Systems Reliability Subcomlittee Report,
P.E. Gannon, "Cost of Electrical Interruptions in
Commercial Buildings", IEEE 1975 I&CPS Conference,
pp. 123-129, 1975.
9. Wojczynski, E., Billinton, R., and Wacker, G.,
"Interruption Cost tMethodology and Results - A
Canadian Ccxnercial and Small Industrial Survey",
IEEE SM 390-2 1983.
10. Beaurecueil, P.S., "Southern California Edison
Company Transmission Line Outages Statistical
with Analysis",, Reliability Conference for the Electric
1980 Power Industry, pp. 27-38, 1980.
11. Fond, C.C., and Le Reverend, B.K., "CaTputerized
Statistical Analysis of Transmission System Outage
Data", Reliability Conference for the Electric
Power Industry, pp. 39-46, 1980.
12. Polena, R.J., 11345 kV Line Outage Data Analysis
for System Planning Application", Reliability
Conference for the Electric Power Industry, op.
45-49, 1981.
 3159

.35

.30 LOAD POINT A .30 WOAD POINT A

EXPONWTAL REPAIR LOGNORMAL REPAIR
S.D.(r)- J9 S.D. (r)../2
AVRA9.8 90 AVERAGE=-.8427
.25 .25

.20 .20

N .15 I .15

.10 .10

.05
o .05

0 0
1 2 3 4 5 1 2 3 4 5
OUTAGE DURATION IN HOURS OUTAGE DURATION IN HOURS
.30

W2IAD POINT N
E2DCPONTMAL REPAIR .25
*25 WAD POINT E
WOGNORMAL REPAIR
S.D. (r)-m S.D. (r)=m/2
AVERACE=1.895 AVERAGE=l.909
.20 .20

k . 15 > . 15

.10 .10
.°S

.05
o .05o

0 0
1 2 3 4 5
OUTAGE DURATION IN HOURS OUTAGE DURATION IN HOURS
.30 .30

.25 I WAD POINT I 25 LOAD POINT I

EXPONENTIL1 REPAIR
5

LOGNORMAL REPAIR
S.D. (r)=m S.D. (r)=u/2
AVERAGE=2.727 AVERAGE=2.761
.20 .20

1>4
4 .15 F .15

A .10 A .10

.05 .05

1 2 3 4 5 0 1 2 3 4
OUTAGE DURATION IN HOURS OUTAGE DURATION IN HOURS

Figure 6 Outage Duration Distributions For An 18 Section System - Case 1.

13. Shannon, R., Systens Simulation; The Art and 17. Wojczynski, R., "Application of Reliabilit Worth
Science, Prentice-Hall Inc., New Jersey, 1975. in Power System Planning", M.Sc. Thesis,
14. Singh, C., and Billinton, R., System Reliability University of Saskatchewan, January 1984.
Modelling and Evaluation, Hutchinson and Campany, 18. Billinton, R., Allan, R.N., "Reliability
London, 1977. Evaluation of Power Systems", Pitman Publishers
15. Billinton , R., Hamud, G.A., Jamali, M.M., Ltd. 1984.
"Reliability Evaluation Using Monte Carlo
Simulation",, CEA Transactions, Spring 1979.
16. Billinton, R., Wojczynski, E.,, Godfrey, M.,
"Practical Calculations of Distribution System
Reliability Indices and Their Probability
Distributions", CEA Transactions, 1981.
3160

.30 .30 .30

.25 LOAD POINT A .25 LOAD POINT A

EXPORNIISAL 100AI0 LOGNORMAL REPAIR
.25 LOAD POINT A
S.D.0(or). S .D. (r) -- I0GNORMAL REPAIR
AVM=AR= 1.576 AVER4CE=1. 50 (0r)/6
SA.DO.
.20 .20 AVERAGE=1 .568
.20

>-
g .15
.15
I FX *
I .10I .10 I°3 l1

.0

AIINAL INlTRRUPTION TIOE IN HOURS

.60 .60 .60

.30 LOAD POIINT C .S0 LOAD POINT C .50 LOAD POINT C

EXPONENTIAL REPAIR LOGNORMAL REPAIR DOGNORMAL REPAIR
S.D.(r)o( S.D.(1r5= S.D. (r)00/6
AVERACE=2.077 AVOERAG=2.061
.40 .40 .40 A01AG:S'2.075

l .30 j .30
l .30
A. .20 .20
A A .20

.10 .10 .10

0 0
A z 3 4 S 6 7 S 0 1 2 3 6 5 6 (7 0 1 2 3 4 S A ?
ANNUAL INTEIuOION TIME II NOURS A1NUAL INIT11UPTIOR TIME IN HDt0S
A
ANNUAL INTERRUPIION TIME Il HDURS

Figure 7 Distribution of Annual Interruption Tines - Case 1.

I
.30 .30

.25
I EPOINENTIAL REPAIR .2 !5 EXPONENTIAL REPAIR .25 EXPOENTUI REP0I0
_S.'D.(r)-a 0.1D(r).a S.D.(r)..
AVERAGE=1.762 AVER00=1 .224 AVERACE=1 .332
.20
S.DI.(SA.DI).2.318
.21 0
S.D.(SAIPI)=.974
.20
S.D. (CAIDi0=1.199

N .15 5 .15
3-
13

I .10 0 .10

.05 .00 .05

0
0.3 1.0 1.5
SAIDI- HDRS/CUMOIERN CADI- HDURSAINlTUoTION
.30 .30 .30

.25 lONORRAL REPAIR .25 iDCNORltAL REPAiIR

S.D.1(r)
_ AVER0E=1.761
.25 S.D. (r)--
AVERAGE=1.346 Figure 8 Distribution of System
S.D.(OSAIDI.)2.292 S.D.(CAII)=l .257 Performance Indices Case 1.
.20 .20 .20

I3 .15 rt .15

1 .10

.05
,, .15 .05
i .10

0
0 1 2 3 O 0 015 1.0 1.5
SAIDI- iV01S/a=00RME1 SAIFI- INTERRUPTIONS/CUSTOME:R C1ID1. HDURS/IIPIrRRUPTION
.10
,Wv Tr .30

.25 ICNORMAL REPAIR .25 10S0.1D1.(r) 006 .23

2 LOGNORMAL1*1
S.D (r) ./6
REPAIR

S.D. (r)-u/6
AVERA0 R=1.757 A0ERA1E01.224 A0E0AG016=.333
S.D.(SAIDI).1.729 .20
S.D.(SAIPI).97?4
.20
S.D.(CAIDI ).589

1 A~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
N .15
3 *iS .15

.10
I.10

.05 .05
.05

°S

1 2 3 0 10 0.5 1.5

S11DI- i0URS0/CJIoMER INTEoRRUPTIONSR SI/OOT ERo CAIRO- H00010/IIIT1I UPTION