704 IEEE Transactions on Power Systems, Vol. 13, No.

2, May 1998

Distribution System Reliability:

Default Data and Model Validation

R.E. Brown J.R. Ochoa

Member Member

Transmission Technology Institute Distribution Information Systlems Development

ABB Power T&D Company Inc. ABB Power T&D Company Inc.
1021 Main Campus Drive 110 Corning Road, Suite 101
Raleigh, NC 27606 Cary, NC 2751 1

Abstract-Distribution system reliability assessment is Since a majority of customer reliability problems stem
able to predict the interruption profile of a distribution from distribution systems [l],utilities must focus on distribu-
system based on system topology and component reliabil- tion systems if substantial improvement in customer reliability
ity data. Unfortunately, many utilities do not have enough are to be gained. Deregulation of the industry has also made it
historical component reliability data to perform such an critical for utilities to provide this level of reliability at the
assessment, and are not confident that other sources of lowest possible cost. To do this, predictive reliability assess-
data are representative of their particular system. As a ment methods are needed [ 2 ] .
result, these utilities do not incorporate distribution sys- Distribution system reliability assessment is a quickly
tem reliability assessment into their design process and maturing field. It has evolved from the first EPRI program in
forego i t s significant advantages. This paper presents a 1978 [ 3 ] , to programs developed and used in-house by utili-
way of gaining confidence in a reliability model by devel- ties [4-71, to commercially available software products. These
oping a validation method. This method automatically planning tools are able to predict the reliability of a distribu-
determines appropriate default component reliability data tion system based on system topology and component reh-
so that predicted reliability indices match historical val- ability data. Unfortunately, these products will never gain
ues. The result is a validated base case from which incre- widespread use until utilities are confident that available data
mental design improvements can be explored. is representative of their actual system.
Ideally, a utility will have a large amount of historical
data from which it can determine the reliability of various
I. INTRODUCTION components such as lines, prlotection devices, and switches.
Most utilities, however, do not have this information avail-
Customer demands for reliable power are quickly able. Values may be obtained from published data corre-
changing. Not only is more energy being demanded, but this sponding to other systems, but this data may not be represen-
energy must be provided at increasing levels of service reli- tative of the system under consideration. This data discrep-
ability. A sustained interruption can cost certain customers ancy is most evident when predicted system reliability indices
hundreds of thousands of dollars per hour. Even a momentary do not agree with historically computed reliability indices.
interruption can cause computer systems to crash and indus- Most utilities do not have a substantial amount of histori-
trial processes to be ruined. To many customers with sensitive cal component reliability data. Nearly all utilities, however,
electric loads, reliability as well as the cost of energy may have historical system reliability data in the form of reliability
drive decisions such as: where a new plant is to be located, indices (e.g., SAIFI, SAIDI-see Section 2.1). When a sys-
whether an existing plant will be relocated, or whether a tem is modeled, the reliability indices predicted by the as-
switch to a new energy provider will be pursued. sessment tool should agree with these historical values. If so,
a certain level of confidence in the model is achieved and
more specific reliability results (e.g., the reliability of a spe-
PE-870-PWRS-2-06-1997 A paper recommended and approved by
the IEEE Power System Planning and Implementation Committee of cific load point or the impact of a design change) can be
the IEEE Power Engineering Society for publication in the IEEE trusted to a higher degree. When this confidence has been
Transactions on Power Systems. Manuscript submitted December 27, achieved and predicted results match historical results, the
1996; made available for printing June 9, 1997. reliability model is said to be validated.
This paper presents a new method of distribution system
reliability model validation. It first identifies which default
component reliability parameters should be modified by per-
forming a sensitivity analysis on a test system. It then presents
a method of computing these parameter values so that pre-

0885-8950/98/$10.00 0 1997 IEEE

 705
dicted system index values match historically computed index ties know how their systems are performing, but is less useful
values. when the specific impact of various design improvement op-
tions wish to be quantified and compared. To do this, a
11. PREDICTIVE RELIABILITY ASSESSMENT method must be developed which is capable of predicting
reliability measures based on system topology, component
Predictive reliability assessment can be broken down into reliability data, and operational data.
three basic steps: defining meaningful measures of reliability,
developing a method to compute these measures, and judi- 2.2 Reliability Assessment Method
ciously choosing the best available data to be used by the
developed method. This section discusses each of these steps The basic methods capable of producing the reliability of
in brief. a distribution system are: network modeling, Markov model-
ing, Monte Carlo simulation, and state enumeration. This pa-
2.1 Measures of Reliability per uses state enumeration due to its computational efficiency
and its ability to model complex system behavior.
In the context of distribution systems, reliability has his- In the state enumeration method, system states are dy-
torically been associated with sustained customer interrup- namically generated by considering each possible primary
tions (interruptions lasting more than a few minutes). This is contingency (including sustained faults, self-clearing faults,
reflected in the predominant use of the reliability indices and passive outages). For each primary contingency, many
SAIFI and SAIDI [8]. These values represent the number of possible system responses are considered. System responses
sustained interruptions and the number of interruption hours include reclosing and sectionalizing behavior, imperfect pro-
that an average customer experiences in a given year (the reli- tection systems, upstream fault isolation, and load restoration.
ability indices ASAI and CAIDI are also widely used, but The result is a sophisticated reliability model which can accu-
they can be directly computed from SAIFI and SAIDI and rately model rare events which can potentially have a large
offer no new information). impact on the reliability of the system. For example, if a mo-
Although sustained interruptions have historically re- mentary interruption occurs on a feeder and the recloser at the
ceived the most attention, the growing sensitivity of electronic beginning of the feeder becomes stuck, the substation bus
loads has made the inclusion of other voltage disruptions nec- protection must clear the fault. This will result in a sustained
essary when considering customer reliability. The first of interruption for the entire bus. Such a situation is rare, but has
these to emerge is momentary interruptions, which is already a high impact to the system and should therefore be included
an indispensable aspect of distribution reliability [9]. Voltage in the reliability assessment.
sags are quickly making the transition from a power quality The above reliability assessment method has been im-
issue to a reliability issue, and voltage spikes and voltage plemented in the engineering tool DISTREL (DISTribution
flicker may be soon to follow. RELiability), which is currently under development. This is
The reliability aspects considered in this paper are: mo- an engineering tool which will eventually be an add-on mod-
mentary interruption frequency, sustained interruption fre- ule for ABB’s distribution analysis package CADPAD, and is
quency, and sustained interruption duration. The system reli- the radial complement of ABB’s network reliability assess-
ability indices that correspond to these measures are: MAIFI ment package NETREL (NETwork RELiability).
(the Momentary Average Interruption Frequency Index),
SAIFI (the System Average Interruption Frequency Index), 2.3 Component Reliability Data
and SAIDI (the System Average Interruption Duration Index:
These indices reflect the overall reliability of a specified area Before the reliability assessment described in Section 2.2
(e.g., an entire system, a single feeder, a region). Equations can be performed, each component must be assigned reliabil-
corresponding to these indices are: ity data. Descriptions of the various parameters are:

# of Customer Momentary Interruptions h,: Momentary failure rate-this is the frequency of

MAIFI = # of Customers Served
per year (1) faults that will clear themselves if the fault site is de-
energized and then re-energized.
# of Customer Sustained Interruptions
SAIFI = per year (2) Is: Sustained failure rate-this is the frequency of faults
# of Customers Served
that requires a crew to be cleared and repaired.
Sum of Customer Interruption Durations hp: Passive failure rate-this is the frequency of compo-
SAIDI = hours/ year (3)
## of Customers Served nent outages that interrupt the flow of current, but do
not cause fault current to flow.
Reliability indices are typically computed by utilities at MTTR: Mean Time To Repair-this is the expected time it
the end of each year by using historical outage data recorded will take a crew to repair a component outage and
in distribution outage reports. This is important so that utili- restore the system to its normal operating state.
706

MTTS: Mean Time To Switch-this is the expected time it ground line sections might be assigned a different failure rate
will take after a failure occurs for sectionalizing per unit length. In this way, the number of reliability parame-
switch to be toggled. ters used in a given system is greatly reduced. There are still,
PSS: Probability of Successful Switching: the probability however, many more default parameters than can be effec-
of a switching device (a protection device, a load- tively adjusted with the limited amount of historical data that
break switch, or a no-load break switch) of actually is typically available. To help decide which default parame-
switching if it is supposed to switch. The comple- ters will be adjusted, a sensitivity analysis will be performed.
ment of PSS is switch unreliability:

For a reliability model to produce representative results,

component reliability data must be representative of the sys-
tem being modeled. Ideally, a utility will have a large amount
of historical data for each component type under considera-
tion. Unfortunately, most utilities are still many years away Switches* I 0.01 0 00 4 I 0.95
from having such data. The alternative is to use previously STS I n ni I o no I 4 I 099
published data that has been collected for other systems. This *Switching times are assumed to be 1 hour
data may or may not be representative of the system to be
studied. In fact, if predicted system indices do not match his-
torically computed system indices, it is likely that this data is
not appropriate.
The remainder of this paper discusses a method of de-
termining appropriate component reliability values so that
predicted indices match historical indices. This method will
then be applied to the test system described in Section 111.
LP7
111. TEST SYSTEM

The test system used in this paper is Bus 2 of the RBTS

as described in [lo]. This system consists of four overhead
11-kV feeders (FLF4) fed from a distribution substation and
serves 22 load points (LP1-LP22). F1 and F2 are operated in 9
a looped configuration as are feeders F3 and F4. The test
system topology is shown in Figure 1 and the component reli-
ability data initially assigned is shown in Table 1.
The reliability data of Table 1 will result in certain reli-
ability index values. It is likely that these values will not
match historically computed values if the component data is
not representative of the equipment being modeled. Predicted
and historical index values for this system can be seen in Ta-
ble 2.
As can be seen from Table 2, predicted reliability index
values do not match historically computed index values. To
validate the system model, certain component parameters
should be adjusted so that all values match. Unfortunately,
there are hundreds of component parameters to choose from.
Section 4 describes how to judiciously choose which parame-
ters will to adjust when validating a distribution system reli-
ability model.

IV. SELECTING PARAMETERS TO ADJUST

When first creating a predictive reliability model, default

values are typically assigned to various classes of compo-
nents. For example, all overhead lines sections might be as-
1
LP22
signed the same failure rate per unit length, and all under- Figure 1. Test System Topology
 707
4.1 Sensitivity Analysis 4.2 Selection of Default Component Parameters

The sensitivity of a function to a parameter is defined as Because system reliability indices are most sensitive to
the partial derivative of the function with respect to that pa- changes in overhead line default values, overhead line hM,
rameter. This is a measure of how much the value of the overhead line is ,
and overhead line MTTR will be the default
function will change if the parameter is perturbed. This can be values that will be adjusted when validating a reliability
approximated by actually perturbing the parameter (keeping model. For a given area under consideration, these results in
all other parameters fixed) and measuring how much the three degree of freedom in a search space looking for three
function changes. For example, if the default overhead line values: MAIFI, SAIFI, and SAIDI.
MTTR is increased by 1%, and the resulting SAIDI value There is yet another justification for choosing overhead
increases by O S % , the sensitivity of SAIDI to default over- line parameters for adjustment. Since line failure rates and
head line MTTR is (0.5 / 1) * 100% = 50%. A sensitivity repair rates are largely dependent upon vegetation, tree-
analysis of MAIFI, SAIFI, and SAIDI to all default compo- trimming, and weather, there will likely be significant varia-
nent parameters has been performed and the results are sum- tion in line failure rates from utility to utility. This means that
marized in Table 3. finding representative data from previously published data
will likely be more difficult for overhead lines than for other
types of components.
Since MAIFI and SAIFI are predominantly affected by
overhead line failure rates, model validation will usually re-
sults in representative values for these failure rates. SAIDI,
however, can be significantly influenced by other component
types. For model validation to obtain a representative value of
d Failure Rate line MTTR, it is essential to obtain the best possible values
Mean Time To Repair 0% 0% 6% from other sources for parameters with a high SAIDI sensi-
Unreliability * -1% 7% 12% tivity.
Reclosers
Recloser FX I 0% I 4% I
Recloser MTTR I 0% I 0% I 6% 4.3 Default Component Parameter Multipliers and
Unreliablity * I -8% I 15% I 26%
I Sectionalizing Switches I I I I Customized Component Values
Sustained Failure Rate 0% I 4% I 3%
Mean 'I'ime To Repair 0% I 0% I 3%
Switching Time 0% I 0% I 22% Although a utility may not have enough historical data to
, ... ., -,"
determine the reliability parameters for all components on its
*Unreliability is defined as (1 - PSS). system, it may know the relative reliabilities of certain com-
ponents. For example, it may be known that the overhead line
As can be seen in Table 3, MAIFI is much more sensitive
failure rate in a certain heavily treed area is twice as much as
to the default overhead line momentary failure rate (AM) than
in a lightly treed area. Similarly, the MTTR of a component
to any other reliability parameter. This is a good indication far away from a crew dispatch location may by 50% longer
that if the predicted value of MAIFI does not agree with the than a component close to the dispatch location. This type of
historical value of MAIFI, default overhead line A M should be information can be incorporated into the reliability model in
adjusted. the form of default component parameter multipliers.
Table 3 also shows that SAIFI is much more sensitive to A default component parameter multiplier is simply a
the default overhead line sustained failure rate (As) than to number that corresponds to a component parameter. The
any other reliability parameter. This implies that the default number that will be assigned to the component parameter is
overhead line As should be adjusted if the predicted value of equal to the default component parameter multiplied by the
SAIFI does not agree with the historical value of SAIFI. corresponding default component parameter multiplier. For
The last set of reliability index sensitivities correspond to example, if a specific line section has a MTTR multiplier of
SAIDI. Table 3 shows that SAIDI is predominately effected 1.5 and the default line MTTR is 4 hours, then the line will be
default overhead line hs and default overhead line MTTR. It assigned a MTTR of (1.5 x 4)= 6 hours.
should be pointed out that the sensitivity of SAIDI to compo- A utility may also have specific component reliability
nent default parameters is more distributed than MAIFI and data for certain components on its system, but not for all
SAIFI. This is reflected in relatively high sensitivities to re- components. To accommodate this information, each compo-
closer unreliability and switching time. Therefore, if predicted nent parameter is allowed to be customized. If a value is cus-
SAIDI values do not match historical SAIDI values, any one tomized, is not allowed to be changed during the model vali-
or all of these parameters could be undesirable. dation process.
708
V. DETERMINING DEFAULT PARAMETER VALUES
4.5 , I

This section develops a method of determining overhead

line section default parameters so that predicted values of
MAIFI, SAIFI, and SAIDI agree with historically computed
values. Before this method is presented, however, is advanta-
geous to examine how these index values actually change as
line section default parameters change.
The first default line section parameter that will be exam-
ined is AM. MAIFI, SAIFI, and SAIDI as a function of hMcan
O J I
be seen in Figure 2. The figure reveals that each index varies 0 02 04 06 08 1
linearly with failure rate. This makes sense because each mo- Overhead Line Momentary Falure Rate (kmlyr)
mentary failure affects the system in the same manner, regard-
less of how many occur. Figure 2. Effect of on ReliabilityIndices
The next default line section parameter that will be exam-
ined is As. MAIFI, SAIFI, and SAIDI as a function of hs can
be seen in Figure 3 . As was seen with AM, all indices vary
linearly with As. This also makes sense because our model
assumes that each sustained interruption impacts the system in
the same way, regardless of the frequency. This assumption is
reasonable except during major events when a common mode
(e.g., wind, earthquake, icing) is causing many failures to
happen at the same time. The method used in this paper does
not consider major events.
The last default line section parameter that will be exam-
ined is MTTR. MAIFI, SAIFI, and SAIDI as a function of
MTTR can be seen in Figure 4. Two interesting features can
be observed from this figure. First, SAIFI and MAIFI do not 0 02 04 06 08 1
change as MTTR changes. Second, SAIDI is piecewise linear Overhead Line %stained Failure Rate (kmlyr)
with respect to MTTR. The first result is expected-
increasing the repair time of a component will not cause any Figure 3. Effect of hs on ReliabilityIndices
customers to experience more or less interruptions. The sec-
ond results is due to switching times. After the MTTR of a
component rises above certain levels (corresponding to the
MTTS of switches), it will become advantageous to isolate
the fault and restore service to some customers. This results in
the a piecewise linear relationship between SAIDI and
MTTR. Once the MTTR is above the longest MTTS, the re-
lationship remains linear.
Determining the appropriate values of default overhead
line AM, As, and MTTR (hereafter referred to as h M , hs, and
MTTR) can be broken down into two steps. This is possible
because MTTR has no effect on MAIFI and SAIFI. Conse-
quently, XM and As can be chosen so that predicted MAIFI and
SAIFI values match historical values. After this has been
done, MTTR can be adjusted to control SAIDI without im-
pacting MAIFI or SAIFI.
t
0 2 6

Overhead Line Momentary Falure Rate (kmlyr)

8 10

The relationship of SAIFI and MAIFI to adjustments of

L M and As (AhM and A&) is given in (4): Figure 4. Effect of MTTR on Reliability Indices
 709
default parameters so that predicted values of MAIFI, SAIFI,
and SAIDI match historical values. The choice for overhead
line default parameters for adjustment is based on a sensitivity
analysis.
After a distribution system reliability model has been
SAIFI, : Target SAIFI value validated, utilities can be reasonable confident that the model
MAIFI, : Target MAIFI value is representative of its actual system. This will facilitate the
SAIFI, : Initial SAIFI value incorporation of the reliability into design and planning pro-
MAIFIi : Initial MAIFI value cedures which will aid in providing high levels of reliability
AS : Sustained Line Failure Rate (/km/yr) for the lowest possible cost.
hM : Momentary Line Failure Rate (/km/yr)
VII. REFERENCES
In equation (4), the partial derivatives can be approxi-
R. Billinton and S. Jonnavitihula, ‘A Test System for Teaching Overall
mated by perturbation methods. This allows AhM and Ahs to Power System Reliability Assessment,’ IEEWPES 1996 Winter Meet-
be solved for. The new values of are then found by adjusting ing, Baltimore, MD, IEEE, 1996.
the old value by the computed change: Le,= Lola+ AL. Once R. Brown, S. Gupta, S.S. Venkata, R.D. Christie, and R. Fletcher,
‘Distribution System Reliability Assessment: Reliability and Cost Op-
these new values are computed, the relationship between timization,’ 1996 IEEE Transmission and Distribution Conference
SAIDI and MTTR can be used to find the new value of Proceedings, Los Angeles, CA, September, 1996.
MTTR. This relationship is: EPRI Report EL201 8, Development of Distriburion Reliability and
Risk Analysis Models, Aug. 1981. (Report prepared by Westinghouse
Corp.)
SAIDI, =- AMTTR + SAIDI, R. Brown, S. Gupta, S.S. Venkata, R.D. Christie, and R. Fletcher,
dMTTR ‘Distribution System Reliability Assessment Using Hierarchical
SAIDZ, : Target SAIDI value Markov Modeling,’ IEEE PES Winrer Meeting, Baltimore, MD, Janu-
ary, 1996.
SAIDI, : Initial SAIDI value S. R. Gilligan, ‘A Method for Estimating the Reliability of Distribution
MTTR : Mean Time To Repair (hours) Circuits,’ IEEE Transactions on Power Delivery, Vol. 7 , No. 2, April
1992, pp. 694-698.
If all parameterhndex relationships are linear within the G. Kjfille and Kjell Sand, ‘RELRAD - An Analytical Approach for
Distribution System Reliability Assessment,’ IEEE Transacrions on
ranges considered, equations (4) and ( 5 ) will results in pre- Power Delivery, Vol. 7, No. 2, April 1992, pp. 809-814.
dicted MAIFI, SAIFI, and SAIDI values which exactly match Yuan-Yih Hsu, Li-Ming Chen, Jiann-Liang Chen, et al., ‘Application
historical values. If an exact match is not found, the process of a Microcomputer-Based Database Management System to Distribu-
can be repeated iteratively until a certain error tolerance is tion System Reliability Evaluation,’ IEEE Transactions on Power De-
livery, Vol. 5, No. l , Jan. 1990, pp. 343-350.
reached. R. Billinton and J. E. Billinton, ‘Distribution System Reliability Indi-
This model validation method has been applied to the test ces,’ IEEE Transactions on Power Delivery, Vol. 4, No. 1, Jan 1989,
system described in Section 3 on a feeder by feeder basis. For pp. 561-568.
the reliability data given in Table 2 and Table 3, solutions R. Brown, S. Gupta, S.S. Venkata, R.D. Christie, and R. Fletcher,
‘Distribution System Reliability Assessment: Momentary Intemptions
converged for each feeder in a single iteration. After model and Storms,’ IEEE PES Summer Meeting, Denver, CO, June, 1996.
validation, all predicted reliability indices matched historical [IO] R. N.Allan, R. Billinton, I. Sjarief, L. Goel, and K. S. So, ‘A Reliabil-
values to within 0.1%. This method was also been tested for ity Test System for Educational Purposes - Basic Distribution System
a wide variety of other reasonable data sets, and converged in Data and Results,’ IEEE Transactions on Power Systems, Vol. 6, No.
2, May 1991, pp. 813-820.
a single iteration for all cases.
The only difficulty encountered when using this model VIII. BIOGRAPHIES
validation method occurs when the system is not completely
controllable. An example of when this occurs is when there Richard E. Brown received his Ph.D. in electrical engineering from the
are no reclosers on the system. In this situation, no momen- University of Washington in 1996. Before beginning graduate studies, Dr.
tary interruptions can occur and MAIFI is not controllable. To Brown worked as a consulting engineer designing industrial distribution
systems. He is currently a senior engineer at ABB’s Transmission Technol-
handle this, equation (4) should be reworked to only consider ogy Institute and specializes in the areas of distribution systems, reliability,
SAIFI. power quality, and design optimization.

J. Rafael Ochoa received his Ph.D. from the University of Manchester

VI. CONCLUSIONS Institute of Science and Technology (UMIST) in 1986. He is currently the
manager of ABB’s Distribution Information System’s development group.
Distributions system reliability models can be validated Prior to this position, Dr. Ochoa was the Reliability and Power Quality
by adjusting default component reliability parameters so that Group’s team leader at the Transmission Technology Institute. From 1987 to
1992, Dr. Ochoa was a Lead Engineer at Ferranti Intemational Controls
predicted results match historical results. This has been dem- Corporation (HCC). From 1978 to 1987, Dr. Ochoa was a Senior Engineer
onstrated on a 4 feeder test system by adjusting overhead line at the Electrical Power Research Institute of Mexico.