Reliability Evaluation of Distribution Networks

Using Fuzzy Logic

M.R Emjedi, Non-Member, K.Awodele, Member, IEEE, S.Chowdhury, Member, IEEE and
S.P.Chowdhury, Member, IEEE
The reliability of the sub-systems, constituting the
Abstract--Reliability is a measure of performance. This
power system, i.e. generation, transmission, distribution,
measure can be used to help systems meet performance criteria, to can be analyzed separately [3] so each subsystem can be
help quantify comparisons between various options, and to help analyzed and the sum of the individual analysis
make economic decisions. This paper presents a fuzzy knowledge combined in order to provide an indication of the extent
based approach for reliability evaluation of a distribution
of the reliability of the power system.
network. This approach makes a direct assessment of the network
configuration, distribution substation, the maintenance policy and The distribution network is that portion of the power
the weather and assigns to the network under consideration a
reliability index by expressing those variables mathematically system which links the bulk power source or sources to
using fuzzy logic. Results obtained using the fuzzy model on actual the consumer’s facilities [2]. Sub transmission circuits,
utility data are presented and compared to historical reliability distribution substations, primary feeders, distribution
indices. The flexibility of the fuzzy model allows for its application transformers, secondary circuits and consumer’s
in simplification of complex concepts into easily handled models.
Because of the simplicity of the model, the speed of computation is connections all form different parts of the distribution
very fast. Thus, large systems could be evaluated easily with network.
minimal computation time.
Distribution network reliability is defined as the ability
Key Words. Distribution network reliability, Fuzzy logic, to deliver uninterrupted service to customers. The
Differential evolution interruption is a direct consequence of the power system
protecting itself and the public from dangerous or
I. INTRODUCTION damaging conditions. It is impractical to deliver
A distribution network is a subsystem of a larger uninterrupted power to customers over a long period of
entity known as the power system. The power system time as the cost associated with this would not make it
has many sub-systems integrated together, which profitable for the utility to operate.
interact in order to provide the utility’s customers
(residential, commercial and industrial) with energy to Presently, electric utilities are facing an increasing
meet their demands. number of complaints about distribution network
Modern society has come to expect that the supply of reliability. This is due to the growth of sensitive loads in
electrical energy will be continuously available on all customer sectors (residential, commercial and
demand. This is not possible due to random failures of industrial) [4]. The consequence of an interruption
equipment and their behavior governed by stochastic depends significantly on the type of customer
and probabilistic laws. As a result, power system interrupted, and is affected by the frequency of the
behavior (reliability, adequacy and security) are difficult interruptions and the duration of each interruption [5].
to analyse and predict [1]. Failure in any part of the Due to this, utilities must maximize reliability to ensure
system can cause interruptions, ranging from that customer reliability requirements are satisfied while
inconveniencing a small number of local residents, to a incurring the lowest possible costs.
major and widespread catastrophic disruption of supply. The objective of evaluating the distribution network
It is stated the 80% of all interruptions occur due to reliability is therefore to:
failures and breakdowns in the distribution network [2]. (i) identify subsystems and circuits with
Since a power system represents one of the critical substandard performance and to ascertain the
layers or infrastructure of a modern society, the ability causes, this aids the utility in taking steps for
to predict the behavior of the power system given the prevention of serious interruptions.
state (or states of various subsystems) is vitally (Improvements are made before a fault occurs
important for the electric utility, in order to reduce rather than waiting for a problem to occur and
probability, frequency and duration of these events and then making necessary changes)
to ameliorate their effects. (ii) reduce the frequency, and the duration of
___________________________________________ interruptions
M.Emjedi is with Electrical Engineering Department, University (iii) minimize costs associated with an interruption
of Cape Town, Cape Town 7701, South Africa (e-mail: both from the utility and customer point of
rustumemjedi@gmail.com). view. Minimization of cost associated with the
K.Awodele is with Electrical Engineering Department, University
of Cape Town, Cape Town 7701, South Africa (e-mail: loss of the ability to sell power and costs
kehinde.awodele@uct.ac.za) associated with the loss of productivity or
S.Chowdhury is with Electrical Engineering Department, inconvenience to customers allows for a
University of Cape Town, Cape Town 7701, South Africa (e-mail: desirable level of reliability to be achieved [6].
sunetra.chowdhury@uct.ac.za).
S.P.Chowdhury is with Electrical Engineering Department, (iv) evaluate all contingencies and their effects on
University of Cape Town, Cape Town 7701, South Africa (e-mail: the distribution network in order to mitigate
sp.chowdhury@uct.ac.za) and find a reliable solution.

978-1-4244-6551-4/10/$26.00 ©2010 IEEE

Distribution network reliability evaluation can be
divided into two basic segments of measuring past
performance and predicting future performance. Most
utilities collect data on past performance and display the
results using a range of statistics.

Reliability indices are typically computed by utilities at

the end of each year by using historical outage data
recorded in distribution outage reports [7]. This is
important so that utilities know how their systems are
performing. Such reliability indices only show a
system’s past performance, i.e. it is impossible to use
those techniques to make predictions on the system’s
future performance when changes on the topology,
protection schemes and switching are undertaken in the
network [8].

Though historic methods are widely used, their

drawback is that the data may not be readily available or
accurate. Also, the approach is reactive in a sense that
the utility waits for the problems to occur and bases the
necessity of improvements on past performance. Thus
this method does not provide improvement for the
system in a consistent attempt to avoid future problems.

Fuzzy approach is used in this paper to evaluate the

reliability of the distribution network. Fuzzy logic (FL)
is a tool used to examine the influence of the various
subsystems that comprise the distribution network in Figure 1: Fuzzy model overview
terms of reliability. The reliability of the subsystems is
Level 1 consists of four separate independent fuzzy
complex in nature as there are many overlapping
inferencing systems namely:
conditions which affect the reliability of that subsystem.
Since many of these conditions can be expressed in (i) Reliability due to network configuration
linguistic terms, the reliability of a subsystem is easily
evaluated using fuzzy based approach. (ii) Reliability due to distribution substation
(iii) Reliability due to maintenance policy and
The basic concept underlying FL is that of a linguistic network topology
variable, that is, a variable whose values are words
rather than numbers. In effect, much of FL might be (iv) Reliability due to weather
viewed as a methodology for computing with words The model assumes that each Fuzzy Inference System
rather than numbers. The use of fuzzy logic as an (FIS) is independent from the other, which simplifies
evaluation tool has distinguished characteristic, this the model greatly. Since each FIS is handled and
technique aims at exploiting the tolerance for evaluated separately this reduces the amount of inputs
imprecision and uncertainty to achieve tractability, into the FIS which results in fewer IF-THEN
robustness and a low solution cost [9]. In doing so, it statements, and reduced complexity.
mimics the ability of the human mind to learn and make
rational decisions in an environment of uncertainty and All the FIS use Mamdani's fuzzy inference method,
imprecision. since we expect the output membership functions to be
fuzzy sets, which means that the reliability of the
The determination of the actual reliability of the system distribution network is not defined by crisp boundaries
may not be as valuable as the determination of where to between the systems rather it consists of overlapping
spend money to obtain the most improvement in boundaries which define the reliability.
reliability.
II. FUZZY ASSESMENT MODEL
2.1 Reliability due to Network Configuration
The approach described in this paper examines the
This fuzzy system makes a direct assessment of the
influence of several configuration variables on
circuit configuration and hazards, and assigns to each
reliability. Because of the difficulty in explaining the
feeder a relative reliability index (between 0 and 1) by
influence of each of these variables on reliability, fuzzy
expressing the configuration variables mathematically
logic is used to mathematically describe them. The
using fuzzy logic. This is useful as the determination of
fuzzy model consists of two levels of inferencing
the highest risk feeders and the feeders which benefit
systems as shown in Fig 1.
the most from a given reliability improvement project,
can be compared to the reliability of the distribution
network as a whole, to see how viable the improvements Protected fuses are described as fuses protected from
are. momentary faults by the use of an upstream recloser
with an overreach setting. Thus when a fault takes
A. Age
place, the recloser opens the circuit. After a time delay,
The performance of distribution components as they the recloser closes to energize the circuit. If the fault is
reach the end of their life may be significantly degraded. temporary, the circuit returns to normal operation.
Considering the age is valuable for a successful However, if the fault is permanent, the fast acting
reliability assessment. From conversations with utility characteristics of the recloser are disabled and the fuse
engineers it was found that the performance of is allowed to melt. If the fuse is unprotected, it may melt
distribution components deteriorates as they reach thirty even during temporary faults. Thus there is an increased
years of age [6]. The linguistic variable Age, is risk in sustained interruptions because of the time
represented by the term set. required to replace the fuse. If the fault is beyond the
recloser but before the fuse, the recloser opens and
Age = [Very Young, Young, Middle-Aged, Old, Very
recloses whenever there is a fault. If the fault persists,
Old] the recloser is locked out in the open position after a
The universe of discourse is defined over [0, 30] years few reclose operations. A circuit breaker is a device that
of age. Although the performance of a distribution will open the circuit whenever there is a fault. It is
component only deteriorates when it is in the Very Old assumed that it does not have the reclosing operation.
category, other categories are included for complete Therefore, the circuits fed by it will have a higher risk
representation of the age variable. of interruption compared to a recloser. However the
total interruption caused by a breaker is less than a fuse
B. Exposure because it can be opened and closed remotely [6].
The exposure of a distribution circuit is defined as the E. Conductor Type
number of trees per kilometer with the potential to touch
exposed overhead conductor. The motivation for the use The conductor type variable describes the
of this variable is the large number of momentary and susceptibility of the conductors to storm or wind
sustained interruptions caused by tree limbs contacting damages. Smaller conductor sizes will be more
the conductor. The linguistic variable Exposure, is vulnerable to damages by storms and wind than larger
represented by the term set: conductors. Underground cables are not subjected to
wind damage, but there is still some risk associated with
Exposure = [Very Low, Low, Average, High, Very damage of an underground cable to lightning and
High] insulation deterioration. This variable is handled the
The universe of discourse for this variable is defined same as for sectionalizing devices where a bell shaped
over [0, 200] trees per kilometer of conductor. membership function is tuned in order to obtain a
Boolean fuzzy set. The Boolean fuzzy set is described
C. Length by:
The length of the distribution circuit section in Conductor type = [underground cable, 636, 394, 336,
kilometers is another variable included in the fuzzy 4/0, 3/0, 2/0, 4, 6]
model. The risk of interruption is assumed to be
increased by an increased length of the section or Except for the underground cable, the other sizes are
circuit. The linguistic variable Length is represented the either area in MCM or conductor number of ACSR
term set: conductors
Length = [Very Short, Short, Medium, Long, Very F. Rule Base (If – Then statements)
Long] The basic function of the rule base is to represent the
The universe of discourse for the length variable is knowledge of an experienced engineer. The derivation
defined over [0, 2] kilometers of primary conductor. of the rules is accomplished by examining the
experience based knowledge in linguistic variable based
D. Sectionalizing Devices terms. The compound rules developed for the system
Within the fuzzy model, this variable is handled relating the five inputs to the output reliability are not
differently, because it is not continuous over the shown.
universe of discourse. Therefore, the representation of G. Output (Reliability due to network
the sectionalizing devices within the Fuzzy Logic configuration)
Toolbox (FLT) is accomplished with the use of Boolean
fuzzy sets. The Boolean fuzzy set for sectionalizing The result of the defuzzification in the FLT is the
devices is described as: reliability of the network as a result of the network
configuration. The output consists of five triangular
Sectionalizing device = [Circuit breaker, Recloser, membership functions. The linguistic variable is
Protected fuse, Unprotected fuse] represented by the term set:
A generalized bell shaped membership function is used Reliability = [Very Low, Low, Average, High, Very
as it can be tuned to represent the Boolean fuzzy set. High]
2.2 Reliability due to Distribution Substation substation reliability. The linguistic variable is
represented by the term set:
A distribution substation is where a power system
transitions from a networked topology to a radial MTTR = [Very Quick, Quick, Average, Long, Very
topology. As such, it can have a significant influence on Long]
distribution system reliability and, ultimately, customer
D. Mean Time Between Repairs (MTBR)
satisfaction [10]. From a reliability point of view, this is
significant because any outage that occurs on a radial This is the expected time between occurring faults, or
system will cause certain customers to experience a the rate/frequency that faults occur in a substation. It
service interruption. can be seen that the longer the time between faults, the
lower the rate of interruptions, which has the effect of
There are various types of distribution substation
increasing substation reliability. The linguistic is
arrangements, ranging from very simple to very
represented by the term set:
complex. This fuzzy system makes a direct assessment
of the distribution substation and its components, by MTBR = [Very Short, Short, Average, Long, Very
considering the substation in isolation, the system Long]
assigns a reliability index (between 0 and 1) by
expressing the distribution substation variables The universe of discourse is [0, 1] where 0 represents a
mathematically using fuzzy logic. very short time between repairs and 1 represents a very
long time between repairs
A. Age
E. Mean Time To Switch (MTTS)
Most utilities have a variety of substation ages in
their systems. Some may be very new with automation This is the expected time it will take after a failure
incorporated into their design and the latest technology. occurs for a sectionalizing switch to be toggled. Much
Others may be relatively old without automated newer substations have very fast control systems in
switching and mechanical relays still in use. These place which greatly reduces this switching time. The
substations require crews to be dispatched for switching faster the switching time, the fewer customers are
when a failure occurs. The linguistic variable is affected by the interruption, this increases the substation
represented by the term set: reliability and vice versa. The linguistic variable is
represented by the term set:
Age = [Very Young, Young, Middle-Aged, Old, Very
Old] MTTS = [Instantly, Average, Prolonged]
The universe of discourse is defined over [0, 30] years The universe of discourse is [0, 1] and a Gaussian curve
of age. is used as the membership function (MF).
B. Automation 2.3 Reliability due to Weather
A major problem encountered in the reliability area of The weather environment can severely impact the
distribution systems is how to reduce the number of performance of an overhead distribution system and an
interruptions experienced by customers. These electric utility's operational ability. The likelihood of
reductions can be obtained with the substitution of system failure increases due to enhanced line failure
equipment with high failure rates and or by increasing rates during bad weather periods. Reliability appraisals
maintenance staff. These are expensive solutions. The without incorporating weather conditions can be quite
automation of protection has proven to be a feasible and optimistic and affect planning and design decisions.
efficient alternative to solve the problem [11].The Recognition of various weather contributions to the total
linguistic variable is represented by the term set: system performance indices help to pinpoint situations
Automation = [No automation, Semi-automation, Full where investment may provide maximum reliability
automation] improvement.

This fuzzy system makes a direct assessment of the

The substation automation has an effect on reliability, if
the substation is fully automated the switching time and reliability due to the effects of weather; the system
fault isolation time is greatly reduce, which has the assigns a reliability index (between 0 and 1) by
expressing the weather variables mathematically using
effect of increasing the reliability of the distribution
substation and has an overall effect of increasing fuzzy logic.
distribution network reliability. A. Wind
C. Mean Time To Repair (MTTR) Weather phenomenon, lightning, wind and icing
This is the expected time it will take a crew to repair cause most of the weather related failures. Although
a component outage and restore the system to its normal lightning, wind and icing can directly cause failures in
overhead feeders, they mostly cause problems by
operating state. It can be seen that the longer the crew
takes to repair the faulty substation equipment; the affecting trees around feeders, such as branches
longer the duration of the interruption, thus lowering the dropping and then touching the feeders. Since weather-
related failures take up a large percentage of the total
failures, it is useful to develop good models for Policy 3: This is the far superior maintenance policy. It
prediction of weather-related failures [12]. The involves reliability centered maintenance where
linguistic variable is represented by the term set: equipment is repaired before it fails, this improves
network reliability.
Wind = [No wind, Calm, Breezy, Strong, Gale force]
The linguistic variable is represented by the term set:
B. Lightning
Maintenance = [Policy 1, Policy 2, Policy 3]
Lightning is another important input variable;
overhead lines situated in areas that have a high B. Network Topology
lightning activity are more susceptible to being struck
causing line damage which in turn reduces the network The network topology is either radial or meshed, where
reliability. The linguistic variable is represented by the radial is the least reliable as there are no other sources
term set: of energy if there is a fault on a feeder.

Lightning = [No lightning, Lightning] 2.5 Distribution Network Reliability

The MF used is a trapezoidal curve. The universe of This level consists of a single inferencing system; it
discourse ranges from [0, 1] where 0 represents no takes as its inputs the outputs of level 1.
lighting and 1 represents severe lightning storm.
All the inputs are triangular membership functions, with
C. Rule Base (If – Then statements) a universe of discourse ranging from [0, 1]. All inputs
consist of the same linguistic variable set namely:
As earlier stated, the basic function of the rule base is
to represent the knowledge of an experienced engineer. Input = [Very Low, Low, Average, High, Very High]
The derivation of the rules is accomplished by
examining the experience based knowledge in linguistic Also, the output function is the same as the input
variable based terms. Each rule has the same weighting. function, so it is also a triangular membership function,
The compound rules developed for the system relating with the same universe of discourse and the same
the two inputs to the output reliability are: linguistic variable set.
(i) If (Wind is No wind) and (Lightning is No III. APPLICATION TO UTILITY DATA
lightning) then (Weather reliability is Very When a system is modelled, the reliability predicted
High) by the assessment model should agree with the
(ii) If (Wind is Gale force) and (Lightning is historical values. If so, a certain level of confidence in
Lightning) then (Weather reliability is Very the model is achieved and more specific reliability
low) results (e.g., the reliability of a specific load point or
impact of a design change) can be trusted to a higher
(iii) If (Wind is Gale force) then (Weather degree. When the confidence has been achieved and
reliability is Low) predicted results match historical results, the reliability
(iv) If (Wind is Breezy) then (Weather reliability is model is said to be validated. This section presents the
High) method of using the constructed fuzzy model to make a
predictive assessment of the Phillipe 1 distribution
(v) If (Lightning is Lightning) then (Weather network reliability.
reliability is Low)
The predicted results are compared to the expected
2.4 Reliability due to Maintenance Policy and Network historical values, the smaller the error between the two
Topology values, the greater the confidence when using the model
to predict.
Maintenance is the activity of restoration where an
unfailed device has its deterioration arrested, reduced or 3.1 Reliability from Historical data
eliminated [13]. Its goal is to increase the duration of
For this network section the system indices are
useful component life and postpone failures that would
calculated using the historical data (Table 1)
require expensive repair thus increasing the reliability of
the distribution network. Table 1: Historical data for Phillipe 1
Total Interruption
Loss Interruption
A. Maintenance Policy Year
(KVA)
load Duration
number
(KVA) (hrs)

There are three maintenance policies namely: 2003 4000 34533 21.63 4
2004 2400 34533 8.7 3
Policy 1: No maintenance carried out at all; this has the
2005 900 34533 5.18 2
effect of reducing the network reliability
2006 24495 34533 2.66 3
Policy 2: Maintenance is carried out in a reactive 2007 17450 34533 16.12 5
manner, or when equipment is close to failure. This
2008 14618 34533 69.04 9
does not improve the network reliability but keeps it
constant Total 63863 207198 123.33 26
The system indices ASIFI (Average System Interruption A. Reliability due to Network Configuration
Frequency Index) and ASIDI(Average System
Interruption Duration Index) were calculated for the The input of fuzzy system 1 consists of five variables:
period from July 2003 to July 2008 using the formulae exposure, length, age, sectionalizing device and
below: conductor type. Fuzzy system 1 maps these five input
variables to the reliability due network configuration.
The reliability due to network configuration is
represented in Table 4.
B. Reliability due to Distribution Substation
The input consists of five variables: Age,
Automation, MTTR, MTBR and MTTS. the outputs are
shown in Table 5.
The results are shown in Table 2. Fuzzy logic can be
used in various applications; it can be used to make Table 4: Reliability due to Network Configuration
predictions or as a functional mapping between inputs Year Reliability
nd outputs. The two indices were fed into the fuzzy
2003 0.609
model which outputted the reliability of the network in
terms of the reliability indices as shown in Table 3 2004 0.594
below. 2005 0.572
Table 2: System reliability indices 2006 0.547
ASIDI (hrs
ASIFI 2007 0.523
per load
Year (Interruptions
KVA
per year) 2008 0.508
interrupted)
2003 0.12 0.39
Table 5: Reliability due to Distribution Substation
2004 0.07 0.2
Year Reliability
2005 0.03 0.04
2003 0.5
2006 0.71 0.75
2004 0.481
2007 0.51 1.26
2005 0.442
2008 0.42 2.04
2006 0.415
2007 0.379
Table 3: Phillipe 1 network reliability
2008 0.314
Year Reliability
2003 0.70
2004 0.75
C. Reliability due to Maintenance Policy and
2005 0.77 Network Topology
2006 0.43
The input consists of two variables: Maintenance Policy
2007 0.40 and Network Configuration. Table 6 below shows the
2008 0.38 outputs.
Table 6: Maintenance Policy and Network Topology
3.2 Predictive Fuzzy assessment Year Reliability

The fuzzy model of section 2 is now used to predict 2003 0.499

the reliability of Phillipe 1 network. The model cannot 2004 0.499
predict the reliability directly. The reliability of four
2005 0.499
separate subsystems which comprise the distribution
network of Phillipe 1 is predicted first. This is done as it 2006 0.499
is assumed that the four subsystems are independent of 2007 0.499
each other. The model could be manipulated to predict
the reliability directly but this would increase the 2008 0.499
number of fuzzy rules and it would make for a less
tractable solution. The reliability of the subsystems is D. Reliability due to Weather
then used as inputs into level 2 of the model which
maps the sub-system’s reliability to the network The reliability due to the Weather is calculated next,
the output of this fuzzy system is used as an input into
reliability. Due to the unavailability of real input data
many assumptions were made about the input variables level two. The input consists of two variables: Wind and
in A-D below. Lightning. The output is shown in Table 7 below.
 Table 7: Reliability due to weather Using the predicted values for 2003 - 2007 and the
Year Reliability expected values for the same period, we try to find the
simplest mapping. We test this mapping on the data for
2003 0.75
2008.
2004 0.75
This method proved fruitless to optimize the fuzzy
2005 0.75 model to increase the predictive power. This method
2006 0.75 will work extremely well with this problem, but it falls
short due to an important constraint, the lack of
2007 0.75
insufficient data points.
2008 0.75
The problem is that those data points are far too few to
give a realistic distribution. The five points could be
E. Distribution Network Reliability strays or outliers and tell us nothing about the true
All of the outputs from level one are fed into level distribution.
two and the output of level two is the reliability of the To solve this problem, data for the last 50 years need to
entire distribution network. be collected, the expected values and the predicted
From Table 8 it can be seen that the error between the values for those 50 years have to be plotted to give a
predicted values and the expected values are not more realistic distribution between predicted values and
acceptable. This signifies that our model is incomplete expected values. We could use those data points to find
and should be optimised in an error reducing manner the functional mapping.
TABLE 8: Predicted reliability for Phillipe Network B. Differential Evolution

3.3 Optimization Techniques Differential evolution is a very powerful optimization

technique. In this method, a weight is assigned to each
Error input. The error between the predicted values and the
Year Historical
Predicted between Percentage expected values is reduced by changing the input
Reliability Historical error weights. The task is to find the set of values for the
Reliability
and predicted
variable that minimizes the error
2003 0.70 0.75 -0.05 6.70
Error = f (w0, w1, w2......w18)
2004 0.75 0.68 0.07 9.37
There is no reason to expect the error to be everywhere
2005 0.77 0.53 0.24 31.20
smooth and continuous in the space – there may well be
2006 0.43 0.49 -0.06 14.79 jumps and discontinuities at specific locations. But if
2007 0.40 0.41 -0.01 2.74 the function models real world phenomenon we expect
that the space will be dominated by regions of
2008 0.38 0.35 0.03 6.90
continuity or by connected regions where error is mostly
large or small [14].
Since many assumptions were made about the inputs
there exists an error between the predicted values and The fuzzy model would have to be programmed, as the
the expected values. Since the assumptions are kept FLT does not contain a method of changing the input
constant there must exist a functional mapping between weights. This optimization method Will be the subject
the predicted value and the expected value. Once this of another paper.
mapping is found the predicted value can be fed into IV. CONCLUSION AND RECOMMENDATIONS
this mapping and the output will be the expected value
or a value very close to it. 4.1 Conclusion
A. Regression
The paper represents a method for predicting the
This is a very simple yet powerful method of reliability of the distribution network, by estimating the
optimization. Given a set of training data of input reliability of the subsystems which comprise the
vectors (predicted values) and a vector of target values distribution network. Fuzzy logic was used to predict
(expected values) the aim is to find a function F (as the reliability as there was no mathematic function
simple as possible) which approximately maps the describing the relationship between the subsystems and
predicted values to the expected values. We will then the network reliability.
predict the likely output expectednew from the new input
predictednew. The proposed method has the advantage of enabling the
possibility of adding new fuzzy systems to the existing
The simplest function F would be a linear weighted sum model, in order to account for factors that were not
of the input variables. Various general functions can be included in the original model. These new fuzzy
used for the functional mapping: systems can be added into fuzzy level 1 or 2. The reason
(i) Linear function is that all subsystems are assumed to be independent
from one another.
(ii) Cubic function
The model is capable of providing answers to such
(iii) Quadratic function questions as what type of conductor should be used,
what is the lowest risk length of feeder, what is the most 8. C.L.C de Castro, A. B Rodrigues and M.G Silva, “Reliability
evaluation of radial distribution systems considering voltage
efficient sectionalising device to use, etc as their effects drop constraints in the restoration process”, 8th international
on the network reliability can be assessed which allows conference on probabilistic methods applied to power systems,
for the most cost effective solution to be chosen that Iowa state university, Ames, Iowa, September 12-16, 2004
provides the highest reliability 9. M. Jamshidi, N Vadiee and T.J Ross, “Fuzzy logic and control:
software and hardware applications”, Prentice Hall, New Jersey,
1994
A very important conclusion drawn is the lack of
10. R.E Brown and T.M Taylor, “Modeling the impact of
accurate data. This is most evident when the predicted substations on distribution reliability”, IEEE Transactions on
reliability does not agree with the historical expected Power Systems, Vol. 14, No. 1, pp 349 – 354, February 1999
values. Most utilities do not have substantial amount of 11. A.M.L da Silva, A.M Cassula, L.C Resende, R.M Almeida, M.T
Yamasaki and M.F.P Gustavo, “Reliability evaluation of
historical component reliability data. Nearly all utilities
distribution systems considering automatic reclosers”, 8th
however, have historical system reliability data. international conference on probabilistic methods applied to
power systems, Iowa state university, Ames, Iowa, September
The flexibility of the fuzzy model allows for its 12-16, 2004
application in simplification of complex concepts into 12. Y. Zhou, A. Pahwa and S.Das, “Prediction of weather related
easily handled models. Because of the simplicity of the failures of overhead distribution feeders”, 8th international
conference on probabilistic methods applied to power systems,
model, the speed of computation is very fast. Thus, Iowa state university, Ames, Iowa, September 12-16, 2004
large systems could be evaluated easily with minimal 13. J. Endrenyi, G.J Anders, L. Bertling and B Kalinowski,
computation time. “Comparison of two methods for evaluating the effects of
maintenance on component and system reliability”, 8th
4.2 Recommendations international conference on probabilistic methods applied to
power systems, Iowa state university, Ames, Iowa, September
12-16, 2004
There exists a great need for the utility to collect 14. J. Greene, Neural Networks and Fuzzy logic class notes, EEE
more accurate component data which could be used 4096S, 2008
more easily for the fuzzy inputs to increase the accuracy
of the model. If there is a very accurate component VII. BIOGRAPHIES
reliability data base, a neuro-fuzzy inference system can
be used where the rules are generated from the data, M.R.Emjedi is currently a student of Electrical Engineering
these rules can be appended with rules from an expert, Department of University of Cape Town, South Africa. Email:
rustumemjedi@gmail.com)
which would increase the prediction accuracy of the
model. The optimisation techniques could still be K.Awodele received her M.Sc Eng (Electrical Power & Machines) in
applied to the neuro-fuzzy system which would increase 1991. After working for several years in industry, she currently
the accuracy. Lectures in the Electrical Engineering Department of University of
Cape Town, South Africa. Her interests are in power system reliability
and demand side management Email: kehinde.awodele@uct.ac.za
V. ACKNOWLEDGMENT S.Chowdhury received her BEE and PhD in 1991 and 1998
respectively. She was connected to M/S M.N.Dastur & Co. Ltd as
The Authors gratefully acknowledge the management Electrical Engineer from 1991 to 1996. She served Women’s
of the Faculty Engineering and the Built Environment of Polytechnic, Kolkata, India as Senior Lecturer from 1998 to 2006. She
the University of Cape Town for providing necessary is currently the Senior Research Officer in the Electrical Engineering
infrastructure and support for undertaking the research Department of The University of Cape Town, South Africa. She
became member of IEEE in 2003. She visited Brunel University, UK
and The University of Manchester, UK several times on collaborative
research programme. She has published two books and over 55 papers
VI. REFERENCES mainly in power systems. She is a Member of the IET (UK) and IE(I)
and Member of IEEE(USA). She is acting as YM Coordinator in
Indian Network of the IET(UK). Email:
1. A.I Petroianu, “Definition and salient features of a power sunetra.chowdhury@uct.ac.za
system”, EEE 4090F, UCT, 2009
2. R. Billinton and J.E Billinton, “Distribution system reliability S.P. Chowdhury received his BEE, MEE and PhD in 1987, 1989 and
indices”, IEEE Transactions on Power Delivery, Vol. 4 , No.1, 1992 respectively. In 1993, he joined E.E.Deptt. of Jadavpur
pp 561 – 568 , January 1989 University, Kolkata, India as Lecturer and served till 2008 in the
3. J. Paska, “Methodologies and tools for electric power system capacity of Professor. He is currently Associate Professor in Electrical
reliability assessment on HL I and HL II levels” 9th international Engineering Department in the University of Cape Town, South
conference on Electrical power quality and utilisation, Africa. He became IEEE member in 2003. He visited Brunel
Barcelona, October 9-11, 2007 University, UK and The University of Manchester, UK several times
4. M.G Da Silva, A.B Rodrigues, C.L.C de Castro, A.C Neto, E.A on collaborative research programme. He has published two books
Moutinho, N.S.A Neto and A. B Cavalcante, “Reliability and over 110 papers mainly in power systems and renewable energy.
assessment of distribution networks in Brazil’s Northeast”, 8th He is a fellow of the IET (UK) with C.Eng. IE (I) and the IETE (I) and
international conference on probabilistic methods applied to Member of IEEE (USA). He is a member of Knowledge Management
power systems, Iowa state university, Ames, Iowa, September Board and Council of the IET (UK).Email: sp.chowdhury@uct.ac.za
12-16, 2004
5. C.T. Gaunt and K. Awodele, ‘”Power system reliability and
adequacy”, EEE 4089F class notes, 2009
6. B.P Lang and A. Pahwa, “Power distribution system reliability
planning using a fuzzy knowledge-based approach”, IEEE
Transaction on Power Delivery, Vol. 15, NO. 1, pp 279 – 284,
January 2000
7. R.E Brown and J.R Ochoa, “Distribution system reliability:
default data and model validation”, IEEE Transactions on Power
Systems, Vol. 13, NO. 2, pp 704 – 709, May 1998