See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/320556464

Fast and Efficient Placement of Fault Indicators Based on the Pattern Search
Algorithm

Conference Paper · October 2017

DOI: 10.1109/PEE.2017.8171695

CITATIONS READS

5 157

2 authors:

Cedomir Zeljkovic Predrag Mršić

University of Banja Luka University of Banja Luka
46 PUBLICATIONS 156 CITATIONS 29 PUBLICATIONS 90 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Cedomir Zeljkovic on 17 March 2021.

The user has requested enhancement of the downloaded file.

 Fast and Efficient Placement of Fault Indicators
Based on the Pattern Search Algorithm
Čedomir Zeljković, Predrag Mršić
Faculty of Electrical Engineering
University of Banja Luka
Banja Luka, Bosnia and Herzegovina
[email protected]

Abstract—This paper proposes a fast and efficient strategy to Supplied) or SAIDI (System Average Interruption Duration
determine the number and locations of fault indicators (FI) in Index) are directly included in the objective functions to be
medium voltage distribution networks. The objective function minimized [7]-[11]. The second type is based on an indirect
takes into account the most important characteristics of the approach as the authors assemble auxiliary objective functions
network such as topology, non-uniform failure rates of the lines that are much more simpler for optimization while, on the
as well as the power demand and number of customers. For other hand, the obtained results are still located in the vicinity
seeking the minimum of the objective function, a straightforward of the optimal solution [12]-[14]. The methodology proposed
algorithm based on the Pattern Search (PS) is developed. The in this paper is built by following the latter principle.
methodology is tested on a real distribution network, showing its
great potential to improve the reliability indices at the lowest The research in this paper uses the idea presented in [14]
investment costs. as the starting point. Our improvements went into two
directions, both in terms of the objective function and the
Keywords—distribution network; reliability indices; fault optimization method. The objective function is modified in
indicators; pattern search; order to include influential input parameters such as non-
uniform failure rates across the network and the impact of the
I. INTRODUCTION reliability of lateral branches. The results show that, by these
Occurrences of faults in the power distribution system is modifications, the reliability indices ENS and SAIDI may be
unavoidable due to many reasons. The situations such as reduced up to 15%. As the second improvement, a fast and
severe weather, contact with trees, human errors or aging of efficient method to minimize the objective function is
the equipment may lead to interruptions in the power supply, proposed. The method is based on the Pattern Search (PS)
thus decreasing the reliability offered to the customers [1]. method containing adaptations to the specific problem.
One of the mechanisms for enhancing the reliability indices is Finally, the proposed methodology is tested on a real medium
the acceleration of the process of finding the faults [2]. voltage distribution network.

Fault indicators (FIs) represent a cost effective measure for II. PROBLEM DEFINITION AND SOLUTION METHODOLOGY
improving the process of fault localization. Their task is to
detect the passage of fault current through a section of a feeder A. Problem Formulation and Basic Assumptions
on which they are installed. By placing these devices at
suitable locations in a distribution network, it is possible to Fault indicators are devices which provide visual or remote
accelerate the process of finding the fault, which in this case is information that they detected a fault current. They therefore
between the last indicator that detected the passage of fault do not help in reducing the frequency of the outages in the
current and the first following that did not [3]-[6]. network but represent a good support for shortening the
interruption duration. By analyzing which FI is tripped and
In this paper we propose a methodology for determining which is not, the area to be searched by the repairing crew
the number and positions of fault indicators in medium voltage may be substantially diminished and thus the outage time
distribution networks. The main goal is to achieve the techno- appropriately shortened. In three examples shown in Fig. 1, it
economic balance, that is to obtain the maximum can be seen that the utilization of three FIs helps in reducing
improvement of reliability indices while using minimum the fault searching zone by approximately 66%. Consequently,
number of installed fault indicators. the process of fault isolation and supply restoration to healthy
According to our literature review, the principles of feeder sections are accelerated which improves the system
optimal placement of FIs can be classified into two distinct reliability. The system average outage duration and the total
types. The first type includes direct optimization methods, energy which is not supplied to the customers are most
where the reliability indices such as ENS (Energy Not commonly assessed by using the SAIDI and ENS reliability
indices defined as follows
This work has been financially supported in part by the Ministry of
Science and Technology of the Republic of Srpska under the project INGRID
8300575.
 therefore to determine how many fault indicators to place into
 ENS  U i Li  
a distribution network and where they have to be located in
i
order to sufficiently improve the reliability indices at a
 U i N Li reasonable investment cost.
 SAIDI  i  
 N Li The main target for the methodology are the medium
i voltage overhead distribution networks, having the radial
topology and one supply point. The initial assumption is that
where Ui is the annual outage time, NLi is the number of the placement space is limited to a set of lines called the main
customers, and Li is the average load determined for each load feeder. The main feeder is a backbone of the system, defined
point i in the network. as a set of lines from the supply point in the substation to the
electrically farthest customer [14]. An illustration of a test
Substation system with a main feeder identified is given in Fig. 2.
Main feeder Main feeder Network
The logic of the existence of the optimal FI investment
Lateral solution is illustrated in Fig. 3. Line BFI represents an
Substation economic value of the benefits of improved reliability, while
Main feeder Main feeder Network line CFI represents the investment costs. The profit, defined as
the difference between the benefits and costs, is depicted by
Lateral the line YFI. While investment costs increase approximately
Substation linearly with the number of indicators, the progress of benefits
Main feeder Main feeder Network of improved reliability relatively quickly enters saturation. It
turns out that, for instance, with the installation of the fifth
Lateral indicator, no further noticeable improvement is achieved [3].
That is the reason why there exists the optimum number of FIs
Circuit breaker Tripped FI Non-tripped FI (Nopt) which guarantees the maximum profit YFImax. The
methodology proposed in this work was designed to examine
Fig. 1. The role of the fault indicators in a process of fault localization. the variations of the number of indicators from 1 to 10. For
each of the ten options, the optimal locations are determined
and benefits and costs are computed. The most profitable case
from this finite set of variants is declared to be the optimum
solution.

Fig. 3. The principle of finding the optimum number of fault indicators.

B. Starting Point of the Approach

The initial idea is introduced in [14] and the main facts
necessary for further understanding will be summarized in this
section. The optimization is based on the indirect principle. An
Fig. 2. Test distribution network with the main feeder highlighted. auxiliary objective function is created to be simple enough for
performing a fast search optimization process while keeping a
It is neither economical nor necessary to install the FIs on good correlation with the target reliability indices. It is
each bus on the main feeder and its laterals. Our goal is recognized that load, number of customers and mutual
distances between the fault indicators are the inputs of the x max
crucial importance. The following two principles are noticed:  f obj   1l x    2cx  d x   dx  
0
 If the load and the number of customers located
downstream from some feeder bus are high, then the
bus is a good candidate to install a fault indicator; 0.25
 If another fault indicator has already been installed in 0.20
the vicinity, the considered bus becomes less favorable
position.

(α1 l(x)+α 2 c(x))d(x)

0.15
For modeling a first principle, a curve (α1·l(x)+α2·c(x))
0.10
presented in Fig. 4 is constructed. It represents a combination
of the normalized load l(x) and the normalized number of
0.05
customers c(x) in terms of location on the main feeder x. The
constants α1 and α2 are the weighting coefficients for the load 0.00
and number of customers, respectively. If the load profile and 0 2 4 6 8 10 12 14 16 18 20
the number of customers have the same level of priority, the Location on the Main Feeder (km)
weighting coefficients are set to 1.0. On the other hand, if one
of the two criteria is more important, the corresponding Fig. 6. Example of a fitness function (α1·l(x)+α2·c(x)) computed for the test
weighting coefficient is set to an appropriate higher value. network with one FI installed at an arbitrary location 5 km away from the
substation. The goal is to minimize area under the curve.
The second principle is taken into account by forming a
graph of the relative distance from the location of the fault to
the nearest device able to indicate that fault. This distance, C. Proposed Modification of the Objective Function
d(x), is also computed in terms of location on the main feeder. The main advantage of the objective function defined
An illustration is given in Fig. 5, at the assumption that a according to equation (3) is its simplicity. On the other hand,
circuit breaker is installed at the sending end of the feeder and the number, load and location of the customers are not the
a fault indicator is located 5 km downstream. only parameters that are influential for the efficient placement
of fault indicators. In this paper, therefore, two slight
2
modifications of the objective function are proposed, trying to
1.5 achieve a noticeable improvement at a reasonable increasing
of complexity.
α1 l(x)+α2 c(x)

1 The first modification is introduced in order to take into

account the impact of lateral branches. It is obvious from the
0.5
Fig. 5 that the value of the function d(x) in some point of a
0 potential fault x, expresses a relative distance to be searched
0 2 4 6 8 10 12 14 16 18 20 by the crew to discover the fault. In that distance, a fraction of
Location on the Main Feeder (km) the main feeder is included but the lateral branches are not,
even though being a part of the network between two fault
Fig. 4. Combination of normalized load profile and the number of customers. indicators. We propose that half lengths of the laterals to be
superimposed to the function d(x) at the points of their
0.8 connection to the main feeder. The half lengths are used since
the expected time to find a fault equals one half of the time
0.6 needed to examine the lateral branch in total. After a proposed
Circuit modification, the distance function d*(x) should look like as an
d (x)

0.4 Fault Indicator

Breaker example shown in Fig. 7.
0.2 0.80
Normalized Distance, d *

0.0 0.60
0 2 4 6 8 10 12 14 16 18 20
(p.u.)

Location on the Main Feeder (km) 0.40

Fig. 5. Example of a distance function d(x) computed for a test network. FI is 0.20
installed at an arbitrary location 5 km away from the substation.
0.00
The fitness function is then defined as a product of 0 2 4 6 8 10 12 14 16 18 20
functions presented in Fig. 4 and Fig. 5, which is Location on the Main Feeder (km)
(α1·l(x)+α2·c(x))·d(x). An example is shown in Fig. 6. The aim
Fig. 7. Example of a distance function d*(x) after a modification with lateral
is to minimize the area under the fitness function that is to find branches taken into account. FI is still located 5 km away from substation.
the minimum of the following objective function:
 The second important fact to be considered is the non- The value of the objective function is calculated by
uniform failure rate across the network. Primarily, the failure numerical integration. Within the integration procedure, a
rate changes along the main feeder as we move from urban to careful attention should be paid to the spikes arisen due to
rural areas. Additionally, the failure rate of the lateral branches inclusion of lateral branches. The superimposed length and
are greater that the failure rate of the main feeder. That is the failure rate of the laterals, included in functions d*(x) and f*(x),
reason why we propose a failure rate function to be included should not depend on an integration step. For instance, the
in the objective function. Generally, the failure rate along the spike in the failure rate function is modeled as a rectangle
main feeder is expressed as a function of location - f(x). which area is equal to flateral and width equal to the step of
Lateral branches are modeled as the discontinuities that are integration. Theoretically, when the integration step tends to
superimposed to the failure rate of the main feeder. The failure zero, the value of superimposed failure rate of the lateral
rate of the laterals is computed by multiplying lengths (L) and connected to the point x, would be expressed by using the
failure rates (f) of all branches, i.e. Dirac delta function δ(x):

 f lateral   Li fi    *
f lateral x   f lateral   x   
i
The same principle is used also for the spikes which exist
The final failure rate function, which combines impacts of in the distance function d*(x).
both main feeder and its laterals, is designated with f*(x). An
example of the failure rate function determined for the test D. Optimization Method
network from Fig. 2 is shown in Fig. 8. Note that the For the minimization of the modified objective function a
logarithmic scale is used for y-axis, since the failure rates of simple method is proposed. It is based on a traditional pattern
the laterals are significantly higher than the failure rates search algorithm including some adaptations appropriate to the
describing the main feeder. assigned task.
1) The Principle of Pattern Search Algorithm
Failure Rate, f* (1/m·year)

1.0E+00 The pattern search is a single-point search optimization

method, applicable for a minimization of various non-
1.0E-01
continuous and non-differentiable functions [15]. Its working
1.0E-02 principle can be illustrated by using a two-dimension problem
shown in Fig. 10. At the very beginning, an initial guess of the
1.0E-03 optimal point is made. This point is designated as X0. In the
following step, a mesh around initial point is constructed
1.0E-04
(points marked with white circles). The values of the objective
0 2 4 6 8 10 12 14 16 18 20
Location on the Main Feeder (km) function is then evaluated for each point of the mesh. If some
points having smaller value of objective function are found,
Fig. 8. A failure rate function f*(x) determined for the test system. there obviously exist better points. One of them becomes a
new guess of the optimal point and the step of the mesh is
After all the proposed modification are taken into account, doubled. Otherwise, the chosen point is the best guess at the
a new objective function is assembled as follows: moment and the mesh step is halved to increase accuracy.
Such iterative procedure continues until the search converges
x max
to the optimal solution with desired accuracy. In some cases,
 1l x    2cx   d x  f x   dx 
* * *
 f obj   however, the procedure may diverge so the optimum point
0 cannot be found within the allowed number of iterations.

A graphic illustration of the modified objective function

calculated for the test network is provided in Fig. 9.
1.0E+00
1.0E-01
(α1 l(x)+α2 c(x))d* (x)·f*(x)

1.0E-02
1.0E-03
1.0E-04
1.0E-05
1.0E-06
1.0E-07 Fig. 10. Illustration of the pattern search on a two-dimension problem.
1.0E-08
0 2 4 6 8 10 12 14 16 18 20 2) Adaptations to the Specific Task
Location on the Main Feeder (km) The following assumptions are added to the traditional
Fig. 9. Example of the final fitness function with all modifications applied. pattern search to make it suit better to the given problem of
The area under the curve should be minimized. optimal placement of fault indicators:
  Initial value of mesh step is chosen so that the area 10
between the two neighboring indicators may be 9
searched in a couple of iterations. A good empirical

Number of Fault Indicators

8
value is LMF/(5NFI), where LMF is the length of the main
feeder and NFI is the number of the fault indicators. 7
6
 Due to limited search space, doubling the mesh during
the iteration procedure is rather avoided. On the other 5
hand, halving of the mesh step is necessary in order to 4
achieve an adequate accuracy.
3
3) Method Validation and Computational Requirements 2
A software solution is implemented on a personal Original objective function
1
computer (CPU: Intel Core i5-4300M 2,6 GHz RAM: 8 GB). Modified objective function
The procedure for a test network converges after several tens 0
of iterations which takes just a couple of seconds, regardless 0 5 10 15 20
of the number of FIs (from 1 to 10). Location on the Main Feeder (km)

The method converges to the global optimum point which Fig. 11. The suggestion for placement of the fault indicators obtained with the
is validated by the exhaustive search through all possible original objective function (black circles) and the modified objective function
combinations. The search resolution is assumed to be (white circles).
50 meters, as this is an average distance between poles in the 11
test distribution network. The good efficiency of the proposed

Location on the FI (km)

algorithm becomes more and more significant as the number 10
of FIs increases, since the time needed for the exhaustive 9
search grows exponentially. Testing all possible combinations 8
takes 1.5 seconds for one FI, while in the case of three FIs, the
procedure lasts more than 6 hours. For the cases with more 7
than three FIs, the "curse of dimensionality" makes the 6
application of the exhaustive search practically impossible. 5 0 5 10 15 20
Number of Iterations
III. RESULTS
Fig. 12. Illustration of the iterative procedure for placement of one FI.
A. Test System
One-line diagram of the test network is shown in Fig. 2.
This is a real 20 kV distribution network located in the C. Reliability Indices
Municipality of Banja Luka (Bosnia and Herzegovina), which In order to justify the proposed modification of the
supplies the customers both in urban and rural areas of the objective function, the reliability indices such as SAIDI and
municipality. The total length of all lines in the network is ENS are computed for both original and modified objective
65 km and the main feeder is 20.14 km long. There are 124 function. It was found that, by using the modified objective
20 kV buses and the total load in average is 1.92 MW. The function that takes into account the existence of lateral
other parameters are not given here due to a limited available branches and non-uniform failure rates across the network,
space. The results are therefore presented in a relative form, SAIDI and ENS can be reduced up to 15%. Fig. 13 shows the
showing the differences between the original and modified comparison of ENS indices computed for original and
methodology. modified objective function, depending on the number of FIs.
Fig. 14 presents the achieved improvement expressed in a
B. Placement of the Fault Indicators form of percentage.
The modified pattern search is conducted on the test 250
Energy not supplied, ENS

f obj
network. The objective function is minimized both in the 200 f * obj
original and modified form while the number of indicators is
(MWh/year)

varied from 1 to 10. The obtained results are summarized in 150

Fig. 11. It is obvious that the proposed modification of the 100
objective function considerably affect the placement of the
FIs. In a special case of using only one FI, the modification of 50
the objective function causes a FI location shifting from 0
7.3 km to 7.9 km. Changes in the location guesses during the 0 1 2 3 4 5 6 7 8 9 10
iteration process for a system with one FI are presented in Fig. Number of FIs on the main feeder
12. The initial guess at iteration 0 is 10.07 km which is half of
the main feeder length. Fig. 13. Energy not supplied (ENS) depending on number of FIs for the
original (fobj) and modified objective function (f*obj).
 15 enable placement of fault indicators on the locations that do
Decrease of ENS (%)

not belong to a main feeder.

10
V. REFERENCES
5 [1] J. Northcote-Green and R. Wilson, Control аnd Automation оf Electrical
Power Distribution Systems, CRC Press: Boca Raton, FL, 2007.
0 [2] J. Teng, W. Huang, and S. Luan, “Automatic and fast faulted line-
0 1 2 3 4 5 6 7 8 9 10 section location method for distribution systems based on fault
indicators,” IEEE Transaction on Power Systems, vol. 29, no. 4, pp.
Number of FIs on the main feeder 1653-1662, July 2014.
[3] H. Falaghi, M. -R. Haghifam, and M. R. Osouli Tabrizi, “Fault
Fig. 14. Relative improvement in ENS due to objective function modification. Indicators Effects on Distribution Reliability Indices”, CIRED 2005 -
18th International Conference and Exhibition on Electricity
Distribution, Turin, June 2005.
D. Techno-Economic Optimal Number of Fault Indicators
[4] E. Vidyasagar, P. V. N. Prasad, and A. Fatima, “Reliability
A techno-economic analysis is summarized in Fig. 15. The Improvement of a Radial Feeder Using Multiple Fault Passage
great improvement is achieved by installation of only one FI. Indicators,” Energy Procedia14 (2012): 223-228.
If the number of FIs is increased to two or three, the profit still [5] D. J. Krajnak, “Faulted circuit indicators and system reliability,” IEEE
grows although at a slower rate. We do still improve the Rural Electric Power Conference, pp. A4/1-A4/4, Kentucky, USA, May
2000.
system since the annual benefit grows due to decrease of ENS.
If the number of FIs is four or greater the improvement of [6] F. M. Angerer, “New developments in Faulted Circuit Indicators help
utilities reduce cost and improve service,” IEEE Rural Electric Power
reliability is not significant any more so the profit starts to Conference, pp. B4-B4/3, Charleston, South Carolina, USA, April 2008.
decrease. As a final decision, the variant with three indicators [7] Ž. Popović, S. Knežević, and B. Brbaklić, “Optimal number, type and
is chosen. The installation locations are read from Fig. 11. location of remotely controlled and supervised devices in distribution
networks,” IEEE PowerTech, Eindhoven, pp. 1-6, June-July 2015.
[8] C. Y. Ho, T. E. Lee, and C. H. Lin, “Optimal Placement of Fault
8000 A Indicators Using the Immune Algorithm,” IEEE Transaction on Power
Systems, vol. 26, pp. 38-45, February 2011.
5000
[9] А. Shahsavari, S. M. Mazhari, A. Fereiduisn, and H. Lesani, “Fault
Indicator Deployment in Distribution Systems Considering Available
2000 Control and Protect Devices: A Multi-Objective Formulation
Number of FIs
Approach,” IEEE Transaction on Power Systems, vol. 29, pp. 2359-
-1000 0 3 6 9 2369, September 2014.
[10] D-P. Cong, B. Raison, J-P. Rognon, S. Bonnoit, and B. Manjal,
-4000 annual cost of FIs, CFI( $/year) “Optimization of Fault Indicators Placement with Dispersed Generation
annual benefit of FIs, BFI($/year) Insertion,” IEEE Power Engineering Society General Meeting, vol. 1,
-7000 annual profit of FIs,YPI($/year) pp. 355-362, June 2005.
[11] R. Dashti and J. Sadeh, “Fault Indicator Allocation in Power
Fig. 15. Selecting the optimum number of FIs. Distribution Network for Improving Reliability and Fault Section
Estimation,” International Conference on Advanced System Automation
and Protection, pp. 1406-1411, October 2011.
IV. CONCLUSION [12] D. M. B. S. de Souza, A. F. de Assis, I. N. da Silva, and W. F. Usida,
In this paper, a fast and efficient method for placement of “Efficient fuzzy approach for allocating fault indicators in power
fault indicators was presented. The research is inspired by the distribution lines,” IEEE Transmission and Distribution Conference and
Exposition: Latin America, pp.1-6, August 2008.
indirect principle introduced in the literature. Instead of direct
[13] D. M. B. S. de Souza, I. N. da Silva, V. Ziolkowski and R. A. Flauzino,
optimization of the reliability indices, an auxiliary objective “Efficient allocation of fault indicators in distribution circuits using
function is defined which is correlated with the reliability fuzzy logic,” IEEE Power & Energy Society General Meeting, pp. 1-6,
indices but being significantly simpler to enable a faster search Calgary, Alberta Canada, July 2009.
optimization. The contribution is made both in terms of [14] W. F. Usida, D. V. Coury, R. A. Flauzino and I. N. da Silva, “Efficient
defining the auxiliary objective function and developing the Placement of Fault Indicators in an Actual Distribution System Using
search method. The proposed methodology was implemented Evolutionary Computing,” IEEE Transaction on Power Systems, vol. 27,
no. 4, pp. 1841-1849, November 2014.
to a real-life distribution network located in the northwest of
[15] A. K. Al-Othman and K. M. El-Naggar, “Application of pattern search
Bosnia and Herzegovina. Achievable improvements of the method to power system security constrained economic dispatch with
reliability indices in some cases reached 15%. Further research non-smooth cost function,” Electric Power Systems Research, vol. 78,
might be conducted to enhance the methodology in order to no. 4, pp. 667–675, April 2008.

View publication stats