Measurement 102 (2017) 150157

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

Fault location method for distribution networks using smart meters

Sadegh Jamali a,, Alireza Bahmanyar a, Ettore Bompard b
a
Centre of Excellence for Power System Automation and Operation, School of Electrical Engineering, Iran University of Science and Technology, P.O. Box: 16846-13114, Tehran, Iran
b
Dipartimento Energia, Politecnico di Torino, Corso Duca degli Abruzzi, 24, 10129 Torino, Italy

a r t i c l e i n f o a b s t r a c t

Article history: The ongoing deployment of smart meters, with data processing and communication features, has pro-
Received 1 November 2016 vided the opportunity to improve distribution systems performance. This paper presents a state
Received in revised form 1 February 2017 estimation-based method for fault location in distribution networks using the measurements provided
Accepted 6 February 2017
by the smart meters. During the normal operation of the system, state estimation methods can handle
Available online 9 February 2017
the errors in real or pseudo-measurements to give the best estimate of the system state and to identify
large measurement errors, called bad data. This concept is extended in this paper under fault conditions
Keywords:
where the fault is considered as an unknown and temporarily connected load which can be dealt with as
Bad data identification
Distribution networks
bad data. The proposed method uses the changeable weighting matrix bad data identification technique
Fault location to find the fault location. The method is algorithmically simple and does not require the fault type. The
Smart meters fault location accuracy is verified by extensive simulation tests on a real 13.8 kV, 134-node distribution
State estimation network for different fault scenarios. The results indicate that the method has a good performance under
measurement and load data errors and for different number and locations of the smart meters in the
network.
2017 Elsevier Ltd. All rights reserved.

1. Introduction ensuring safety prior to re-energizing the interrupted section of the

It is known that over 60% of customer power outages are due to
the faults on medium voltage distribution networks. In contrast to
transmission lines, distribution networks have several branches
and tapped laterals dispersing over vast rural and/or urban areas
with much fewer measuring and relaying points. These networks
are vulnerable to different types of faults arising from a variety
of causes such as adverse weather conditions, equipment failure,
bird contacts and vegetation growth [1]. In European networks,
durations of supply outages are generally low, ranging from about
15 min to 400 min per customer in a year [2]; however, as societies
become more dependent on electrical energy, a higher level of sup-
ply continuity is demanded. In this context, fault location offers
significant benefits for distribution system operator (DSO) by nar-
rowing down the search area to find the fault point. This poten-
tially improves the continuity of supply and lowers the network
operation costs.
Traditional fault location techniques for distribution networks
are based on activities such as grouping of customer trouble calls.
Then, a repair crew patrols the area looking for fault evidence and
Corresponding author.
E-mail addresses: sjamali@iust.ac.ir (S. Jamali), bahmanyar@iust.ac.ir (A. Bah-
manyar), ettore.bompard@polito.it (E. Bompard).
feeder. The whole restoration process can take a long time from
tens of minutes to several hours. Therefore, many research works
have been devoted to designing more efficient fault location
methods.
Deployment of fault indicators can be considered as a reliable
solution [3,4]. However, a large number of them are required to
provide accurate results. Impedance-based fault location algo-
rithms are a less expensive option which is based on the calcula-
tion of the impedance to the fault from the main substation and
finding the points with the same impedance as possible fault loca-
tions [59]. These methods are easy to be applied in real systems,
but they usually lead to multiple locations for a single fault and, as
shown in [7], impedance-based algorithms are also sensitive to
load data errors. The travelling wave-based methods in [10,11] cor-
relate the fault location with some characteristic frequencies asso-
ciated with specific travelling wave paths. These methods are
difficult to be applied in distribution networks with several laterals
and load taps [10]. They also require measurements at very high
sampling frequency, which limits their practical application.
Learning-based methods proposed in [1215] use a large set of
input-output fault data for training from which any fault data
can be mapped to a fault location. The input comprises of some
characteristic features extracted from the measured voltage and
currents and the output is usually the distance to the fault.

http://dx.doi.org/10.1016/j.measurement.2017.02.008
0263-2241/ 2017 Elsevier Ltd. All rights reserved.
 S. Jamali et al. / Measurement 102 (2017) 150157 151

Learning-based methods present very short execution times, but tification and hence the fault location [35,36]. Compared to previ-
they require a large set of training data and retraining following ously proposed methods in the literature, this method can handle
any changes in distribution network topology. The integrated the measurement and load data errors effectively by exploiting a
methods proposed in [1618] combine different types of fault loca- limited number of smart meters. It does not require fault type
tion methods to overcome their limitations. These methods should and can use different types of measurements, i.e. voltage, current
meet all the requirements of the individual methods. or power.
The introduction of smart meters has created the opportunity to Extensive simulation tests are carried out on a 134-node, radial
improve distribution systems operation. Several research projects distribution network. The results show that the proposed method
have reported different system architectures for installation and correctly identifies the fault location for different fault types, fault
better exploitation of smart meters [19,20]. For example, the impedances, and fault positions. The results also validate the per-
authors are involved in the Flexmeter project as one of the ongoing formance of the method against load data inaccuracies, and for dif-
European Union Horizon 2020 projects. The aim of Flexmeter is ferent number and location of the smart meters.
development and demonstration of a flexible smart metering The paper is organised as follows: Section 2 presents the math-
architecture using different types of off-the-shelf smart meters ematical background of the state estimation theory. Section 3
for different general-purpose services [19]. In [20] authors describe describes the details of the proposed fault location method. Sec-
a research project to develop a network of high-precision phasor tion 4 gives the simulation results for different fault scenarios
measurement units, termed micro-synchrophasors or lPMUs, pro- and finally, Section 5 concludes the paper and lists its
viding high-resolution angle measurements at comparatively low contributions.
cost with easy plug-and-play installation. Motivated by the recent
advances in distribution networks measurement and communica- 2. Mathematical background
tion infrastructures, several research studies are now exploring the
applications of the emerging metering and communication tech- In order to describe basic concepts and notations of the new
nologies for electric power distribution systems [21,22]. algorithm, a brief description of the state estimation equations is
One of the most interesting applications is in the fault location required. The non-linear equations relating the measurements vec-
area. The integrated methods proposed in [2325] use an tor z and the system state vector x are:
impedance-based fault location algorithm to estimate the fault dis-
tance and find all possible fault locations. Then, they exploit the z hx e 1
smart meters with voltage monitoring capability to identify the where h(x) is the vector of nonlinear functions and e is the vector of
correct solution. Although such integrated methods can overcome measurement errors.
the multiple estimation problem of the impedance-based algo- Branch-current-based State Estimation (BSE), commonly used
rithms, the problem of sensitivity to load data inaccuracies still in distribution networks, uses the real and imaginary parts of the
remains unsolved. The voltage sag-based methods proposed in branch currents (Ir, Ix) as state variables:
[2628] are based on the fact that each fault causes voltage sags
with different characteristics at different nodes. These methods x Ir ; Ix  2
consider the fault at each network node, one at a time, and calcu- The estimate of the unknown state vector x is obtained by min-
late voltage sags at all nodes having voltage measurements. Then, imizing the least-squares function below:
by comparing the measured and calculated voltage sags, the node
with the maximum similarity is identified as the fault location. X
m
2 T
jx wi zi  hi x z  hx Wz  hx
Although these methods provide acceptable results under ideal i1
conditions, they are sensitive to measurement inaccuracies and 2 31
load data errors [29]. In [30] the voltages measured by the smart r21 0  3
6 . 7
6
meters installed along the feeder are used to estimate different W 6 0
4
r22 .. 7 7
5
fault current values for each node. The faulted node is then identi-
.. ..
fied by comparing the calculated currents. This method requires a .  .
large number of meters to provide accurate results and it is sensi-
tive to measurement errors. The authors in [31] use the compres- where W is the (m  m) weighting diagonal matrix whose elements
sive sensing signal processing technique and pre- and during- determine the influence of measurements based on their variances
fault voltages measured by smart meters or phasor measurement rj2, and m is the number of measurements.
units to locate distribution network faults. In [32], the authors In order to minimise j(x) with respect to the state vector, the
extend the application of the idea proposed in [31] to locate simul- first-order optimality conditions need to be satisfied:
taneous faults. Compared to the voltage sag-based algorithms [26 @jx
28] and the method proposed in [30], these methods show better gx HT xWz  hx 0 4
@x
performance under measurement inaccuracies and load data
errors. However, they require a minimum number of meters to where H(x) = @h(x)/@x is the Jacobian matrix.
provide acceptable results and cannot be implemented with lim- The estimate is obtained by solving the non-linear optimality
ited measurements. The method proposed in [33] assumes the conditions of Eq. (4) through an iterative process:
fault as a current injection. It considers a fault at each network 1
xk1 xk  Gxk  gxk 5
node, performs state estimation and calculates the fault current.
Finally, it selects the faulted node by investigating the obtained where G(x) is the gain matrix:
state estimation residuals for all the nodes. In [34], the authors
@gx
review and compare different fault location methods, in terms of Gx HT xWHx 6
their advantages, limitations, and requirements. @x
This paper presents a new state estimation-based fault location Starting from an initial guess, the state estimation algorithm
method by considering the fault as an unknown and temporarily iteratively updates the state variables for k = 1, 2, 3 . . . until appro-
connected load which can be dealt with as a bad data. The method priate convergence is attained [37]. Details of BSE are described
uses the changeable weighting matrix technique for bad data iden- in [38,39].
152 S. Jamali et al. / Measurement 102 (2017) 150157

After the convergence, estimation residuals can be obtained as and current, available load demands and during fault voltage mea-
follows: sured at node 4, results in highly erroneous estimates and hence
large normalised residual values. As shown in Fig. 2a, the large nor-
r i zi  hi ^x; i 1; . . . ; m 7
malised residuals clearly indicate the presence of a gross anomaly;
where x is the estimated state vector and zi is the measured value of however, it is usually not convenient or even possible to find the
the ith measurement. bad data (i.e. load pseudo-measurement of the faulted node) only
Residuals indicate the difference between the measured values by inspection of the normalised residual.
and the corresponding estimated values. When there are no bad As pointed out in the previous section, the diagonal elements of
measurement data, the estimation residuals will be small; how- the weighting matrix determine the influence of measurements on
ever, in the case of gross measurement errors, the difference state estimates. Therefore, once the measurement residuals
between the observed measurements and their corresponding esti- increase due to gross measurement errors, the weighting matrix
mates becomes large. The conventional methods to flag on the bad can serve as an effective curb to unexpected influence [35,36]. In
data are mostly based on the calculation of the normalised residu- such a situation, reducing the weight of the bad measurement
als rN
i . can help to suppress its effect. As shown in Fig. 2b, compared to
Fig. 2a, assigning a small weight to the load pseudo-
jri j
r Ni p ; i 1; . . . ; m 8 measurement of node 2 curbs its influence and significantly
Xii decreases the normalised residual values. This indicates that node
where Xii is the ith diagonal element of the residual covariance 2 is the node with bad pseudo measurement (i.e. the faulted node).
matrix: The main idea of the proposed fault location algorithm is based
on the fact that decreasing the weight associated with the bad data
X W 1  HG1 HT 9 reduces its impact on state estimates. Therefore, the bad data
rN
Comparing values with a threshold, the measurement having would simply be the one which decreasing its weight decreases
i
the normalised residual with the largest value more than the the normalised value of residuals. Fig. 3 depicts the building blocks
threshold is identified as bad data [37].

3. Fault location algorithm (a)

The proposed algorithm assumes that the during-fault voltage
and current measurements are available at the main substation
200
NORMALIZAD RESIDUAL

and some points along the feeder. The measurements need to be

synchronised using the global positioning system (GPS) or a com-
puter network. The system data and forecasted or measured load
demands are also assumed to be available, which are used as
pseudo-measurements for the state estimation. Because of the lim-
ited number of real measurements in the distribution system, load 100
pseudo-measurements are necessary to ensure the observability
[38,39].
If we consider the fault as a load temporarily connected at the
faulted node, this gross anomaly appears as a bad data in
pseudo-measurements. As described in the previous section, the
presence of this bad data results in large normalised residual val- 0
ues, which can help to identify it. However, the least-squares esti- 1 2 3 4 5 6
mator may, unfortunately, be quite sensitive to such gross errors NODE
and inspection of the normalised residual usually does not reveal
which data is bad. As an example, consider the simple system (b)
shown in Fig. 1 at which a fault happens at node 2. The fault
appears as a bad data in load pseudo-measurements. Therefore,
8
performing state estimation using the main substation voltage
NORMALIZAD RESIDUAL

0
1 2 3 4 5 6
NODE

Fig. 2. Normalised residuals of load pseudo-measurements. (a) When the same

weights are assigned to all load pseudo-measurements. (b) When a small weight is
Fig. 1. Simple distribution network with a voltage measurement at node 4. assigned to the load pseudo-measurement of the faulted node.
 S. Jamali et al. / Measurement 102 (2017) 150157 153

method are the same for all different fault types to eliminate
Start
delays and errors caused by possible fault type misidentification
[40].
Get the network data,
real measurements and
pseudo measurements 4. Case study and results

An overhead, three-phase, 13.8 kV, 134-node real life distribu-

Select the first node (k=1)
Assign a small weight to the load pseudo
measurement of the kth node
Perform state estimation
Calculate the sum of the normalised residuals
(k) using Eq. (10)
Calculate the index value
(k) using Eq. (11)
k=k+1
No Is k equal to the
number of nodes?
Report the node with the
maximum index () as
the faulted node
Fig. 3. Flowchart of the proposed fault location method.
of the proposed method. After a fault is detected, the fault location
process starts from the first node. For each node, the algorithm
considers it as the faulted node and assigns a small weight to its
load pseudo-measurement. Then, the state estimation process is
performed to calculate the normalised residuals using the new
weighting matrix. In the next step, the sum of the normalised
residuals for all load pseudo-measurements is calculated:
X 3 
n X 
 N
rk r i;j 
i1 j1
where rN i,j is the normalised residual for phase j of node i when a
small weight is assigned to the kth load pseudo-measurement.
Considering a small weight for the load pseudo-measurement
of the faulted node restricts its influence on state estimates and
results in the least differences between the calculated and mea-
sured values (i.e. residuals). Thus, the following index is calculated
for each node and in the final step, the node with the largest value
of the index is identified as the faulted node.
1 5X applied at all 134 nodes (i.e. 134 fault scenarios), for 109 fault
gk
rk e
where rk is the sum of the normalised residuals and e is a small
number to avoid division by zero.
As can be seen in Fig. 3, the proposed algorithm is very simple
and efficient. It only investigates the effect of changing the ele-
ments of the weighting matrix on the normalised value of residuals
to locate the fault. Therefore, in contrast to the voltage sag-based
methods proposed in [2628], it does not need to calculate the
fault impedance or its current to apply the fault at each network
node. Moreover, the steps and the equations of the proposed
tion feeder, shown in Fig. 4, is selected for the studies [26,30].
Loads are modelled as constant impedances and voltage measure-
ments are supposed to be available at nodes 20, 30, 45, 51, 57, 75,
87, 111, 118, 127 and 134. The proposed method aims to use the
already installed smart meters to enhance the fault location perfor-
mance. However, the optimal location of meters can be determined
using a method similar to the optimal meter placement algorithms
proposed in [41,42].
The proposed fault location method is tested through extensive
simulated fault scenarios. For each considered case, the simulated
network provides the measurements and the proposed method,
developed in MATLAB, uses these values to estimate the fault loca-
tion. It is noteworthy mentioning that the proposed algorithm is
able to handle the errors caused by measurement inaccuracies or
noise. However, gross measurement errors should be identified
and eliminated from the measurement set using a bad data filter-
ing method [35].
Consider a three-phase fault at node 105. Upon receiving the
measured values, the fault location algorithm starts from the first
node. It assigns a small weight to the load pseudo-measurement
of the node and calculates the index values (g) using Eq. (11).
Fig. 5 shows the calculated index values for all network nodes,
where the horizontal axis shows the number of the investigated
nodes, and the vertical axis represents the g value. The figure
clearly indicates that node 105 has the largest index value that
means it is the actual location of the fault.
Several other fault scenarios are considered for different fault
types, locations, and resistances. The impact of load estimation
inaccuracies, measurement errors, and variation of meters quantity
and location are also studied to demonstrate the performance of
the proposed method.
4.1. Effect of fault resistance and fault type
The results of the proposed method for single-phase faults (AG)
10
at all nodes, with four different fault resistance values (536 differ-
ent scenarios) are presented in Fig. 6. For each considered case, the
distance between the faulted node and the node selected by the
fault location algorithm is estimated as the fault location error.
The numbers of fault locations estimated with an error between
0 and 50 m, 50 and 100 m, 100 and 200 m, 200 and 400 m and
higher than 400 m are counted separately and reported in the fig-
ure. The horizontal axis shows the considered error classes and the
vertical axis represents the number of faults located with an error
within each error class. For example, for AG faults with Rf equal to
11 scenarios the estimation errors are less than 50 m.
The obtained results indicate that for all considered fault resis-
tance values, most of the faulted nodes are located with distance
errors from 0 to 50 m. Moreover, comparison of the results shows
that the proposed method provides almost similar results for dif-
ferent fault resistances, i.e. its accuracy is not affected by the fault
resistance.
Fig. 7 presents the obtained results for four different fault types
at all nodes (536 different scenarios). It can be seen that the pro-
posed method presents satisfactory results for different fault types
since most of the estimation errors are within 0 to 50 m.
154 S. Jamali et al. / Measurement 102 (2017) 150157

60

54 59 127 126
55 53 52 56
57 88 125
124
58 123
122
50 49 61 62 63 78 89 90
51 119
48 76 77
120
Feeder 31 64 79
30 29 27 26 33 121
65
1
28 32 34 66
2 3 47
25 68 67 69 70 80
46 114
24 71 81 91
83 82 93 94
5 6 7 8 9 10 22 35 37 39 72 92 115 116 117 118
4
73 113
23 36 38 84 99
74
129 85 95 97
96 100 101
14 13 12 40 42 43 128 112
11 41 86 98
75 87 102
44
15 45 131
16 132 104 103
20 19 17 21
133 130 107
105 106
18 134 109 110 111
108

Fig. 4. Topology of the 134-node distribution network [26].

Fig. 5. Ranking of the nodes for a three-phase fault at node 105.

120
112
111
109
109

Rf =1
111
110

110
110

Rf =5 Single-phase to ground fault (AG) with Rf =10

100 Rf =20 Phase to phase fault (AC) with R fa=Rfc=5
NUMBER OF NODES WITHIN

Rf =50 100 Two-phase to ground fault (BCG) with Rfb=Rfc=5 and Rf =10
NUMBER OF NODES WITHIN
EACH ERROR CLASS

Three-phase to ground fault with Rfa=Rfb=Rfc=5 and Rf =10

80
EACH ERROR CLASS

80

60
60
Rfa Rfb Rfc
40
Rf
40

20
14
14
13

13

20
15
14
9
8
8
8

11
10
3
3

9
1
1

0
3
2
1

> 400 m 200-400 m 100-200 m 50-100 m 0-50 m 0

ERROR CLASS > 400 m 200-400 m 100-200 m 50-100 m 0-50 m
ERROR CLASS
Fig. 6. Number of nodes within each error class for single-phase faults (AG) at all
nodes, with different fault resistances. Fig. 7. Number of nodes within each error class for different fault types at all nodes.
 S. Jamali et al. / Measurement 102 (2017) 150157 155

4.2. Effect of load data inaccuracies
In distribution networks, due to customers behaviour uncer-
tainties, load data are usually inaccurate and it is difficult to esti-
mate the exact values at the fault instant. In order to test the
influence of the load data inaccuracies on the performance of the
proposed method, the data employed in the simulations is taken
as the actual data and the load data is created by random variation
of this data within specified ranges, i.e.:
Pi;j Pai;j 1 dL
Q i;j Q ai;j 1 dL
where Pai,j, Qai,j are the actual load data for ith node and jth phase and
dL is randomly generated by a normal distribution with zero mean
and standard deviation of 20%.
Fig. 8 presents the results for single-phase and three-phase
faults with two different fault resistances at all nodes (536 differ-
ent scenarios). It is interesting to note that even for high fault resis-
tance values, load data inaccuracies do not have a significant
impact on the performance of the proposed method. By incorporat-
ing all available measurements into calculations, the proposed
state estimation-based algorithm reduces the effect of load data
errors and ensures the accuracy even under uncertain conditions.
This is an individual property that state estimation brings to the
proposed fault location algorithm by providing the most optimal
estimate under imperfect conditions.
4.3. Effect of measurements inaccuracies
Smart meters provide an opportunity to improve distribution
systems operation. However, meters precision may be influenced
by errors. Gross measurement errors due to bad calibration or
unreliable communications can affect any fault location method.
Therefore, before performing the process of fault location, bad
measurements have to be identified and eliminated from the mea-
surement set. This issue can be addressed by inspection of pre-
fault data and measurements using a bad data filtering method
[35,36]. However, a fault location method should be able to handle
small measurement errors caused by meter inaccuracies or noise.
In order to study the effect of measurement errors on the proposed
fault location method, random errors are added to voltages gener-
ated during simulation:
V Mk V aMk 1 dV
111
110
109
where VMk, VaMk are the measured and the actual voltage values of
the kth meter and dV is randomly generated by a normal distribu-
tion with zero mean and standard deviations of 0.5% and 1%.
Fig. 9 summarises the obtained results for single-phase and
three-phase faults at all nodes (536 different scenarios). A compar-
ison with the ideal condition (Fig. 7) reveals that random errors
within 0.5% of the measured voltages do not affect the fault loca-
tion results significantly, but increasing the standard deviation to
1% may slightly decrease the method performance. However, the
results are still acceptable since most of the faulted nodes are
12 located with distance errors lower than 200 m. It can also be seen
that the method provides slightly better results when it is used to
estimate the location of three-phase faults.
Besides the inaccuracies in measured magnitudes, the effect of
synchronization error on signals phase angle deviations can also
affect the results. There are two main techniques for measurement
synchronization: global positioning system (GPS) and computer
network. Accuracy limits for synchrophasor measurements defined
by IEEE Standard specify the theoretical accuracy of synchroniza-
tion to be better than 1 ms (0.02 degree in 60 Hz) [43]. On the other
hand, the method proposed in [44] for network synchronization
can provide time synchronization with an accuracy of better than
4 ms (0.08 in 60 Hz). In order to study the effect of signal phase
angle deviations, the phasors of the measurements generated dur-
ing the simulation are rotated by random angles given by a normal
distribution with zero mean and standard deviations of 0.1 and
0.5. Fig. 10 presents the obtained results for single-phase and
three-phase faults at all nodes (536 different scenarios). As can
be seen, even with relatively large phase angle errors, the proposed
method provides acceptable results. Therefore, it can be imple-
mented with inexpensive network synchronization method.
4.4. Effect of meters quantity and location
In all of the previously studied scenarios, it is assumed that 11
voltage measurements are placed throughout the distribution net-
work providing the required real measurements. In this section,
five other cases are considered to evaluate the performance of
the proposed method with fewer meters on different locations:
Case (1) Nine meters are placed at nodes 20, 30, 45, 57, 75, 87,
118, 127, and 134.
Case (2) Six meters are placed at nodes 20, 30, 57, 87, 118, 127.
13

109

Single-phase to ground fault (AG) with Rf =10 Measurement errors with zero mean and standard deviation of 0.5%
120
109

Single-phase to ground fault (AG) with Rf =10

107

Single-phase to ground fault (AG) with Rf =50

106

104

100 Three-phase to ground fault with Rfa=Rfb=Rfc=5 and Rf =10 Three-phase to ground fault with Rfa=Rfb=Rfc=5 and Rf =10
NUMBER OF NODES WITHIN

NUMBER OF NODES WITHIN

Three-phase to ground fault with Rfa=Rfb=Rfc=20 and Rf =30 100

Measurement errors with zero mean and standard deviation of 1%
EACH ERROR CLASS

EACH ERROR CLASS

Single-phase to ground fault (AG) with Rf =10

80
80 Three-phase to ground fault with Rfa=Rfb=Rfc=5 and Rf =10

60
60

40
40
16

20
15

20
15
12
12

12

12

12

12
11

11
9
8

7
6
6

6
5
3
3

3
3

3
2

2
1

0 0
> 400 m 200-400 m 100-200 m 50-100 m 0-50 m > 400 m 200-400 m 100-200 m 50-100 m 0-50 m
ERROR CLASS ERROR CLASS

Fig. 8. Number of nodes within each error class (Effect of load data inaccuracies). Fig. 9. Number of nodes within each error class (Effect of measurement errors).
156 S. Jamali et al. / Measurement 102 (2017) 150157

110
110
Phase angle errors with zero mean and standard deviation of 0.1 degree 120 Load data errors with zero mean and standard deviation of 20%

111
111

111
106
103
Single-phase to ground fault (AG) with Rf =10

107
Single-phase to ground fault (AG) with Rf =10
Three-phase to ground fault with Rfa=Rfb=Rfc=5 and Rf =10 Three-phase to ground fault with Rfa=Rfb=Rfc=5 and Rf =10
100
NUMBER OF NODES WITHIN

NUMBER OF NODES WITHIN

100 Measurement errors with zero mean and standard deviation of 1%
Phase angle errors with zero mean and standard deviation of 0.5 degree
EACH ERROR CLASS

EACH ERROR CLASS

Single-phase to ground fault (AG) with Rf =10 Single-phase to ground fault (AG) with Rf =10
80 Three-phase to ground fault with Rfa=Rfb=Rfc=5 and Rf =10
Three-phase to ground fault with Rfa=Rfb=Rfc=5 and Rf =10 80

60 60

40 40

19
20 20

15
14

14

14
13

13
12
10
9
8

8
7

7
6
5
4
3

3
2

2
1
0 0
> 400 m 200-400 m 100-200 m 50-100 m 0-50 m > 400 m 200-400 m 100-200 m 50-100 m 0-50 m
ERROR CLASS ERROR CLASS

Fig. 10. Number of nodes within each error class (Effect of measurements phase Fig. 12. Number of nodes within each error class using different measurements.
angle errors).

able to incorporate all of the available measurements (voltage, cur-

rent or power) into calculations. In order to test this ability, it is
106

Case 1 supposed that in addition to the 11 voltage measurements

98

100 Case 2 employed for the previous tests, the currents of lines 1011, 48
Case 3
NUMBER OF NODES WITHIN

52, 7778, 9597 and 90119 are also available.

87

Case 4
Fig. 12 presents the obtained results for faults at all nodes under
EACH ERROR CLASS

80 Case 5
the four different considered cases (536 different scenarios). Com-
73
72

pared to Figs. 8 and 9, the performance of the algorithm is

improved when all measurements are incorporated into the fault
60
location calculations. The ability of the proposed method to exploit
different measurements helps to maintain its accuracy.
40
33

5. Conclusions
27

19
18
17

20
16

16

16
14

A new state estimation-based method is described for fault

13

13

13
9

location in distribution networks. Several test studies validate

5
3

the accuracy of the proposed method for different fault types, posi-
1
1

0
> 400 m 200-400 m 100-200 m 50-100 m 0-50 m tions, and resistances. The results also indicate that the method is
ERROR CLASS able to handle load data errors and measurements inaccuracies for
different number and location of the smart meters. The main con-
Fig. 11. Number of nodes within each error class for three-phase faults (Effect of tributions of the paper can be summarised as follows:
meters quantity and location).

(1) The proposed algorithm does not require the fault type.
Case (3) Nine meters are placed at nodes 21, 31, 43, 58, 70, 111,
Therefore, it eliminates the possible errors caused by fault
124, 129, and 131.
type misidentification. As discussed in Section 3, in contrast
Case (4) Six meters are placed at nodes 3, 20, 30, 111, 118 and
to the voltage sag-based methods [2628], which need to
134.
calculate the fault impedance or its current when applying
Case (5) Four meters are placed at nodes 20, 57, 118, and 127.
a fault at each network node, the proposed method does
not require such calculations.
Results under the considered cases for three-phase faults at all
(2) The accuracy of fault location methods generally decreases
nodes (670 different scenarios) are summarised in Fig. 11. Compar-
when the fault resistance increases. The proposed method
ing the results with the ideal condition indicates that increasing
gives almost similar results for different fault resistances.
the number of meters improves the accuracy of the proposed
(3) Most previous fault location methods are sensitive to load
method. The obtained results show that the location of the meters
data errors. The simulation results indicate that the pro-
influences the method performance. However, when the meters
posed method is insensitive to load data errors.
are well distributed over the distribution network, the method pre-
(4) In contrast to the methods proposed in [3032], the pre-
sents satisfactory results even with 4 m. This is a remarkable char-
sented method can be implemented using a limited number
acteristic of the proposed method, especially from an economic
of meters. Moreover, compared to the previously proposed
point of view.
methods in [2332], which rely on meters with voltage mea-
suring capability, the proposed method is able to use all dif-
4.5. Performance of the algorithm using different available ferent types of available meters (current, voltage or power)
measurements for fault location.
(5) For this study, the meters are arbitrarily placed in the net-
All of the previous tests are performed using voltage measure- work. The simulation results indicate that the method pre-
ments. However, the proposed state estimation-based method is sents satisfactory results with a limited number of meters,
 S. Jamali et al. / Measurement 102 (2017) 150157
as long as they are well distributed over the distribution net-
work. However, it is planned to enhance its performance by
optimal placement of meters in a future work.
(6) The method is algorithmically simple and can be easily
implemented in any distribution network and using differ-
ent state estimation algorithms. This ability provides a good
potential for its future application.
References
[1] E. Bompard, T. Huang, Y. Wu, M. Cremenescu, Classification and trend analysis
of threats origins to the security of power systems, Int. J. Electr. Power Energy
Syst. 50 (2013) 5064.
[2] CEER, 5th Benchmarking Report on the Continuity of Electricity Supply,
(Council of European Energy Regulators, 2011).
[3] K. Sun, Q. Chen, Z. Gao, An automatic faulted line section location method for
electric power distribution systems based on multi-source information, IEEE
Trans. Power Del. 31 (2016) 15421551.
[4] J. Teng, W. Huang, S. Luan, Automatic and fast faulted line-section location
method for distribution systems based on fault indicators, IEEE Trans. Power
Syst. 29 (2014) 16531662.
[5] R.H. Salim, M. Resener, A.D. Filomena, K.R.C.D. Oliveira, A.S. Bretas, Extended
fault-location formulation for power distribution systems, IEEE Trans. Power
Del. 24 (2009) 508516.
[6] R.H. Salim, K.C.O. Salim, A.S. Bretas, Further improvements on impedance-
based fault location for power distribution systems, IET Gener. Transm. Distrib.
5 (2011) 467478.
[7] C. Orozco-Henao, A. Bretas, R. Chouhy-Leborgne, A. Herrera-Orozco, J. Marn-
Quintero, Active distribution network fault location methodology: a minimum
fault reactance and Fibonacci search approach, Int. J. Electr. Power Energy Syst.
84 (2017) 232241.
[8] C. Grajales-Espinal, J. Mora-Flrez, S. Prez-Londoo, Advanced fault location
strategy for modern power distribution systems based on phase and sequence
components and the minimum fault reactance concept, Electr. Power Syst. Res.
140 (2016) 933941.
[9] J.J. Mora-Flrez, R.A. Herrera-Orozco, A.F. Bedoya-Cadena, Fault location
considering load uncertainty and distributed generation in power
distribution systems, IET Gener. Transm. Distrib. 9 (2015) 287295.
[10] A. Borghetti, M. Bosetti, C.A. Nucci, M. Paolone, A. Abur, Integrated use of time-
frequency wavelet decompositions for fault location in distribution networks:
theory and experimental validation, IEEE Trans. Power Del. 25 (2010) 3139
3146.
[11] M. Goudarzi, B. Vahidi, R. Naghizadeh, S. Hosseinian, Improved fault location
algorithm for radial distribution systems with discrete and continuous
wavelet analysis, Int. J. Electr. Power Energy Syst. 67 (2015) 423430.
[12] D. Thukaram, H. Khincha, H. Vijaynarasimha, Artificial neural network and
support vector machine approach for locating faults in radial distribution
systems, IEEE Trans. Power Del. 20 (2005) 710721.
[13] H. Zayandehroodi, A. Mohamed, M. Farhoodnea, M. Mohammadjafari, An
optimal radial basis function neural network for fault location in a distribution
network with high penetration of DG units, Measurement 46 (2013) 3319
3327.
[14] A. Rafinia, J. Moshtagh, A new approach to fault location in three-phase
underground distribution system using combination of wavelet analysis with
ANN and FLS, Int. J. Electr. Power Energy Syst. 55 (2014) 261274.
[15] L. Ye, D. You, X. Yin, K. Wang, J. Wu, An improved fault-location method for
distribution system using wavelets and support vector regression, Int. J. Electr.
Power Energy Syst. 55 (2014) 467472.
[16] R.H. Salim, K.R.C. De Oliveira, A.D. Filomena, M. Resener, A.S. Bretas, Hybrid
fault diagnosis scheme implementation for power distribution systems
automation, IEEE Trans. Power Del. 23 (2008) 18461856.
[17] D.S. Gazzana, G.D. Ferreira, A.S. Bretas, A.L. Bettiol, A. Carniato, L.F.N. Passos, A.
H. Ferreira, J.E.M. Silva, An integrated technique for fault location and section
identification in distribution systems, Electr. Power Syst. Res. 115 (2014) 65
73.
[18] R. Dashti, J. Sadeh, Fault section estimation in power distribution network
using impedance-based fault distance calculation and frequency spectrum
analysis, IET Gener. Transm. Distrib. 8 (2014) 14061417.
157
[19] A. Bahmanyar, A. Estebsari, E. Pons, S. Jamali, E. Bompard, E. Patti, A.
Acquaviva, Emerging smart meters in electrical distribution systems:
opportunities and challenges, in: 24th Iranian Conference on Electrical
Engineering (ICEE), Tehran, 2016.
[20] A. Von Meier, D.E. Culler, A. McEachern, R. Arghandeh, Micro-synchrophasors
for distribution systems, in: ISGT, 2014, pp. 15.
[21] G. Sanchez-Ayala, J.R. Agerc, D. Elizondo, M. Lelic, Current trends on
applications of PMUs in distribution systems, in: Innovative Smart Grid
Technologies (ISGT), 2013, pp. 16.
[22] M. Majidi, M. Etezadi-Amoli, H. Livani, M.S. Fadali, Distribution systems state
estimation using sparsified voltage profile, Electr. Power Syst. Res. 136 (2016)
6978.
[23] A. Bahmanyar, A. Estebsari, E. Pons, S. Jamali, E. Bompard, An improved fault
location method for distribution networks exploiting emerging LV smart
meters, in: IEEE Workshop on Environmental, Energy, and Structural
Monitoring Systems (EESMS), Bari, 2016, pp. 16.
[24] S.F. Alwash, V.K. Ramachandaramurthy, N. Mithulananthan, Fault-location
scheme for power distribution system with distributed generation, IEEE Trans.
Power Del. 30 (2015) 11871195.
[25] F.C.L. Trindade, W. Freitas, Low voltage zones to support fault location in
distribution systems with smart meters, IEEE Trans. Smart Grid (2016) 110.
[26] R.A.F. Pereira, L.G.W.D. Silva, M. Kezunovic, J.R.S. Mantovani, Improved fault
location on distribution feeders based on matching during-fault voltage sags,
IEEE Trans. Power Del. 24 (2009) 852862.
[27] S. Lotfifard, M. Kezunovic, M.J. Mousavi, Voltage sag data utilization for
distribution fault location, IEEE Trans. Power Del. 26 (2011) 12391246.
[28] S.M. Brahma, Fault location in power distribution system with penetration of
distributed generation, IEEE Trans. Power Del. 26 (2011) 15451553.
[29] P.C. Chen, V. Malbasa, Y. Dong, M. Kezunovic, Sensitivity analysis of voltage sag
based fault location with distributed generation, IEEE Trans. Smart Grid 6
(2015) 20982106.
[30] F.C. Trindade, W. Freitas, J.C. Vieira, Fault location in distribution systems
based on smart feeder meters, IEEE Trans. Power Del. 29 (2014) 251260.
[31] M. Majidi, A. Arabali, M. Etezadi-Amoli, Fault location in distribution networks
by compressive sensing, IEEE Trans. Power Del. 30 (2015) 17611769.
[32] M. Majidi, M. Etezadi-Amoli, M.S. Fadali, A novel method for single and
simultaneous fault location in distribution networks, IEEE Trans. Power Del. 30
(2015) 33683376.
[33] S. Jamali, A. Bahmanyar, A new fault location method for distribution networks
using sparse measurements, Int. J. Electr. Power Energy Syst. 81 (2016) 459
468.
[34] A. Bahmanyar, S. Jamali, A. Estebsari, E. Bompard, A comparison framework for
distribution system outage and fault location methods, Electr. Power Syst. Res.
145 (2017) 1934.
[35] T. Van Cutsem, M. Ribbens-Pavella, Bad data identification methods in power
system state estimation-a comparative study, IEEE Trans. Power App. Syst. 104
(1985).
[36] J.-M. Lin, H.-Y. Pan, A static state estimation approach including bad data
detection and identification in power systems, in: Power Engineering Society
General Meeting, Tampa, 2007, pp. 17.
[37] A. Abur, A.G. Exposito, Power System State Estimation: Theory and
Implementation, Mercel Dekker, 2004.
[38] W.-M. Lin, J.-H. Teng, S.-J. Chen, A highly efficient algorithm in treating current
measurements for the branch-current-based distribution state estimation,
IEEE Trans. Power Del. 16 (2001) 433439.
[39] J.-H. Teng, Using voltage measurements to improve the results of branch-
current-based state estimators for distribution systems, IEE Proc. Gen. Transm.
Distrib. 149 (2002) 667672.
[40] B. Das, Fuzzy logic-based fault-type identification in unbalanced radial power
distribution system, IEEE Trans. Power Del. 21 (2006) 278285.
[41] R. Singh, B.C. Pal, R.B. Vinter, Measurement placement in distribution system
state estimation, IEEE Trans. Power Syst. 24 (2009) 668675.
[42] V. Basetti, A.K. Chandel, Optimal PMU placement for power system
observability using Taguchi binary bat algorithm, Measurement 95 (2017)
820.
[43] IEEE Standard C37.118.1.2011. IEEE Standard for
Synchrophasormeasurements for Power System, 2011.
[44] S. Johannessen, Time synchronization in a local area network, IEEE Control
Syst. Mag. 24 (2004) 6169.