This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2018.2874879, IEEE
Transactions on Power Delivery
1

A Searching Based Method for Locating High

Impedance Arcing Fault in Distribution Networks
S. Hamid Mortazavi, Zahra Moravej, Senior Member, IEEE, and S. Mohammad Shahrtash1

l LM Mutual-inductance per unit length of a three-phase

Abstract--Locating high impedance fault (HIF) in a distribution
line.
network enhances the reliability of the system and improves the
quality of the service for the customers. Due to low magnitudes of LE Rem( X ) Length of equivalent three-phase line of the
fault current and nonlinearity of fault impedance it is a very hard downstream network from X.
task and less considered by the distribution automation system. Rem( X )
R Load Resistance of phase-a equivalent load of the
a
This paper introduces a novel method for estimating the
location of high impedance arcing faults in radial distribution downstream network from X.
systems. The proposed method searches the feeders of the LRem( X )
Load a
Inductance of phase-a equivalent load of the
network at different locations, to find where the estimated downstream network from X.
parameters of the assumed HIF would result the most accurate Estimated HIF current for the assumed HIF at X.
estimation of the faulty phase voltage. The fault current and the
I HIF( X )
arc parameters are estimated at each probable location employing UP Inception voltages of arc during positive half cycles.
an appropriate model to estimate the faulty phase voltage. Absolute value of Inception voltages of arc during
UN
As the proposed method uses the time domain analysis, it is
applicable for nonhomogeneous distribution networks with negative half cycles.
unbalanced loading as well as symmetrical systems. R HIF Resistance of Arc.
The results of simulations of IEEE 34-node test feeder with TF Inception time of the HIF.
various HIF parameters by EMTP-RV show that the proposed
method is well capable of locating high impedance faults despite NF Inception samples of the HIF.
uncertainty in the load data. N Sample numbers per cycle.
t Sampling time interval.
Index Terms-- Distribution network, Fault location, Arcing PhF The faulty phase.
fault, High impedance fault (HIF), Fault current estimation.
Er V The error between the calculated values of the faulty
(X )

phase voltage and the measured values at the source

NOMENCLATURE node assuming an HIF at (X).
(p)
V Fault (X )
Calculated voltage at (p) assuming an HIF at (X).
VABC (p)
Three-phase voltage location (p).
(p)
V Fault (X )
Estimated voltage at (p) assuming an HIF at (X).
p , ( p 1))
I (ABC The three-phase current flows from the (p) to the
(p+1). X HIF The most possible location of HIF.
R ABC( p , ( p 1 ))
The resistance matrix of the line sections between
(p) and (p+1).
 )
The inductance matrix of the line sections between I. INTRODUCTION
L(ABC
p, p 1 )
(

(p)
R Load-a
(p) and (p+1).
Resistance of load at node (p).
H IGH impedance fault (HIF) detection is now available
as a protection function in power distribution
network protection IEDs (intelligent electronic devices) [1].
L(Load-
p)
a
Inductance of load at node (p). HIF can be defined as a fault that draws insufficient current for
rLL Self-resistance per unit length of a three-phase line. the conventional protection devices such as overcurrent relays
l LL Self-inductance per unit length of a three-phase line. or fuses to pickup [2]. Due to low magnitude of fault current,
HIFs do not damage the system components, but they are a
rLM Mutual-resistance per unit length of a three-phase
danger to public safety. In addition HIFs often involve arc that
line. increases risk of fire [2]. There is a wide range of methods
proposed for HIF detection, but HIF location in the
S. H. Mortazavi and Z. Moravej are with the Faculty of Electrical and distribution system is less considered. The reason for this is
Computer Engineering, Semnan University, Semnan, Iran (e-mail: that in addition to the common challenges for the fault location
[email protected]).
(FL) in the distribution network (DN) like multi-phase lateral
S. M. Shahrtash is with the Electrical Engineering Department, Center of
Excellence for Power System Automation and Operation, Iran University of and load taps along the main feeder, unbalanced loading,
Science and Technology, Tehran, Iran (e-mail: [email protected]). variety of conductors and configuration [3]-[4], the
nonlinearity of arc resistance and low magnitudes current of

0885-8977 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2018.2874879, IEEE
Transactions on Power Delivery
2

fault, expose additional challenges for the HIF location. model of HIF at unknown location in a distribution feeder and
Nonlinearity of fault impedance makes the fundamental phasor the fault distance and the parameters of HIF are obtained
based analysis inefficient. Moreover, low magnitude of fault employing curve fitting algorithm. The proposed method is
current in conjunction with the considerable change in the load interesting but it has been evaluated only for the first section of
current (during post-fault time with respect to pre-fault time, the network, without any laterals.
due to the change in voltage at the connection point of loads), In this paper a new time domain algorithm for HIF location
make it very hard to obtain HIF current exactly. in distribution systems is presented. The method is a searching
Fault location is needed by the distribution automation method that considers fault at different locations of the system
system to meet the requirements for the reliability indices for through a set of probable locations of fault. According to the
the smart distribution system. An automatic FL in electricity voltage and current at each location, an equivalent model of
distribution networks improves the reliability indices of the the downstream network for each of the assumed fault
system by outage management, reducing restoration time, locations is considered. The equivalent model is utilized for
decreasing power disruption, and improving quality of service fault current and arc parameters estimation. Then the voltage
provided by utilities [5]-[6]. Locating high impedance arcing of faulty phase is computed at the source node that IED is
faults introduces additional benefits that can be stated as the mounted, employing the estimated HIF parameters. The error
optimal outage management. As HIFs have low current between the computed voltage and the measured voltage is
magnitudes, the HIF detection methods are affected by the used as a criterion to estimate the location of the HIF.
fault characteristics and the system operating conditions. To investigate the performance of the proposed method,
Therefore, no method can detect all of the HIF’s with 100% IEEE 34-node test system was simulated using EMTP-RV
security against the mal-operations. Utilities prefer to utilize program considering high impedance faults with various
the HIF detection function as an alarming function [7]. parameters at different locations of the system. In each case
Therefore, if the approximate location of the possible fault can the voltage and current signals at the main source node were
be estimated, the operator of the system will make better stored with the sampling rate of 7680 Hz, which is equivalent
decision considering the priority of the loads in the possible to 128 samples per cycle in a 60 Hz system [1]. Applying the
fault area and risk evaluation analysis of the hazards on that proposed HIF location method to the simulated cases
area, sending the service crew for site visiting while being presented promising results even for unbalanced and
confident about fast fault clearing specially in the case of non- nonhomogeneous distribution system.
broken conductors. These benefits of HIF location, motivated The rest of this paper is organized as follows. In Section II,
researchers and utilities in recent years to develop algorithms the problem of the HIF location and the proposed solutions are
for HIF location. presented and the proposed algorithm for the HIF location is
Moreover, arcing fault resistance in distribution networks is described, in sequel. The selected test system and simulations
nonlinear and the fault current is approximately zero whenever are detailed in Section III. The results of implementing the
the voltage magnitude is less than a threshold. The HIF current proposed method for HF detection in test system are also
magnitude changes about four times in each fundamental presented in Section III. This section is followed by a
frequency cycle, thereby the voltage magnitudes and the loads summarized conclusion.
also present a dynamic behavior in each cycle. Therefore, the
fundamental phasor based methods that present promising II. THE PROPOSED METHOD FOR THE HIGH IMPEDANCE FAULT
results for the locating of non-arcing fault, may fail. LOCATION
In [8] power line communication (PLC) has been proposed
for HIF location. But employing PLC for distribution systems A. The problem of HIF location in distribution network
needs extra components and communication system for each There are many algorithms that have been proposed for the
lateral and it is not applicable for the complex system. bolted fault location in DN [14]-[16]. Generally, the one-
In [9] a method for HIF location based on the measurement ended measurement FL process in DN starts from the
of harmonics in different locations of the distribution system measurement node and the distance between the measurement
and performing harmonic power flow has been presented. But, node and the fault is estimated by the FL algorithm through a
the results of simulations for harmonics do not agree with series of algebraic equations. If the estimated distance is
other realistic reports. smaller than the length of the first section of network, it
To overcome the problem with the non-stationary signals, represents the location of fault but if the obtained distance is
the time domain method can be applied for the arcing fault beyond the next load tap or lateral, another FL process must be
location [10]-[12] for both transmission and distribution executed. For the next FL process, the voltage and current
systems. These methods consider a voltage-current model for signals at the beginning of the next section are calculated
the arc in time domain, and by presenting the time domain according to the available data of loads and characteristics of
voltage-current relation in the circuit that fault current flows, the system. The load data means the load values at each node a
the locations of fault can be estimated. few cycles before the inception time of fault. The load data can
A time domain method for HIF location in distribution be obtained through load flow or state estimation method like
networks has been introduced in [13]. The method considers a [16]-[18]. The FL process is iterated until the obtained

0885-8977 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2018.2874879, IEEE
Transactions on Power Delivery
3

distance to fault is smaller than the length of the examined line location in any large radial distribution network is presented in
section. Fig. 1.
But as aforementioned, HIF location in distribution network
presents more problems, which can be summarized as follows: The proposed algorithm includes three main stages:
1- HIFs are usually with arc that is a nonlinear impedance 1. Initializing and data preparation;
in nature. Therefore, additional unknown parameters (in 2. Computing the error in the estimated voltage for each
addition to fault resistance and distance to fault) get involved of the assumed fault locations;
in the fault loop impedance equations, which makes the 3. Making the decision on the location of the HIF.
solving methods to be inefficient. The processes in each stage are described in sequel.
2- In the case of bolted fault, the fault current compared to
the load current is large enough so the error in the fault current 1) Initializing and data preparation
estimation due to variation in the load current, after fault Start: Triggering by HIF detection unit, as soon as an HIF
inception, can be ignored. But HIFs are in order of load is detected [20], the estimated fault inception time (TF) and the
current and the change in the load current during the post-fault faulted phase (PhF) are sent to the fault location unit.
must be compensated properly otherwise the error in the Data Windowing: According to the fault inception time,
estimated location of fault increases considerably. the pre-fault and the post-fault data windows are constructed
3- The FL process is finished whenever the estimated from the recorded data of the measured signals. Pre-fault data
distance to fault become smaller than the length of the window consists of one period of the fundamental frequency
examined section [16]. As the measurement is available only at data and the post-fault one includes two periods. As there may
the source node, the voltage and current at each node are be some uncertainty about the estimated fault inception time
calculated according to the available data of loads and [20], a half cycle margin is considered for the pure pre-fault
characteristics of the system. But there is always inaccuracy in and post fault data window. The pre-fault data window is
these calculated values compared to the true ones. The main employed for the load estimation in the normal conditions and
sources of error are uncertainty in the load current at each tap the post-fault data window is used for the fault location
and error of the discrete formulation of differentiating the algorithm.
sampled signals. Due to nonlinearity of the fault impedance
and low magnitude of fault current for the HIFs, the From the Fault Detection Unit
TF , PhF (Fault Inception Time
uncertainty in the estimated distance to HIF is considerable and Faulty Phase)

compared to the bolted FL, in such a way that HIF location

1) Initializing and Data Preparation

Start
may result wrong section far from the true faulty section.
To overcome these problems and develop a method for HIF Measured Data
Pre & Post-Fault
(Voltage and Data Windows Preparation
location in distribution network, a searching based algorithm Current Signals)
presented in this paper that is described in the next sub-
Initializing: ErHIF = ∞ , i =1
section. The main remedies to the above problems, proposed
in this paper, can be summarized as follows: Network Data Xi Set of the Possible
(Topology, Calculate The Voltage and
Fault Locations
1- To deal with the non-stationary characteristic of the fault Load Data, ...)
Current at Xi
{X1 , X2 , … , XN-1 , XN }
impedance, time domain analysis is employed (that has been
for Each of the Assumed Fault Locations
2) Computing the Error in the Estimated Voltage

used by other researchers for arcing FL in a line section [10]- Estimate the Parameters
[12]) and curve fitting method is used to find unknown values of the Remainder of the
System Equivalent From Xi
(same as [19]).
2- To compensate the load current during the post-fault Estimate HIF Parameters
data windows, at each possible location a model for the load in I HIF( X i ) , UP , UN , RHIF
the downstream network is considered and the post-fault load
(X i )
current is estimated in order to extract the fault current more Calculate V Fault ( X i ) with The
Estimated HIF Parameters
accurately.
3- To reduce the uncertainty in the estimated location of Calculate Er
V (X i )
HIF, the FL process is executed for a set of the probable
locations of HIF at different feeders of the network. At each
3) Making the Decision on the

ErHIF = Er Yes
location a model for the HIF and downstream network is V (X i ) Er < ErHIF
Location of the HIF

V (X i )
XHIF = Xi
considered and voltage at the measurement node is estimated. No
A location that results the most accurate estimation of the No
i=N i=i+1
faulty phase voltage at the measurement node, is nominated as Yes
the most possible location of the HIF. Output : XHIF
The Most Possible Fault Location)
B. The proposed HIF location algorithm
Fig. 1 The overall flowchart of the proposed HIF location algorithm in the
The overall flowchart of the proposed algorithm for HIF distribution system.

0885-8977 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2018.2874879, IEEE
Transactions on Power Delivery
4

Set of Probable Locations of HIF: Locations along all of [16]-[18] can be employed to estimate the loads of the system.
faulted phase (PhF) feeders at fixed intervals (for example These methods can estimate load values with a high level of
1km) construct the set of probable location of HIF. All of accuracy. In this paper it is assumed that the load data are
locations in this set are examined by the proposed algorithm to available, but a reasonable uncertainty is considered in load
find the most possible location of HIF. values for evaluation.
Calculating the voltage and the current at the assumed 2) Computing the error in the estimated voltage for each of
fault location: The voltage and current at each assumed the assumed fault locations
location are determined during pre-fault and post-fault time. Estimating the parameters of the downstream
Starting from the source node, the voltage at each node is equivalent system: To estimate the fault current at each of the
calculated by subtracting the voltage drop between the assumed location of the fault, an equivalent model of the
predecessor node to the next successor node in the fault downstream network must be derived. The downstream
current path (the path between the source node and assumed network from X is modeled by an equivalent line connected to
fault location). equivalent load admittance as illustrated in Fig. 3 [19].
The general three-line model of a section between two There are many types of load in the distribution system that
nodes of the distribution system as depicted in Fig. 2 can be are connected to MV feeders through MV/LV transformers.
considered for calculating the voltage and current along the Considering every single load at the LV side of the
system. The voltage drop along the line section between two transformers is not practically possible as there are so many
locations of the system is obtained by [10]: loads with different models. But according to the studies done
in [23], the aggregation of different types of loads at the MV
(p, p 1 )
VABC (
(t )  VABC
 ) (p)
 VABC
( p 1)
(t ) side of the network, will presents constant impedance behavior
(1)
 R (ABC
p, p 1 )  )
. I (ABC
( p, p 1 )
(t )  L(ABC
(  )
p, p 1 )
. (  )

d I (ABC
p, p 1 )
(t )
(  )
 in the first five cycles after voltage sag, if the voltage does not
fall below 60% of the pre-fault values [23]-[24]. Therefore, the
dt
loads can be considered as constant impedance during a few
where V  [V V (p)
ABC
V ] is the three-phase voltage at the
a
(p)
b
(p)
c
(p) T
cycles after the fault occurrence [16], [23]-[24].
location (p), and I (p, p 1 )
(t )  [I a( p ,( p 1)) I b( p ,( p 1)) I c( p ,( p 1)) ]T is the
ABC
(  )
The parameters of the equivalent model are estimated
three-phase current flows from the (p) to the (p+1). employing the calculated voltages and currents at X during the
 R aa( p , p 1 ) R ab( p , p 1 ) R ac( p , p 1 ) 
(  ) (  ) (  )
 Laa( p , p 1 ) Lab( p , p 1 ) L(acp , p 1 ) 
(  ) (  ) (  ) pre-fault period (t<TF). Considering the circuit model as
 ( p , p 1 ) ( p , p 1 ) ( p , p 1 )  and ( p , p 1 )  ( p , p 1 ) ( p , p 1 ) ( p , p 1 )  depicted in Fig. 3, assuming a fault has occurred at time TF on
R (ABC  )
p, p 1 )
(
  R ba (  )
R bb (  )
R bc (
 L ABC   Lba
 ) (  )
Lbb (
Lbc 
 ) (  ) (  )

R ( p , p 1 ) R ( p , p 1 ) R ( p , p 1 ) 
(  ) (  ) (  )  L ( p , p 1 ) L ( p , p 1 ) L( p , p 1 ) 
(  ) (  ) (  )
phase-a, the voltage of faulty phase at the assumed fault
 ca cb cc   ca cb cc  location X can be calculated as:
represent resistance and the inductance of the line sections
between (p) and (p+1) respectively. 
V a( X ) (t )  LE Rem( X ) . rLL .I a( X ) (t )  rLM .I b( X ) (t )  rLM .I c( X ) (t )
(3)

For the sampled values the (t) is replaced by (nΔt) and d (X ) d d
 l LL . I a (t )  l LM . I b( X ) (t )  l LM . I c( X ) (t )
derivative terms of current signals in (1) turn into numerical dt dt dt
differentiation. Several methods are available for numerical X ) d

differentiation [21]-[22]. The accuracy in the calculations is  R Load

Rem( X )
a
.I a( X ) (t )  LRem(
Load a
I a( X ) (t )
dt
very important issue in the performance of the algorithm as the
error in the estimated value is the criterion for the proposed where rLL, lLL, rLM, and lLM are self-resistance, self-inductance,
decision making scheme to locate HIF. Therefore, the Gear 2nd mutual-resistance, and mutual-inductance per unit length of the
order method (described in Appendix) that presents a three-phase line, respectively. Although the downstream lines
reasonable accuracy and a wider region of stability was at each node of the system may have different characteristics,
employed for calculating. this equivalent model for the downstream which is found by
The current flows from the node (p), to the next applying curve fitting, provide enough accuracy in the
downstream node (p+1), is obtained by subtracting the load estimation of the impedance of the downstream system
irrespective of non-homogeneity of the real distribution
current I (Load-ABC
p)
(t ) , lateral current I ( p ) (t ) and shunt Lat-ABC
network. LE Rem( X ) , R Load
Rem( X )
a
, and LRem( X )
Load a
are length of the
capacitance current I (Cap-ABC
p)
(t ) from the current flows to its
equivalent line, resistance and inductance of the equivalent
predecessor branch ((p-1), p): loads orderly that are unknown. The trapezoidal method is
utilized to calculate the numerical differentiations (as
p , ( p 1))
I(ABC (t )  I (ABC
( p 1), p )
(t )  I Cap-ABC
(p)
(t )  I (Load-ABC
p)
(t )  I (Lat-ABC
p)
(t ) (2)
described in Appendix).
Similarly, to the impedance-based methods, the loads are The R.H.S. of (3) includes three unknown parameters
converted to impedances and are assumed to remain identical H Rem( X )  [LRem( X )
Load a
Rem( X )
, R Load a
, LE Rem( X ) ] . The values of V a( X ) (n )
as the system moves to the post-fault time [12], [23]. The and I (ABC
X )
(n )  [I a( X ) (n ), I b( X ) (n ), I c( X ) (n )] are obtained for the
equivalent impedances of loads are obtained from the available
data of loads at each node and the calculated voltages during samples in the pre-fault data window as described earlier in
the pre-fault data window. The state estimation method like this section and rLL, lLL, rLM, and lLM are constant parameters

0885-8977 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2018.2874879, IEEE
Transactions on Power Delivery
5

Distribution System From the Source The Assumed Fault The Downstream
The Source Node Node to the Assumed Fault Location X Location Network
p 1)
I (Lat-ABC
2)
I (Lat-ABC
p)
I (Lat-ABC
(1) (2) (p) (p+1)
Equivalent
(1 )
VABC IED ( p , ( p 1))
Load Model
(2 ) (( p 1), p ) ( p 1) (( p 1), ( p  2 )) (X )
VABC I ABC (p)
VABC I ABC VABC I ABC I ABC VABC (X )

p ,( p 1))
Z (ABC
1 2),
Z (ABC
2 ... )
,
Z (ABC
I (ABC
1 2)
,

IHIF(X)
Source Equivalent
Line

C(ABC

C(ABC
C(ABC
C(ABC

HIF
2)

p 1)
1)

p)
p 1)
I (Load-ABC
2)
I (Load-ABC
p)
I (Load-ABC

Measured Values Calculated Values Estimated Values

Lateral Branch
(p) (p+1)
a
b
c

(p)
(p)

(p)

I Lat
I Lat
I Lat

( p , ( p 1 ))
Ia(p,(p+1)) R aa( p ,( p 1 ))
((p-1),p)
Ia Va(p) L aa

Lab
R ac( p ,( p 1 ))

Lac
I Load

R ab( p ,( p 1 )) R bc( p ,( p 1 ))
I Cap

( p , ( p 1 ))

( p , ( p 1 ))
(p)
(p)

( p , ( p 1 )) ( p , ( p 1 ))

a
a

Ib((p-1),p) Vb(p) Ib(p,(p+1)) R bb Lbb

Lbc
( p , ( p 1 ))
Vc(p) Ic(p,(p+1)) R cc( p ,( p 1 )) Lcc( p ,( p 1 ))
Ic((p-1),p)
R Load

R Load
R Load
(p)

(p)
(p)
a

c
b

(p, p+1 )
Ca(p,(p+1)) Cb(p,(p+1)) Cc ( )
L(Load

L(Load
L(Load
p)

p)
p)
a

c
b

Fig. 2. General single-line model of the distribution system assuming the fault at the location X and the three-line circuit model of a branch section between two
adjacent nodes considered for calculating voltage and current.

N F  (T F / t )  (N / 2) represents the fault inception sample

The Assumed Fault Location number that is half cycle delayed compared to the estimated
Rem ( X )
(X ) (X )
V Fault
Rem ( X )
I Load Rem ( X )
a R aa
Laa Rem ( X )
R Load LRem (X )
fault inception time (TF) to compensate the uncertainty in the
Va(X ) Ia X
( )
a Load a

(X ) estimation. For this estimation whenever needed in each

Vb(X ) Ib
(X )
procedure, the nonlinear least-squares curve fitting algorithm
Vc(X ) Ic embedded in MATLAB Optimization Toolbox™ [25] has
TF IHIF(X) been employed.
HIF parameters estimation: The post-fault load current of
UN UP
Downstream Network the downstream network (Rem(X)) is obtained by employing
From X
the post-fault voltage of the faulty phase and the downstream
RHIF
equivalent network. Then, the fault current, considering fault
Fig. 3 The equivalent circuit of the downstream network from the assumed at the point X is simply obtained:
location of HIF for estimating the fault current.
I HIF( X ) (n )  I a( X ) (n )  I Load-
Rem( X )
a (n ) , n NF (5)

According to the circuit model of HIF [26]-[28] as depicted

that their values depends on the line characteristics of the in Fig. 3, the relation between the fault current and the voltage
system. The V a( X ) (n ) in (3) is a function of unknown vector (X )
at the fault location (V Fault (X )
) can be expressed as:
(n ) which is named FI .
(X )
H Rem( X ) and the known data I (ABC
X )

(V Fault
(X )
(t ) U P ) / R HIF , V Fault
(X )
(t )  U P
The least-square error based algorithms can be employed to 
(X ) (X )
 (X ) (6)
estimate the unknown vector for samples in the pre-fault data I HIF( X ) (t )  (V Fault ( X ) (t )  U N ) / R HIF , V Fault ( X ) (t )  U N
(X )

window. The least-square error problem can be written as [19]: 


 0 , U N V Fault (X )
(X )
(t )  U P
N F N

 H 
2
H Rem( X )  arg min  FI (X )
, I (ABC
X )
(n ) V a( X ) (n )  (4)
n N F  2 N 1
 Rem( X )  where, UP and -UN represent inception voltages of arc during
positive and negative half cycles, and RHIF is the fault
where n is the sample number, N is the sample numbers per resistances. For the sampled value n, this equation can be
cycle (128 samples according to the 7680 Hz sampling written as:
frequency) and ∆t is the sampling time interval.

0885-8977 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2018.2874879, IEEE
Transactions on Power Delivery
6


I HIF( X ) (n )  sgn V Fault
(X )
(X )
(n ) U P . (V Fault 
(X )
(X )
(n ) U P ) / R HIF  (7) 3) Making the decision on the location of the HIF
The result of (11) is the criterion used for decision making

sgn U N V Fault (X )
(X )
(X )
(n ) . (V Fault(X )
(n ) U N ) / R HIF about the location of fault. The closer the assumed fault
location is to its actual location, the error in the calculated
where sgn function is defined as:
faulty phase voltage compared to the measured values
1 , n 0 becomes smaller. The assumed location that results the least
sgn(n )   (8)
0 , n 0 error is the most possible fault location. The proposed
algorithm can be presented in a single formula like:
The R.H.S. of (7) includes the calculated values of FarthestNode

(X )
V Fault ( X ) (n ) and three unknown parameters HHIF(X) = [UP, UN,
X HIF  arg min  [Er V ]
X 1( km ) (X )
(12)

RHIF]. Therefore, the I HIF( X ) (n ) in (7) can be considered as a or in a more completed formula like as:
function FIHIF( X ) of unknown vectors HHIF(X) and the known data. N F  2 N 1

 [ 2N  V ] (13)
FarthestNode
1
The least-square error based algorithms is employed to X HIF  arg min (1 )
(n ) V Fault
(1 )
(X )
(n )
X 1( km ) n N F
estimate the unknown vector for samples in the post-fault data
window: (1 )
where V Fault (n ) is obtained through the procedure described
(X )
N F  2 N 1

  
(n )  I HIF( X ) (n )  (9) previously.
2
H HIF( X )  arg min  FIHIF( X ) H HIF( X ) ,V Fault
(X )

n N F
 (X )
The proposed method investigates locations along all of the
feeders of the system, to find the most possible location of
and I HIF( X ) (n ) is obtained at each sample employing (5).
HIF. Except in the rare cases where there are multi branches
Calculating the voltage of the faulty phase at the source along the main feeder with the same characteristics and similar
node: After obtaining the parameters of the assumed HIF, the loads, it is capable to detect the actual faulted branch,
voltage of the faulty phase at the assumed fault location employing the difference in load of each branch. Generally, if
(X )
V Fault (X )
for the n  N F is estimated as follows: more than one possible location of fault exists, same as the
other fault location methods that uses the measuring only in the
I HIF( X ) (n ).R HIF  U P , I HIF( X ) (n ).R HIF  U P main node, the HIF indicators [28] will be needed to specify

 (10)
V Fault ( X ) (n )  I HIF( X ) (n ).R HIF U N , I HIF ( X ) (n ).R HIF  U N
(X )
the actual faulty feeder.


 V ( X ) (n ) , U N <I HIF( X ) (n ).R HIF  U P
III. SIMULATIONS AND RESULTS
The ~ symbol means that it is an estimated value.
The voltage at the source node where the protection and the A. Simulations
measuring device is installed, can be obtained through a In order to investigate the performance of the proposed
(X )
backward power flow starting from V Fault . If the assumed algorithm for HIF location, the IEEE 34-node test system was
(X )
selected and simulated in EMTP-RV. Fig. 4 illustrates the
HIF location sites on the actual location of the fault, the
(1 )
modified single line diagram of this distribution feeder. There
calculated voltage at the source node V Fault (X )
and the are 5 line configurations with 4 different conductors in this
measured voltage V (1 ) will be similar, as well. Inaccuracy in system. There are single-phase (phase-a or phase-b) and three-
the calculated voltage indicates there might be an error in the phase lateral feeders in this system. The farthest distance from
estimated HIF parameters including the fault location. This a node to the source node is about 58 km. The line
process is repeated for all of the possible locations of fault. configurations and the characteristics of the conductors were
The most accurate estimation of the faulty phase voltage at the modeled as presented in [30]. Loads (single-phase or multi-
source node reveals the location of the HIF. phase) of the network were modeled as spot loads by series
impedances. The impedances of the loads at each node were
Calculation of the error: The error between the calculated obtained with the nominal voltage of the system (24.9 kV).
values of the faulty phase voltage and the measured/assigned HIFs were modeled by the two-diode model as illustrated in
values at the source node (V (1 ) (, for the post-fault data Fig. 3 [26]-[28]. This model is a simple model that can
present all characteristics of HIF like nonlinearity, asymmetry
window, can be calculated by:
and randomness in different cycles [2], [27]-[28].
N F  2 N 1
T F N (11)

The values of fault resistance (RHIF) were set in the range of
Er  V (1 )
(n ) V Fault
(1 )
(n ) , N F  
t 2
(X )
V (X ) n N F 100 to 1000 ohm and assumed to be constant during the two
cycles after inception of fault. UP were set between 2.1 to 7.5
where n is the sample number, N is the sample numbers per kV and UN in the range of 2.1 to 8.85 kV randomly in different
cycle and ∆t is the sampling time interval. It should be simulation while UP ≤ UN.
emphasized that the estimated fault inception time (TF) is half The simulation time-step was set to 868 nanoseconds and
cycle delayed to compensate the uncertainty in the estimation. the waveforms were stored with the rate of 150, which makes

0885-8977 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2018.2874879, IEEE
Transactions on Power Delivery
7

the sampling frequency as 7.68 kHz or representing 128

samples per cycle that is in the range of available commercial
distribution IEDs. The fault inception time was set from 100 to
140 milliseconds randomly in different simulations. High
impedance faults were simulated at different locations and
phases of the network and the required data were stored and
prepared for further processing.
An example of phase current at the substation before and
after occurrence of an HIF has been shown in Fig. 5 and the
HIF current has been illustrated in Fig. 6. The fault location is
Fig. 5. An example of the phase current at the source node (substation) in the
31.63 km from the source node. Estimated fault currents, with simulated distribution system before and after an HIF occurrence at node 816,
the procedure described in Section II, assuming fault locations phase-a, with RHIF=750 Ω.
at the distances of 10 and 32 km from the source node, are also
shown in Fig. 6. As it can be seen, assuming the fault location
next to the actual location of fault, the similarity of the
estimated fault current to the actual values is resulted. It is
worth to note that the error between the estimated fault current
and its real values, leads to error in HIF parameter estimation.
As an example, in Table I, the real values and the estimated
HIF parameters assuming HIF at different locations, have been
displayed. Assigning the fault location closer to its actual
location, more accurate estimation will be obtained and vice
versa. Higher accuracy in the HIF parameters leads to smaller
error between the calculated voltage and the measured value at
the source node. These errors for the HIF at different locations
along the network feeder, considering 1km distance between Fig. 6. HIF current at the fault location and the estimated values with the
fault current estimation procedure assuming HIF at two different locations.
adjacent locations, have been illustrated in Fig. 7.
(The exact fault location was at 31.63 km).
As it can be observed in Fig. 7, as the distance of the
assumed fault location from the actual fault location increases,
the error value increases and vice versa and the calculated TABLE I
TRUE AND ESTIMATED PARAMETERS OF AN HIF OBTAINED BY THE FAULT
error as a function of distance from the source node is a PARAMETER ESTIMATION PROCEDURE ASSUMING HIF AT DIFFERENT
convex function. Therefore, the gradient based algorithm can LOCATIONS
be applied to optimize the searching process and reduce the
HIF parameters
number of possible locations and the computations burden for Assumed Fault Distance
UP (v) UN (v) RHIF (Ω)
the fault location algorithm [31].
Exact values: 31.6 km
5250 6000 750
B. Results (used in the simulation)
10 km 5458.79 6264.17 883.88
The results of the proposed algorithm for HIF location in the 20 km 5405.86 6182.69 816.73
test distribution system (Fig. 4), for different fault resistance at 32 km 5273.74 6013.73 746.7
various locations of the system, obtained by assuming HIF 40 km 5166 5891.19 703.27
55 km 4889.81 5594.33 638.34
along the feeders of the network at 1 km interval, are presented
in Table II. The fault location error was calculated by:
Etimated fault location -Actual fault location (14)
FLError (%)=100 
Length of the longest feeder

848

822 846

820 844
864
818 842
800 802 806 808 812 814 824 826 834 860 836
858 840
816
69/24.9 kV

832
888 890 838
810 852

Fig. 7. Error in the estimated voltage values obtained by (10), considering HIF
828 830 854 856
at different locations for a HIF with RHIF=750Ω at 31.6 km distance from the
Fig. 4. Single line diagram of the test system (modified version of the IEEE source node (the HIF that has been presented in Fig. 6).
34-node test system).

0885-8977 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2018.2874879, IEEE
Transactions on Power Delivery
8

All calculations have been carried out in a desktop

computer with Intel core i3 2.4 GHz CPU and 4 GB RAM. TABLE III
The average computational time for each case of the proposed HIF LOCATION ERROR FOR THE HIGH LOAD OPERATING CONDITION WITH
±10% UNCERTAINTY IN THE LOAD VALUES
HIF location algorithm (investigating about 80 possible
locations of HIF) was about 90 seconds in the MATLAB Location of Distance from HIF Location Relative Error (%)
HIF Metering Point for Different Fault Resistances
environment. (Node No.) (km) R=250 Ω R=500 Ω 1000 Ω
The results presented in Tables II are obtained assuming the 806 1.31 0.54 -1.19 0.53
data of loads at each node are available, but the load 808 11.14 0.24 -1.47 1.97
impedances that were used for current-voltage calculation 814 27.1 2.67 6.18 -0.75
along the distribution network have been modified between - 816 31.63 9.76 2.77 2.82
824 34.74 1.28 4.67 -0.45
10% up to +10% randomly. 830 41.32 2.25 2.29 9.22
To investigate the effect of loading, the performance of the 852 52.55 2.68 7.88 7.74
proposed algorithm, were also evaluated for high load 834 55.82 -0.32 -0.31 4.88
operating condition (increasing the total system loads about 820 46.83 -2.03 -0.30 -0.30
810 12.91 5.79 9.25 7.39
30%) and low load operating condition (decreasing the loads
Standard Deviation of Errors 3.37 3.88 3.76
about 30%). The results of HIF location considering different Mean of Errors 2.76 3.63 3.61
loading conditions and for different values of fault resistances Mean of Mean of Errors 3.33
have been shown in Table III and Table IV. As it can be
observed, increasing the load values while the fault current is
the same, decreases the accuracy of the fault location
algorithm due to the uncertainty exposed to the load values TABLE IV
HIF LOCATION ERROR FOR THE LOW LOAD OPERATING CONDITION WITH
and numerical differentiation, but the results are still
±10% UNCERTAINTY IN THE LOAD VALUES
acceptable.
The results of the proposed HIF location with 1% white Location of Distance from HIF Location Relative Error (%)
HIF Metering Point for Different Fault Resistances
noise added to the voltage and current signals at the source (Node No.) (km) R=250 Ω R=500 Ω 1000 Ω
node have been presented in Table V. The accuracy of the 806 1.31 0.54 -1.19 -1.17
algorithm decreases in the noisy condition due to numerical 808 11.14 0.24 -1.47 1.97
differentiation in the calculation of voltage along the system. It 814 27.1 0.97 -0.75 -0.75
is recommended that in case of receiving noisy signals, 816 31.63 1.09 1.07 -0.64
824 34.74 1.28 1.26 4.75
appropriate suppression measures like signal de-noising must 830 41.32 0.55 -1.18 4.02
be taken. 852 52.55 2.68 -0.79 -5.88
834 55.82 3.15 -0.31 6.62
TABLE II 820 46.83 -0.30 1.43 -0.30
HIF LOCATION ERROR FOR THE NORMAL LOAD OPERATING CONDITION 810 12.91 4.05 4.05 3.98
WITH ±10% UNCERTAINTY IN THE LOAD VALUES Standard Deviation of Errors 1.33 1.01 2.33
Mean of Errors 1.49 1.35 3.01
Location of Distance from HIF Location Relative Error (%)
Mean of Mean of Errors 1.94
HIF Metering Point for Different Fault Resistances
(Node No.) (km) 100 Ω 250 Ω 500 Ω 750 Ω 1000 Ω
806 1.31 0.54 -1.17 -1.19 -2.92 -1.17
Middle of
6.23 0.39 0.39 0.38 2.12 -4.81 TABLE V
806-808
HIF LOCATION ERROR FOR THE NORMAL LOAD OPERATING CONDITION
808 11.14 0.23 0.24 -1.50 1.94 -1.50
WITH ±10% UNCERTAINTY IN THE LOAD VALUES AND 1% ADDITIVE WHITE
Middle of NOISE IN THE MEASUREMENT SIGNALS
16.85 -0.26 -0.25 -0.26 -0.26 -1.95
808-812
812 22.57 0.98 0.98 -0.74 2.72 -2.48 Location of Distance from HIF Location Relative Error (%)
814 27.1 0.17 1.90 -1.56 -1.54 -1.56 HIF Metering Point for Different Fault Resistances
816 31.63 -0.64 1.07 1.09 -0.64 -5.74 (Node No.) (km) R=250 Ω R=500 Ω 1000 Ω
824 34.74 -0.45 1.28 -0.44 -0.45 4.75 806 1.31 -1.17 -1.19 -2.87
854 41.32 0.55 0.56 -1.18 -1.16 -1.18 808 11.14 0.24 -1.50 -1.50
Middle of 814 27.1 0.98 -0.74 -0.75
46.93 1.62 1.59 1.62 -0.11 -0.11
854-852 816 31.63 2.77 2.82 -0.63
852 52.55 0.95 4.42 2.63 2.68 -0.79 824 34.74 1.28 -0.44 -12.58
834 55.82 1.39 1.42 3.15 4.80 4.88 830 41.32 0.56 -1.18 4.02
846 57.34 4.05 3.98 4.05 7.52 7.39 852 52.55 4.42 6.04 7.88
836 57.25 0.43 0.43 3.83 5.63 7.37 834 55.82 -0.32 8.35 6.62
820 46.83 3.17 -2.03 -0.30 -0.30 -0.30 820 46.83 -0.30 -0.30 8.37
810 12.91 -0.16 1.54 1.54 -1.86 3.25 810 12.91 1.54 1.54 1.54
Standard Deviation of Errors 3.21 1.11 1.21 1.22 2.11 Standard Deviation of Errors 1.32 2.67 4.02
Mean of Errors 2.18 1.00 1.45 1.59 2.29 Mean of Errors 1.36 2.41 4.68
Mean of Mean of Errors 1.856 Mean of Mean of Errors 2.8147

0885-8977 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2018.2874879, IEEE
Transactions on Power Delivery
9

IV. CONCLUSIONS (p+1) at time instant (nΔt) for the phase-a according to (1),
In this paper a time domain algorithm for HIF location in can be obtained as (15).
distribution systems has been introduced that only requires the B. Calculating load current
measurements provided by an IED at the main substation. The
The voltage across phase-a load impedance, considering
method estimates the most possible location of HIF based on
notation in Fig. 2, is obtained by:
the degree of accuracy in the estimated voltage of the faulty
d (p)
phase compared with the measured value. V a( p ) (t ) R Load-
(p)
a I Load-a (t )  L Load-a
(p) (p)
I Load-a (t ) (16)
dt
The simulations performed on a modified version of IEEE
34-node test feeder with various HIF parameters, have shown Employing the trapezoidal method [21], the load current at
that the proposed method is well capable of locating high each sample is obtained by (17).
impedance faults despite an uncertainty exposed in the load
C. Downstream Network Equivalent Calculation
data. As the proposed method employs time domain analysis, it
is applicable for unsymmetrical and nonhomogeneous, To estimate the equivalent model of the downstream
distribution networks with unbalanced loading as well as the network from the assumed fault location, the circuit model
balanced symmetrical system. presented in Fig. 3 and the (3) is considered. Applying the
Exact load modelling is one of the challenges for trapezoidal method to (3) for the sampled signals in the pre-
distribution network modeling. In this paper, based on fault data window, yields (18).
practical measurements of load behavior and works done by REFERENCES
other researchers, the constant impedance loads have been
used. Including other load models is planned in the future [1] “SEL-451-5 Relay Protection, Automation, and Control System
work. Fault location algorithms in distribution system with Instruction Manual,” Schweitzer Engineering Laboratories, Pullman,
Washington 2013.
DGs are different compared to FL algorithms in the
[2] C. Benner and B. Russell, “Practical High-Impedance Fault Detection
conventional distribution system. In the case of distribution On Distribution Feeders,” IEEE Trans. Ind. Appl., vol. 33, no. 3, pp.
system with DGs, it is required that their contribution on fault 635–640, May/Jun. 1997.
condition be measured and processed in an aggregated [3] IEEE Guide for Determining Fault Location on AC Transmission and
algorithm with the others and the main source, as well. As the Distribution Lines, IEEE Standard C37.114-2014.
usage of DGs are expanding in distribution system, developing [4] A. Bahmanyar, S. Jamalia, A. Estebsarib, and E. Bompardb, “A
comparison framework for distribution system outage and fault
an HIF location algorithm in distribution system with DGs, is location methods,” Electric Power Sys. Res., vol. 145, pp. 19-34,
considered as future work of the authors. Apr.2017
[5] M. Kezunovic, “Smart fault location for smart grids,” IEEE Trans.
Smart Grid, vol. 2, no. 1, pp. 11-22, Mar. 2011.
APPENDIX
[6] S. Kazemi, M. Lehtonen, and M. Fotuhi-Firuzabad, “A comprehensive
approach for reliability worth assessment of the automated fault
A. Calculating voltage drop along a series impedance of a management,” IEEE/PES Transmiss. Distrib Conf., New Orleans-LA,
line: 2010.
[7] A. C. Depew, J. M. Parsick, R. W. Dempsey, C. L. Benner, B. D.
Employing the Gear 2nd order method for calculating Russell, and M. G. Adamiak, “Field experience with high-impedance
voltage drop along the line section between nodes (p) and fault detection relays,” in Proc. EEE/PES Transmiss. Distrib. Conf.
Exhibit, pp. 868-873, 2006.

V a( p ,( p 1)) (n )  R aa( p ,( p 1)) .I a( p ,( p 1)) (n )  R ab( p ,( p 1)) .I b( p ,( p 1)) (n )  R ac( p ,( p 1)) .I c( p ,( p 1)) (n ) 
1  ( p ,( p 1))
. Laa
2t 

. 3I a( p ,( p 1)) ( n )  4 I a( p ,( p 1)) ( n  1)  I a( p ,( p 1)) ( n  2)   (15)
( p , ( p 1))
Lab 
. 3I b( p ,( p 1)) ( n )  4 I b( p ,( p 1)) ( n 1)  I b( p ,( p 1)) ( n  2)  
( p , ( p 1))
Lac .  3I c ( p , ( p 1))
( n )  4 I c( p ,( p 1)) ( n 1)  I c( p ,( p 1)) (n  2)  

t
2

V a( p ) (n ) V a( p ) (n  1) R Load-
(p)
a . I Load-a ( n  1)  L Load-a . I Load-a ( n  1)
(p) (p) (p)
 (17)
I Load-a (n ) 
(p)

t
(p)
R Load-a.  L(Load-
p)
a
2

   
V a( X ) (n )  LE Rem ( X ) .  rLL . I a( X ) (n )  I a( X ) (n  1)  l LL .  2 / t  . I a( X ) (n ) I a( X ) (n  1) 

r LM .I
( n )  I ( n 1)   l .  2 / t  .  I (n ) I (n  1)  
b
(X )
b
(X )
LM
(X )
b
(X )
b (18)
r LM .I
(n )  I (n  1)   l .  2 / t  .  I (n ) I (n  1)   
c
(X ) (X )
c LM
(X )
c
(X )
c

Rem ( X )
R Load a
.  I ( n )  I ( n  1)   L
(X )
a
(X )
a
.  2 / t  .  I ( n )  I ( n  1)  V ( n  1)
Rem ( X )
Load a a
(X ) (X )
a a
(X )

0885-8977 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2018.2874879, IEEE
Transactions on Power Delivery
10

[8] A. N. Milioudis, G. T. Andreou, and D. P. Labridis, “Detection and [29] M. Sarlak, and S.M. Shahrtash, “High-Impedance Faulted Branch
location of high impedance faults in multiconductor overhead Identification Using Magnetic-Field Signature Analysis,” IEEE
distribution lines using power line communication devices,” IEEE Trans. Power Delivery, Vol. 28, No. 1, pp. 67-74, Jan. 2013.
Trans. Smart Grid, vol. 6, no. 2, pp. 894-902, Mar. 2015. [30] IEEE PES Distribution System Analysis Subcommittee.
[9] J. Zhou, B. Ayhan, C .Kwan, S. Liang, And W. Lee, “High- Distribution Test Feeders, [Online]. Available:
http://www.ewh.ieee.org/soc/pes/dsacom/ testfeeders.
performance arcing-fault location in distribution networks,” IEEE
[31] C. Orozco-Henaoad, and et al, “Active distribution network fault
Trans. Industry Applications, vol. 48, no. 3, pp. 1107–1114,
location methodology: a minimum fault reactance and fibonacci
May/Jun2012.
search approach,” Int. Journal of Elect. Power & Energy Syst., vol.
[10] M. Mohammadi Alamuti, and et al, “Intermittent fault location in
84, pp. 232-241, 2017.
distribution feeders,” IEEE Trans. Power Del., vol. 27, No. 1, pp.
96-103, Jan. 2012.
[11] W. Zhang, and et al,“ Multi-cycle incipient fault detection and
location for medium voltage underground cable,” IEEE Trans. S. Hamid Mortazavi received the B.Sc. degree from
Power Del., vol. 32, no. 3, pp 1450-1459, Jun. 2017. Ferdowsi University of Mashhad, Mashhad, Iran in
[12] A.R. Herrera-Orozco, and et al, “Incipient fault location 2002, the M.Sc. degree from Iran University of
formulation: a time-domain system model and parameter estimation Science and Technology, Tehran, Iran in 2007, and
approach,” Electrical Power and Energy Systems, vol. 90, pp. 112- the Ph.D. degree from Semnan University, Semnan,
123, Sep. 2017. Iran in 2018 all in electrical engineering. His research
[13] F. C. L. Iurinic, and et al, “Distribution systems high impedance interests include power system protection power
fault location: a parameter estimation approach,” IEEE Trans. system restoration, and application of artificial intelligence and digital
Power Del., vol. 31, no. 4, pp. 1806-1814, Aug. 2016. signal processing techniques in power system.
[14] S. Lotfifard, M. Kezunovic, and M. J. Mousavi, “A systematic
approach for ranking distribution systems fault location algorithms Z. Moravej (SM’11) received the B.E and M.E
and eliminating false estimates,” IEEE Trans. Power Del., vol. 28, degrees in electrical engineering from Bangalore
no. 1, pp. 285-293, Jan. 2013. University, India, in 1985 and 1991, respectively, and
[15] M. M. Saha, J. J. Izykowski, and E. Rosolowski, Fault Location on the Ph.D. degree from Banaras Hindu University,
Power Networks, 1st ed. London, U.K.: Springer, 2010. India, in 2001. Currently, she is an Associate
[16] R. H. Salim, and et al, “Extended Fault-Location Formulation for Professor with the Electrical and Computer
Power Distribution Systems,” IEEE Trans. Power Del., vol. 24, no. Engineering Faculty, Semnan University, Semnan,
2, pp. 508-516, Apr. 2009. Iran. Her areas of research interest include the application of artificial
[17] “Distribution fast simulation and modeling technical update: first intelligence and machine learning in power system protection, power-
sensor location strategy for the enhancement of distribution state quality monitoring, and substation automation systems. Dr. Moravej is a
estimation,” EPRI, Palo Alto: CA and EDF R&D, Clamart: France, member of CIGRE and a member of IAEEE of Iran.
1013831, 2008.
[18] H. Cifuentes-Chaves, J. Mora-Flórez, and S. Pérez-Londono, “Time S. Mohammad Shahrtash was born in Tehran, Iran,
domain analysis for fault location in power distribution systems 1960. He received the B.Sc. degree in electrical
considering the load dynamics,” Elect. Power Syst. Res., vol. 146, engineering from Abadan Institute of Technology
pp. 331–340, May. 2017. (AIT), Abadan, Iran in 1980, the M.Sc. degree in
[19] W. Zhang, Y. Jing, and X. Xiao, “Model-based general arcing fault electrical engineering from UMIST, England in 1985,
detection in medium-voltage distribution lines,” IEEE Trans. and the Ph.D. degree in electrical engineering from
Power Del., vol. 31, no.5, pp. 2231-2241, Oct. 2016. Sharif University of Technology, Iran in 1995.
[20] S. H. Mortazavi, Z. Moravej, and S. M. Shahrtash, “A hybrid Currently, he is a Professor and the Head of Center of Excellence for
method for arcing faults detection in large distribution networks,” Power System Automation and Operation (CEPSAO), Iran University of
Int. Journal of Elect. Power & Energy Syst., vol. 94, pp. 141-150, Science and Technology. His main research areas are condition
Jan. 2018. monitoring, protection, electromagnetic transient analysis, and power
[21] C. A. Thompson, “A study of numerical integration techniques for system studies.
use in the companion circuit method of transient circuit analysis,”
Purdue Univ. School of Elec. Eng., IN: USA, Apr. 1992. [Online].
Available: http://docs.lib. purdue.edu/ecetr/297/.
[22] J. Arrillaga and N. Watson, Power Systems Electromagnetic
Transient Simulation, London, U.K.: The Institution of Engineering
and Technology (The IET), (reprint) 2007.
[23] C.A. Reineri and C. Alvarez, “Load research for fault location in
distribution feeders,” IEE Gener. Transm. Distrib., vol. 146, no. 2,
pp. 115–120, Mar. 1999.
[24] “Distribution fault location: field data and analysis,” EPRI, Palo
Alto: CA, 1012438, 2006.
[25] Statistics and Machine Learning Toolbox™ User's Guide, Natick-
MA, USA: The Math Works Inc., 2016.
[26] A. E. Emanuel, and E. M. Gulachenski, “High impedance fault
arcing on sandy soil in 15 kV distribution feeders: contribution to
the evaluation of the low frequency spectrum,” IEEE Trans. Power
Del., vol. 5, no. 2, 676–686. Apr. 1990.
[27] Z. Moravej, S. H. Mortazavi, and S. M. Shahrtash, “DT-CWT
based event feature extraction for high impedance faults detection
in distribution system,” Int. Trans. Elect. Energy Syst., vol. 25,
no.12, pp. 3288-3303, Dec. 2015.
[28] Y. Sheng, and S. M. Rovnyak, “Decision Tree-Based Methodology
for High Impedance Fault Detection,” IEEE Trans. Power Del.,
vol. 19, no. 2, 533–536. Apr. 2004.

0885-8977 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.