IEEE POWER & ENERGY SOCIETY SECTION

Received 9 July 2024, accepted 11 August 2024, date of publication 15 August 2024, date of current version 27 August 2024.
Digital Object Identifier 10.1109/ACCESS.2024.3443917

Detection and Location of Open Conductor Faults

for Power Distribution Networks Using a
Contingency Analysis Approach
GI-DO SIM , (Student Member, IEEE), HYO-SEOP IM, (Student Member, IEEE),
JOON-HO CHOI, (Senior Member, IEEE), SEON-JU AHN , (Member, IEEE),
AND SANG-YUN YUN , (Member, IEEE)
Department of Electrical Engineering, Chonnam National University, Gwangju 61186, South Korea
Corresponding author: Sang-Yun Yun ([email protected])
This work was supported in part by Korea Electric Power Corporation (KEPCO) through the KEPCO Research Institute under Grant
R22DA08, and in part by the Regional Innovation Strategy (RIS) through the National Research Foundation of Korea (NRF) funded by the
Ministry of Education (MOE) under Grant 2021RIS-002.

ABSTRACT This paper presents a method for the detection and location of open conductor faults (OCFs)
in power distribution networks using contingency analysis (CA) based on unsymmetrical power flow
calculations. The proposed method is designed for use in distribution management systems and broadly
comprises two stages. The first stage is CA in study mode. This study presents a CA technique for detecting
events different from those of the transmission network to evaluate network security. Accordingly, OCFs
that can occur in the target network are analyzed using the unsymmetrical power flow method regarding
the sections of the feeder remote terminal units (FRTUs), and the results are organized as a dataset for
OCF CA (DOCA). The DOCA for the distribution lines (D/L) where topology changes have occurred
is partially changed, and a detailed fault analysis is performed. The second stage is OCF detection and
location in a real-time operating mode. This study proposes a method that compares the mean square error
between real-time FRTU measurements and DOCA. The reliability of the proposed method is verified using
MATLAB/Simulink to conduct various case studies on the IEEE 33 bus and 118 bus standard test models. The
advantages of the proposed method can be summarized as follows: First, by comparing the distance between
the entire set of measurements and DOCA, the proposed method achieves a considerably higher detection rate
than those achieved by existing methods that use a single measurement element. Second, when calculating
the distance between the two sets, the proposed method uses weights based on the magnitude variation of
the comparison elements, which results in almost no false positives, making it robust to realistic conditions
such as measurement sensor errors and communication delays. Third, fast responses during operations are
also possible because detection and location are performed simultaneously.

INDEX TERMS Open conductor fault, contingency analysis, unsymmetrical power flow method, fault
detection and location method, power distribution network.

I. INTRODUCTION conductor faults (OCFs) [1]. Furthermore, the detection and

The extreme climate fluctuations caused by abnormal location of OCFs at power distribution networks are becom-
weather and global warming are causing severe socio- ing increasingly complex because of various reasons, such
economic damage, including large-scale fires due to open as an accelerated connection of distributed energy resources
(DERs), the expansion of covered lines and reinforced-
The associate editor coordinating the review of this manuscript and concrete poles, and the use of various controllers for the
approving it for publication was Sarasij Das . efficiency of network operations.

2024 The Authors. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.
114304 For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ VOLUME 12, 2024
G.-D. Sim et al.: Detection and Location of OCFs for Power Distribution Networks

In power distribution networks, faults have traditionally methods because they use certain detection elements of field
been detected on the basis of momentary fluctuations in the devices, and they are unable to consider the effects of false
measurement values of field terminal units and protective positives (FPs) and false negatives (FNs) in field devices
equipment. However, OCFs are accompanied by dynamic when the DMS determines the final location.
changes after the instantaneous disconnection of conductors Third, OCF detection research was also conducted using
or connections. In addition, various fault behaviors (high- artificial intelligence [11], [12], [13], [14]. One proposed
impedance faults, overcurrent faults, arcing, etc.) can occur method used an artificial neural network-based algorithm to
depending on whether there is contact with a medium after learn actual faults and the voltage and current of an EMTP
the fault, which can lead to secondary damage such as fires model and made comparisons with relay measurement val-
and electric shock injuries. Therefore, there is a need for a ues [11]. Another method performed detection by removing
detection and location technology that is specific to OCFs and voltage and current noise and learning features by the discrete
different from existing methods. wavelet transform [12]. Further, a technique that extracted
Various studies have been conducted on OCF detection fault features using the variational mode decomposition tech-
and location. First, attempts were made to detect OCFs using nique and ultimately determined faults via a support vector
feeder remote terminal units (FRTUs) and their measurement machine was also presented [13]. Another study performed
values [2], [3], [4], [5], [6]. A voltage-based OCF detection detection by using deep learning to extract voltage and cur-
method using voltage outage [2] and OCF detection methods rent features from the frequency band directly after OCFs
using the current unbalance of negative-sequence current and applied them to a field phasor measurement unit [14].
compared to positive-sequence current were proposed [3], In addition, a method that extracted features from sensor
[4]. These deterministic detection methods have the advan- signals and detected faults using a graph-embedded low-rank
tage of easy applicability to actual networks because the tensor learning machine was proposed [15]. In another study,
fluctuations of a certain element are set as the detection the characteristics of three-phase current waveform data were
criteria; however, they are unavoidably influenced by the analyzed using convolutional neural networks to estimate the
network environment and fault location. To resolve these type, phase, and location of faults, including high-impedance
problems, detection methods utilizing multiple detection ele- faults [16]. The abovementioned methods perform learning
ments as open fault condition (OFC) were proposed [5]. using instantaneous value-based data; however, they are not
However, this approach has the limitation of being vulnerable realistic because the historical stored data of the actually
to any increase in the DER connection and end-of-line faults operating DMS is based on root-mean-square values. In addi-
that produce little fluctuation. A method was proposed to tion, there is a problem in that the detection criteria can vary
calculate the impedance during low-impedance ground fault according to the environment of the target network.
(LIGF) and high-impedance ground fault (HIGF) conditions Fourth, studies were conducted on the detection and loca-
after OCF. This impedance was then reflected in the operating tion of OCFs by contingency analysis (CA) [17], [18].
range of distance relays for detection [6]. This method has One proposed method calculated the Thevenin equivalent
limitations in that because of the range of fault impedance impedance for each symmetrical component assuming a
considered, it cannot consider dynamic changes in the OCF single-phase OCF in a line section and used the load-side
aspect. zero-sequence component current fluctuation rate for the
Second, studies were conducted on primary detection using assumed substation secondary-side zero-sequence compo-
field-installed devices and determining final location either nent voltage as the detection criterion [17]. This method is
by collecting data from these devices in the distribution limited by the problem of setting the voltage fluctuation cri-
management system (DMS) or via communication between terion and the fact that it requires a measurement environment
them [7], [8], [9], [10]. Research was conducted on meth- that guarantees simultaneity. Further, a method that estimated
ods that detect OCFs using voltage outage information from the voltage of fault candidate points and ultimately performed
feeder-end voltage sensors and identify fault sections by power flow calculations to select the final fault locations as
topology analysis [7]. A method was proposed to detect OCFs the locations with the smallest deviation between the mea-
by measuring voltage unbalance with smart meters at the end sured voltage and the calculated voltage at the substation
of a line and identifies the fault section by a common path secondary side was also studied [18]. This method can be
search between detection devices [8]. Furthermore, a method applied to common network topologies but has problems
was proposed to detect faults from the voltage unbalance in in that it does not consider DER and takes a long time to
the secondary side of the distribution transformer and iden- determine the location because of its repeated power-flow
tify OCF locations via device-to-device communication [9]. calculations.
In addition, a method was proposed to detect OCFs by calcu- To summarize the above literature review, methods that
lating transient zero-sequence admittance at the FRTU and are based on field devices and methods that make a final
fault location from the difference in zero-sequence admit- OCF decision and perform location via cooperation between
tance at the source- and load-side of the fault point [10]. These field devices and DMS have problems in that their detection
methods have the same limitations as the abovementioned criteria are set according to the target network and they are

VOLUME 12, 2024 114305

 G.-D. Sim et al.: Detection and Location of OCFs for Power Distribution Networks

vulnerable to operational noise (FPs, FNs, communications II. OCF CHARACTERISTIC ANALYSIS AND REVIEW OF
delays and failures, etc. in field devices). Methods that use CA PREVIOUS STUDIES
have problems in that they do not consider the various OCF OCFs occur because of factors such as mechanical impacts
behaviors, can only be applied to certain network topologies, from the external environment, mechanical defects in con-
and do not have a clear basis for setting the detection criteria. nections, and corrosion due to deterioration; they are defined
In addition, only methods that apply to real-time operational as unintentional single-phase or double-phase openings in
situations are mentioned, and insufficient consideration is energized conductors. In OCFs, the measurement values of
given to the time spent and computations performed by FRTUs fluctuate during faults because of two reasons:
DMSs that manage hundreds or thousands of distribution 1) Changes in the fault itself: In an OCF, the conductor
lines (D/Ls). itself is mechanically separated. Therefore, the aspects of the
To resolve these problems, this paper proposes a real-time OCF can dynamically change, commencing directly after the
OCF detection and location method that performs offline CA, fault. As such, OCFs cannot simply be defined as openings
which considers various behaviors of OCF and compares the in an electrical circuit but can be broadly classified according
results with real-time FRTU measurement values. CA, which to three aspects: (1) a high-resistance state is maintained,
is a detailed fault-analysis method that requires consider- in which the power line does not contact a low-resistance
able computation and time, is performed separately offline. medium after the fault; (2) a separated power line contacts
During real-time operations, OCF detection and location are a low-resistance medium such as the earth or a pole, and
performed by comparing the mean squared error (MSE), the fault changes into a ground fault (GF); (3) the situation
considering the FRTU measurement error. The contributions changes into an OCF after the GF occurs.
of this paper can be summarized as follows: 2) Fluctuations in measurement values due to the network
1) A CA method based on an unsymmetrical power flow structure: If an OCF occurs in a power distribution network
method is proposed, considering OCF characteristics and that comprises simple loads, the voltage fluctuations on the
DERs. To overcome the limitations of certain detection ele- source side of the fault point are minimal, and the current is
ments, CA is employed to detect fluctuations in voltage, affected by the load drop caused by the fault [2], [4]. How-
current, and power factor (PF) angle, which are organized ever, various voltage and current fluctuations can occur on
into a dataset for OCF CA (DOCA). By comparing DOCA the load side of the fault point depending on the transformer
with the overall measurement set, the detection rate of the connection [5]. When a 1 load is connected to the primary-
proposed method is significantly higher than those of existing side 1 connection transformer and the Yg-Yg connection,
methods using a single measurement element. it acts as the current source after the OCF occurs, and a
2) When comparing the distance between the FRTU voltage close to a maximum of 0.5 pu can be maintained.
measurement set and DOCA, the proposed method applies However, the maintained voltage can fluctuate because of the
weights based on the magnitude variation of comparison internal voltage drop caused by the load current. In the case
elements (calculated in advance), thus resulting in almost no of a Yg-1 connection to supply an industrial high-voltage
FPs. The measurement error of sensors and the sensitivity customer (using a motor) or a transformer to connect a DER,
are also analyzed in advance, thereby affording OCF decision a maximum voltage of 1.0 pu can be maintained. The voltage
and location criteria suitable for the network. Therefore, the and current fluctuation patterns vary because of the distri-
method is robust to realistic conditions, such as measurement bution of the load and the amount of distributed generation
errors and communication delays that may occur in the oper- below the fault point due to the transformer connection.
ation of the distribution system. To determine whether an OCF has occurred, it is necessary
3) The dataset of DOCA is composed of the values of mea- to detect instantaneous changes. Therefore, OCF detection
surement elements when OCF occurs in a specific section. methods that use fluctuations in the measurement values
Detection and location are performed simultaneously when of field terminal devices have traditionally been employed.
determining the event by a comparison with the measure- Typical examples of these include a method that uses current
ment values of FRTU, which enables quick response during unbalance (the ratio of the positive phase sequence compo-
operation. nent to the negative phase sequence component) changes [3],
4) The proposed method of performing CA along with a method that detects OCFs by combining multivariate detec-
detection and location of OCFs uses only data and measure- tion elements into a decision tree [5], and methods that
ment infrastructure that are commonly possessed by most detect OCFs using voltage drop and voltage unbalance and
power companies today. Therefore, this study is highly appli- determine the locations of the OCFs by a path search [7],
cable to actual network operations. [8]. To examine the limitations of existing OCF detection
The remainder of this paper is organized as fol- methods, the test power distribution network shown in Fig. 1
lows. Section II analyzes the characteristics of OCFs. was used. In Fig. 1, the impedance of M.Tr. is 30% based
Section III describes the CA-based OCF detection and loca- on 100 MVA, and each load is assumed to be 0.5 MVA
tion algorithm. Section IV verifies the algorithm by case (lagging PF of 0.95). The distribution transformer for the load
studies. Finally, Section V concludes this paper. is 6% of 1-MVA capacity, and Yg-Yg and 1-Yg connections

114306 VOLUME 12, 2024

G.-D. Sim et al.: Detection and Location of OCFs for Power Distribution Networks

TABLE 2. Fluctuation in voltage and current measurements due to a

change in PV output and load demand.

FIGURE 1. Test network for a comparison of previous studies.

TABLE 1. Comparison of OCF detection methods using FRTU

measurement values.

connection feeder investment and radial feeder operation plan

based on N-1 CA to examine security when expanding power
distribution networks [19]. Zou et al. presented a method
that performed N-1 CA to examine the capacity limits of
main transformers and connection lines and find vulnerable
connection points during power distribution planning [20].
However, the detection of certain faults such as OCFs is
becoming increasingly difficult because of increasing opera-
tional complexity caused by the increase in DER connections;
therefore, the need for CA and its application in real-time
operations is emerging. Real-time OCF detection that is based
on contingency analyses of power distribution networks may
encounter problems in terms of the computation and analysis
time. In general, the number of D/Ls that are managed by a
single distribution control center range from at least several
hundred to over a thousand. Therefore, performing a detailed
analysis for each contingency case requires an enormous
are assumed. The capacity of photovoltaics(PV) is set at amount of computation and time and is almost impossible
0.5 MW and connected as Yg-1. The line is ACSR 160 mm2 to perform during real-time operations. Accordingly, it is
and has a total length of 5 km. The fault simulated a situation necessary to separate time slots for analysis and the use of
in which a high resistance was maintained after the OCF, analysis results. However, this process also has a problem
and the fault locations were set as the source and load sides in that the network situation during analysis, specifically
with the PV. Table 1 compares the detection results of each the load and PV generation, may differ from the real-time
of the existing methods when an OCF occurs. In the case of operational situation. To examine how this problem affects
fault location ‘‘F2’’, all methods succeeded at OCF detection; OCF detection, the fluctuations of voltage and current mea-
however, in the case of fault location ‘‘F1’’, there were some surement are simulated according to the DER output and
detection failures. Fault location ‘‘F1’’ revealed that FNs load demand fluctuations in the network shown in Fig. 1.
occurred because there was little fluctuation in the detection Table 2 presents the results when the same OCF as that
elements (Table 1) due to the effect of the voltage source that presented in Table 1 is simulated in fault location ‘‘F1’’.
was maintained by the DER connection transformer below As revealed by Table 2, changes in PV output did not have
the fault point. a significant effect on voltage fluctuations, and the current
In this paper, field measurements and analytical situation fluctuation rate according to load demand fluctuation was not
predictions are compared to resolve the problems in previous large either. Thus, a method that performs CA using only the
studies that depended solely upon measurement values for static information that is possessed offline and applies the
OCF detection. CA is a method for making analytic situation results to real-time operations can be proposed.
predictions regarding power networks. The CA that is used in
transmission network operations has been employed to eval- III. CONTINGENCY ANALYSIS-BASED OCF DETECTION
uate the operating vulnerabilities of the current situation by AND LOCATION ALGORITHM
implementing a power flow method assuming the occurrence A. OVERALL STRATEGY
of single or multiple network faults and the situations that The overall strategy for OCF detection and location is pre-
occur after they are removed. Because the fault in the distribu- sented in Fig. 2. As shown in Fig. 2, the proposed method
tion network has a small impact, research has been conducted comprises two stages: (1) a stage that creates a DOCA, which
on using it to plan objectives for certain target networks rather is a comparison dataset for OCF detection, by CA in a study
than real-time operations. Lin et al. presented an optimal mode, (2) a stage that performs OCF detection and location

VOLUME 12, 2024 114307

 G.-D. Sim et al.: Detection and Location of OCFs for Power Distribution Networks

the situation after the OCF behavior changes is performed.

In the case of a GF after an OCF, a fault analysis is performed
using a power flow method for a situation that reflects both
disconnections and GFs in the fault model, and the same pro-
cedure is performed for an OCF after a GF. By comparing the
CA results and measurement values after all aspect changes
were complete by data polling, this study aims to mitigate the
problem of FPs and FNs. Finally, when the MSE is lower than
the detection criterion, an OCF is detected, and its location is
estimated by minimum MSE estimation.

B. PERFORMING CA AND CREATING DOCA

IN STUDY MODE
The proposed OCF CA is divided into the process of screen-
ing and selecting OCF contingency cases and the process
FIGURE 2. Overall flowchart of the proposed OCF detection and location of creating the DOCA by detailed fault analysis. CCSS is
strategy.
performed to derive the contingency cases, which are the
targets of the detailed fault analysis; this process is shown
in the flowchart in Fig. 3. This process is performed using a
by comparing the minimum MSE of the DOCA and the
breadth-first search of the target D/Ls [21]. First, the network
measurements of FRTUs in a real-time operation mode. The
data are read by finding all D/Ls when CCSS is initially
summary of each process is as follows:
performed or the relevant D/Ls when a topology change has
1) CA stage: The processes are aperiodically or periodi-
occurred. In the case of an OCF, changes in the fault location
cally performed in study mode using the static data possessed
between the measurement equipment on both the source and
by the power company and the present topology data at the
load sides do not have a significant effect on measurement
time it is performed; the results of performing these processes
values. Therefore, contingency case selection is performed
are stored in the database of DMS for real-time operations.
using sections consisting of FRTUs. To achieve this, the lines
During the initial startup of the system or database update, all
and switches (SWs) are searched starting at the substation
cases are generated; in addition, a contingency case selection
secondary-side breaker. In the case of the lines, the search
and screening (CCSS) process is performed for the D/Ls with
continues, and in the case of the SWs, the section between
topology changes. In the contingency cases produced by the
two SWs is designated as a contingency case. If the path ends
CCSS process, a single-phase OCF in an N-1 contingency
situation is assumed for each type for all FRTU sections of
each of the D/Ls. FRTU measurements are predicted for each
fault case by a detailed fault analysis of each contingency
case. Each FRTU measurement that is calculated for the i-th
fault is defined as (1):
DOCAi = Vi1 , Ii1 , θi1 , . . . Vij , Iij θij

(1)
where V , I , and θ are the voltage, current, and PF angle mea-
surement values of the j-th FRTU. Fault analysis is repeated
for the n contingency cases derived from CCSS, and the
results are stored in the memory database of the operating
system.
2) OCF detection and location stage: The processes are
performed if an event detection alarm occurs in one or
more FRTUs during real-time operations. The DMS uses
data-polling commands to collect the present FRTU mea-
surements from the D/L where the alarm occurred. The
MSE between DOCAi and the present measurement dataset
is calculated by a Euclidean distance comparison between
the calculated value of the j-th measurement of DOCAi and
the present measurement of the j-th FRTU. The DMS data
polling of the FRTUs to apply the proposed method may
have a problem of FPs or FNs due to changes in the fault
behavior. To resolve this problem, a fault analysis based on FIGURE 3. Flowchart of contingency case selection and screening.

114308 VOLUME 12, 2024

G.-D. Sim et al.: Detection and Location of OCFs for Power Distribution Networks

after an SW, that section is designated as a contingency case.

If there are no more lines or D/Ls to search, the process is
finished, and fault points are designated within the line sec-
tions of the contingency case. The final contingency cases are
generated by making assumptions about the OCF behaviors at
the designated fault points. The final contingency cases store
the information given by (2):

casem,i = {FNm,i , FTm,i } (2)

where casem,i is the i-th contingency case of the m-th D/L,

FNm,i is the faulted node of that case, and FTm,i is the OCF
behavior. For a detailed analysis, previous studies used a
method that calculated the zero-sequence fault voltage and
current using a voltage and current relational formula based
on the equivalent impedance for each sequence [18] and a
method that estimated voltage using the power flow method
starting at the fault point [19]. Nevertheless, these methods
have problems in that they do not reflect the control char-
acteristics of DERs during faults and do not consider the
various OCF behaviors. Grainger et al. presented an OCF
analysis method that used the Thevenin equivalent resistance
in a situation where one or two phases are connected in a
three-phase open situation [22]. However, this method is dif-
ficult to apply to power distribution systems in which DERs
and loads are dispersed. In the present work, detailed fault
analysis is performed using an unsymmetrical three-phase
power flow method that reflects the control characteristics
of the power-conversion device of DER during faults and
the various behaviors of OCFs and can also consider the
characteristics of distributed loads and their connection trans- FIGURE 4. Flowchart of the detailed fault analysis for the contingency.

formers [23]. Fig. 4 shows the flowchart of the detailed fault

analysis, which comprises the following steps: ③ Calculation of the voltage for nodes using the voltage of
① Selecting the contingency case: The contingency cases the substation primary side: To do so, the current is calculated
that were previously selected via CCSS are selected, and a by only activating the voltage source of the substation (slack
fault model is applied according to the fault type. In this work, bus) in a state where the distribution generator and load are
several fault models are used. First, when a high resistance open, and the voltage of each node is calculated using the
is maintained after the OCF, it is modeled as an infinite calculated current as given by (3):
impedance inserted in the faulty line in a series. Second, cases    
Vs,1   Is
of GFs after OCFs are divided into source- and load-side GFs. Z 11 · · · Z n1
..   0 
 Vs,2 
  . ..
 
To model the open conductor itself, an infinite impedance is  ..  =  .. . .  ..  (3)

.  . 

inserted in series, and it is modeled in the form of the source-
Z1n · · · Znn
and load-side nodes grounded with a fault point resistance. Vs,n 0
Third, the case of an OCF after a GF is a situation in which
where IS is the current vector that flows from the substation,
a GF occurs because of contact with objects such as trees
Znn is the network impedance matrix, and Vs is the voltage
and then changes into an OCF owing to a physical impact.
vector due to the voltage source at the substation side.
Such faults are modeled in the same way as when a high
④ Calculation of the voltage change due to DERs and
resistance is maintained after an OCF, but it is assumed that
loads: The substation node is eliminated, and the voltage
protective equipment is operating near the GF fault point and
fluctuation in a situation in which only the DER and load are
the load-side DERs of the fault point are eliminated (due to
connected is calculated using (4):
low voltage).
② Creation of the Zbus matrix for the selected case: The 1V k1
   
  0
network admittance data and the fault model of the contin-  1V k  Z 11 · · · Z n1  k
2  . .. ..   I2 
 ..  =  .. . .  ..  (4)

gency case are applied to create an admittance matrix for the
. . 


three-phase unsymmetric power flow method. Finally, this Z1n · · · Znn
matrix is inverted to create the impedance matrix. 1V kn Ink

VOLUME 12, 2024 114309

 G.-D. Sim et al.: Detection and Location of OCFs for Power Distribution Networks

where k is the iteration number of power flow calculation. I k

is the network current vector and is calculated by dividing the
DER current and the load current values by the corresponding
node voltage. 1V k is the voltage vector change due to DER.
Next, 1V k is added to calculate the voltage using (5):

1V k1
 k+1     
V1 Vs,1
 V k+1   Vs,2   1V k 
 2   2
.  =  ..  +  .. (5)
 
 ..  .  .


Vnk+1 Vs,n 1V kn

where V k+1 is the node voltage vector.

⑤ Reflection of the current limitation of DERs and whether
the voltage has converged: The DER output current is cal-
culated using the connection point voltage. It is determined
whether this current exceeds the maximum current. If the
DER current is greater than the limit value, the current is FIGURE 5. Overall database of DOCA for OCF detection and location.

limited as given by (6):

SDG SDG and after the i-th OCF occurs, respectively. The weights are


 < IDG,max calculated using the relevant values as given by (8):
IDG = VDG VDG
SDG (6)
 IDG,max
 ≥ IDG,max 1Vi,j ; 1Ii,j ; θi,j
VDG ωVi,j ;Ii,j ;θi,j = (8)
1Vi,j + 1Ii,j + 1θi,j
where IDG is the DER output current, VDG is the connec-
where ωVi,j , ωIi,j , and ωθi,j are the voltage, current, and PF
tion point voltage, and SDG is the DER supplied power.
angle weights of the j-th FRTU of the i-th CA results,
IDG,max is the maximum current limit according to the
respectively.
power-conversion equipment capacity. If the current of all
The fault analysis results are used to create the DOCA
DERs is less than the limit value, a decision is made regarding
and stored in the database for DMS operations. Fig. 5 shows
whether the voltage has converged. If it has not converged
an example of the DOCA database. The DOCA comprises
entirely, the process returns to ④ and repeats.
three tables: D/Ls, FRTUs, and fault analysis results. The
⑥ Calculating the voltage, current, and PF angle of the
D/L table stores information on the number of FRTUs and
FRTU installation point: The voltage and phase of each node
contingency cases, and the FRTU table stores information
are estimated using the power flow method, and then the
on the affiliated D/Ls. The fault analysis results table stores
current and PF angle of the FRTU installation points are
the FRTU measurement values that were analyzed for each
calculated. The current is calculated using the voltage at
contingency case of D/L.
both ends of the FRTU and the three-phase line impedance
between them, and the PF angle is calculated using the volt-
C. OCF DETECTION AND LOCATION IN REAL-TIME
age and the current phase.
OPERATION MODE
⑦ Computing the weights for the voltage, current, and PF
The OCF detection and location process consists of FRTU
angle: There are differences in the amounts of fluctuation
event detection, polling the present data of the FRTUs of
in the voltage, current, and PF angle according to the OCF
the relevant D/Ls, calculating error by a comparison of the
fault type and detection location. When the OCF decision is
DOCA and the present measurement data, and deriving con-
made by comparing the DOCA set and the current measure-
clusions. The flowchart for this process is shown in Fig. 6 and
ment values presented in Section III-C, weights are assigned
described in detail below:
according to the amount of fluctuation to prevent bias in
1) FRTU event detection and collection: Collection of
the results for variables with small amounts of fluctuation.
FRTU data by the DMS is broadly divided into peri-
Equation (7) expresses the fluctuation calculation for calcu-
odic/aperiodic data polling and aperiodic report by exception
lating the weights:
(RBE). RBE is a method in which the FRTU transmits sig-
kj − ki,j nals without a request by the DMS when an event occurs.
1ki,j (PU ) = (7) In this study, the FRTU alarm is used for suspected abnor-
kj
mal situations as a preprocessing step for OCF detection.
where 1ki,j is the amount of fluctuation in each measurement To implement the preprocessing step, the voltage fluctua-
element (k = V ; I ; θ), and it indicates the i-th case, j-th tion (1Vx−cycle ) and current unbalance (1Iub_x−cycle ), which
FRTU voltage, current, and PF angle. kj and kij are the analy- commonly occur in almost all abnormal network situations,
sis values of each measurement element of the FRTU before are used. In the IEEE C57.13 and IEC 61869 protection

114310 VOLUME 12, 2024

G.-D. Sim et al.: Detection and Location of OCFs for Power Distribution Networks

values of FRTUs in the case of source-side GFs or the

present source-side measurement values of FRTUs in the case
of load-side GFs. In the case of GFs, the changes in the
magnitude of current according to the fault point resistance
are relatively large. Therefore, comparisons are only made
between values for the load-side of the faulted point, which
are not affected by overcurrent. The comparison of the DOCA
data and the present measurement data is obtained using (9):
|kj,Meas (tj ) − ki,j |
dk,ij = (9)
kj,Meas (tj )

where dki,j is the error (pu) that is found by dividing the

Euclidean distance between the measurement values of FRTU
and the estimation values (CA results) of DOCA by the mea-
surement values, and kj,Meas (tp ) are the measurement values
of the j-th FRTU, which are measured and sent after the data-
polling command. tj is the time taken for the j-th FRTU to
receive the data polling and transmit the measurement values
FIGURE 6. Overall flowchart of the OCF detection and location strategy.
to the DMS. Finally, the MSE is found from the Euclidean
error and weights as given by (10), and detection and location
are performed:
class 3 standards, the voltage error is defined as 3%, and
in IEEE 1547, the instantaneous voltage fluctuation rate Pn q
2
2
2
due to DERs is limited to 3% [24], [25], [26]. Therefore, j=1 ωVi,j dVi,j + ωIi,j dIi,j + ωθi,j dθi,j
di = (10)
in this work, the value of the voltage error plus the volt- n
age fluctuation rate is set as the minimum criterion (α)
where di is the MSE value (pu) of the Euclidean distance error
for event detection. The current unbalance criterion (β) is
of the voltage, current, and PF angle, and n is the total number
based on a previous study [3]. The suspected abnormal sit-
of FRTUs. Finally, an OCF is decided to have occurred if
uation alarm uses the information detected by the FRTU in a
there is a case that satisfies the condition given by (11):
high-impedance state immediately after OCF occurs. In the
case of an OCF after a GF, an alarm does not occur in the di < ε, i = 1, 2, . . . ..m (11)
FRTU in cases accompanied by an overcurrent (1 |I | > γ )
to avoid using information that is detected in situations that where m is the total number of contingency analyses, and ε
are not high-impedance states. is the detection criterion. The minimum criterion is chosen
2) DMS data polling: If an event is detected, the present according to the voltage and current measurement error of
measurement values of FRTUs for the relevant D/Ls are col- the FRTU, and then it is set up considering the margin.
lected via polling commands from the DMS. In the duration If the minimum MSE is greater than the criterion value,
in which the polling command and data collection are carried it means that the event does not fully match the contingency
out, the fault behavior changes, and stabilization is performed case analysis results of the relevant D/L, and it is decided
to change to a state in which fixed measurement values are as ‘‘OCF not detected.’’ If there are one or more cases in
maintained. In the case of an OCF after a GF or a GF after which the MSE is less than the criterion value, it is decided
an OCF, protective equipment may operate because of an as ‘‘OCF detection.’’ Finally, the fault location is estimated
overcurrent. However, in an overcurrent situation in the case by estimating the minimum MSE. If there is one case that
of an OCF after a GF, an alarm does not occur. Therefore, the satisfies the condition given in (10), the FNmi of that case is
polling command is executed after the protective equipment estimated as the fault location. If there are two or more cases,
recloses. In the case of a GF after an OCF, there is time to the FNmi with the minimum MSE is estimated as the final
change to a GF and for the protective equipment to operate. fault location; this can be expressed as (12):
Therefore, the measurement values can be collected in a state


where fixed measurement values are maintained. FN OCF = argmin(di FNm,i ) (12)
3) Fault detection and location: The data that were pre-
viously collected via data polling are compared with the where FN OCF is the location where the first-priority fault
DOCA results that were stored in advance. The DOCA stores occurred. In the case of OCFs, there is no significant differ-
the calculation datasets for each OCF type, and the present ence according to the section during fluctuation, Therefore,
FRTU measurement values are compared with all presumed if second- and third-priority FN mi are designated, the field
fault types. The stored dataset of GF faults after OCFs in crew can respond quickly if an OCF is not found at the first-
DOCA is compared with the present load-side measurement place location.

VOLUME 12, 2024 114311

 G.-D. Sim et al.: Detection and Location of OCFs for Power Distribution Networks

FIGURE 7. Modified IEEE 33 bus test system.

IV. CASE STUDIES

Case studies were performed on the proposed contingency
analysis-based OCF detection and location method using
modified IEEE 33 bus and 118 bus test networks [27], [28].
The procedure for the case studies is summarized below:
1) The networks are implemented in MATLAB/Simulink
and used to generate true-value measurements during faults.
Measurement noises are added to the true value to create field
measurement values.
2) A program for detailed fault analysis is implemented in FIGURE 8. 118-bus distribution test system.
MATLAB. This program is used to calculate a DOCA for the
same contingencies as those modeled in MATLAB/Simulink. TABLE 3. Overall fault scenarios for case studies.
3) The data from 1) are used for obtaining the occurrences
of event detection alarm of FRTUs and data polling measure-
ments and compared with DOCA, which is the result of 2).
With this process, the final determinations regarding whether
OCFs occurred and their locations are made.
The specific configurations of the test networks for
MATLAB/Simulink modeling are as follows:
1) Network configuration: The 33-bus network consists of
one D/L, and the 118-bus network comprises three D/Ls. The
original models have an open-loop topology; however, for
this work, the connection points are removed because they
are used for fault detection only.
2) Fault location: As shown in Figs. 7 and 8, four and nine
fault locations are modeled for the 33- and 118-bus networks,
respectively.
3) Fault types: The fault types are chosen as given in
Table 3, considering the OCF aspects mentioned in Section II.
All faults are assumed to be single-phase faults. For the
faults with arcing, a numerical model that was implemented considering a previous study’s model of resistance during a
as per the results of actual experiments performed in pre- GF [32], the minimum fault point resistance is calculated as
vious studies is reflected in the MATLAB/Simulink fault 15 , and testing is performed up to 50  to validate that
model [29]. For the impedance of an HIGF after an OCF, this the algorithm is valid even when the fault point resistance
study refers to the impedance of various materials measured increases.
during verification experiments (asphalt: ∼1500 , sidewalk The assumptions about fault point resistance in Table 3
block: ∼2000 ) [30]. For the scenarios in which an OCF are made to calculate the comparison target measurement
occurs after a GF (type 6–8), the fault point resistance is values. However, the fault point resistance during all real-time
set at 30 , which is the maximum impedance considered operation scenarios cannot be said to match the assumptions
when setting up protective equipment [31]. However, in fault in Table 3. Therefore, the measurement difference values
type 11 wherein only a single-line GF occurs, 50  is addi- according to the fault point for each fault type are examined
tionally tested to examine whether an FP occurs as the fault as presented in Tables 4 and 5. These tests are performed
point resistance increases. In the case of a GF after an OCF at the same fault locations as the IEEE 33 bus test network,

114312 VOLUME 12, 2024

G.-D. Sim et al.: Detection and Location of OCFs for Power Distribution Networks

TABLE 4. Average of difference in measurement values of the FRTU load. Moreover, to consider characteristics similar to those of
according to fault point resistances (HIGF after OCF).
actual power distribution operations, a voltage ride-through
function is incorporated from among DER grid codes [33].
6) Protective equipment: For the line protective equip-
ment, it is assumed that an overcurrent relay-based breaker is
installed at the feeder start point, and reclosers are placed in
the middle of feeders. The protective equipment standards of
Korea Electric Power Corporation are used for the minimum
pick-up current, time dial setting, and protection coordina-
tion [31].
TABLE 5. Average of difference in measurement values of FRTU according
to fault point resistances (LIGF after OCF). 7) Load and lines: The load is connected to all buses, and
total active and reactive powers of 3.715 MW and 2.3 MVAr,
respectively, are set up for the 33-bus network, while total
active and reactive powers of 22.709 MW and 172.041 MVAr
are set up for the 118-bus network, respectively. D/L #1, D/L
#2, and D/L #3 are set at 10.28 MW and 7.97 Mvar, 7.38 MW
and 5.11 Mvar, and 5.05 MW and 3.96 Mvar, respectively. For
the line impedance data, the data that are presented in the test
network are used.
8) Transformer: The existence of a Yg-1 connection,
and the measurement data of the FRTU connection point are which is generally used for DER network connections to
used. In Table 4, the voltage, current, and PF angle that occur prevent the spread of low-voltage-side GFs and suppress
according to fault resistance when there is an HIGF after harmonics, is assumed, and the transformer for low-voltage
an OCF are compared with the case where the maximum loads is set up with a Yg-Yg connection.
resistance of 5000  occurs. The test results revealed that 9) Measurement error consideration: To reflect the mea-
in the case of voltage, a maximum difference of 0.191% surement error of the FRTU during faults, an error is assumed
occurred, and in the case of current and PF angle, differences according to IEEE C57.13 [24] and IEC 61869 [25]; it is
of 5.13% and 5.95% occurred, respectively. Table 5 presents distributed randomly (Gaussian distribution) and used.
an analysis of the differences in voltage, current, and PF
when comparing a certain fault point resistance and other A. TESTS USING THE 33-BUS NETWORK
fault point resistances when an LIGF occurs after an OCF. Table 6 presents a comparison of the proposed OCF detection
In the case of 15 , the difference in current with 30  and location method and methods used in previous studies.
was not large at 3.77%; however, when compared to 50 , The proposed method is compared with a current unbal-
there was a relatively large error of 8.07%. In the case of ance method [3], multivariate method [5], voltage outage
30 , there was a maximum difference of 6.85% in phase method [7], and voltage unbalance method [8]. The methods
compared to 50 , and the error was smaller than in the case used in previous studies are expressed in (13)–(15).
of 15 . In Tables 4 and 5, the error in voltage according
to the fault resistance is relatively small, and ultimately it
TABLE 6. Results of OCF detection and location for the IEEE 33 test
does not exceed the detection criterion when calculating MSE system.
including voltage. Accordingly, it was assumed in this work
that the high resistance during the CA was the medium value
of 2000 , and the low resistance was 30 , which had the
smallest error.
4) FRTU: FRTUs are installed in each piece of protective
equipment and SW, and the sampling period of FRTUs is set
by setting the simulation time step to 130 µs. In addition, it is
assumed that it takes a maximum of 2 s to transmit data to
the front-end processor (FEP) after the polling command is
received.
5) DER: As for the capacity of the 33-bus DER, 0.8 MW,
which is the maximum value among the values suggested
by the reference literature [27], is used as the capacity of
each DER. In the case of the 118-bus network, the con-
nection varies according to the D/L. For D/L#1 and D/L#2,
9 and 3 MW are connected, assuming 1.5 and 1 MW per 1) The current unbalance method [3] detects the ratio of
PV, respectively. D/L#3 comprises a network with only a the positive phase sequence component to the negative phase

VOLUME 12, 2024 114313

 G.-D. Sim et al.: Detection and Location of OCFs for Power Distribution Networks

sequence component as follows: that the proposed method is effective at detecting faults in
I2 networks connected to multiple DERs and at the ends of lines.
Iub = , (13) The FP rate given in Table 6 is the rate at which fault type 11
I1
(GF) was incorrectly judged as an OCF. The proposed method
where Iub is the ratio of the positive phase sequence compo- produced only one FP in the single-line GF scenario. As for
nent to negative phase sequence component, I1 is the positive the other methods, the single-element methods had FP rates of
sequence of current, and I2 is negative sequence of current. 30% or more when there was a GF fault, and the multivariate
An OCF is detected if this ratio is greater than 15%. method had an FP rate of more than 15%. From these results,
2) The multivariate detection element method [5] detects it was found that the proposed method had a high detec-
OCFs by combining various OFCs such as overvoltage, volt- tion rate improvement of 27.3%–50.8% and dramatically
age outage, and current unbalance fluctuations into a decision improved the FN rate (30.2%–65.3%) for faults at the ends
tree. of lines and in lines with a high DER connection capacity,
3) The voltage outage method [7] uses the voltage drop rate which is a weakness of the existing methods. In particular,
for OCF detection as follows: it was verified that there was a 9.2%–31.6% improvement in
Vpost − Vpre FPs for GFs with more than a certain fault resistance that pro-
1V = , (14)
Vpre duces fault point load-side measurement value fluctuations
similar to those of OCFs. In addition, location rates were
where Vpost is the voltage after the event and Vpre is the
compared. Previous studies [3], [5] only proposed detection
voltage before the event. An OCF is detected if 1V is 30%
methods, while some studies [7], [8] estimated locations by
or more.
common topology searches in the source-side direction of
4) The voltage unbalance method [8] uses the voltage
OCF detection equipment. The location rates were calculated
unbalance ratio for detecting an OCF.
as the ratio of successful locations out of all OCF cases. The
s √ 4 + V4 + V4
1 − 3 − 6β VAB proposed method showed good results with a location rate of
K= √ , β= BC AC
(15) 92.9%. The previous studies [7], [8] showed poor location
1 − 3 − 6β 2 + V2 + V2 2


VAB BC AC rates of 8.6% and 10.3% because many topology searches
where K is voltage unbalance rate, VAB , VBC , Vcb is the mag- failed because of the DERs and the FNs of FRTUs.
nitude of line voltages. An OCF is detected if K is 30% or To determine whether fault locations can be classified as
more. per MSE, Fig. 9 presents the measurement values during
In the proposed method, the individual event alarm of the OCFs in the source- and load-side sections of the SW (#1,
FRTU criterion for detecting OCFs is set at 6% for voltage by #2, #4, #6) installation point nearest to each fault location.
adding the measurement error and the voltage fluctuation rate Fig. 9 shows the voltage (■), current (◆), and PF (●) when
as mentioned previously. In the case of current, it is set at 15% a fault occurs in the source-side section of the SW as well
as per a previous study [3]. Finally, OCF detection based on as the voltage (■), current (◆), and PF (●) when a fault
minimum MSE is set at 7%, considering measurement error occurs in the load-side section. When high-resistance faults
and the margin. occurred after OCFs in the source- and load-side sections
Table 6 lists the detection rate, FP rate, and location rate in of SW#1, the voltage, current, and PF patterns were similar
the previous methods and the proposed method. The detection at other SWs, but current value differences occurred in the
rate refers to the ratio at which OCFs are classified and FRTUs of CB, RC#2, and SW#6. Furthermore, in the case
detected for all faults. Table 6 reveals that the proposed of GFs after OCFs, there was a difference in the current
method improves the detection rate compared to the previous measurements at CB, SW#1, RC#2, and SW#6 as well as a
methods regardless of the fault location. In the case of the difference in the PF measurement values at RC#2 and SW#6.
voltage-based method, when a fault occurred at the end of In the other graphs as well, if the fault section changes, there
the line, the detection rate decreased because the voltage was is certainly at least one FRTU that shows a difference in at
maintained owing to the DERs. In the case of the current least one of the three measurement elements. Therefore, it is
unbalance-based method, when the difference between the possible to perform location because a difference occurs in
PV output and the load below the fault point was small, the MSE values at both ends of the FRTU section when a fault
as in the case of fault locations ‘‘F3’’ and ‘‘F4’’ in Fig. 7, occurs.
the detection rate decreased because the unbalance was low To verify the possibility of distinguishing OCFs from other
owing to the current supplied by the DERs. The multivariate events, the FRTU measurement values during GFs and OCFs
detection-based method was also found to have a sharply at designated fault locations in the IEEE 33 bus are compared
reduced detection rate when the fault occurred at the end of as shown in Fig. 10. In the graph, the light-colored marks
the line because most OFCs had little fluctuation. Particularly show the voltage, current, and PF during an OCF, and the
in the case of location ‘‘F3’’, which is on the shortest branch regular marks show the measurement values during a GF.
line, all the previous methods showed low detection rates, but In the case of fault location ‘‘F1’’ in Fig. 7, the voltage
the proposed method achieved a high detection rate with little measurement values of the FRTU connected to the load side
difference from those for other locations. This result indicates of the fault point showed similar patterns during the OCF

114314 VOLUME 12, 2024

G.-D. Sim et al.: Detection and Location of OCFs for Power Distribution Networks

FIGURE 9. Comparison of measured values of faults at both sides of the SW (From the left: SW#1, RC#1, SW#2, SW#3, SW#4, SW#5, RC#2, and SW#6).

FIGURE 10. Comparison of measured values of each FRTU when OCF or GF occurs.

and the GF. However, there were differences in current and small. Most of the location attempts were successful when
PF measurement values at the FRTU connected to the source an OCF was detected. The FP rate for GFs was also low, and
side of the fault point. In the case of fault location ‘‘F4,’’ FPs only occurred in 2 out of 36 scenarios.
there was no considerable difference in PF, but there were To verify reliability in an actual network, tests were per-
large differences in the current measurement values at CB and formed by changing the polling time of the FRTUs as given in
RC#2. In addition, at all fault locations, there was at least one Table 9. Maximum delay times were assumed, and an FRTU
FRTU where differences in the OCF and GF measurement that transmitted polling commands randomly was selected.
values occurred. Maximum delay times of 1.5, 2, 3, and 4 s were assumed,
given the processing times from the FRTU to the central office
B. PERFORMING TESTS USING THE 118-BUS NETWORK terminal (COT), from the COT to the FEP, and from the FEP
This study analyzes the detection rate and FP rate according to the DMS. The results are presented in Table 9. When the
to adjustments in detection criterion (ε) of in. (11) for the communication delay was 2 s, the detection rate was high and
network as given in Table 7. In the test results, at 6% or less, showed no difference compared to the existing results. When
there were no FPs, but the detection rate was relatively low. the communication delay was 3 s or more, the detection rate
The FP rate was the same at 7% and 8%, but the detection decreased slightly because of the difference in measurement
rate increased at 8%. At 9% and above, the detection rate times, data transmissions in an outage due to the operation of
increased, but the FP rate increased sharply. On the basis of the protective equipment, etc. However, a high detection rate
these results, the detection criterion was set at 8%; Table 8 was maintained.
presents the results of the analysis of each feeder. As the To verify the effects of communication failures, the detec-
DER connections became more numerous, the detection rate tion rate according to the wired/wireless ratio of FRTUs is
decreased, but the overall detection rate for each D/L was examined; the corresponding results are given in Table 10.
high, and the difference due to the DER connections was In FRTU wired communications, fiber-optic cables were

VOLUME 12, 2024 114315

 G.-D. Sim et al.: Detection and Location of OCFs for Power Distribution Networks

TABLE 7. Detection and FP rate according to detection criteria. the proposed algorithm was installed. All simulations were
performed on a PC with an Intel core i7 3.4-GHz CPU and
32-GB RAM. When CA was performed with the algorithm,
it took 0.77 and 9.04 s in the case of the IEEE 33 bus and
118 bus, respectively. As mentioned previously, the CA is
performed in the non-real-time study mode, and other than
the initial startup, it is only performed for D/Ls that have
TABLE 8. Results of OCF detection and location for the 118-bus test undergone topology changes. Therefore, this level of time
system. consumption will not be a significant issue when applying
the proposed method.

V. CONCLUSION
This study proposes a contingency analysis-based OCF detec-
tion and location method for power distribution networks.
Following are the main conclusions of this study:
1) By comparing the distance between the entire mea-
surement set and DOCA, the proposed method significantly
TABLE 9. Test results considering communication delay. improves the OCF detection rate by 27.3%–50.8% compared
to existing methods that use a specific detection element.
In addition, the proposed method improves FNs for faults
in lines with a high DER connection capacity and line ends,
which is a weakness of existing methods.
2) Using the weights based on the magnitude variation
of the comparison elements when calculating the distance
between the DOCA and FRTU sets, the proposed method
significantly reduces the occurrence of FPs in other events
such as GFs by 9.2%–31.6%. Improving upon FPs, which
are a type of false alarm, is very important for operations.
used, and it was assumed that almost no communica- Repeated false alarms are significant in that they can lead
tion failures occurred. In FRTU wireless communications, to a ‘‘Cry Wolf Effect’’ in which alarms for actual events
frequency-based methods and wireless local area network are ignored because operators and field crews are repeatedly
were used, and a communication failure rate of ∼10% was overworked and ignore repeated alarms. In addition, tests that
assumed. In Korea, most communications are performed via reflect realistic conditions such as FRTU measurement errors,
wires, but in countries with inadequate infrastructure invest- time delays due to polling commands, and communications
ment, the ratio of wireless communications is higher, and the failures due to the wired/wireless communication ratio are
effect of communication failures is greater. To examine this, performed. Accordingly, the proposed method is confirmed
Korea’s wired/wireless communication ratio of 85:15 was to be robust under a variety of realistic conditions, proving
compared with ratios of 50:50 and 15:85. Although commu- its adaptability to actual power-distribution networks.
nication failures occurred as the wireless ratio increased, the 3) In the location of OCFs, the proposed approach shows
detection rate did not change significantly, and a single com- better results compared by 82.6%–84.3% to previous meth-
munication failure had almost no effect. This result implies ods. OCFs are permanent faults, and they must be dealt
that the proposed method is robust against communications with by dispatching field crews, which makes it crucial to
failures. locate them in comparison with other events. Therefore, the
Finally, the time taken to create contingency cases from an proposed algorithm, which performs detection and location
actual DMS was measured to determine the time taken when simultaneously, will be able to increase reliability by enabling
rapid responses during actual operations.
TABLE 10. Test results considering communication failures.
4) The proposed algorithm uses data that are com-
monly possessed by power companies and the typical FRTU
measurement values. Therefore, it is practical in that there
is no need for special investments in facilities, and it can be
universally applied to power distribution network operation
systems that are equipped with basic infrastructure. This
paper proposes an aperiodic screening method. However,
additional research is required on the criteria for selecting
the target D/Ls that need CA. It is possible to determine the
risk of forest fires from weather data and predict situations in

114316 VOLUME 12, 2024

G.-D. Sim et al.: Detection and Location of OCFs for Power Distribution Networks

which damages increase when OCFs occur. In addition, the [15] H. Xu, X. Wang, J. Huang, F. Zhang, and F. Chu, ‘‘Semi-supervised
accident rates in a distributed power network increase when multi-sensor information fusion tailored graph embedded low-rank tensor
learning machine under extremely low labeled rate,’’ Inf. Fusion, vol. 105,
the network is newly installed or when its lifespan is expiring May 2024, Art. no. 102222.
and the time for replacement is approaching [34]. These facts [16] J. B. Thomas, S. G. Chaudhari, K. V. Shihabudheen, and N. K. Verma,
could be used to adjust the period and scope of the analysis ‘‘CNN-based transformer model for fault detection in power system net-
works,’’ IEEE Trans. Instrum. Meas., vol. 72, pp. 1–10, 2023.
and reduce the overall time taken by contingency analysis. [17] X. Wang, T. Ding, and W. Xu, ‘‘An open conductor condition monitoring
As DER connections rapidly increase, fault detection and scheme using natural voltage and current disturbances,’’ IEEE Trans.
location are becoming increasingly complex. However, the Power Del., vol. 34, no. 3, pp. 1193–1202, Jun. 2019.
[18] S. H. Mortazavi, Z. Moravej, and S. M. Shahrtash, ‘‘A searching based
lack of timely and commensurate investments in facilities is method for locating high impedance arcing fault in distribution networks,’’
a common problem that must be resolved to protect power IEEE Trans. Power Del., vol. 34, no. 2, pp. 438–447, Apr. 2019.
distribution networks. Therefore, it is important to devise [19] Z. Lin, Z. Hu, and Y. Song, ‘‘Distribution network expansion planning
considering N − 1 criterion,’’ IEEE Trans. Power Syst., vol. 34, no. 3,
methods that can handle power distribution networks that pp. 2476–2478, May 2019.
create changes within the current infrastructure and systems, [20] Q. Zou, F. Luo, and T. Zhang, ‘‘An incidence matrix based analytical
and the use of the proposed method will be effective in reduc- method of N-1 contingency parallel analysis of main transformers in distri-
bution networks,’’ CSEE J. Power Energy Syst., early access, May 6, 2022,
ing damages caused by OCFs in current power-distribution doi: 10.17775/CSEEJPES.2021.01490.
networks. [21] S. J. Simmons, ‘‘Breadth-first trellis decoding with adaptive effort,’’ IEEE
Trans. Commun., vol. 38, no. 1, pp. 3–12, Jan. 1990.
[22] J. J. Grainger, W. D. Stevenson, and G. W. Chang, Power System Analysis,
REFERENCES 2nd ed., New York, NY, USA: McGraw-Hill, 2016, pp. 422–430.
[1] G. Bade. (Nov. 2018). PG&E Reports Second Line Fault in Camp Fire [23] J.-S. Hong, G.-D. Sim, J.-H. Choi, S.-J. Ahn, and S.-Y. Yun, ‘‘Fault location
Area. [Online]. Available: https://www.utilitydive.com/news/ method using phasor measurement units and short circuit analysis for
[2] Y.-S. Ko, ‘‘A self-isolation method for the HIF zone under the network- power distribution networks,’’ Energies, vol. 13, no. 5, p. 1294, Mar. 2020.
based distribution system,’’ IEEE Trans. Power Del., vol. 24, no. 2, [24] IEEE Standard Requirements for Instrument Transformers, Standard
pp. 884–891, Apr. 2009. C57.13, 2016.
[3] D. K. J. S. Jayamaha, I. H. N. Madhushani, R. S. S. J. Gamage, [25] IEC Instrument Transformers—Part 2: Additional Requirements for Cur-
P. P. B. Tennakoon, J. R. Lucas, and U. Jayatunga, ‘‘Open conductor fault rent Transformers, Standard 61869, IEC, 2012.
detection,’’ in Proc. Moratuwa Eng. Res. Conf. (MERCon), Moratuwa, Sri [26] IEEE Standard for Interconnection and Interoperability of Distributed
Lanka, May 2017, pp. 363–367. Energy Resources With Associated Electric Power Systems Interfaces,
[4] A. Wontroba, A. P. Morais, G. J. Cardoso, J. P. A. Vieira, P. E. Farias, IEEE Standard IEEE 1547, 2018.
M. Gallas, and J. P. Rossini, ‘‘High-impedance fault detection on downed [27] S. H. Dolatabadi, M. Ghorbanian, P. Siano, and N. D. Hatziargyriou,
conductor in overhead distribution networks,’’ Electr. Power Syst. Res., ‘‘An enhanced IEEE 33 bus benchmark test system for distribution sys-
vol. 211, Oct. 2022, Art. no. 108216. tem studies,’’ IEEE Trans. Power Syst., vol. 36, no. 3, pp. 2565–2572,
[5] J.-S. Hong, S.-Y. Hyun, Y.-W. Lee, J.-H. Choi, S.-J. Ahn, and S.-Y. Yun, May 2021.
‘‘Detection of open conductor fault using multiple measurement factors [28] D. Zhang, Z. Fu, and L. Zhang, ‘‘An improved TS algorithm for loss-
of feeder RTUs in power distribution networks with DGs,’’ IEEE Access, minimum reconfiguration in large-scale distribution systems,’’ Electr.
vol. 9, pp. 143564–143579, 2021. Power Syst. Res., vol. 77, nos. 5–6, pp. 685–694, Apr. 2007.
[6] V. H. Makwana and B. R. Bhalja, ‘‘A new digital distance relaying [29] S.-Y. Hyun, J.-S. Hong, S.-Y. Yun, C.-H. Kim, and Y. Lee, ‘‘Arc modeling
scheme for series-compensated double-circuit line during open conductor and kurtosis detection of fault with arc in power distribution networks,’’
and ground fault,’’ IEEE Trans. Power Del., vol. 27, no. 2, pp. 910–917, Appl. Sci., vol. 12, no. 6, pp. 2777–2793, Mar. 2022.
Apr. 2012. [30] Y.-B. Jung, W.-S. Choi, Y.-S. Jeon, and B.-S. Lee, ‘‘Development of
[7] L. Garcia-Santander, P. Bastard, M. Petit, I. Gal, E. Lopez, and H. Opazo, method for detecting high impedance fault,’’ KEPCO, Naju, South Korea,
‘‘Down-conductor fault detection and location via a voltage based method Tech. Rep. R11DA01, Sep. 2013, pp. 90–93.
for radial distribution networks,’’ IEE Proc., Gener., Transmiss. Distrib., [31] Guidelines for Distribution Network Protection, KEPCO, Naju,
vol. 152, no. 2, p. 180, 2005. South Korea, Oct. 2008.
[8] F. L. Vieira, P. H. M. Santos, J. M. C. Filho, R. C. Leborgne, and [32] V. D. Andrade and E. Sorrentino, ‘‘Typical expected values of the fault
M. P. Leite, ‘‘A voltage-based approach for series high impedance fault resistance in power systems,’’ in Proc. IEEE/PES Transmiss. Distrib. Conf.
detection and location in distribution systems using smart meters,’’ Ener- Expo., Latin Amer. (TD-LA), Sao Paulo, Brazil, Nov. 2010, pp. 602–609.
gies, vol. 12, no. 15, p. 3022, Aug. 2019. [33] Technical Standards of the Distributed Energy Resource Connection in
[9] E. M. Esmail, M. M. Elgamasy, T. A. Kawady, A.-M.-I. Taalab, Power Distribution System, KEPCO, Naju, South Korea, 2021.
N. I. Elkalashy, and M. A. Elsadd, ‘‘Detection and experimental investiga- [34] G. Klutke, P. C. Kiessler, and M. A. Wortman, ‘‘A critical look at the
tion of open conductor and single-phase Earth return faults in distribution bathtub curve,’’ IEEE Trans. Rel., vol. 52, no. 1, pp. 125–129, Mar. 2003.
systems,’’ Int. J. Electr. Power Energy Syst., vol. 140, Sep. 2022,
Art. no. 108089.
[10] J. Li, Y. Liu, C. Li, D. Zeng, H. Li, and G. Wang, ‘‘An FTU-based
method for locating single-phase high-impedance faults using transient
zero-sequence admittance in resonant grounding systems,’’ IEEE Trans.
Power Del., vol. 37, no. 2, pp. 913–922, Apr. 2022.
[11] M. Michalik, M. Lukowicz, W. Rebizant, S.-J. Lee, and S.-H. Kang,
‘‘New ANN-based algorithms for detecting HIFs in multigrounded MV GI-DO SIM (Student Member, IEEE) received
networks,’’ IEEE Trans. Power Del., vol. 23, no. 1, pp. 58–66, Jan. 2008.
the B.S. and M.S. degrees in electrical engineer-
[12] H.-G. Yeh, S. Sim, and R. J. Bravo, ‘‘Wavelet and denoising techniques
ing from Chonnam National University, Gwangju,
for real-time HIF detection in 12-kV distribution circuits,’’ IEEE Syst. J.,
vol. 13, no. 4, pp. 4365–4373, Dec. 2019.
South Korea, in 2018 and 2020, respectively,
[13] B. K. Chaitanya, A. Yadav, and M. Pazoki, ‘‘An intelligent detection of where he is currently pursuing the Ph.D. degree
high-impedance faults for distribution lines integrated with distributed with the Department of Electrical Engineering.
generators,’’ IEEE Syst. J., vol. 14, no. 1, pp. 870–879, Mar. 2020. His research interests include distribution net-
[14] J. Che, T. Kim, S. Pyo, J. Park, B. An, and T. Park, ‘‘Prevention of wildfires work protection technologies, fault analysis, fault
using an AI-based open conductor fault detection method on overhead detection and location, and protection coordina-
line,’’ Energies, vol. 16, no. 5, p. 2366, Mar. 2023. tion optimization.

VOLUME 12, 2024 114317

 G.-D. Sim et al.: Detection and Location of OCFs for Power Distribution Networks

HYO-SEOP IM (Student Member, IEEE) SEON-JU AHN (Member, IEEE) received the
received the B.S. degree in electrical engineering B.S., M.S., and Ph.D. degrees in electrical engi-
from Chonnam National University, Gwangju, neering from Seoul National University, Seoul,
South Korea, in 2022, where he is currently pur- South Korea, in 2002, 2004, and 2009, respec-
suing the integrated Ph.D. degree with the Depart- tively. He is currently a Professor with Chonnam
ment of Electrical Engineering. His research National University, Gwangju, South Korea. His
interests include distributed generation and oper- current research interests include power quality,
ating distribution systems. distributed energy resources, microgrids, smart
grids, and real-time simulation.

SANG-YUN YUN (Member, IEEE) received

the B.S., M.S., and Ph.D. degrees in electri-
cal engineering from Soongsil University, Seoul,
South Korea, in 1996, 1998, and 2002, respec-
tively. From 2002 to 2009, he was a Senior
JOON-HO CHOI (Senior Member, IEEE) Researcher with the Electrotechnology Research
received the B.S., M.S., and Ph.D. degrees in and Development Center, LSIS, Cheongju,
electrical engineering from Soongsil University, South Korea. From 2009 to 2016, he was a Senior
Seoul, South Korea, in 1996, 1998, and 2002, Researcher at the KEPCO Research Institute, Dae-
respectively. Since 2003, he has been a Profes- jeon, South Korea. He is currently a Professor with
sor with Chonnam National University, Gwangju, the Department of Electrical Engineering, Chonnam National University,
South Korea. His research interests include opera- Gwangju, South Korea. His research interests include the design of EMS,
tion and control strategies of distributed generation DMS, and MGMS as well as protection technologies for active distribution
and operation algorithms of smart grids. networks.

114318 VOLUME 12, 2024