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.2016.2585380, IEEE
Transactions on Power Delivery
1

Novel Protection Scheme of Single-phase

Earth Fault for Radial Distribution Systems
with Distributed Generators
Yuanyuan Wang, Member, IEEE, Gen Wei, Hongming Yang, Member, IEEE, Haowei Chen, Zi Ouyang
and David J. Hill, Fellow, IEEE
 protection of distribution networks must be able to reliably
Abstract —This paper presents a new method based on relay detect, locate and isolate single-phase earth fault before it
agent for single-phase earth fault protection in radial develops into multiple-phase faults with more severe hazards.
distribution systems with distributed generators. In order to get Therefore, the single-phase earth fault protection is a hot topic
the most informative fault features to discriminate fault for this
system, traditional protection schemes based on different fault
for research.
features are analyzed. The selection of fault features is discussed. The increasing number of distributed generators (DGs)
Moreover, the cluster center of historical fault data is calculated including the inverter-based distributed generator (IBDG) and
to analyze the space distribution of fault data for each feeder by the synchronous-based distributed generator (SBDG) has been
applying fuzzy clustering algorithm. The space relative distance connected to the network recently which introduces lots of
between the on-line sample data and the cluster center and the new challenge to detect and locate the single-phase earth fault.
minimum value of the space relative distance among all the For example, DGs can reduce the fault current drawn from the
relays at relay agent are obtained. Finally, the coordination substation at certain locations, which makes that the fault
strategy of relay agent is proposed to discriminate the faulty
information becomes more difficult to extract [3]. Moreover,
feeder. The proposed protection scheme is evaluated in a radial
distribution network with distributed generations using PSCAD the types of DGs can also affect the protection of distribution
simulation. networks. IBDG make use of renewable. However,
renewables (such as wind energy and solar energy) are
Index Terms—Radial distribution systems, distributed intermittent in power supply, which may change the fault
generations, fuzzy clustering algorithms, single-phase earth fault contribution in the line with time and cause the misjudging of
protection. the protection device [4].
Traditionally, radial networks are protected by coordinated
over-current relays whereas meshed networks are protected by
I. INTRODUCTION directional over-current relays (DOCRs) [5]. In [6], the

I N traditional distribution networks, the neutral point with

non-solid earthing such as ungrounded and grounding via
Petersen-coil are commonly used. For such systems,
DOCRs are improved to protect the networks with DGs.
However, those approaches are only applicable on the faults
with high fault currents which are hard to be applied in the
single-phase earth fault is the most common fault type [1][2]. protection of high-resistance single-phase earth fault. To solve
However, the single-phase earth fault is very difficult to detect the problem, the inverse time admittance (ITA) relay is
because of its low magnitude of the fault current. The presented in [7]. Here, the ITA relay utilizes the measured
admittance of the protected line to identify a faulted condition
in a network. The ITA relay is insensitive to fault current level.
This work is supported by funding from the Faculty of Engineering & However, an excessive fault resistance can cause the error on
Information Technologies, The University of Sydney, under the Faculty the measured admittance which makes the protection
Research Cluster Program, National Natural Science Foundation of China (No. coordination fail.
71371065, 71331001, 71420107027, 91547113, 51207014), Hunan Province
100 Talent Program Fund, State Scholarship Fund of China
In order to improve the reliability of traditional protection
(No.201308430317), Scientific Research Fund of Hunan Provincial Education methods above, several methods based on communication
Department (No.15B006), and Natural Science Foundation of Hunan Province network were previously proposed in [8], [9] and [5]. In [8]
(No. 2015JJ4001). and [9], a protection scheme measures current contribution
Y. Y. Wang, G. Wei, H. M. Yang, H. W. Chen and D. J. Hill are with the
Hunan provincial Engineering Research Center of Electric transportation and from each source to identify fault sections then isolates faulted
smart distributed network, the Hunan Provincial Key Laboratory of Smart zone. In this scheme, the neural network algorithm is used to
Grids Operation and Control, College of Electrical and Information train the current contribution to corresponding output fault
Engineering, Changsha University of Science and Technology, Changsha, sections under different fault conditions. This scheme is
410114, China (e-mail: yuanyuan.wang.1980@ieee.org; 413225364@qq.com;
yhm5218@hotmail.com; 445909300@qq.com; david.hill@sydney.edu.au). affected by the penetration of DGs. In [5], a method based on
Z. Ouyang is with the School of Photovoltaic and Renewable Energy relay agent (RA) is proposed to isolate a faulted segment by
Engineering, the University of New South Wales, Sydney, NSW 2052, analyzing the sign of wavelet coefficients of the fault transient
Australia (e-mail: ouyangzi@gmail.com). current. However, the actual performance of this method is
D. J. Hill is also with the School of Electrical and Information Engineering,
The University of Sydney, Sydney NSW 2000, Australia, and with the
still unpredictable. The above methods only use the single
Department of Electrical and Electronic Engineering, The University of Hong fault information like fault current which are difficult to
Kong, Hong Kong.

0885-8977 (c) 2016 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.2016.2585380, IEEE
Transactions on Power Delivery
2

effectively restrain the effects of different fault conditions [10].

Therefore, the protection of single-phase earth fault for radial
distribution systems with DGs is necessary to develop more
sensitivity and reliability.
In this paper, a novel single-phase earth fault protection
based on RA is presented for distribution systems with DGs. Fig. 1. The basic structure of RA
Section II briefly introduces the proposed protection
framework. Section III presents the proposed feature
extraction including the selection of fault feature and the data
normalization. Section IV illustrates the fuzzy c-means
clustering (FCM) and discusses the space relative distance.
Section V demonstrates the coordination strategy of RA.
Section VI presents the case study and simulation results.
Section VII concludes the paper.

II. PROPOSED PROTECTION FRAMEWORK

RA is a computer system that is capable of managing and
controlling the local relay, and as well as a reactive agent that
senses the fault and sends the status signal to other RA
[11][12]. The basic structure of RA is shown in Fig. 1. In the
distribution systems, the RA is installed at the bus and controls
at least two local relays. According to the position of other RA, Fig. 2. The proposed protection framework based on RA
RA can be divided into the upstream RA (URA) and the
downstream RA (DRA). Based on the status signal of local comparability measurement between the sample data and the
relays and other RA, the RA can judge whether a fault occurs fault data. Here, the fault types including feeder fault and bus
in the section where is protected. fault can be judged due to the obvious differences in the value
The proposed protection framework based on RA is of the space relative distance between the feeder fault and bus
introduced in Fig. 2. It has four functions including fault fault, which will be analyzed in detail in the section VI. For
detection, feature extraction, fault discrimination and RA feeder fault, generally speaking, the smaller space relative
coordination. When the protection starts, the local information distance has higher possibility that the sample data belongs to
of relays including voltage and current is collected for RA. the fault data. Based on this theory, comparing the distance in
Fault detection is capable of detecting the earth fault with all relays at the bus, the smallest distance can be obtained to
high resistance. Traditionally, the measurement of the identify the section where the fault occurs.
zero-sequence voltage is suggested to be used for fault However, for feeder fault, the fault section determined by
detection because it is very sensitive to earth faults [13]. The the RA itself may be not the most accurate one. Therefore, the
value of zero-sequence voltage is equal to zero in normal specific fault line cannot be found. In order to solve this
operation but rise to a very high level during a single-phase problem, the coordination strategy of RA is proposed to
earth fault. This method can detect the earth fault if the determine the fault line in section V. According to the results
amplitude of the voltage exceeds a threshold value which is 15 of the fault discrimination, the coordination of RAs can
percent of the amplitude of normal voltage. effectively find out the specific fault line.
The feature extraction is executed after the fault has been
detected. Here, the local information can be handled to extract
all the fault features by using fast fourier transform algorithm III. FEATURE EXTRACTION
(FFT), and then all fault features are normalized by the z-score
method. A. Fault Feature Selection
In the fault discrimination, the fault features are divided into The proposed method is integrated with all kinds of fault
the sample data and the historical fault data. The sample data is features to provide an effective solution for distribution
the real time data from each relay at the bus. The historical systems protection. However, different approaches based on
fault data is the fault data from the historical fault accident different fault features have variable applicable ranges, fault
which is recorded in the tripped relay. The historical fault data detection robustness and accuracy [14]. Each single scheme
can be considered a fault cluster. The fault cluster center of the has its own advantages and disadvantages. Therefore, for the
historical fault data can be calculated off-line by FCM proposed method, the most informative fault features are
algorithms. The fault cluster center can describe the space selected to discriminate fault.
distribution of fault data under different fault condition for For the radial distribution systems with DGs, the fault
each feeder. The space relative distance between the sample environments are more complex than the traditional
data and the fault cluster center is calculated by distance distribution systems. In addition, more impact factors
calculation methods for each relay, which shows the including fault resistance, fault initial angle, fault position and

0885-8977 (c) 2016 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.2016.2585380, IEEE
Transactions on Power Delivery
3

the control strategy of IBDG, etc. have to be considered after the effects of the historical fault data X f . The data set X 
the occurrence of single-phase earth fault. For this system,
including sample data X s and the historical fault data X f is
different protection approaches based on different fault
features are analyzed as follow: defined to as follows:
1) The approaches based on the steady-state components  xs1 , xs 2 , , xst 
such as the negative and zero sequence components including
 X s   x , x , , x1t 
current and power. Due to the appearance of the negative X       11 12 (3)
sequence grid voltage during the unsymmetrical fault, the  X f       
negative-sequence current and harmonics generated from  
IBDG can adversely affect the reliability and accuracy of the  xn1 , xn 2 , , xnt 
protection approaches using those fault features [15]. In For different fault features, the above data have different
addition, the approaches using zero sequence components can scales and different effects on the process of fault analysis. In
only be applied in the non-grounding system. These methods order to guarantee the reliability of the results, it is necessary
cannot detect fault in the Petersen-coil-grounded system due to to carry out the normalization of the raw data. Data
the compensation of Petersen coil. normalization can transform the raw data into the normalized
data with same proportional scale. There are many methods for
2) The approaches based on the transient components. The
data normalization. The z-score method, a most common
compensation of Petersen coil in a charging and discharging
normalization method, is very useful when the minimum and
are not fully worked in the first 0.5 to 1 cycles when the fault maximum values among data are unknown [18]. The z-score
occurs [16]. The voltage and current signals have different method based the mean and standard deviation of data is
characteristics between the fault feeder and sound feeder in defined as follows:
this period. Therefore, the transient components such as the xkj  x j
zero-sequence transient component, the transient energy and xkj  , 1  k  n, 1  j  t (4)
the transient feature using wavelet transformation, etc. are S ( xj )
widely used to overcome the effects of Petersen coil. In 1 n
addition, the zero-sequence component is mainly affected by
x j   xkj
n k 1
(5)

the distributed capacitance of feeder and the fault resistance,

1 n
etc. and other parameters (such as output power of DG, etc.)
have relatively little impact on the distribution of transient zero
S ( xj )  
n  1 k 1
( xkj  xkj ) 2 (6)

sequence current [17]. where xkj is the normalized pattern value, x j is the mean
In this paper, the transient components based on the
zero-sequence component including the magnitude and phase value of the jth fault feature, S ( xj ) is the standard deviation of
angles of zero-sequence current, the phase angle difference the jth fault feature.
between the zero-sequence voltage and current, and the The historical fault data X f is normalized using the
transient zero-sequence energy are selected as fault features to
analyze the fault. Those fault features not only have little function (4)-(6) and then transformed into the normalized
effects on the compensation of Petersen coil, but also can historical fault data X f . Similarly, considering the effects of
avoid the interference of DG. historical fault data X f , the normalized sample data X s will
B. Data Normalization be obtained by the above normalization process of data set X  .
When a single phase-earth fault occurs, the selected fault
features as the inputs are extracted in real time for each relay.
This online data is named as sample data X s as follows: IV. FAULT DISCRIMINATION
X s  ( xs1 , xs 2 , , xst ) (1) Intelligent algorithms such as Neural Network algorithm
[8][9], are widely applied for fault discrimination to
where t is the number of fault feature for each data. discriminate between the fault data and the non-fault data for
The historical fault data are recorded off-line for each relay, the real-time sample data. In the proposed method, the spatial
and defined as follows: distribution regularities of historical fault data are analyzed by
 x11 , x12
 , , x1t  FCM algorithm. The sample data are an unknown data. It
 x , x ,  , x  could be a fault data or a non-fault data. The sample data can
X f   21 22 2t  (2) be judged as a fault data if it is followed the distribution
　 　  　    regularities. In this section, the space relative distance as a
  comparability measurement (or similarity measurement)
 xn1 , xn 2 , , xnt 
between the sample data and the fault data will be introduced.
where n is the number of data.
In the proposed method, the sample data X s calculates the A. FCM Algorithm
space relative distance with historical fault data X f for fault Fuzzy clustering plays an important role in pattern
recognition and data analysis. The FCM algorithm is one of
discrimination. Therefore, the sample data X s need consider the best known fuzzy clustering methods. In FCM, each data is

0885-8977 (c) 2016 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.2016.2585380, IEEE
Transactions on Power Delivery
4

assigned a membership to represent the degree of belonging to

a certain class (or cluster) [19]. The normalized historical fault
data set X is given as a certain cluster to calculate the cluster
center by the FCM algorithm. The objective function of the
clustering optimization is shown as below:
n c
J m U , P    ikm xk  pi , m  1,  
2
(7)
k 1 i 1
where U  [  ik ]c n is membership function matrix and ik is
the membership value of data xk for cluster i,
pc   pc1 , pc 2 ,..., pct  is cluster center and c  1, n is the
number of clusters, m, greater than 1, is fuzzifier that controls
the amount of fuzziness in fuzzy classification, k is the number
of data.
The FCM algorithm can be mainly divided into three steps:
Step 1: Give the number of clusters c and the fuzzifier m. c
is set as 1 and m is set as 2. Then membership function matrix
U is initialized and satisfied as follows:
c n
ik  0,1i, k ;  ik  1, k ; 0   ik  n, i (8) Fig. 3. Structure of coordination strategy
i 1 k 1

Step 2: Calculate a new cluster centers pi and membership

distance methods for fault discrimination. The distance
 ik as below: methods are introduced as follows:
1) Euclidean distance:
    x
n m

pi  k 1 ik k (9) t (11)
  
n
d kj2    xkj  pkj 
m 2
k 1 ik

 ik 
1 / x  p 
k i
2 1 /( m 1)

, 1  i  c, 1  k  n (10) 2) City block metric distance:

j 1

 1 / x  p 
c 2 1 /( m 1)
t
d kj   xkj  pkj
k k i
(12)
Step 3: According to the new cluster centers pi and j 1

membership  ik , calculate the value of the objective function 3) Chebyshev distance [21]:
t 1/ z

J m U , P  based on (7). When the obtained value is less than

z
d kj  lim( xkj  pkj )
z 
j 1
(13)
the minimum amount of improvement (set as 1e-5), stop the
 max xkj  pkj
process of FCM algorithm, and otherwise go back to the step Due to different calculation principles, the space relative
2. distance based on different distance methods have different
Through the above processes of FCM algorithm, the cluster results for fault discrimination. Therefore, it is necessary to
center of historical fault data set can be calculated. In the compare all distance methods to find out the optimal one for
proposed method, the cluster center has a clear physical the proposed method.
meaning. It comprehensively describes the distribution of fault
data under different condition for each feeder [20].
V. RA COORDINATION
B. Space Relative Distance
RA is a protection system that not only autonomously
Similarity measurement is an important indicator to
collects the local information to detect fault but also cooperate
describe the closeness degree between the simple data and the
with other RAs for fault discrimination. When a feeder fault
fault cluster center. The distance methods are one of the most occurs, the results of fault discrimination are divided into the
common calculation methods. They indicate the space relative fault section (FS) and sound section (SS) for each feeder which
distance between the nearby data and the remote data. The are connected with the RA. The fault section is defined as the
nearby data has similar characteristics and the remote data has section protected by the relay with the minimum value of
great difference. When a fault occurs, the space relative space relative distance. The sound section is defined as the
distance between the normalized sample data X s and the section protected by the relay with the greater value of space
cluster center pi of historical fault data is calculated by relative distance. The coordination strategy of RA is shown as
Fig. 3. According to the different sections, the RA will sent
different status signals to the adjacent RA including URA and

0885-8977 (c) 2016 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.2016.2585380, IEEE
Transactions on Power Delivery
5

TABLE II
CLUSTER CENTER OF HISTORICAL FAULT DATA FOR ALL RELAYS
Cluster center Fault feature
pf1 -0.2892 0.1427 0.1663 0.2625
pf 2 -0.2951 0.2091 0.1896 0.3633
pf 3 -0.2225 0.1574 0.1797 0.1358
pf 4 -0.0835 0.4027 -0.5974 0.1739
pf 5 -0.2323 0.0412 -0.593 0.1803
pf 6 0.2361 -0.2056 -0.0222 0.2906
pf 7 0.2401 -0.5818 0.0405 0.2891
pf 8 0.2511 0.4643 -0.174 0.1898
pf 9 -0.4853 -0.1385 0.0448 0.0489
p f 10 -0.3513 -0.1052 0.0443 0.0021
p f 11 0.2669 0.0219 -0.1892 0.1523
p f 12 -0.3336 -0.0199 0.0803 0.2487
p f 13 -0.3248 -0.0527 0.0698 0.3382

Fig. 4. The Studied System Model for Simulation

control and manage the local relays centrally. The
communication lines (the dash lines shown in Fig. 4) assumed
TABLE I as a perfect state in this paper, connect the RA and relay as
FEEDER PARAMETERS
well as the RA and other RAs. In addition, the interconnection
Overhead line Cable line of IBDG and SBDG to the system is considered as Fig 1. In
R XL XC R XL XC this system, a 3-MW IBDG using PQ control strategy is
Positive sequence 0.45 0.5385 34979 0.27 0.08 7981.74 connected to the bus 2, and is considered the random change of
its supply. A 2.7-MW SBDG is connected to the bus5 on a far
Zero sequence 0.7 1.2273 83766 2.7 0.32 11373.98
right end of the system, and is assumed as full-power operation.
Note：the unit of feeder parameters is Ω/km
The loads are modeled by constant impedances, which are
represented by arrows shown in Fig. 4.
DRA. The signals include trip signals and block signals. The This paper only considered the fault occurs in the feeder and
trip signals are sent to adjacent RA where a fault may occur. In the bus. The situation of fault occurred in the inside of DG is
contrast, the block signals denote that the section protected by out of the scope of this paper.
RA is sound and the RA needs to block all relays. B. Generation of Historical Fault Data
In the coordination strategy, the RA closed to the main grid
In this simulation test, the historical fault data is generated
has a priority to execute the coordination strategy which can
by simulating different fault conditions beforehand. All the
send the signal to DRA. Then the DRA also sent the signal to
fault conditions are assumed to be occurred at phase A of each
the URA due to the results of fault discrimination. RA can be
feeder. The beforehand fault conditions are shown as follows:
cooperated to locate the exact fault line from top to bottom of
a) fault resistance Rf : 1Ω, 50Ω, 150Ω, 300Ω and 500Ω;
the main grid.
b) fault initial angle  : 0°, 45° and 90°;
c) the percent of the position Lf from the fault point to the far
left end of each feeder: 10%, 50% and 90%.
VI. SIMULATION ANALYSIS
d) the percent of the output power S of IBDG: 50% and
A. Simulation Model 100%.
A radial distribution systems (10 kV and 50 Hz) grounded Under above fault conditions, the data of each phase voltage
by Petersen-coil as shown in Fig. 4 is simulated using PSCAD. and current for all relays are recorded by PSCAD where the
The feeders including the overhead line (the fine lines such as data sampling rate is set as 4000Hz. The recorded data are
L1), the cable line (the bold lines such as L2) and the hybrid input into the Matlab software to normalize by z-score method
transmission line including overhead line and cable line (such and extract the fault features which constitutes the history fault
as L3) are considered in this system. The feeder parameters data of each relay. Then the fault cluster centers of each relay
including resistance R , inductive reactance X L and shown in Table II are calculated by FCM algorithm in Matlab.
capacitive reactance X C for positive sequence and zero In Table II, the cluster centers p f 1 - p f 13 are corresponding
sequence are shown in Table I. The Petersen coil grounding to the historical fault data of the relays R1-R13 respectively.
method with 110% degree of compensation is used, and the Each cluster enter has four variables such as the phase angle of
Petersen coil inductance is set as 0.83 H. zero-sequence current, the phase angle difference between the
There are 14 relays in this system used to protect the feeder. zero-sequence voltage and current, the magnitude of
And the system has four RAs installed at different buses to zero-sequence current and the transient zero-sequence energy.

0885-8977 (c) 2016 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.2016.2585380, IEEE
Transactions on Power Delivery
6

TABLE III
THE SPACE RELATIVE DISTANCE USING C ITY B LOCK METRIC DISTANCE METHOD UNDER DIFFERENT FAULT CONDITIONS FOR ALL RELAYS
Fault conditions RA1 RA2 RA3 RA4
Fault
Rf (Ω)  (°) Lf (%) S (%) R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13

F1 1 0 5 0 4.32 16.7 18.03 18.76 19.79 - - - - - - - -

F2 50 15 10 10 20.8 1.83 18.41 18.81 19.97 - - - - - - - -

F3 100 30 20 20 20.86 16.92 0.89 18.8 19.96 - - - - - - - -

F4 200 60 30 40 20.80 16.78 18.34 1.18 19.96 3.19 20.03 - - - - - -

F5 400 90 40 50 20.62 16.37 17.92 1.45 19.77 23.02 3.08 - - - - - -
F6 600 95 50 60 20.56 16.37 17.79 18.63 2.82 - - 3.14 21.12 12.31 - - -
F7 800 105 60 70 20.66 16.35 17.83 18.73 4.04 - - 15.15 4.89 12.39 - - -
F8 1000 120 70 80 20.43 15.65 17.44 18.54 4.09 - - 14.54 20.94 3.63 5.94 20.54 21.01
F9 1200 150 80 90 20.16 15.73 17.03 18.39 4.66 - - 14.25 20.7 4.43 16.97 4.5 20.85
F10 1500 180 90 100 20.32 16.61 17.39 18.85 4.32 - - 14.58 21.07 3.38 17.12 20.66 3.55
Note: there are only shown the data of the RA which are needed to coordinate the URA or DRA for fault location.

block metric distance method and Chebyshev distance method

can obtain different results shown in Fig. 5 for the space
relative distance.
From Fig. 5, the space relative distances of the different
relays (such as R1, R2,…, R5) to RA1 are compared when the
fault F1 happens. For fault F1, the fault feeder is L1. The space
relative distances of relay R1 is the minimum value among the
relays connected the RA1, which can identify the faulty feeder
reliably. However, the space relative distances using City
block metric distance method have the bigger difference
between the fault feeder and the healthy feeder, which means
that it has enough protection capability. This result also applies
to other fault conditions in other feeders.
Fig. 5. The space relative distance based on different distance calculation Considered the single-phase earth faults F1-F10 with
methods during the occurrence of fault F1
different fault conditions occurred in the feeders L1-L10
respectively, the space relative distance using City block
TABLE IV metric distance method for all relays are shown in Table III. As
THE SPACE RELATIVE DISTANCE USING C ITY B LOCK METRIC D ISTANCE
METHOD U NDER B US FAULT AT BUS1 FOR RA1 seen in Table III, all the minimum value among the relays
Fault conditions RA1 connected each RA are marked as bold one. In addition,
according to the coordination strategy of RA shown in Fig. 3,
S (%) R1 R2 R3 R4 R5 d RA
the tripped relays are marked with an underline during the
100 20.66 17.01 17.87 18.71 19.81 3.65 appearance of fault.
50 20.7 17.09 17.95 19.04 19.81 3.61 Based on the simulation results shown above, we can see
0 20.66 17.01 17.87 18.99 19.77 3.65 that the proposed method can reliably identify and isolate the
100 - 17.09 17.99 18.92 19.83 2.74 fault line by use of relay agent. For instance, when the fault
Switching off F10 occurs, the RA1 firstly starts the process of fault
50 - 17.16 18.06 18.96 19.86 2.7
Feeder L1
0 - 17.1 17.99 18.92 19.83 2.73 discrimination. The result shows that the space relative
distance of R5 is the minimum value and the fault section is
identified as the feeder L5 or its downstream area. Then the
RA1 sends the block signal to RA2 and the trip signal to RA3.
C. Identification for Feeder Fault According to the received signal, the RA2 will block all relays
When a single-phase earth fault F1 (Rf =1Ω,   0 and under its control, and the RA3 will execute the same process
S=0%) occurs at the 5% length of the feeder L1, the real-time what RA1 performed. Through above the process, the fault
and local fault data collected by the relay R1 are considered as line is accurately determined as the feeder L10 and the relay
a sample data Xs1. The space relative distance between the R13 is tripped.
sample data Xs1 and the cluster center p f 1 is calculated by
D. Identification for Bus Fault
distance calculation methods. The different distance
calculation methods based on Euclidean distance method, City Table IV shows the space relative distance of the relays
connected to RA1 when the single-phase earth faults

0885-8977 (c) 2016 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.2016.2585380, IEEE
Transactions on Power Delivery
7

more historical fault data and the better the proposed method
performed. In addition, the protective margins of the proposed
method in practical situation are enough to meet the needs with
a few data.
F. Comparison
In radial distribution systems, the DG’s fault contribution
may be same when the faults occur in the different feeders that
are connected in parallel with same parameters. For this fault
scenario, the methods (in [8][9]) based on current contribution
of sources may be misjudged the fault line. In contrast, the
proposed method using four fault features to determine fault
can avoid it. Considering feeder L2 with the same parameters
as feeder L1, the simulation results are shown in Table V when
Fig. 6. The space relative distance using City block metric distance method the faults happen in L1 and L2 , respectively, for the proposed
for relays R1-R5 which are considered the different amount of historical method. In Table V, the difference of the space relative
fault data during the occurrence of fault distance among the relays (R3, R4, R5) is small under different
fault points, and it can deem that the current contribution from
each source is nearly equal for R1 and R2. According to the
(Rf=500Ω and   0 ) occurred at the bus 1. The bus faults
simulation results, comparing the method based on the current
consider the variation of loading condition (switching off
contribution of sources, the proposed method has higher
feeder L1) and output power S of IBDG. Compared with the
reliability for the faults happened in the different feeders with
data of all feeder faults shown in Table III for RA1, Table IV
same parameters.
has the smaller difference among the data during the bus fault.
TABLE V
Thus, the difference d RA of the space relative distance can be THE SPACE R ELATIVE DISTANCE USING CITY BLOCK METRIC DISTANCE
defined by METHOD UNDER DIFFERENT FAULT POINTS
Fault conditions RA1
d RA  d max  d min (14) Fault
Rf (Ω)  (°) Lf (%) S (%) R1 R2 R3 R4 R5
where the d max is the maximum value among the relays point
L1 200 0 50 100 2.31 20.87 18.14 18.89 20.08
connected RA and the d min is the minimum value among the L1 400 0 50 100 2.91 19.72 18.23 18.84 20.18
relays connected RA. L1 800 0 50 100 3.38 21 18.09 19.04 20.18
L2 200 0 50 100 20.75 2.31 18.14 18.89 20.08
From the Table III and Table IV, for RA1, the range of L2 400 0 50 100 20.81 2.96 18.13 18.93 20.15
d RA is from 2.7 to 3.65 during the bus faults and the range of L2 800 0 50 100 20.75 3.48 18.09 19.06 20.18
Note: there are only shown the data of the RA which are needed to
d RA is from 8.83 to 15.23 during the feeder faults. There are coordinate the URA or DRA for fault location.
obvious differences in d RA between the bus fault and feeder
fault. Thus, a threshold d threshod is defined to identify bus fault. VII. CONCLUSION
When the d RA for RA is less than the d threshod , the fault A novel protection method based on RA has been proposed
type is identified as bus fault for RA, and all relays connected in this paper to improve the protection performance of
to RA are tripped. On the contrary, the fault type is identified traditional methods in the single-phase earth fault protection
as feeder fault when the d RA is more than the d threshod . for radial distribution systems with DGs. Its specific merits are
as follows:.
E. Effect of Historical Fault Data 1) The space relative distance based on different distance
The fault cluster center can effectively reflect the distributed methods are compared to find out the optimal one for the
situation of fault data for each feeder. In addition, the proposed method. The result shows that the space relative
reliability can be improved further with increasing the amount distances using City block metric distance method is the best
of the available history fault data. However, in some practical method to find the difference between the sample data and the
situations, the amount of the available history fault data is not historical fault data.
enough to support the need of the existing protection methods 2) The simulation results show that the proposed protection
based on the intelligent algorithm such as neural network method has desired performances under the line and bus faults
algorithm which need to have enough training samples. with variable fault condition including the variation of loading
Therefore, the effects of the amount of historical fault data condition and output power of DG.
need to be considered in this paper. 3) The more historical fault data are collected, the better the
Fig. 6 shows the space relative distance using City block proposed method performed. The fault cluster of historical
metric distance method for relays R1-R5 which are considered fault data can better under the more fault data reflected
the different amounts of historical fault data when the fault (Rf different fault conditions for each feeder. In addition, the
=1Ω,   0 , Lf=5%, S=100%) occurs in L1. It reflects that the proposed method can discriminate fault under new fault
conditions, which is not included in historical fault data.

0885-8977 (c) 2016 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.2016.2585380, IEEE
Transactions on Power Delivery
8

Series,” Neural Networks (IJCNN), The 2010 International Joint

Conference on, Barcelona, Spain, Jul. 2010, pp. 1-8.
[19] YIN Zhonghang, TANG Yuangang, SUN Fuchun and SUN Zengqi,
REFERENCES “Fuzzy Clustering with Novel Separable Criterion,” Tsinghua Science
[1] Zhang Zhixia, Liu Xiao, Piao Zailin., “Fault line detection in neutral and Technology , vol. 11, iss. 1, pp. 50-53, Feb. 2006.
point ineffectively grounding power system based on phase-locked [20] Yuanyuan Wang, Xiangjun Zeng, Zhaoyang Dong and Yue Huang,
loop,” IET Gener. Transm. Distrib., vol. 8, iss. 2, pp. 273-280, 2014. “Novel protection scheme of stator single-phase-to-ground fault for
[2] Claudio S. Mardegan, Rasheek Rifaat, “Insights Into Applications of powerformers,” Int. J. Electr. Power Energy Sys., vol.53, pp. 321-328,
IEEE Standards for Ground-Fault Protection in Industrial and 2013.
Commercial Power Systems,” IEEE Trans. Ind. Appl., vol. 51, no. 4, pp. [21] Torleiv Kløve, Te-Tsung Lin, Shi-Chun Tsai and Wen-Guey Tzeng,
2854-2861, Jul/Aug. 2015. “Permutation Arrays Under the Chebyshev Distance,” IEEE
[3] Mohammed A. Haj-ahmed and Mahesh S. Illindala, “The Influence of Transactions on Information Theory, vol. 56, no. 6, pp. 2611-2617, Jun.
Inverter-Based DGs and Their Controllers on Distribution Network 2010
Protection,” IEEE Trans. Ind. Appl., vol. 50, no. 4, pp. 2928-2937,
Jul/Aug. 2014. Yuanyuan Wang (S’12-M’15) received the B.S. and M.Sc
[4] Xiangning Lin, Zhimin Zhao, Zhiqian Bo, “Fault Analysis on degree in electrical engineering from Changsha University
Distribution Feeders With Distributed Generators,” IEEE Trans. Power of Science and Technology, China, in 2004 and 2007
Syst., vol. 20, no. 4, pp. 1757-1764, Nov. 2005 respectively, the Ph.D. degree in electrical engineering
[5] N. Perera, A. D. Rajapakse and T. E. Buchholzer, “Isolation of Faults in from the College of Electrical Engineering, Guangxi
Distribution Networks With Distributed Generators,” IEEE Trans. University, Guangxi, China, in 2012.
Power Del, vol. 23, no. 4, pp. 2347-2355, Oct. 2008. Now she is an Associate Professor with the College of
[6] H. H. Zeineldin, Hebatallah M. Sharaf, Doaa K. Ibrahim and Essam Electrical and Information Engineering, Changsha University of Science and
El-Din Abou El-Zahab “Optimal Protection Coordination for Meshed Technology. Her research interests are power system protection and control.
Distribution Systems With DG Using Dual Setting Directional
Over-Current Relays,” IEEE Trans. Smart Grid., vol. 6, no. 1, pp. Gen Wei received B.S degree in electrical engineering
115-123, Jan. 2015. from Xi’an Polytechnic University, Xi’an, China, in 2014.
[7] M. Dewadasa, A. Ghosh, G. Ledwich and M. Wishart, “Fault isolation in He is currently working toward the M.S. degree in
distributed generation connected distribution networks,” IET Gener. electrical engineering in the College of Electrical and
Transm. Distrib., vol. 5, iss. 10, pp. 1053-1061, 2011. Information Engineering, Changsha University of Science
[8] Hadi Zayandehroodi, Azah Mohamed,Hussain Shareef and Masoud and Technology, Changsha, China.
Farhoodnea, “A novel neural network and backtracking based protection His research interests are power system protection and
coordination scheme for distribution system with distributed control.
generation,” Int. J. Electr. Power Energy Sys., vol.43, pp. 868-879,
2012. Hongming Yang (M'12) received her M.Sc. Degree in
[9] N. Rezaei and M.-R. Haghifam, “Protection scheme for a distribution electrical engineering from Wuhan University in 1997, and
system with distributed generation using neural networks,” Int. J. Electr. received her Ph.D. degree in electrical engineering from
Power Energy Sys, vol. 30, pp. 235-241, 2008. Huazhong University of Science and Technology in 2003.
[10] Ying N. Wang, Jin F. Ye, Guo J. Xu, Qing M. Chen, Hai Y.Li and Xin She was a Research Associate at the Hong Kong
R.Liu, “Novel Hierarchical Fault Diagnosis Approach for Smart Power Polytechnic University during 2009–2010, a visiting
Grid with Information Fusion of Multi-data Resources Based on Fuzzy scholar in North Carolina State University during
Petri Net,” Fuzzy Systems (FUZZ-IEEE), 2014 IEEE International 2010–2011, and a Research Fellow at the University of
Conference on, Beijing, China, Jul. 2014, pp. 1183-1189. Newcastle during 2013–2014. Currently, she is a fully Professor at Changsha
[11] Hui Wan, K. K. Li and K. P. Wong, “An Adaptive Multiagent Approach University of Science and Technology. Her research interests include power
to Protection Relay Coordination With Distributed Generators in system analysis and power markets.
Industrial Power Distribution System,” IEEE Trans. Ind. Appl., vol. 46,
no. 5, Sep/Oct. 2010.
[12] P. C. Maiola and J.G. Rolim, “A Multi-Agent System for protection Haowei Chen received B.S degree in electrical
coordination of radial systems in the presence of distributed generation,” engineering from Hubei Institute for Nationalities,
Developments in Power Systems Protection, 2012. DPSP 2012. 11th Huanggang, China, in 2015. He is currently working
International Conference on., Birmingham, UK, Apr. 2012, pp. 1-6 toward the M.S. degree in electrical engineering in the
[13] M.F. Abdel-Fattah and M. Lehtonen, “Transient algorithm based on College of Electrical and Information Engineering,
earth capacitance estimation for earth-fault detection in medium-voltage Changsha University of Science and Technology,
networks”, IET Gener. Transm. Distrib., vol. 6, iss. 2, pp. 161-166, Changsha, China.
2012. His research interests are power system protection and
[14] Yuanyuan Wang, Jiaming Zhou, Zewen Li, Zhaoyang Dong and Yao Xu, control.
“Discriminant-Analysis-Based Single-Phase Earth Fault Protection
Using Improved PCA in Distribution Systems,” IEEE Trans. Power Del, Zi Ouyang (M'11) received his PhD at the University of
vol. 30, no. 4, pp. 1974-1982, Aug. 2015. New South Wales Australia and Bachelor in Physics at
[15] Timbus, A.V., Rodriguez, P., Teodorescu, R., Liserre, M.and Blaabjerg, Nanjing University, China. He is currently Research
F., “Control Strategies for Distributed Power Generation Systems Fellow at School of Photovoltaic and Renewable Energy
Operating on Faulty Grid,” Industrial Electronics, 2006 IEEE Engineering, UNSW Australia. Prior to his current
International Symposium on, Montreal, QC, Canada, Jul. 2006, vol. 2, position, he was a post-doctoral fellow in Swinburne
pp. 1601-1607. University of Technology Australia, conducting research
[16] Nagy I. Elkalashy, Abdelsalam M. Elhaffar, Tamer A. Kawady, Naser G. for Suntech Power.
Tarhuni, and Matti Lehtonen, “Bayesian Selectivity Technique for Earth His research interests are photovoltaic and energy storage technologies for
Fault Protection in Medium-Voltage Networks,” IEEE Trans. Power on-grid and off-grid systems.
Del, vol. 25, no. 4, pp. 2234-2245, Oct. 2010.
[17] QIAN Hong, HUANG Zhengrun, RUAN Dabing, “Single
David J. Hill (M'76-SM'91-F'93-LF'14) received the Ph.D.
phase-to-earth fault location of small current grounding system with
degree in electrical engineering from the University of
distributed generation,” Electric Machines and Control, vol. 18, no. 8, pp.
Newcastle, Australia, in 1976.
17-23, Aug. 2014.
He holds the Chair of Electrical Engineering in the
[18] Eduardo Ogasawara, Leonardo C. Martinez, Daniel de Oliveira, Geraldo
Department of Electrical and Electronic Engineering at the
Zimbrão, Gisele L. Pappa and Marta Mattoso, “Adaptive Normalization:
University of Hong Kong. He is also a part-time Professor
A Novel Data Normalization Approach for Non-Stationary Time
of Electrical Engineering at The University of Sydney,

0885-8977 (c) 2016 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.2016.2585380, IEEE
Transactions on Power Delivery
9

Australia. During 2005-2010, he was an Australian Research Council

Federation Fellow at the Australian National University. Since 1994, he has
held various positions at the University of Sydney, Australia, including the
Chair of Electrical Engineering until 2002 and again during 2010–2013 along
with an ARC Professorial Fellowship. He has also held academic and
substantial visiting positions at the universities of Melbourne, California
(Berkeley), Newcastle (Australia), Lund (Sweden), Munich, Hong Kong (City
and Polytechnic). His general research interests are in control systems,
complex networks, power systems and stability analysis. His work is now
mainly on control and planning of future energy networks and basic stability
and control questions for dynamic networks.
Prof. Hill is a Fellow of the Society for Industrial and Applied Mathematics,
the Australian Academy of Science, and the Australian Academy of
Technological Sciences and Engineering. He is also a Foreign Member of the
Royal Swedish Academy of Engineering Sciences.

0885-8977 (c) 2016 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.