ArticlePDF Available

Protection framework for microgrids with inverter‐based DGs: A superimposed component and waveform similarity‐based fault detection and classification scheme

Authors:

Abstract and Figures

Since the penetration level of the inverter-based distributed generations (IBDGs) into microgrids (MGs) is increasing, the protection issues of such networks have become more challenging. The present work aims to design a protection framework for fault detection and classification in IBDG dominated MGs. In the proposed approach, the current waveforms measured from one end of the protected lines are processed using superimposed components. Then, the faults are identified using the Euclidean distance concept. A Pearson correlation coefficient-based approach is also developed to characterize the faulty phases in the proposed fault classification unit. Furthermore, an auxiliary index is defined to discriminate load change conditions from the faults. In addition to being simple and efficient, the proposed technique is highly capable of functioning in different MG operating modes and configurations without changing its settings. Also, it demonstrates robust performance against IBDG impacts, noises, and stimulating fault scenarios. To assess the scheme proficiency, a modified IEEE 15-bus system with highly penetrated IBDGs is simulated. The results confirm the quickness and high accuracy of the proposed technique. Moreover, the method is validated in an experimental laboratory test bench. Finally, the superiority of the method is confirmed by a comparative study with other recent methods.
This content is subject to copyright. Terms and conditions apply.
Received: 9 June 2021 Revised: 7 January 2022 Accepted: 3 February 2022 IET Generation, Transmission & Distribution
DOI: 10.1049/gtd2.12438
ORIGINAL RESEARCH
Protection framework for microgrids with inverter-based DGs: A
superimposed component and waveform similarity-based fault
detection and classification scheme
Mohammad Amin Jarrahi1Haidar Samet1,2Teymoor Ghanbari3
1School of Electrical and Computer Engineering,
Shiraz University, Shiraz, Iran
2Department of Electrical Engineering, Eindhoven
University of Technology, Eindhoven, The
Netherlands
3School of Advanced Technologies, Shiraz
University, Shiraz, Iran
Correspondence
Haidar Samet, Namazi Sq., School of Electrical and
Computer Engineering, Shiraz University, Shiraz
7134851154, Iran.
Email: samet@shirazu.ac.ir
Abstract
Since the penetration level of the inverter-based distributed generations (IBDGs) into
microgrids (MGs) is increasing, the protection issues of such networks have become more
challenging. The present work aims to design a protection framework for fault detec-
tion and classification in IBDG dominated MGs. In the proposed approach, the current
waveforms measured from one end of the protected lines are processed using superim-
posed components. Then, the faults are identified using the Euclidean distance concept. A
Pearson correlation coefficient-based approach is also developed to characterize the faulty
phases in the proposed fault classification unit. Furthermore, an auxiliary index is defined
to discriminate load change conditions from the faults. In addition to being simple and
efficient, the proposed technique is highly capable of functioning in different MG operat-
ing modes and configurations without changing its settings. Also, it demonstrates robust
performance against IBDG impacts, noises, and stimulating fault scenarios. To assess the
scheme proficiency, a modified IEEE 15-bus system with highly penetrated IBDGs is sim-
ulated. The results confirm the quickness and high accuracy of the proposed technique.
Moreover, the method is validated in an experimental laboratory test bench. Finally, the
superiority of the method is confirmed by a comparative study with other recent methods.
1 INTRODUCTION
1.1 Problem statement and motivations
Microgrids (MGs) are developed as one of the main parts of
modern power systems and smart grids. Generally, MGs are
defined as medium or low-voltage systems with various kinds
of distributed generators (DGs), loads, and energy storages
[1]. Extensive application of cost-effective power-electronic
interfaces enables MGs to widely utilize renewable-energy-
based DGs. These types of DGs, so-called inverter-based DGs
(IBDGs), are the backbone of MGs. However, one of the main
concerns regarding the expansion of MGs with highly pene-
trated IBDGs is their protection challenges. Since IBMGs have
various behaviours under fault conditions, the protection sys-
tem designers will be faced to some new challenges in such net-
works [2].
This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the
original work is properly cited, the use is non-commercial and no modifications or adaptations are made.
© 2022 The Authors. IET Generation, Transmission & Distribution published by John Wiley & Sons Ltd on behalf of The Institution of Engineering and Technology
From the protection point of view, IBDGs can impact magni-
tude, waveform, direction, and duration of the fault current [3].
According to IEEE Std. 1547.2 and reports of Sandia National
Laboratories, the fault current of IBDGs are limited to 1.2–2
times of the rated current due to the current limiting strategy
in the inverter control systems for preventing damages to semi-
conductor components [4–6]. Moreover, IBDGs can affect the
output waveform to be non-sinusoidal and as a result, differ-
ent harmonic contents can be produced. These factors make
protection schemes unreliable and inefficient. Among these
schemes, fault detection and classification are more prone to be
influenced by the IBDGs [7]. The fault detection is the first sub-
routine of different MG protection schemes. So, it must operate
correctly with the highest possible speed. Also, fault classifica-
tion is necessary to start the fault isolation procedure. Motivated
from these issues, the fault detection and classification in MGs
in presence of IBDGs are studied in this paper.
2242 wileyonlinelibrary.com/iet-gtd IET Gener. Transm. Distrib. 2022;16:2242–2264.
JARRAHI ET AL.2243
1.2 The literature survey
Up to now, numerous schemes have been developed for fault
detection and classification in MGs. These methods can be
broadly classified into two general groups named single-end
measurement-based techniques and double-end measurement-
based algorithms. In the following, some of the well-known
techniques of each group are discussed in more detail.
The method proposed in [8] utilizes the voltage signals of the
single-end of MG lines along with a decision-tree to detect and
classify the faults. This method mainly suffers from shortcom-
ings related to close-in faults in which the voltage magnitudes
are near zero. Application of current and voltage sequences
has been presented in [9] to identify the faults in IBDG dom-
inated MGs. This approach does not consider the impacts of
limited fault currents in islanded (IL) operation mode of MGs
which is a common phenomenon. In [10], a technique based
on angles of voltage sequences has been developed, calculated
from one end of the line to classify the faults in MGs with
photovoltaic (PV) DGs. Although this technique has promis-
ing results, high computational burden and low speed can be
considered as its downsides. The algorithm built on three-phase
current waveforms processed with wavelet transform and deep-
learning has been reported in [11]. This approach generally suf-
fers high complexity and the need for a huge amount of data.
The transient monitoring function concept has been applied to
current waveforms to distinguish the faults in islanded MGs in
[12]. The approach is very sensitive to measurement noises and
large load connections, so it cannot be the ultimate choice for
protection of MGs. In [13], the short-time correlation trans-
form of superimposed voltage and current signals have been
utilized for the protection of MGs. This method is also not
practically feasible as the dynamic IBDG characteristics and
faults with high resistances are not taken into consideration. A
mathematical morphology-based fault detection and classifica-
tion technique has been developed in [14] for IBDG enriched
MGs. This method is system-dependent which means it is case
sensitive and its settings must be changed for each studied net-
work. In [15], the traveling waves of current signals has been
utilized. The accuracy and speed of this approach are suitable
for MG protection, however, it requires ultra-high sampling
rate devices, which are very expensive. The application of over-
current protection in MGs is reviewed in [3]. Although this
type of protection offers a simple and economical method, low
sensitivity and the vital need for extra devices are their main
problems. Moreover, they can be easily affected by network
disturbances and noise interferences. Different threshold set-
tings for various MGs with different configurations are also
another shortcoming. What is more important is that the fault
current generated by the IBDGs is limited which the overcur-
rent based method either do not respond or if they do respond,
the operating time is much larger. Regardless of the existing
shortcomings in the one-end protection approaches, they are
preferred over double-end or differential-based techniques yet.
They do not require high-cost and problematic communication
links.
A differential protection strategy using voltage signals along
with a sensitivity analysis approach has been developed in [16].
The method is not cost-efficient due to its high dependence
on the required complex communication infrastructures. The
application of time-frequency transform has been presented in
[17] as the differential current-based technique. Noise sensitiv-
ity, low speed, and requirement of different communication cen-
tres are the limitations of this technique. In [18], the applicability
of differences in the current frequency components measured
from both-end of the protected line has been investigated for
identification of faults in IBDG dominated MGs in IL oper-
ation mode. Since this technique injects a frequency through
the DG interface during fault conditions, its implementation
in practice is under-question. Active power, current magnitude,
and voltage sags are inputs of the differential-based method in
[19]. Although this approach models the control algorithm of
inverters and current limiters, it does not consider the effects of
failure and/or delays in communication links which is common
in differential algorithms. That is why a backup protection is
necessary to ensure the protection reliability, which is not cost
efficient. Also, this may increase the total cost and size of the
protection system and limits its application in MG. An adaptive
differential intelligent protection scheme has been proposed in
[20] utilizing the machine learning models. The main drawbacks
of this scheme are the critical need for setting readjustments
when MG switches between different operation modes and the
huge data required for the training stage. The ensemble empir-
ical mode decomposition has been applied to current signals
in [21] to compute the differential energy to discriminate the
faults. The impacts of noises and faults with high resistance are
not studied in this technique. Moreover, the performance of this
approach has not been assessed in presence of IBDGs consid-
ering that the fault current level is limited in such conditions.
1.3 The contributions and organization
In this paper, a simple and efficient protection framework is
proposed for fault detection and classification in MGs with
highly-penetrated IBDGs. The proposed technique utilizes the
superimposed components of one-end current signals for the
fault detection unit. In this unit, a well-known waveform sim-
ilarity technique named Euclidean distance (ED) is developed
to characterize the faults. Then, the Pearson correlation coef-
ficients (PCC) of fault-imposed current waveforms are com-
puted as the fault classification indices. Based on some defined
thresholds and rules, the faults are precisely identified and cat-
egorized in half cycle after the fault inception. Moreover, an
auxiliary index is developed to differentiate the load change
from fault occurrence conditions. The performance of the pro-
posed method is evaluated in a network derived from IEEE 15-
bus system simulated in MATLAB/Simulink considering vari-
ous challenging scenarios. Also, experimental verification is per-
formed using the gathered data from a laboratory small-scale
test system. The results indicate that the proposed protection
framework has a promising capability to identify faults with
2244 JARRAHI ET AL.
FIGURE 1 Typical configuration of an IBDG
model [19]
a high degree of speed and accuracy. Furthermore, to verify
the superiority of the developed scheme, it is compared with
some other recent techniques from different aspects using some
quantitative and qualitative indices.
The main contributions of this method can be summarized
as follows:
An efficient, fast and accurate approach based on ED and
PCC for detection and classification of faults in MGs.
No need for communication channels as the method uses the
local one-end current measurement.
High efficiency in both operation modes and configurations
as it does not need to readjust its settings.
Robustness against IBDGs impacts, measurement noises,
load changes etc.
The rest of the paper is organized as follows: Section 2dis-
cusses the IBDG fault characteristics. The proposed method is
presented in Section 3. Section 4dedicates to simulation studies
and results. The experimental verification is presented in Sec-
tion 5. The comparative study is revealed in Section 6and at the
end, the conclusions are being listed in Section 7.
2IBDGS SPECIFICATIONS
2.1 IBDG modelling
Typically, an IBDG unit included a primary source of generation
and an energy conversion stage, as observed in Figure 1. Vari-
ous kinds of technologies can be used as a primary source in
these units such as microturbines, fuel cells, solar panels, batter-
ies, and wind generators. The energy conversion stage presented
in Figure 1represents the most common configuration for this
type of application, composed of a DC/DC converter, a DC
link capacitor, and a DC/AC converter [19]. The link capacitor
is usually designed with high energy storage capacity in order
to meet sudden changes in demand and afford the operational
dynamics of the technology used as a primary generation source.
In these conditions, the link capacitor voltage tends to be con-
stant during the operation of the presented IBDG configura-
tion. Accordingly, the IBDG configuration shown in Figure 1
can be simplified to the model presented by Figure 2, with con-
stant voltage input Vdc. Therefore, the fault behaviour of an
IBDG relies essentially on the DC/AC converter characteristics
and its control strategies.
Studies have shown that the fault response of IBDGs is
intrinsically related to its control strategy [22–26]. The control
FIGURE 2 Simplified IBDG model [19]
strategy used in an IBDG unit is defined according to its opera-
tional condition and the nature of its interaction with MG. The
strategies applied to IBDGs are commonly based on a multi-
loop structure, composed of an Inner-Loop and an Outer-Loop
control. Several methods can be used to achieve the purpose of
these loops, and its determination relies on the IBDG adopted
mode of control, which can be classified as grid-forming or
grid-following. Figure 3shows numerous possible combinations
that can be employed in an IBDG control strategy regarding its
mode of operation and methods that can be used [19].
The IBDG model developed for the simulation studies is
modelled considering the grid-following mode of control, aim-
ing to afford microgrid islanded and grid-connected mode of
operation. Figure 4shows a typical control scheme used for
IBDGs in grid-following mode.
Note that the presented diagram is composed of distinct
blocks, with different objectives, so that the desired control can
be achieved. Each of these blocks has an extensive variety of
internal methods that can be employed to perform its function.
The blocks presented in the diagram are listed below:
Current limiter: Responsible for limiting the output current
magnitude of the inverter.
Current controller: Responsible for controlling the current
quality.
Power controller: Responsible for controlling the active and
reactive power injection in the grid.
PLL: Responsible for obtaining the phase of the measured
voltage at the IBDG connection point.
dq0/abc: Responsible for converting the reference frame of
the measured parameters.
The wide variety of control strategies presented in Figure 3,
along with the operational characteristics presented in Figure 4,
elucidates the vast number of techniques and methods com-
binations that can be used to implement the control scheme
of IBDGs. Currently, there are no established norms or stan-
dardized design rules for the development of IBDGs control
strategies [5]. Consequently, IBDG’s short-circuit contribution
JARRAHI ET AL.2245
FIGURE 3 IBDGs control strategies and possible methods
[19]
FIGURE 4 Typical control scheme used for IBDGs in grid-following mode [19]
varies according to its technical characteristics. Therefore, fault
detection and classification in IBDG-dominated MGs becomes
a complex issue.
Based on the studies presented in [22–26], the IBDG model
used in this work is developed regarding three main characteris-
tics:
Use of natural reference frame (NARF): Allows independent
control of each of the IBDG phases.
Use of RMS current limiter: Avoids current waveform output
distortion when limiting the fault current contribution.
Use of proportional-resonant (PR) controller: Enables the
use of NARF and RMS current limiter.
Thus, the developed IBDG model can be represented by the
diagram of Figure 5. Therefore, during the normal operation
of MG, IBDG model is controlled based on the voltage of
its point of connection, in order to achieve the power supply
reference. On the other hand, during the fault condition, the
IBDG model operates as a current source, providing a maxi-
mum current magnitude value of 2 p.u.
2.2 IBDG fault characteristics
IBDGs generally employ power-electronics-based interfaces
between the sources and MGs. As mentioned, high penetra-
tion of IBDGs may lead to various problems for MG protection
schemes [22–24]. That is why; their fault characteristics must be
investigated. In the following, a brief explanation is given in this
regard:
As mentioned, IBDGs fault behaviours are characterized by
their control technique, the current limiting method, and the
implemented reference frame [19, 25]. More detail in this
regard can be found in [5].
2246 JARRAHI ET AL.
FIGURE 5 IBDG model in normal and fault
condition [19]
IBDGs are generally aimed at limiting the current to certain
permissible values during the fault inception [3]. Hence, the
protection methods based on exceeding current signals from
the relevant thresholds, generally entitled over-current tech-
niques, may be faced to malfunctioning because the threshold
settings should be readjusted.
When a fault occurs, the current of the inverters experiences
a very short transient and then it is limited to 1.2–2 times of
its rated value in steady-state fault conditions. Based on this
specification, it is proved that IBDGs behave as a controlled
current source during the faults in MGs [22].
Since IBDGs have power-electronics components, they can
affect the current waveforms to be non-sinusoidal with
unwanted harmonics in the fault situations [2].
Negative or zero-sequence components of the current might
not be created in IBDG dominated MGs in fault conditions
because of the design specifications of the inverters [26].
The inertia of MGs with IBDGs is much lower than the
synchronous-based ones [23]. So, their rate of change of fre-
quency is significantly high. Moreover, it is expected to have
faster power swings in MGs enriched with IBDGs.
The phase angle of IBDGs dynamically changes based on
their utilized control scheme and inverter terminal voltage
amplitude [24].
The above-mentioned specifications of IBDGs may result in
different challenges for various MG penetration aspects includ-
ing protection frameworks. So, their effects must be considered
in the relevant protection methods.
3THE PROPOSED PROTECTION
FRAMEWORK
The proposed protection framework has two consecutive units.
In the first stage, the fault inception moment is determined by
the fault detection unit (FDU). When it senses an abnormal sit-
uation, an activator switch triggers the second unit entitled the
fault classification unit (FCU). Each unit is discussed in detail as
follows.
3.1 The fault detection unit
FDU utilizes the three-phase instantaneous currents to form the
superimposed component-based fault detection (SFD) signal.
This signal is derived from the normalized superimposed three-
phase currents of single-end of the MG lines. Since different
types of faults may occur, it is necessary to monitor all three-
phase currents. So, in this paper, SFD signal which includes
desirable information of all three phases is utilized as the main
signal in the fault detection unit. This signal has all signatures
of the three-phase currents and reduces the overall processing
time and memory requirements. Steps of the proposed scheme
are described in the following.
Step 1: Normalizing the current signals
In order to generalize the proposed scheme to be robust
in different systems with various configurations and modes,
the current signals are firstly normalized using their base
values:
ipu (n)=i(n)
ibase
(1)
Equation (1) describes the normalized values for each
phase current in which ndenotes the nth sample of the cur-
rent signal and i,ipu and ibase are the sampled, normalized,
and base values, respectively. It should be mentioned that
the base values are selected using the specifications of each
system.
Step 2: Defining the superimposed component signals
Having the per-unit values of the phase current signals from
the first step, the superimposed component (SIC) signal for each
phase is computed as follows:
SICa(n)=ia,pu (n)ia,pu (nNs)
SICb(n)=ib,pu (n)ib,pu (nNs)
SICc(n)=ic,pu (n)ic,pu (nNs)
(2)
As described in (2), SIC signals are defined as the absolute val-
ues of the cycle-by-cycle comparison approach for each phase
current signal. In (2), SICa,SICb, and SICcare the proposed SIC
signals for the phases a,b, and c, respectively and ia,pu,ib,p u, and
ic,pu are normalized currents at the relay location. Also, Nsis the
number of samples per cycle.
JARRAHI ET AL.2247
Step 3: Outlining the fault detection signal
In the normal operation of MG, the magnitude for SIC sig-
nals is very close to zero. The minor deviations of SIC signals
from zero can be referred to the presence of different noises in
the measured current signals and unbalanced loads. But after a
fault occurrence, magnitude of SIC signals will increase. Consid-
ering this increase, a unique signal for fault detection purposes
can be defined. To do so, the summation of SIC signals is con-
sidered as follows:
SFD (n)=SICa(n)+SICb(n)+SICc(n)(3)
The value of SFD signal is near zero during normal operation,
but it experiences abrupt change after the fault occurrence.
Step 4: Proposing the fault detection index
To make a suitable criterion from SFD, a well-established
waveform-similarity-based approach named ED is utilized. This
approach measures the waveform similarity thus the normal
operation of MG can be easily distinguished from the faults.
In digital signal processing, the similarity between two discrete
sampling signals, for example, x(n), y(n), is commonly measured
by ED, which is defined as (4):
EDxy =
N
i=1
(x(i)y(i))2(4)
In (4), EDxy is the average of ED between x(n) and y(n) in a
data window with the number of Nsampling points and irepre-
sents the number of the sampling point. The calculation of ED
is simple and intuitive. If two time-series data are the same, ED
is calculated to be zero, otherwise, there is a large value of ED
[27]. In order to recognize the fault, the ED concept is applied
to SFD signal with the following procedure:
EDSFD (i)=
N
i=1
(SFD (i)SFD (iNs))2(5)
As presented in (5), the proposed EDSFD calculates ED
between the current sample of SFD signal (SFD(i)) and the cor-
responding sample in the last cycle (SFD(i-Ns)). The number of
samples in the calculated EDSFD equals the number of samples
in a cycle of MG. It is evident that the value of EDSFD is very
near to zero in the normal operation of MG but it increases to
higher values after fault incidence. However, since the ampli-
tude of the EDSFD is different for MGs with various modes
and structures, it is not easily possible to identify the fault by
comparing the EDSFD with a general threshold. To significantly
characterize the change in MG with all possible conditions, a
fault detection index (FDI) based on EDSFD is defined as fol-
lows:
FDI (i)=11
1+EDSFD (i)(5)
From (5), it is evident that the possible range of the FDI is
[0,1]. In the normal operation of the MG, the value of FDI is
very close to zero since the value of EDSFD is very small in
such conditions. However, it increases to a higher value (near
to one) due to the significant values of EDSFD.IfFDI exceeds
a predefined threshold, the occurrence of a fault in the MG is
characterized. In this paper, the threshold is determined using a
well-known technique named Otsu thresholding method (OTS)
[28]. The implementation procedure of this approach can be
found in the following subsection [29]. Using this technique,
the value of the threshold is derived 0.56 1/A2using various
simulation and experimental case studies. It should be noted
that the selected threshold is constant for different MGs struc-
tures and operation modes since the FDI value varies only
between 0 and 1. In order words, the threshold setting is not
case sensitive and no need to readjust for different MGs struc-
tures or operating modes. The application of this threshold
helps to have desirable results regarding accuracy and operating
time.
The step-by-step operation of the proposed FDU is
depicted in the schematic shown in Figure 6. As presented
in this figure, FDU has a low computational burden and
as a result, it improves the speed of the whole protection
framework.
3.2 The fault classification unit
Recognizing the fault type and fault location are the duties of
FCU. This unit starts operating right after FDU characterizes
the fault inception moment. The proposed FCU utilizes the
samples of SIC signals calculated in the last unit. So, it would not
impose any additional burden on the relay. The proposed algo-
rithm is based on the utilization of PCCs between the samples of
SIC signals of each phase. The idea behind this algorithm came
from the similarity seen in the SIC signals in the first moments
after any fault type as they behave with a specific shape with a
pattern. This unique characteristic makes the SIC signals suit-
able for defining some indices to classify the faults. Before pre-
senting the proposed algorithm, the required concepts of PCCs
are presented.
Fundamentals of Pearson correlation coefficient
PCC is a powerful statistical tool to determine the type and
degree of relationship between one quantitative variable with
another one. The PCC indicates the intensity of the relationship
as well as the type of relationship (direct or inverse) between the
variables. The value of PCC is a number between 1 and 1and
if there is no relationship between the two variables, this value
will be zero. The closer value of this coefficient to one refers
to more similarity of them, while negative value of this coeffi-
cient means that the relationship between the two variables is
inverse. The PCC for the two signal samples x={x1,x2,…,xn}
and y={y1,y2,…,yn}, represented as the PCC(x,y), is defined
2248 JARRAHI ET AL.
FIGURE 6 Step-by-step operation of the proposed FDU
as [30]:
PCC (x.y)
=n
i=1(xi1
nn
j=1xj)(yi1
nn
j=1yj)
n
i=1(xi1
nn
j=1xj)
2n
i=1(yi1
nn
j=1yj)
2
(6)
The value +1 for PCC(x,y) indicates a perfect positive linear
correlation, and the value 1 indicates perfect negative linear
correlation. Using the concept of PCC, the fault classification
indices are presented in the next step.
Step 1: The proposed fault classification indices
The first task in the proposed FCU is to differentiate the
grounded faults from ungrounded ones. To do so, a ground
fault index is proposed considering the ground current or zero
sequence current:
i0(t)=1
3(ia(t)+ib(t)+ic(t)) (7)
The ground current i0is computed and based on the spec-
ification of this signal, the feature for detecting the grounded
faults is defined as follows, named ground fault index (Cg):
Cg=max(i0hal f cycle a fter fault (8)
As mentioned, the proposed FCU is based on the PCC
between SIC signals after fault detection. PCCs between SIC sig-
nals are defined in some pairs between each phase and the other
ones. In order to illustrate the idea of the proposed algorithm,
some typical waveforms of SIC signals for different faults are
presented in Figure 7. Considering the mentioned explanations
and the resemblances and changes of the SIC signals waveforms,
the PCC concept can be the perfect choice to characterize the
faulty phase/phases. To do so, the following indices are sug-
gested:
Cab =PCC (SI Ca.SICb)
Cca =PCC (SI Cc.SICa)
Cbc =PCC (SI Cb.SICc)
(9)
In (9), Cab,Cca and Cbc are PCCs of SIC signals between
phases aand b, phases cand aand phases band c, respectively.
The window length for calculation of the proposed indices is a
half cycle after fault incipient moment identified in FDU.
To understand the procedure of the proposed indices more
in detail, the proposed Cab, as an example, is expanded similar
to (6) as follows:
Cab =
TH+tf
i=tfSICa1
H
TH+tf
j=tf
SICaSICb1
H
TH+tf
j=tf
SICb
TH+tf
i=tfSICa1
H
TH+tf
j=tf
SICa2
TH+tf
i=tfSICb1
H
TH+tf
j=tf
SICb2
(10)
where tfindicates the sample at which fault is detected, THis
the sample number of SIC signals a half cycle after fault detec-
tion (10ms in 50Hz systems) and Hrepresents the number of
samples in the studied window.
Step 2:The fault classification logics
Based on the indices presented in (7)to(10) and extensive
analysis on values of the proposed indices in different condi-
tions, the methodology for the classification of faults is defined
as follows. First, the value of Cgis monitored to categorize the
JARRAHI ET AL.2249
FIGURE 7 Some typical waveforms for superimposed components in different fault types: (a) AG (b) ABG (c) AB (d) ABC
fault type as grounded or ungrounded based on the following
rule:
Grounded faults Cg>1
Ungrounded faults 0<Cg<1(11)
It is widely accepted that i0is not considerable in grounded
faults [1], so the magnitude of i0can be monitored after
fault occurrence to flag grounded faults. That is why an index
based on this well-known fact is defined. The value of Cgin
ungrounded faults is very small and it is not comparable with
related magnitudes in grounded faults. So, if Cgis more than a
pre-defined value such as 1, the fault will be in the grounded
fault group, otherwise, it is among balance faults. After that, the
fault classification rules are defined as follows:
For grounded faults:
If only Cbc is approximately equal to one and the other
two indices are equal to other values than one, it is an A-G
fault.
If only Cca is approximately equal to one and the other two
indices are equal to other values than one, it is a B-G fault.
If only Cab is approximately equal to one and the other two
indices are equal to other values than one, it is a C-G fault.
2250 JARRAHI ET AL.
If Cab is the minimum value among the indices, it is an A-B-G
fault.
If Cca is the minimum value among the indices, it is an A-C-G
fault.
If Cbc is the minimum value among the indices, it is a B-C-G
fault.
For ungrounded faults:
If only Cab is approximately equal to one and the other two
indices are equal to other values than one, it is an A-B fault.
If only Cca is approximately equal to one and the other two
indices are equal to other values than one, it is an A-C fault.
If only Cbc is approximately equal to one and the other two
indices are equal to other values than one, it is a B-C fault.
If the absolute value of Cab,Cca, and Cbc indices are less than
a threshold (here it is 0.9), it is an A-B-C fault.
Considering that Cab,Cca ,andCbc are coefficients that are
derived from similarity and differences of SIC signals, the con-
figuration and operation mode of the MG will not affect the
algorithm. Also, because the proposed indices select the faulty
phases only if they are equal or opposite to the constant value
(here it is 1), the problems relegated to threshold selection are
resolved. So, the proposed algorithm can be easily implemented
in practical conditions.
Step 3: Load change discrimination
Up to this stage, the faults and faulty phases are identified.
However, some non-fault conditions such as load change may
lead to mal-operation of the method as their impact on mea-
sured signals is similar to the faults [1, 31]. So, to avoid any mal-
function in such conditions and improve the method reliability,
an auxiliary index is considered.
The proposed auxiliary index is found on the behaviours of
SIC signals during load change. These signals have different
characteristics in comparison with the faults. This unique fea-
ture makes it possible to differentiate between the two com-
mon phenomena. To do so, the maximum value of ED of SIC
signals for each phase in the same window of the last step (a
half cycle after fault detection) is considered. Then, the largest
value between them is selected as the proposed criterion which
is formulated as follows. It must be noted that this index is cal-
culated when the values of FCU indices characterize an A-B-C
fault logic.
CL=max
max EDSICa
max EDSICb
max EDSICcin half cycle after fault
(12)
If the value of this index is greater than a predefined thresh-
old, identification of the fault is confirmed. Otherwise, this
change is considered a non-fault situation. The selected thresh-
old for this index is 2.5 p.u. based on Otsu approach. The
flowchart of FCU in detail is depicted in Figure 8.
3.3 Threshold selection approach
As mentioned, the thresholds are determined using a well-
known technique named OTS. This method was originally pro-
posed to separate the pixels of an image. Generally, OTS tries
to intercept a signal into two or more classes and designate an
effective threshold. The implementation steps of OTS in FDI
are as follows:
1. First, the samples related to the non-fault and fault con-
ditions of FDIs are gathered into two different rows of a
matrix. The fault conditions included all possible fault sce-
narios in both operation modes and topologies, faults with
various resistances, faults in different inception angles etc.
Moreover, the non-fault situations contained critical events
such as load change, DG outage, and different levels of
noise.
2. After that, a probability function density (PDF) is assigned
to each row of the developed matrix.
3. In the third step, a normal-function curve is fitted to each
PDF.
4. Finally, the curves related to each PDF are plotted in com-
mon coordinates and the cross-point is chosen as the thresh-
old.
Using this technique, the value of the threshold for FDI is
derived 0.56 1/A2,asshowninFigure9. The procedure of
selecting the threshold for CLis completely similar to what is
stated above which its value is 2.5 p.u.
4SIMULATION RESULTS
In this section, the simulated MG and assessment of the per-
formance of the proposed method in different conditions are
presented. In the end, a sensitivity analysis is also given to illus-
trate the capability of the proposed technique.
4.1 The simulated MG
The studied system is a modified version of IEEE 15-bus net-
work, simulated in MATLAB/Simulink 2018b. The network
contains five IBDGs and 13 loads which operate at 11 kV.
Figure 10 depicts the single line diagram of the system that
can operate in both grid-connected (GC) and IL modes with
radial and mesh configurations using CB1 and CB2 switches
which can be hybrid circuit breakers based on the advantages
stated in [30]. The rated power and voltage of all IBDGs are
250 kVA and 400 V, respectively. Also, they are connected
to the network through 0.4/11 kV transformers. It must be
noted that modelling of IBDGs with detailed specifications
were explained in the last section. The inverters of all IBDG
are modelled as three-level IGBT bridge type, controlled with
PWM approach. Also, in order to remove the unwanted har-
monics produced by the bridge, LC filters are utilized at inverter
JARRAHI ET AL.2251
FIGURE 8 The flowchart of the proposed FCU
FIGURE 9 Otsu thresholding results for FDI
FIGURE 10 The modified IEEE 15-bus system
2252 JARRAHI ET AL.
FIGURE 11 The results of the proposed FDU different operation modes (a) GC mode (b) IL mode
outputs. Furthermore, the upstream network is modelled as a
100 MVA and 60 Hz ideal voltage source in series with an
impedance. The current waveforms are captured at relay loca-
tions with the sampling frequency of 7.68 kHz meaning 128
samples per cycle. More detailed information about the loads,
lines, and other components of the studied MG can be found
in [3].
4.2 The performance assessment
The proposed protection framework is evaluated in the intro-
duced MG by simulating various fault and non-fault scenarios.
To do so, the simulated scenarios and their challenges are firstly
explained, and then some typical results for each case are pre-
sented with a detailed discussion. Based on the results, one can
confirm that the technique can satisfactorily handle MG protec-
tion against different stimulating conditions.
4.2.1 Effects of operation mode
The operation mode can seriously affect the fault behaviour
and eventually, the protection methods. In grid connected (GC)
operation mode, the upstream network dominantly contributes
to the fault that results in higher fault current values and bi-
directionality of the fault. However, this issue is completely dif-
ferent in islanded (IL) mode as the fault current level is lim-
ited. Besides, the presence of IBDGs makes this scenario more
complicated. So, it is vital that the protection scheme is able to
handle both operation modes. To assess this scenario, different
faults are simulated in both operation modes. Figures 11 and 12
and Table 1show some typical results of the proposed FDU
and FCU in this scenario, respectively. The simulated faults in
Figure 11 are different fault types occurred in 50% of C2 with
fault resistance RF=5at fault occurrence time TF=0.5 s
in the radial system. Figure 12 shows the simulated three phase
faults of Figure 11 in one graph for a visual comparison. As
observed in Figure 12,theFDI signal exceeds the threshold
sooner in the GC mode (after 0.65 ms), which means that FDU
senses the faults faster in such conditions. However, it detects
the fault in IL mode after 0.8 ms which is also a very fast detec-
tion. The values of FCU indices presented in Table 1reveal that
the operation mode does not affect the method performance
due to the utilization of waveform similarity concepts. The con-
ditions of the faults were the same as the faults in Figures 11
and 12. For example, the value of indices Cg,Cab,Cbc, and Cca
in IL mode for AG fault which is considered as the worst test
case is 50.35, 0.65, 0.99, and 0.67, respectively. According to the
defined FCU rules, these values correctly categorize the fault
type as an AG fault.
4.2.2 Effects of topology type
The change in configuration of the MG from radial to mesh
and vice versa can impact the current magnitude and direction.
Hence, the protection schemes might be affected due to differ-
ent fault current values and behaviours. To change the topol-
ogy of the simulated MG, the configuration is changed using
CB2. Figure 13 shows the performance of FDU for this sce-
nario in which the MG operation mode is GC and a single-
line to ground fault occurs in 60% of C3 with fault resistance
RF=10 at TF=0.6 s. As presented in this figure, the
proposed FDI identifies the simulated cases in less than 1.8
ms. Moreover, one can find out that the change in the topol-
ogy impacts FDI signals as well as the response time of FDU.
The results for FCU evaluation in this scenario are tabulated in
FIGURE 12 The results of the proposed FDU in different operation
modes for a three-phase fault in line C2
JARRAHI ET AL.2253
TAB LE 1 The results of FCU in different operation modes
Fault type
GC operation mode IL operation mode
CgCab Cbc Cca CLCgCab Cbc Cca CL
AG 137.23 0.78 1 0.76 — 50.35 0.65 0.99 0.67
BG 141.27 0.34 0.37 0.99 — 61.81 0.41 0.49 1
CG 125.66 1 0.56 0.49 44.53 0.99 0.27 0.26 —
ABG 451.27 0.51 0.25 0.58 174.69 0.68 0.12 0.43 —
ACG 384.71 0.21 0.36 0.32 — 126.39 0.61 0.39 0.44 —
BCG 502.13 0.33 0.44 0.46 98.63 0.09 0.28 0.39 —
AB 0.02 1 0.13 0.18 0.01 0.99 0.45 0.39 —
AC 0.08 0.37 0.29 1 0.03 0.28 0.21 1
BC 0.07 0.55 0.99 0.41 — 0.06 0.39 0.99 0.71
ABC 0.004 0.68 0.45 0.39 3.64 0.001 0.59 0.63 0.47 2.78
TAB LE 2 The results of FCU in different topologies
Fault type
Radial topology Mesh topology
CgCab Cbc Cca CLCgCab Cbc Cca CL
AG 201.25 0.32 0.99 0.37 — 181.44 0.30 1 0.24
BG 193.67 0.59 0.61 0.99 — 163.58 0.22 0.28 0.99
CG 145.59 0.99 0.51 0.58 — 176.91 1 0.41 0.43
ABG 286.63 0.42 0.31 0.41 — 121.14 0.41 0.25 0.23 —
ACG 211.13 0.17 0.44 0.29 — 181.23 0.47 0.27 0.36 —
BCG 303.67 0.18 0.29 0.53 — 98.63 0.27 0.39 0.22 —
AB 0.07 0.99 0.26 0.29 0.02 1 0.34 0.29
AC 0.09 0.62 0.59 1 0.04 0.47 0.42 1
BC 0.03 0.49 1 0.51 0.05 0.27 1 0.35
ABC 0.006 0.47 0.62 0.29 4.09 0.008 0.39 0.65 0.58 3.11
FIGURE 13 The results of the proposed FDU in different topologies for
a single-line to ground fault in line C3
Table 2, in which the conditions are the same as the faults in
Figure 13. Based on this table, it is concluded that the proposed
technique successfully classifies the fault type in different
topologies. One important feature of the proposed framework
is that it does not need to change its settings in different situa-
tions. The results of Table 2confirm this statement as the value
of FCU indices precisely select the faulty phases in both topol-
ogy configurations.
4.2.3 Effects of fault resistances
One of the factors that can affect the performance of the
protection scheme is fault resistance. Identification of faults
with high resistances is among the challenges in MG protec-
tion issues [26]. The protection algorithms might not be able to
function properly at a certain value of the fault resistance. So,
this scenario is considered in this paper by simulating the faults
with various resistances. The results of FDU are presented in
Figure 14. In this study, the system is operated in GC mode
with radial configuration in which the faults are single-phase to
2254 JARRAHI ET AL.
FIGURE 14 The results of the proposed FDU in different fault
resistances for a single-line to ground faults in line C4
TAB LE 3 The results of FCU in different fault resistances
Fault type
Fault
resistance
()
FCU indices
CgCab Cbc Cca CL
AG 0.25 364.23 0.47 0.99 0.43 —
25 186.21 0.33 1 0.38 —
250 69.85 0.42 1 0.47 —
ABG 0.25 412.55 0.19 0.38 0.08 —
25 303.19 0.27 0.52 0.17 —
250 194.97 0.22 0.43 0.13 —
AB 0.25 0.07 0.99 0.23 0.41 —
25 0.03 1 0.61 0.37 —
250 0.01 0.99 0.25 0.31 —
ABC 0.25 0.002 0.27 0.55 0.36 4.12
25 0.006 0.44 0.31 0.51 3.87
250 0.008 0.67 0.23 0.33 2.68
ground type occurring in 75% of line C4 at TF=0.7 s. It can be
seen from this figure that the elapsed detection time for faults
with higher resistances is a little bit longer. When a fault with
high resistances occurs, the fault current signal as well as SICs
rise with lower magnitude and higher time than faults with lower
resistances, so the distance between the fault sample and the
corresponding sample in the last cycle within the proposed ED
definition would be larger than a typical fault condition. That
is why; FDI takes more time to rise, as it reaches the threshold
a little bit longer than typical test cases, and consequently, the
fault is identified longer. For instance, the fault with RF=250
is identified after 1.9 ms which is higher than the other test
cases. The results of FCU are presented in Table 3.Asshownin
this table, an increase in the fault resistance does not affect the
algorithm performance.
FIGURE 15 The results of the proposed FDU in different far away faults
for three phase faults in line C4
4.2.4 Effects of fault location
Generally, the lines in MGs can be considered as short ones
from the length perspective, so the proposed method performs
well for far away faults because the current signals (which are
the inputs of the method) will be affected considerably in such
cases. In order to demonstrate the performance of the sug-
gested techniques, some comprehensive simulation studies are
done and the results are given as follows. To do so, the faults are
simulated in different locations (between 85% and 99% away
from the relay location). The results of three phase to ground
far away fault are depicted in Figure 15. In this study, the system
is operated in IL mode with radial configuration in which the
faults are occurring in line C4 at fault inception time TF=0.4 s.
Also, the comprehensive results of fault classification indices are
tabulated in Table 4. From the results, it is obvious that the faults
are successfully detected and classified using the developed FDI
and classification indices.
4.2.5 Effects of fault inception angle
The fault inception time or angle can affect the performance of
fault detection and classification approaches. Faults may occur
at any moment, where they might have different waveforms and
features. To assess the effect of this issue, some different fault
inception angles (FIA) are considered and the results are pre-
sented in Figure 16 and Table 5. The results of Figure 16 are
typical three phase to ground faults for various FIAs occurred
in line C3 when system is in GC mode. Also, the comprehensive
results of fault classification indices are tabulated in Table 5.As
observed, FIA does not affect the efficiency of the proposed
method.
4.2.6 Effects of simultaneous faults
Simultaneous faults are situations that two or more faults occur
at the same time but in different locations. Various kinds of
simulation studies are conducted to evaluate the impact of
JARRAHI ET AL.2255
TAB LE 4 The results of FCU in different fault locations
Fault type
Fault
location
(%)
FCU indices
CgCab Cbc Cca CL
AG 25 297.63 0.51 1 0.49 —
50 196.38 0.42 0.99 0.45 —
75 136.71 0.36 0.99 0.32 —
99 107.93 0.29 1 0.24 —
ABG 25 412.55 0.28 0.31 0.12 —
50 303.19 0.13 0.49 0.07 —
75 194.97 0.31 0.33 0.18 —
99 147.39 0.36 0.56 0.29 —
AB 25 0.06 1 0.24 0.61 —
50 0.02 0.99 0.41 0.28 —
75 0.05 1 0.34 0.29 —
99 0.04 0.99 0.41 0.49 —
ABC 25 0.002 0.37 0.45 0.56 4.12
50 0.006 0.34 0.31 0.66 3.87
75 0.008 0.52 0.28 0.39 2.68
99 0.003 0.38 0.42 0.69 2.61
FIGURE 16 The results of the proposed FDU in different far inception
angles for three phase faults in line C3
simultaneous faults. The results of a typical type of simultane-
ous faults are depicted in Figure 17. For example, an AG fault
in 35% and BG fault in 80% of C3 with fault resistance RF=
10 at TF=0.6 s are depicted in this figure, and it is observed
that the proposed FDI accurately reacts to this faulty scenario
as it reaches the threshold after 1.7 ms of the fault occurrence.
The results of other cases also indicate that the method can
function correctly in such a scenario. Table 6shows the per-
formance of FCU in this test case for more than 200 various
conditions including two LG fault (LG+LG), one LG fault
and another LL fault (LG+LL) and one LG fault and another
LLG fault (LG+LLG). This table also confirms that the method
recognizes the faulty phase/phases with a high degree of
accuracy.
TAB LE 5 The results of FCU in different fault inception angles
Fault type
Fault inception
angle (degree)
FCU indices
CgCab Cbc Cca CL
AG 0 425.27 0.57 0.99 0.33 —
45 298.28 0.24 0.99 0.33 —
90 246.38 0.32 1 0.67 —
ABG 0 446.95 0.17 0.28 0.09 —
45 389.25 0.36 0.27 0.1 —
90 315.34 0.31 0.29 0.21 —
AB 0 0.08 0.99 0.41 0.21 —
45 0.07 1 0.47 0.36 —
90 0.02 1 0.28 0.36 —
ABC 0 0.004 0.24 0.56 0.37 3.18
45 0.005 0.35 0.44 0.55 3.55
90 0.006 0.69 0.24 0.39 3.96
FIGURE 17 The results of the proposed FDU in a typical simultaneous
fault condition
TAB LE 6 Performance of proposed method is various simultaneous fault
conditions
System
operation
mode
Top o l og y
type
Simultaneous fault
LG +LG LG +LL LG +LLG
GC Radial 99.27% 99.39% 99.76%
Mesh 99.29% 99.61% 99.37%
IL Radial 99.76% 99.39% 99.37%
Mesh 99.61% 99.39% 99.76%
4.2.7 Effects of composite faults
The phenomenon is entitled composite fault when a short-
circuit fault occurs in MG as well as a non-fault event such as
DG outage at the same time. This type of event must be identi-
fied by the protective relay so that the other elements of MG did
not get damaged. To simulate this scenario, it is assumed that an
LG fault occurred in the middle of line C3 and simultaneously,
2256 JARRAHI ET AL.
FIGURE 18 The results of the proposed FDU in a typical composite
fault event
TAB LE 7 Performance of proposed method is various composite fault
conditions
System
operation
mode
Top o l og y
type
Composite fault
LG +DG
outage
LLG +
DG
outage
LL +DG
outage
LLLG +
DG
outage
GC Radial 99.61% 99.39% 99.76% 99.61%
Mesh 99.76% 99.61% 99.61% 99.39%
IL Radial 99.29% 99.39% 99.29% 99.29%
Mesh 99.29% 99.61% 99.76% 99.61%
IBDG-4 is disconnected at TF=0.7 s. The results of FDU are
presented in Figure 18. As presented in this figure, the proposed
FDI identifies the simulated case in less than 1.9 ms. The results
for FCU evaluation in this scenario are tabulated for 200 test
cases in Table 7, in which the conditions are the same as the
faults in Figure 18. Based on this table, it is concluded that the
technique successfully classifies the fault type in the occurrence
of composite faults.
4.2.8 Effects of conditions when fault currents
are less than normal current
There are some possible circumstances that the fault current
becomes less than the normal current in MGs. This issue is
deliberated in the development of the proposed method as the
absolute value of SIC of current signals have utilized in both
FDU and FCU to resolve the associated problems. The increas-
ing or decreasing change of the current waveforms in faulty
conditions eventually results in the positive/increasing change
in SICs and then proposed FDI. Also, the input of the FCU
is SICs with the same behaviour. So generally, such mentioned
conditions would not affect the performance of the suggested
protection scheme. In order to show the performance in such
scenarios, some simulation studies are done and a typical FDI
results is depicted in the Figure 19. Based on the given results,
FIGURE 19 The results of the proposed FDU in conditions when fault
current is less than normal
TAB LE 8 The perfor mance of FCU in different conditions with noise
System
operation
mode
Top o l og y
type
Noise SNR (dB)
40 30 20
GC Radial 99.81% 99.43% 99.24%
Mesh 99.80% 99.61% 99.03%
IL Radial 99.61% 99.41% 99.22%
Mesh 99.62% 99.43% 99.23%
it is concluded that the method is highly capable to handle this
type of challenging test case, as well.
4.2.9 Effects of noise
Impacts of the noises on the performance cannot be neglected
in practical applications. Therefore, its effect should be deliber-
ated in the evaluation process. To do so, the proposed scheme
is assessed in a noisy environment by adding Gaussian noises
with various signal-to-noise ratios (SNRs) to the measured cur-
rent signals [32]. Some typical noisy FDIs waveforms (including
20, 30, and 40 dB SNR) are depicted in Figure 20 for a three-
phase fault in line C10 of the system operating in IL mode with
mesh configuration. As observed, the proposed FDU preserves
its high performance in such conditions. In other words, it is
robust against possible noises. Furthermore, the accuracy rate of
FCU in this scenario is tabulated in Table 8which approves the
high capability of the proposed technique to function in noisy
conditions. The presented values in this table are calculated by
considering more than 2000 test cases with all types of fault in
various conditions. For example, the value 99.22% for IL mode-
radial type scenario indicates that FCU misclassifies only 4 cases
out of 517 test cases due to the presence of 20 dB noises.
4.2.10 Effects of load change
MGs might be subjected to load change situations. Such
conditions must be studied in the performance assessment
JARRAHI ET AL.2257
FIGURE 20 The results of the proposed FDU in different noises for a single-line to ground fault in line C3
FIGURE 21 The results of the proposed FDU in different load change conditions
procedure as their probability of occurrence is very high in
comparison with fault incidence. Although signatures of the
load change conditions are similar to the fault conditions,
the protection frameworks must avoid tripping in such condi-
tions and consider them as normal operations. To simulate this
condition, different possible loading scenarios are considered.
Figure 21 depicts the results for 30%, 40% and 50% incre-
mental variation in the connected loads that is somehow the
worst test cases in this regard. As shown in this figure, FDI
exceeds the threshold for the 50% load change variation test
case and flags it as the fault. However, the proposed auxiliary
index CL=1.98 p.u. is lower than the threshold. So, FCU
recognizes them as normal condition. This issue means that
the proposed technique is highly immune to the load change
conditions.
4.2.11 Effects of capacitor banks switching
Capacitor banks are utilized for various purposes in MGs. How-
ever, they might not be always connected to the system and
due to the circumstances, they switched into the system. The
behaviour of such scenarios might be considered as faults in
the view of digital relays. The proposed protection scheme dif-
ferentiates the capacitor banks connection using the suggested
FDI along with fault classification indices and load change dis-
crimination index. In order to assess this condition, some pos-
sible scenarios are studied with details, which one typical worst
FIGURE 22 The results of the proposed FDU in capacitor bank
switching condition
possible test case is presented in the following as an example.
Figure 22 depicts the results for worst possible capacitor bank
switching event in the simulated MG, in which the proposed
method accurately recognizes it as a non-fault cases. Based on
the given results, one can confirm that the suggested methodol-
ogy preserves its high performance in such scenarios.
4.2.12 Effects of power swing
In this scenario, the effect of power swing phenomena on
the performance of the proposed scheme is studied. Power
2258 JARRAHI ET AL.
FIGURE 23 The results of the proposed FDU in power swing condition
swings in MGs may occur because of faults, line switching,
DG outage, and disconnection of large loads. False opera-
tion of relays under the power swing condition is a common
challenge. Therefore, FDU and FCU should discriminate the
power swing phenomenon from the fault condition. In the
simulated system, assume that the voltage angle of IBDG-3
varies up to 45with 1 Hz frequency at time 2 s for emu-
lation of power swing. Since the dynamic of this variation is
low, the proposed method does not have the wrong opera-
tion under this condition, as shown in Figure 23.TheFDU
does not recognize this condition as a fault, so the proposed
FCU, as well as the proposed related indices, would not be
activated.
4.2.13 Effects of system frequency fluctuations
The frequency of MGs might fluctuate due to various reasons.
For example, when a large deviation between the supply and
the demand is happened due to a fluctuation in wind power, an
unexpected frequency fluctuation might occur in MGs. There
are two perspectives in assessing the impact of such scenarios
on the proposed protection scheme. The first viewpoint is that
the frequency of MG is 49.5 Hz, for example from the first
moment of the simulation and then a fault occurred in the sys-
tem. The second standpoint is that the frequency is turned to
49.5 Hz at a specific moment. Both of these cases are stud-
ied and the results are brought as follows. Based on the given
results, it is observed that the method perseveres its accuracy
against frequency fluctuations. The results of the first scenario
for a three phase to ground fault are presented in Figure 24.
The difference in this scenario with conditions when frequency
if system is 50 Hz is that the magnitude of FDI would not be
zero during normal condition, however it is negligible in com-
parison with the threshold. It is seen that the method success-
fully operates in this test case and does not get affected by
such a challenging situation. Furthermore, the results of sud-
den frequency variation impact on the developed technique are
depicted in Figure 25. As observed, the FDU does not iden-
tify this change as a fault and as a result, the FCU does not get
activated.
FIGURE 24 The results of the proposed FDU in off-nominal frequency
condition
FIGURE 25 The results of the proposed FDU in frequency deviation
condition
TAB LE 9 The perfor mance of proposed method in different sampling
frequencies
fs (kHz)
Performance criteria
TD-mean (ms) PFDU (%) PFCU (%)
1.92 4.21 82.4p 83.18
3.84 2.86 92.36 91.87
7.68 1.49 99.86 99.67
15.36 1.35 99.86 99.67
30.72 1.26 99.86 99.81
4.2.14 Effects of sampling frequency variations
The sampling frequency (fs) can affect the performance of the
proposed method. Due to the limitations in the sampling rate
of measurement devices, the chosen fsmust be in the range of
reasonable and common values of these devices to facilitate the
implementation of protective approaches. To assess the impact
of fs, different common sampling frequencies (1.92, 3.84, 7.68,
15.36 and 30.72 kHz) are considered and the method is assessed
in each frequency. Results of this assessment are presented in
Table 9. In this table, the accuracy and speed of the proposed
FDU and FCU in each frequency are given. The mean elapsed
JARRAHI ET AL.2259
FIGURE 26 The results of the proposed FDU in all IBDGs outage
condition
time for fault detection (TD-mean) and the performance percent-
age (PFDU) are the indicators of FDU. The performance per-
centage (PFCU) is only considered for FCU because the con-
sidered window of this unit is constant. As this table shows,
the accuracy and speed of the method were not desirable at
fs=1.92 and 3.84 kHz. But the results obtained by fs=7.68
kHz show that the method has a high degree of accuracy and
speed. Although increasing fsto 15.36 and 30.72 kHz results in
enhancement of the speed, but this increment does not change
the accuracy so much.
4.2.15 Effects of IBDG outage
All of IBDGs in MG might be prone to connect-
ing/disconnecting from the network. This issue can be a chal-
lenge for FDU because it can severely cause disturbances in
the current signals. In order to assess the performance of the
method in this test case, all IBDGs are disconnected from MG.
The relevant results are depicted in Figure 26. In this figure,
it can be observed that this phenomenon cannot jeopardize
the proposed technique functionality as FDI did not reach the
threshold in the worst case that is the connection/outage of
IBDGs.
4.2.16 Effects of change in weather conditions
The operation of wind turbines and PV-based DGs are char-
acterized by intermittency due to the uncertainty in wind speed
and solar irradiance. Under weather intermittency, the design
of a reliable protection scheme becomes challenging. The com-
plexity of the protection task arises due to the necessity of adapt-
ing the algorithm for varying dynamics of the sources. The vari-
ation in power output leads to variable voltage–current profiles
for similar fault scenarios. That is why; the protection schemes
must be assessed in such scenarios. To do this, all of the simu-
lated IBDGs in the modified IEEE 15-bus system are replaced
with PV and wind-based DGs with the specifications gathered
from [33]. In this condition, IBDGs 1, 2 and 3 are replaced
with PV-based DGs as well as IBDGs 4 and 5 are changed to
TAB LE 10 The performance of proposed method in different weather
conditions
Type of weather related change
System
operation
mode
Top o l og y
type
2DG
change
3DG
change
4DG
change
5DG
change
GC Radial 99.61% 99.39% 99.76% 99.61%
Mesh 99.76% 99.61% 99.61% 99.39%
IL Radial 99.29% 99.39% 99.29% 99.29%
Mesh 99.29% 99.61% 99.76% 99.61%
wind-based DGs. To evaluate the impact of weather intermit-
tency, the following related values are considered for the simu-
lated DGs [33]:
Wind speed (ws) range: 7–21 (m/s)
Solar irradiance range (G) range: 600–1000 (W/m2)
In this scenario, more than 225 different test cases are simu-
lated and the method is assessed in each of them. It is tried to
consider various mixed fault cases with numerous magnitudes
of wind speed and solar irradiance. For example in one study
having around 40 test cases, only the solar irradiance of DG1
and wind speed of DG4 are altered and the method is evaluated
for various kinds of fault types with different conditions such
as fault location, fault inception angle, fault resistance etc. The
results of the method performance are tabulated in Table 10.In
this table, the term ‘‘2 DG change’’ means that two of five DGs
can have different weather conditions, and the corresponding
value in the other three DGs is kept constant. Based on the
given results, it is obvious that the method preserves its high
degree of accuracy and speed in varying weather conditions.
4.2.17 Sensitivity analysis
In order to find out the robustness of the proposed approach, a
comprehensive sensitivity analysis is done by considering dif-
ferent conditions. The results for this investigation are tabu-
lated in Table 11, in which the impacts of fault type, fault resis-
tance, fault inception time, topology configuration, load change,
and IBDGs generation contributions are taken into account for
both GC and IL operation modes. It is worth mentioning that
all the simulation test cases (more than 3500 different scenarios)
are considered in this analysis. In Table 11,FT,RF,FIA,TT,LP,
and SDG represent the fault type, fault resistance, fault inception
angle, topology type, loading, and DGs contribution percentage,
respectively. The mean elapsed time for fault detection (TD-mean)
and the performance percentage (PFDU ) are the sensitivity anal-
ysis indicators of FDU. The performance percentage (PFCU)is
only considered for FCU sensitivity analysis because the consid-
ered window of this unit is constant. As observed in this table,
TD-mean increases in case of LG fault types, RFmore than 150 ,
FIA is in the range of 45to 90,TTin the radial topology, LPis
in the range of 110% to 120%, and SDG is in the range of 25% to
2260 JARRAHI ET AL.
TAB LE 11 The sensitivity analysis of the proposed protection framework
Scenario Condition
Sensitivity criteria
TD-mean
(ms)
PFDU
(%)
PFCU
(%)
Fault type LG 1.94 99.38 99.27
LLG 1.34 99.42 99.38
LL 1.71 99.05 99.11
LLL 1.08 99.89 99.89
Fault
resistance
0<RF<75 1.11 99.85 99.85
75 <RF<150 1.46 99.34 99.34
150 <RF<225 1.83 99.05 99.05
225 <RF<300 2.14 98.55 98.55
Fault
incep-
tion
angle
0<FIA <451.25 99.42 99.42
45<FIA <901.78 99.34 99.34
90<FIA <1351.55 99.85 99.85
135<FIA <
180
1.49 99.89 99.89
Topology
type
TT=radial 1.87 99.34 99.34
TT=mesh 1.62 99.55 99.55
Load
change
80% <LP<90% 1.22 99.42 99.42
90% <LP<100% 1.31 99.42 99.42
100% <LP<
110%
1.49 99.34 99.34
110% <LP<
120%
1.64 99.05 99.05
IBDG
penetration
25% <SDG <50% 1.88 98.89 98.89
50% <SDG <75% 1.57 99.05 99.05
75% <SDG <
100%
1.33 99.55 99.55
50%. Also, PFDU and PFCU of the method decrease in the case
of LL fault types, RFmore than 150 ,FIA is in the range of
90to 135,TTin the radial topology, LPin the range of 110%
to 120% and SDG the range of 25% to 50%. It is confirmed that
the proposed method has a sufficiently good performance for
all of the fault conditions.
5EXPERIMENTAL VERIFICATIONS
In order to verify the performance of the proposed technique
in practice, a simple system emulating an MG is implemented
in a low-scale laboratory test bench as shown in Figure 27.The
test bench consists of two pairs of three step-down single-phase
transformers that each pair is connected to the 220 V and 50
Hz upstream network. So, the faults are fed from two sides
to emulate their bidirectional behaviour in real MGs. The lines
are built using rheostats with some capacitors and inductors to
emulate the pi-model of the lines. A fault switch with different
high-power resistances is connected to multiple sections of the
lines to emulate the incidence of different fault types in various
locations. The current waveforms are captured using three
clamp-on current transformers (CTs) connected to a data log-
ger with a sampling rate of 7.8 kHz (156 samples per cycle).
More than 80 test cases are considered to gather the experimen-
tal dataset for assessment of the method in real situations. It
should be noted that the logged current waveforms are trans-
mitted to a personal computer in which the suggested technique
is coded in MATLAB software.
The captured current waveform of the faulted phase and its
FDI for a typical single line to ground fault are presented in
Figure 28. As shown in this figure, the proposed algorithm rec-
ognizes the fault precisely after only 0.7 ms. The indices Cab,Cca,
Cbc, and Cgfor this fault are 0.99, 0.27, 0.56 and 156, respectively
that correctly flag the fault as an AG fault. From these typical
results, it is obvious that the fault is precisely recognized and
classified by the proposed algorithm. The results for the other
cases approve that the method has a high degree of performance
in practical applications.
6COMPARATIVE EVALUATION
In order to validate the effectiveness of the developed pro-
tection framework, a comparison is made between the pro-
posed method and some recent techniques. To have a fair
comparison, some similar protection approaches including the
one-end current signal-based algorithms are selected. Also, it
should be noted that all of the algorithms are assessed using
the same simulation and experimental data. Moreover, their set-
tings are changed to have the best performance. The compared
methods are (A) transient-monitoring based algorithm [12], (B)
mathematical-morphology based scheme [14], (C) harmonic-
component based approach [34] and (D) auto-cosine similar-
ity based technique [35]. In order to explore the proficiency of
the compared methods, some features are considered. For the
sake of having quantitative comparison, the following factors
are deliberated [36]:
Dependability: This factor indicates the scheme reliability in
the identification of the faults numerically:
Dep%=
Total number of detected fault cases
Total number of fault cases (13)
Security: This index represents the false alarm of the method.
In other words, it specifies the number of non-fault cases that
is mis-identified as a fault:
Sec%=
Total number of detected non fault cases
Total number of non fault cases (14)
Accuracy: This feature determines the overall performance of
the method in case of correct decision-making of both fault
and non-fault scenarios:
Acc%= Total number of correclty detected cases
Total number of cases (fault and non fault)
(15)
JARRAHI ET AL.2261
FIGURE 27 The experimental laboratory test bench
FIGURE 28 The captured current waveform of
the faulted phase and its related FDI
FIGURE 29 Quantitative comparison between the proposed FDU and the other recent approaches
In these factors, the number of fault cases, non-fault cases,
and all test cases in both simulation and experimental datasets
are 3169, 425 and 3594, respectively. These numerical factors
are calculated for each method and presented in Figures 29
and 30 for FDU and FCU as bar charts to have a clear visual
comparative presentation. As observed in Figures 29 and 30,the
proposed approach significantly outperforms the algorithms B
and Din all three statistical items. The algorithm Chas shown
better results in the indices compared with the other methods,
but its performance is not desirable for the MG protection.
Algorithm Aexhibits comparable results in dependability fac-
tor for fault detection, but its performance in the security and
2262 JARRAHI ET AL.
FIGURE 30 Quantitative comparison between the proposed FCU and the other recent approaches
accuracy indices is not suitable enough as it fails to operate
correctly in various non-fault scenarios. This quantitative com-
parison shows the high capability of the proposed technique in
the protection of IBDG dominated MGs.
A comprehensive comparison is also done between the men-
tioned methods and the proposed algorithm. The results of
this study are tabulated in Table 12. In this table, eight factors
are considered to compare the methods including speed, sam-
pling rate, complexity, threshold modifications, noise immunity,
non-fault condition discrimination capability, handling IBDG
impacts, and validation with experimental data. The speed fac-
tor shows the mean delay of each method in milliseconds and
the needed sampling rate factor yields the sampling frequency
in which, the method works properly. The ‘Big O’ concept is
utilized to quantize the complexity of the methods. The ‘Big O’
notation is a metric for algorithm scalability. The entire point
of ‘Big O’ could compare how efficiently one algorithm solves
problems compared to another. This approach can be described
TAB LE 12 Comprehensive comparative study
Methods
Criteria A [12]B[14]C[34]D[35]
Proposed
method
Speed (ms) 11.25 20.67 18.44 16.34 10
Required sampling rate
(kHz)
5 7050107.6
Complexity O(n2)O(n
1.62)O(n
2.36)O(n
2)O(n)
Threshold
modification needed
No Yes No Yes No
Noise immunity No Yes No Yes Yes
Non-fault condition
discrimination
No No Yes No Ye s
Handling IBDG
impacts
YesNoNoYesYes
Validation with
experimental data
No No Yes No Ye s
simply as “how quickly the algorithm grows relative to the input,
as the input gets larger”. The ‘Yes’ and ‘No’ terms are utilized
to justify the specification of each index in each indicator. The
superiority of the proposed method in comparison with the
studied techniques is clear from Table 12. The speed of the pro-
posed method which shows the protection delay time is consid-
ered higher than other methods due to flagging the fault and its
type after only half a cycle, however, methods B, C and D are not
fast enough for MG protection application. The needed sam-
pling frequency of the proposed technique is around 7.6 kHz,
which is considered a low value for protective relaying of MGs.
However, methods Band Cneed current signals sampled at 50
kHz or more to function properly which are not desirable. The
proposed method utilizes only simple mathematical procedures,
so its ‘Big O’ related complexity is less than other algorithms.
However, methods Band Cimpose high computation proce-
dures on the relays because of using frequency concepts and
complex mathematical operations. No need for modifying the
threshold settings is another feature of the proposed approach.
As demonstrated in the results section, the performance of the
developed approach won’t change in the presence of noises.
Conversely, methods Aand Care not reliable in noisy situations.
The ability to discriminate the non-fault conditions makes the
suggested scheme superior to other studied algorithms. More-
over, the high performance of the method in handling IBDG
impacts confirms its dominancy. At last, assessment of the pro-
posed approach with experimental data gathered from the lab-
oratory setup highly supports that it can function precisely in
practice.
7 CONCLUSIONS AND FUTURE
WORKS
Protection of IBDG based MGs is a challenging task due to the
unique behaviour of IBDGs during faults. In this paper, a sim-
ple and efficient methodology is suggested for FDU and FCU
of IBDG enriched MGs. The concept of waveform similarity is
JARRAHI ET AL.2263
utilized in both stages of proposed method. To do so, the devel-
oped technique combines the utilization of SICs, ED, and PCC
to discriminate the fault and faulty phases. Unlike the existing
two-terminal methods, the proposed scheme operates based on
the local measurements, hence avoiding communication system
failures. Furthermore, it demonstrates a very fast and reliable
performance in challenging conditions which makes it suitable
for practical applications. The efficacy of the method is evalu-
ated using plenty of simulations and experiments in different
non-fault and fault conditions. Moreover, its performance is
compared with some other similar methods using various quan-
titative and qualitative factors. The results show that the pro-
posed method has desirable speed and accuracy compared with
the other algorithms.
Future work can be focused on implementation of the pro-
posed method in large-scale IBDG dominated MGs. Deriving
the mathematical modelling equations of IBDGs and studying
their control system behaviour can be also done in future. Fur-
thermore, assessing the behaviour of non-linear loads on the
method performance is another interesting line of research.
CONFLICT OF INTEREST
The authors declare that they have no known competing finan-
cial interests or personal relationships that could have appeared
to influence the work reported in this paper.
FUNDING INFORMATION
The authors received no financial support for the research,
authorship, and/or publication of this paper
DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available
from the corresponding author, Haiar Samet, upon reasonable
request.
REFERENCES
1. Yadav, R., Kumar Pradhan, A., Kamwa, I.: Real-time multiple event detec-
tion and classification in power system using signal energy transformations.
IEEE Trans. Ind. Inf. 15(3), 1521–1531 (2018)
2. Samet, H., Ghanbari, T., Ashtiani, H.J., Jarrahi, M.A.: Evaluation of direc-
tional relay algorithms in presence of wind turbines and fault current lim-
iters. Int. Trans. Electr. Energy Syst. 29(3), e2772 (2019)
3. Mahamedi, B., Fletcher, J.E.: Trends in the protection of inverter-
based microgrids. IET Gener. Transm. Distrib. 13(20), 4511–4522
(2019)
4. IEEE Application Guide for IEEE Std 1547 (TM). IEEE Standard
for Interconnecting Distributed Resources with Electric Power Sys-
tems. IEEE Std 1547.2-2008. 1–217 (2009), https://doi.org/10.1109/
IEEESTD.2008.4816078
5. Reno, M.J., Brahma, S., Bidram, A., Ropp, M.E.: Influence of inverter-
based resources on microgrid protection: Part 1: Microgrids in radial dis-
tribution systems. IEEE Power Energy Mag. 19(3), 36–46 (2021)
6. Ropp, M.E., Reno, M.J.: Influence of inverter-based resources on micro-
grid protection: Part 2: Secondary networks and microgrid protection.
IEEE Power Energy Mag. 19(3), 47–57 (2021)
7. Jarrahi, M.A., Samet, H., Ghanbari, T.: Novel change detection and fault
classification scheme for ac microgrids. IEEE Syst. J. 14(3), 3987–3998
(2020)
8. Ranjbar, S., Farsa, A.R., Jamali, S.: Voltage-based protection of micro-
grids using decision tree algorithms. Int. Trans. Electr. Energy Syst. 30(4),
e12274 (2020)
9. Lahiji, H., Badrkhani Ajaei, F., Boudreau, R.E.: Non-pilot protection
of the inverter-dominated microgrid. IEEE Access 7, 142190–142202
(2019)
10. Hooshyar, A., El-Saadany, E.F., Sanaye-Pasand, M.: Fault type classifica-
tion in microgrids including photovoltaic DGs. IEEE Trans. Smart Grid
7(5), 2218–2229 (2015)
11. James, J.Q., Hou, Y., Lam, A.Y.S., Li, V.O.K.: Intelligent fault detection
scheme for microgrids with wavelet-based deep neural networks. IEEE
Trans. Smart Grid 10(2), 1694–1703 (2017)
12. Sadeghkhani, I., Golshan, M.E.H., Mehrizi-Sani, A., Guerrero, J.M.,
Ketabi, A.: Transient monitoring function–based fault detection for
inverter-interfaced microgrids. IEEE Trans. Smart Grid 9(3), 2097–2107
(2016)
13. Allahdadi, K., Sadeghkhani, I., Fani, B.: Protection of converter-interfaced
microgrids using modified short-time correlation transform. IEEE Syst. J.
14(4), 5172–5175 (2020)
14. Gush, T., Basit, S. Bukhari, A., Haider, R., Admasie, S., Oh, Y.-S., Cho, G.-
J., Kim, C.-H.: Fault detection and location in a microgrid using mathemat-
ical morphology and recursive least square methods. Int. J. Electr. Power
Energy Syst. 102, 324–331 (2018)
15. Li, X., Dy´
sko, A., Burt, G.M.: Traveling wave-based protection scheme
for inverter-dominated microgrid using mathematical morphology. IEEE
Trans. Smart Grid 5(5), 2211–2218 (2014)
16. Manditereza, P.T., Bansal, R.C.: Protection of microgrids using voltage-
based power differential and sensitivity analysis. Int. J. Electr. Power
Energy Syst. 118, 105756 (2020)
17. Chaitanya, B.K., Yadav, A., Pazoki, M.: An improved differential protec-
tion scheme for micro-grid using time-frequency transform. Int. J. Electr.
Power Energy Syst. 111, 132–143 (2019)
18. Soleimanisardoo, A., Kazemi Karegar, H., Zeineldin, H.H.: Differential
frequency protection scheme based on off-nominal frequency injections
for inverter-based islanded microgrids. IEEE Trans. Smart Grid 10(2),
2107–2114 (2018)
19. Pinto, J.O.C.P., Moreto, M.: Protection strategy for fault detection in
inverter-dominated low voltage AC microgrid. Electr. Power Syst. Res.190,
106572 (2021)
20. Marín-Quintero, J., Orozco-Henao, C., Percybrooks, W.S., Vélez, J.C.,
Montoya, O.D., Gil-González, W.: Toward an adaptive protection scheme
in active distribution networks: Intelligent approach fault detector. Appl.
Soft Comput. 98, 106839 (2021)
21. Anand, A., Affijulla, S.: Ensemble empirical mode decomposition-based
differential protection scheme for islanded and grid-tied AC microgrid.
IET Gener. Transm. Distrib. 14(26), 6674–6681 (2021)
22. Baeckeland, N., Herteleer, B., Kleemann, M.: Modelling fault behaviour of
power electronic converters. Int. J. Electr. Power Energy Syst. 123, 106230
(2020)
23. Carrasco, E.M., Moreno, M.P.C., Martínez, M.T.V., Vicente, S.B.: Improved
faulted phase selection algorithm for distance protection under high pene-
tration of renewable energies. Energies 13(3), 558 (2020)
24. Keller, J., Kroposki, B.: Understanding fault characteristics of inverter-based dis-
tributed energy resources. No. NREL/TP-550-46698. National Renewable
Energy Lab.(NREL), Golden, CO (2010)
25. Haddadi, A., Farantatos, E., Kocar, I., Karaagac, U.: Impact of inverter
based resources on system protection. Energies 14(4), 1050 (2021)
26. Hooshyar, A., Iravani, R.: Microgrid protection. Proc. IEEE 105(7), 1332–
1353 (2017)
27. Dong, X., Peng, Q., Wu, H., Chang, Z., Yue, Y., Zeng, Y.: New principle for
busbar protection based on the Euclidean distance algorithm. PLoS One
14(7), e0219320 (2019)
28. Otsu, N.: A threshold selection method from gray-level histograms. IEEE
Trans. Syst. Man Cybern. 9(1), 62–66 (1979)
29. Jafari, A., Faiz, J., Jarrahi, M.A.: A simple and efficient current-based
method for interturn fault detection in BLDC motors. IEEE Trans. Ind.
Inf. 17(4), 2707–2715 (2020)
30. Zhang, Y., Li, Y., Song, J., Chen, X., Lu, Y., Wang, W.: Pearson correlation
coefficient of current derivatives based pilot protection scheme for long-
distance LCC-HVDC transmission lines. Int. J. Electr. Power Energy Syst.
116, 105526 (2020)
2264 JARRAHI ET AL.
31. Jarrahi, M.A., Samet, H., Sahebi, A.: An EMD based fault type identifica-
tion scheme in transmission line. In: 2016 24th Iranian Conference on Electrical
Engineering (ICEE). Shiraz, Iran, pp. 422–427 (2016)
32. Jarrahi, M.A., Samet, H.: Modal current and cumulative sum based fault
detection approach in transmission lines. Int. J. Emerging Electr. Power
Syst. 19(6), 20180037 (2018). https://doi.org/10.1515/ijeeps-2018- 0037
33. Gururani, A., Mohanty, S.R., Mohanta, J.C.: Microgrid protection using
Hilbert–Huang transform based-differential scheme. IET Gener. Transm.
Distrib. 10(15), 3707–3716 (2016)
34. Chen, Z., Pei, X., Yang, M., Peng, L., Shi, P.: A novel protection scheme
for inverter-interfaced microgrid (IIM) operated in islanded mode. IEEE
Trans. Power Electron. 33(9), 7684–7697 (2017)
35. Mahfouz, M.M.A.: A protection scheme for multi-distributed smart micro-
grid based on auto-cosine similarity of feeders current patterns. Electr.
Power Syst. Res. 186, 106405 (2020)
36. Kar, S., Samantaray, S.R., Dadash Zadeh, M.: Data-mining model based
intelligent differential microgrid protection scheme. IEEE Syst. J. 11(2),
1161–1169 (2015)
How to cite this article: Jarrahi, M.A., Samet, H.,
Ghanbari, T.: Protection framework for microgrids with
inverter-based DGs: A superimposed component and
waveform similarity-based fault detection and
classification scheme. IET Gener. Transm. Distrib. 16,
2242–2264 (2022). https://doi.org/10.1049/gtd2.12438
... In this paper, the main motivation is to detect the faults on the DC side of DCMG. The fault detection in DCMGs is extremely challenging due to various reasons and a fast and accurate scheme is needed to guarantee a fault-resilient DCMG [4]. T. Ghanbari is with the School of Advanced Technologies, Shiraz University, Shiraz, Iran. ...
Article
This paper develops a transient monitoring function (TMF) based fault detection method for DC microgrids (DCMGs). In this method, the measured currents from one-end of the poles are firstly transformed into modal components. Then, the modal current is estimated using TMF concept. Afterward, the residual signal is defined as the absolute value of the difference between the modal and estimated currents. This signal has negligible fluctuations during the normal operation, while it experiences significant variations after fault occurrence. This feature is selected to reflect the fault incidence. To do so, the teager-kaiser energy operator (TKEO) is applied to the residual signal to amplify the changes, followed by a fault detection index (FDI). Comparing FDI with a threshold can reveal the fault. Furthermore, FDI can discriminate the faulty conditions in different operation modes of DCMG with a wide variation of fault conditions. To assess the developed scheme performance, a DCMG with different kinds of sources and loads is simulated in MATLAB/Simulink. The results confirm the quickness and the high accuracy of the proposed technique. Moreover, the method is validated in an experimental laboratory small-scale test bench. Finally, to demonstrate the superiority of the method, a comparative study with recent methods is presented.
Article
Full-text available
This study presents a differential protection scheme based on ensemble empirical mode decomposition (EEMD) for AC microgrid. The fault current level is significantly lower during the islanded operation of a microgrid, which leads to the malfunction of the traditional over‐current protection scheme. The proposed differential protection scheme uses the spectral differential energy of the first intrinsic mode function extracted from the decomposition of the current signal using EEMD for effective fault detection in the AC microgrid. The proposed EEMD‐based differential protection scheme is validated on 10 bus and modified IEEE 34‐bus AC microgrid test systems during various shunt faults. Moreover, the performance of the proposed EEMD‐based differential protection scheme is evaluated under high fault impedance scenarios. The simulation results reveal that the proposed differential protection scheme can effectively detect the faulty line in an AC microgrid during seamless islanded and grid‐tied operations.
Article
Full-text available
Inverter-based resources (IBRs) exhibit different short-circuit characteristics compared to traditional synchronous generators (SGs). Hence, increased uptake of IBRs in the power system is expected to impact the performance of traditional protective relay schemes-set under the assumption of a SG-dominated power system. Protection engineers need to study these challenges and develop remedial solutions ensuring the effectiveness of system protection under higher levels of IBRs. To address this need, this paper studies the impact of IBRs on a variety of protective relay schemes including line distance protection, memory-polarized zero sequence directional protective relay element , negative sequence quantities-based protection, line current differential protection, phase comparison protection, rate-of-change-of-frequency, and power swing detection. For each protection function, potential misoperation scenarios are identified, and recommendations are provided to address the misoperation issue. The objective is to provide an improved understanding of the way IBRs may negatively impact the performance of traditional protection schemes as a first step towards developing future remedial solutions ensuring effective protection under high share of IBRs.
Article
Microgrids are being deployed at a rising rate, primarily as a means of increasing power system resilience. Commonly, a microgrid today includes at least some inverter-based resources (IBRs), and many microgrids have modes or conditions under which they are entirely energized by IBRs. Most microgrids today are deployed on radial distribution circuits, but it is conceivable that they could also be considered for deployment on secondary network systems.
Article
Secondary networks are deployed when exceptionally high-reliability electric service is required for specific loads. Secondary network protection makes extensive use of the fact that the available fault currents in such systems are typically very high. Also, most faults in secondary networks must be isolated from both sides, which requires some special considerations. Today, microgrids are also being deployed as a means of increasing power system resilience on radial circuits. Many microgrids are energized exclusively by inverter-based resources (IBRs), either all the time or under certain operational conditions. The protection of IBR-sourced microgrids is entirely different from that of secondary networks. IBRs are often distributed throughout a system, their fault currents are generally low, and the current flow is not unidirectional. Thus, if it were ever contemplated to deploy a microgrid on a secondary network, significant system protection challenges would have to be overcome.
Article
Conventional protection schemes have proven insufficient for the protection of Active Distribution Networks (ADN). Novel protection schemes with an adaptive approach should be developed to guarantee the protection of ADN under all their operating conditions. This paper proposes an ADN adaptive protection methodology, which is based on an intelligent approach fault detector over locally available measurements. This approach uses Machine Learning (ML) based techniques to reduce the strong dependence of the adaptive protection schemes on the availability of communication systems and to determine if, over a fault condition, an Intelligent Electronic Device (IED) should operate considering the changes in operational conditions of an ADN. Additionally, the methodology takes into account different and remarkable recommendations for the use of ML techniques. The proposed methodology is validated on the modified IEEE 34-nodes test feeder. Additionally, it takes into consideration typical features of ADN and micro-grids like the load imbalance, reconfiguration, changes in impedance upstream from the micro-grid, and off-grid/on-grid operation modes. The results demonstrate the flexibility and simplicity of the methodology to determine the best accuracy performance among several ML models. Besides, they show the methodology’s versatility to find the suitable ML model for IEDs located on different zones of an ADN. The ease of design’s implementation, formulation of parameters, and promising test results indicate the potential for real-life applications.
Article
This paper presents a protection strategy based on active power flow direction, current magnitude and voltage sags to determine the existence of low impedance faults in inverter-dominated low voltage AC microgrids. Different fault situations are considered, concerning each element of the microgrid (line section, load, node and energy sources). Simulation is done in MATLAB/Simulink software. Microgrid was modelled based on CIGRE benchmark for network integration of renewable and distributed energy resources, considering the low voltage european distribution network characteristics. Inverter-based distributed generation units were modelled considering a control scheme composed by natural reference frame control and RMS current limiter. The results obtained from simulations indicate that the proposed strategy is able to protect the microgrid in any topology configuration or mode of operation, for all type of low impedance faults.
Article
Renewable energy sources primarily employ power electronic converters to deliver their energy to the grid. These converter-based distributed generators (CBDGs) have a fault behaviour different from conventional generators, capable of compromising the reliability of traditional protection schemes. This paper investigates the fault behaviour of CBDGs by presenting both an analytical steady state model and a numerical transient simulation model developed in MATLAB/Simulink. Using these models, the impact of grid code requirements regarding reactive current injection during faults is analyzed. The results underline that different modelling techniques can support the design of grid codes.
Article
Microgrids are emerging as an alternative mode of operation for distribution systems integrated with Distributed Energy Resources (DERs). With appropriate management and control of the DERs, a section of a distribution system can operate isolated from the main grid thereby enhancing the reliability and security of supply to consumers. However, the integration of DERs has raised many technical challenges including protection. The traditional distribution system protection cannot provide reliable protection to the microgrid in the isolated mode due to the limited short-circuit capacities of the converter-interfaced DERs. This paper proposes the application of a new voltage-based relay type for the protection of microgrids. The relay algorithm achieves its protection function through active power differential and sensitivity calculations based on voltage measurements within a specified protection zone. In the paper, the new relay type is modelled in Digsilent PowerFactory software and installed at the nodes of a microgrid test system. The performance of the relay type is investigated under variety of faults. The relay is shown to operate correctly and effectively to detect and identify faults in both radial and meshed microgrids integrated with inverter-interfaced DER technologies.
Article
There are many technical challenges associated with microgrid (MG) operation and reliability. A fast and reliable protection scheme is persistently needed either for a grid-connected or island mode of operation. This paper presents a novel communication-less directional protection scheme for a typical microgrid at the common bus (PCC) supplied by hybrid (AC/DC) renewable distributed generations (DGs). This protection scheme is based on the continuously measuring of each DG phase's current feeder and calculating the auto cosine similarity factors for the sampled current with movable window patterns concept. The selected window width is ¼ cycle and the proposed algorithm is applied between the measured window and its simultaneous stored previous cycle window. Using a movable pattern enables consideration of the intermittent generation of the different DG units. Applying the cosine similarity for the proposed relay action increases the MG reliability. The magnitudes indicate the fault incident while the sign is used for the fault feeder selection. The main advantages of the introduced technique are fast, sensitive and do not need any extra instruments or communication links for practical implementation. Besides, it eliminates the effects of fault impedance and loads changing on the digital relay efficiency.