IET Generation, Transmission & Distribution

Research Article

Islanding detection scheme for converter- ISSN 1751-8687

Received on 26th March 2018
Revised 5th August 2019
based DGs with nearly zero non-detectable Accepted on 9th October 2019
E-First on 31st October 2019
zone doi: 10.1049/iet-gtd.2018.5168
www.ietdl.org

Om Hari Gupta1, Manoj Tripathy2 , Vijay K. Sood3

1Department of Electrical Engineering, National Institute of Technology Jamshedpur, Jamshedpur, Jharkhand, India
2Department of Electrical Engineering, Indian Institute of Technology Roorkee, Roorkee, India
3Department of Electrical, Computer & Software Engineering, University of Ontario Institute of Technology, Oshawa, Canada

E-mail: manojfee@iitr.ac.in

Abstract: This study presents a new method for detecting an islanding event in a microgrid with embedded converter-based
distributed generation (DG). Unlike other schemes, the proposed scheme injects negligible perturbations into the microgrid after
the generation of an alert signal. The proposed scheme then uses another index called superimposed impedance, ΔZ. The ΔZ
is characterised by a low steady-state magnitude during the grid-connected mode and a high magnitude during the islanded
mode. Furthermore, for a fault at the point of common coupling (PCC), the magnitudes of ΔZ and PCC voltage are both very low
– except during the initial transient period, where |ΔZ| momentarily crosses the threshold. Therefore, an islanding event can be
detected if the magnitude of ΔZ is high for some specified time. Moreover, a fault event will not be misdirected as an islanding
event because the steady-state magnitudes of both the ΔZ and PCC voltage are very low in the case of a fault. The robustness
of the proposed detection scheme is evaluated against different islanding conditions and also, for a fault at the PCC, first by
using MATLAB-based simulations and later by using a laboratory-based experimental setup. A comparison with a recently
published detection scheme shows the superiority of the proposed schemes.

1 Introduction based on rate of change of voltage phase angle (ROCOVPA) is

proposed in [16], which has relatively very small NDZ. However,
The presence of distributed generations (DGs) within a microgrid when the inverter control mode is changed from the dependent to
introduces new challenges for the distribution system protection independent mode, this scheme finds its limitation. Further, a fault
due to bi-directional current/power flow. Within the microgrid, event is also misdetected as an islanding event by the ROCOVPA
there are mainly two protection concerns viz., protection against technique.
(internal) faults [1] and microgrid islanding [2–4]. The microgrid Active islanding detection scheme [17] directly interacts with
islanding refers to the condition in which the microgrid is energised the power system by injecting disturbances. Several active
by the DG while the main grid is disconnected. As per IEEE STD islanding detection schemes are available that reduce the amount of
1547-2018 [5] and other industrial standards recommendations, disturbance injection but not the rate of injections. These injections
viz., G83/1, G59/1, G75, ETR-113/1, and CEI 11-20 [6], it is the impact on and largely reduce the power quality. Communication-
responsibility of the DG to detect islanding condition and to based techniques [13] are more reliable but are very expensive in
disconnect itself within 2 s (i.e. 100 cycles for a 50 Hz system). terms of the requirement of peripheral signal equipment and,
However, DGs should be capable of operating in either grid- therefore, are not preferred [16]. A hybrid method, as proposed in
connected mode or islanded mode [7] with the assistance of load [14], is based on the frequency set point variation after observing
shedding, if required [8]. More importantly, DGs should be able to the voltage imbalance at the PCC. However, this scheme finds
switch from power/current control mode to voltage/frequency limitation if there is insufficient voltage imbalance and it may take
control mode in the event of islanding in order to maintain the around 1.5 s (or 75 cycles) to detect an islanding event. Similarly,
point of common coupling (PCC) voltage and frequency at rated in the scheme proposed in [15], there must be some voltage
values [9]. It requires the islanding detection to be faster than given variation at the PCC to apply the rate of power shift. There are a
in [5]; how much faster is still an open question and depends on few recent schemes proposed for the detection of islanding events
many factors. Faster detection provides more time for subsequent [2, 18, 19]. The results presented in [2, 18] are exceptional under
control and other protective actions. Therefore, in practice, faster various system conditions; however, the synchronous-based DGs
islanding detection is always desired. In the proposed work, are used in [2] and DGs used in [18] are wind farm based i.e. both
islanding events (under different system conditions) are detected the authors in [2, 18] use machine-based DGs whereas the scheme
nearly in 0.24 s (i.e. 12 cycles ‒ with an intentional delay of 5 proposed in this paper is intended to be used particularly for
cycles to prevent false detection due to the presence of transients) converter-based DGs. Further, the scheme proposed in [19] gives
which is faster than specified in [5]. Overall, islanding detection is promising results under various distressed conditions; nevertheless,
necessary in both the following cases: to disconnect the DG unit or it has comparatively high NDZ of 1%. In practice, for DGs rated
to switch the mode of operation of DG. under 250 kVA, the conventional under/over voltage (UOV) and
In the literature, there are four types of islanding detection under/over frequency (UOF) relays are used to detect islanding.
techniques identified namely, passive techniques [10, 11], active However, for DGs rated over 250 kVA (fitted with an automatic
techniques [12], communication-based techniques [13], and hybrid voltage regulator), specialised schemes are required as these DGs
techniques [14, 15]. Passive techniques are based on power system can maintain the load frequency and voltage within the specified
parameters (e.g. mismatching of active power ΔP and reactive limits. This introduces new challenges and becomes a costly
power ΔQ) and find limitations when an insufficient mismatch of practice.
parameters exists. The islanding event is not detected if ΔQ and ΔP To address such challenges and limitations of the reported
supplied by the main grid are very small i.e. lie within the non- schemes, this paper proposes the use of rate of change of negative
detectable zone (NDZ). A passive islanding detection technique sequence voltage (ROCONSV) and superimposed impedance, ‘ΔZ’

IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 23, pp. 5365-5374 5365
© The Institution of Engineering and Technology 2019
 Therefore, unlike many islanding detection schemes, the proposed
index (ΔZ) provides uniqueness in differentiating the fault and
islanding events, which is one of the main advantages of the
proposed method. The generation of the AS and calculation of ΔZ
are explained next.

2.1 Generation of the AS

The AS will be a logical 1 if the following is true:

dV PCC −
> V dthN (1)
dt

where dV PCC − /dt is the ROCONSV, the (−) signifies the

Fig. 1  Islanding event at 0.6 s during nearly perfect power matching of
negative sequence voltage, and V dthN is the threshold for
active and reactive powers
(a) Three-phase voltages, (b) ROCONSV
ROCONSV. The selection of VdthN is based on the magnitude of
ROCONSV during the normal operation and can be defined as in
(2) below:

V dthN ≃ 4 × ROCONSVconnected (2)

where ROCONSVconnected is the peak of ROCONSV during the

grid-connected mode.
The conventional features (e.g. voltage, frequency etc.) do not
deviate sufficiently from their steady-state values during an
islanding event if there is a small, or nearly zero, active/reactive
power mismatch between the load and DG. However, it is found
that there is always some deviation in the magnitude of negative
sequence voltage following an islanding event irrespective of the
power mismatch between the load and DG [21]. Moreover, during
any system changes (symmetrical or asymmetrical), there will be
some short-time finite changes in the negative sequence voltage;
even for a system that is perfectly balanced. Consider the case of
an islanding event during the nearly perfect power matching; the
corresponding three-phase voltages (vab, vbc, and vca) and their
ROCONSVs are presented in Fig. 1. It can be seen that the
ROCONSV (dVneg/dt) is sufficient to cross its threshold value
(VdthN). This is because of the marginal mismatching, which yields
Fig. 2  Generic microgrid
a very small negative sequence voltage for a small duration (as
(a) Single-line equivalent circuit, (b) Superimposed equivalent circuit
post-islanding, the microgrid remains balanced). Therefore, the
ROCONSV can be used to generate the AS even if nearly a perfect
as an index for the detection of islanding events for converter- active/reactive power match exists between the load and DG.
based DGs. The performance of the proposed scheme is examined
under different conditions, including UL1741 standard test
conditions for single- and multi-bus microgrid architectures and 2.2 Superimposed impedance (ΔZ) calculation
validated using a laboratory-based experimental setup. The Consider the microgrid shown in Fig. 2a and its superimposed
important contributions of the proposed ΔZ-based scheme are as equivalent circuit shown in Fig. 2b. The grid is said to be available
follows: (connected) when circuit breaker CB1 is closed and it is
unavailable when CB1 is open. The breaker CB2 is used to connect/
• negligibly small NDZ and comparatively very small power
disconnect the DG to the PCC. The complex load ‘ZL’ remains
quality degradations,
• high efficiency and reliability, always connected to the system (complete DG description is given
later in Section 3). Using several features extracted using PCC
• ability to detect islanding even when the inverter is controlled
measurements, the AS is first obtained. If the AS is logical 1, an
independently and does not require any training,
intentional small perturbation is injected into the DG supply and,
• it can be easily implemented and has a low cost. simultaneously, the ΔZ is calculated as per (3) below:

2 Proposed event detection scheme ΔV PCC

ΔZ = AS (3)
In this paper, instead of injecting disturbances at regular intervals, a ΔIdg
new alert signal (AS) is first generated before the injection of a
very small perturbation; typically, the perturbation is about one- Under the influence of different modes, the calculated ΔZ will be
tenth of perturbation used in [20]. First, the AS is generated using different, as given in (4):
ROCONSV. Second, after the perturbation injection, ΔZ is
calculated, which is defined as the ratio of superimposed (or ZL ∥ ZIB, connected mode
incremental) PCC voltage phasor to the superimposed (or ΔZ = Z L, islandedmode (4)
incremental) DG current phasor.
During the connected mode, ΔZ represents the parallel ZL ∥ ZIB ∥ Z f , fault mode
combination of load impedance and grid impedance (i.e. typically
very small in magnitude) and during the islanded mode, it is equal where ZL is the complex impedance of the load.
to the load impedance (i.e. typically large in magnitude). For any As depicted in (4), in connected mode (i.e. CB1 and CB2 closed,
ground fault within the distribution system, the PCC voltage and CBf open), ΔZ is equal to ZL||ZIB. Usually, the modulus of the grid
magnitude of ΔZ (i.e. |ΔZ|) will be very small or close to zero. impedance |ZIB| is very small compared to the modulus of the load

5366 IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 23, pp. 5365-5374
© The Institution of Engineering and Technology 2019
 reported in [20]) but based on performance evaluations, it was
found that the proposed scheme can successfully work with a 1%
perturbation ratio which effectively reduces the power quality
degradation to a minimum. Furthermore, contrary to other reported
schemes (e.g. [20]), which use periodical injections, the proposed
scheme injects the perturbation only when AS is logical 1 which
further improves the power quality. Fig. 3 shows a comparative
Fig. 3  Plot of PPAM versus PDM for proposed scheme and scheme of [20]
analysis of the proposed scheme and that of [20]. It presents the
percentage perturbation amount per minute (PPAM) with PCC
disturbances per minute (PDM) for both the schemes. The PCC
disturbances include load disturbances, capacitor switching events,
islanding events, fault events etc. The PPAM can be calculated as

PPAM = % perturbation × perturbation frequency per minute (7)

The schemes that use periodical perturbation injections have a

constant frequency of perturbation and usually, their minimum
value can be 30/min as per IEEE STD 1547-2018 [5]. Therefore,
for the scheme of [20], the minimum value of PPAM using (7) will
be constant and equal to 3 PU/min (i.e. 0.1 PU × 30/min)
corresponding to 10% (i.e. 0.1 PU) injection. Hence, the PPAM is
independent of the PDM for the case of [20]. On the contrary, in
the proposed scheme, the frequency of perturbation (and thus
PPAM) depends on the PCC disturbances (or PDM). Since only
1% (i.e. 0.01 PU) perturbation is allowed in the proposed scheme,
the PPAM for the proposed scheme can be obtained as

PPAM = 0.01 × PDM (8)

It is clear from (8) that PPAM is directly proportional to the PDM.

From Fig. 3, it can be observed that for a very large value of PDM
Fig. 4  Flowchart of the proposed scheme (up to 90/min), PPAM of the proposed scheme is very small
compared to that of the existing scheme [20].
impedance |ZL|. Therefore, |(ZL || ZIB)| < |ZIB| or in other words, | Fig. 4 presents the flow diagram of the proposed scheme. The
ΔZ| is very small (i.e. less than |ZIB|). Conversely, during the three-phase voltages and DG output currents are first measured at
islanded mode (i.e. CB1 and CBf open, CB2 closed), |ΔZ| is equal the PCC; then using (1), the disturbance at the PCC is detected. If a
disturbance is not detected using (1), the algorithm returns to the
to |ZL| which is very high compared to |ZIB|. Moreover, during the measurement block. Also, if a disturbance is detected, then the AS
fault at PCC (i.e. CB1, CB2 and CBf all closed), ΔZ represents the is generated and a perturbation is caused by reducing the DG
parallel combination of ZL, ZIB, and the fault impedance ‘Zf’ (i.e. output by only 1% after 7 cycles or 0.14 s (a preselected time to let
ZL || ZIB || Zf) so, again, |ΔZ| is very small or close to zero. In any transients to die out). Now, the ΔZ is calculated at PCC using
summary, it can be concluded that |ΔZ| is large and is equal to |ZL| (3) and if its magnitude i.e. |ΔZ| is found to be more than the
when the microgrid is islanded. For the other two cases (i.e. threshold value (|ΔZthr|) for a duration of 0.1 s (i.e. 5 cycles, pre-
connected mode and fault mode), |ΔZ| is very small and is less than selected to avoid any false islanding event detection due to other
|ZIB|. switching events), an islanding event is detected. Else, there is
another type of switching event indicated and the algorithm returns
to the measurement block.
2.3 Threshold selection for |ΔZ|
The |ΔZ|, during the connected mode, nearly reflects the grid 3 DG description and converter control
impedance |ZIB| and during the islanding mode, it reflects the load
impedance |ZL|, which is many times greater than |ZIB|. After A generic microgrid, as shown in Fig. 2a, is firstly used as the test
system: the data is given in the Appendix. The VSC-based DG is
conducting various tests under different conditions, it is observed
connected to the PCC via a breaker CB2. Another breaker CB1 is
that the threshold for |ΔZ| (i.e. |ΔZthr|) to detect an islanding event
included to connect or disconnect the PCC from the main grid.
can be set as given below in (5):
When CB1 is closed, the load is supplied by both the main grid and
ΔZthr = H ⋅ ΔZ connected DG. On the other hand, if CB1 is open, the load is only supplied by
(5)
the DG i.e. an islanded microgrid. The control scheme of the
where |ΔZ|connected = |ΔZ| during the connected mode and H is microgrid of Fig. 2a is shown in Fig. 5. The solar-based DGs are
usually operated in unity power factor mode to supply the active
defined below in (6):
power only. The PCC voltage and DG current outputs are used to
obtain active (P) and reactive (Q) power outputs of DG. The active
ΔZ − ΔZ connected
islanded
H ≃k× (6) power is compared with a reference active power (Pref) and an
ΔZ islanded
error signal (EP) is generated. This error signal is given to an active
power controller (i.e. an integrator) which gives the direct axis
where k defines the margin of safety. From the obtained results for
reference current (Id). The quadrature axis reference current (Iq)
various cases, the best selected value of k is 3.
and zero sequence reference current (I0) are kept zero as only the
2.4 Application of the proposed scheme with improved power active power is to be delivered by the inverter. Then, using the
quality dq0–abc transformation, reference current (iref) is obtained, which
has the frequency and phase identical to that of the PCC voltage.
The proposed scheme utilises a 1% perturbation ratio (i.e. 1% The frequency (f) and phase (θ) of the PCC voltage are obtained by
reduction in rated power injected by the DG) to detect the islanding using a phase-locked loop (PLL). Using iref and DG output current
event. This injection ratio could be kept higher (e.g. 10% as

IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 23, pp. 5365-5374 5367
© The Institution of Engineering and Technology 2019
Fig. 5  Control scheme of converter-based DG

(idg), firing pulses are generated with a hysteresis current controller

to trigger the inverter switches.

4 Results
The proposed scheme is first evaluated using an off-line
environment using MATLAB. Afterwards, the scheme is validated
using a single-phase laboratory prototype setup.

4.1 Off-line verification for single bus test model

The test model (Fig. 2a) is simulated in MATLAB/Simulink with a
step size of 10 μs: the control scheme is shown in Fig. 5. A three-
leg H-bridge inverter, working in hysteresis current controlled
mode, is used to convert the DC into AC. The reference current is
obtained using the error between the DG output power and desired
DG power. Now, based on the error introduced by the difference of
the DG reference current and actual DG current, firing pulses are
generated using a hysteresis current controller. The hysteresis band
is kept small (i.e. 0.1 A or 1.52 × 10−3 PU). The solar-based DG
inverters are usually operated to supply active power only, i.e.
operated at unity power factor. Hence, the reactive power demand
of load to be supplied by the main grid. Therefore, in this work, the
converter-based DG is implemented to provide active power only.
The effectiveness of the proposed scheme has been investigated
under different microgrid conditions including UL1741 test
conditions [22], i.e. load resonates at a nearly fundamental
frequency (50 Hz) with quality factor (QF) equal to 1.8.

4.1.1 Islanding during nearly perfect power-match conditions Fig. 6  Islanding event with nearly perfect power match
including UL1741 test condition: Fig. 6 presents the results (a) ROCONSV, (b) AS, (c) Pdg, (d) |ΔZ|, (e) ID, (f) VPCC, (g) fPCC, (h) PL and ΔP
obtained for an islanding event occurring at 0.6 s during the nearly
perfect power match (i.e. ΔP and ΔQ are almost zero). After an Table 1 Comparison of power quality with [20]
islanding event, the ROCONSV just crosses the threshold VdthN, Parameters Scheme of [20] Proposed scheme
which can be seen in Fig. 6a. It is clear from Fig. 6b that as per the PCC voltage THD, % 0.49 0.24
algorithm, the change in AS from logical zero to one is delayed by fundamental voltage, V 437.34 439.88
0.4 s after detection of a disturbance at the PCC occurring at 0.6 s. DG current THD, % 1.09 1.00
The generation of AS causes a decrement in the DG output power fundamental current, A 64.81 65.55
by 1% of its rated value, as shown in Fig. 6c. Now, the |ΔZ| is
calculated which settles at nearly 12.5 PU (base value is |ΔZ|
connected). Since the |ΔZ| is greater than the threshold value ‘|ΔZthr|’
is detected but the scheme of [20] injects the perturbations at pre-
[i.e. 3 PU; obtained using (5)] for >0.3 s as depicted in Fig. 6d, the
selected regular intervals which cause more degradation of power
islanding event is detected and the islanding detection indicator
quality than the proposed scheme.
(ID) moves from logical 0 to 1, see Fig. 6e. Further, it is also clear
To investigate the effect of load QF, |ΔZ| is calculated for
from Figs. 6f‒h that voltage, frequency, and active power hardly
islanding events with different QFs, i.e. 0.3, 1.8 (UL1741 test
change from their steady-state values after the occurrence of an
condition [22]), and 2.5. The plots of |ΔZ| under different QFs are
islanding event. Therefore, passive techniques (e.g. UOV) find
depicted in Fig. 7 and the values of |ΔZ| are included in Table 2. It
limitations.
can be seen that |ΔZ| is nearly equal in all three cases. Therefore,
To observe the power quality degradation due to perturbation
the proposed technique is able to detect the islanding event during
injection, two cycles (one pre-injection and one post-injection) are
the nearly perfect power match conditions including UL1741 test
considered. A comparison of various features with that of [20] is
condition and has a negligible NDZ, as depicted in Fig. 8. The
presented in Table 1. It is found that the proposed scheme is better
scheme has been tested for all the possible/practical conditions.
in all aspects such as PCC voltage THD, DG current THD, and
However, next sections report results/plots of some representative
fundamental magnitudes of voltage and current. Furthermore, the
cases only.
proposed scheme injects the perturbation only when a disturbance
5368 IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 23, pp. 5365-5374
© The Institution of Engineering and Technology 2019
 Table 3 Comparison with voltage/frequency-based scheme
Mismatch during islanding, VPCC, PU fPCC, Hz |ΔZ|, PU
PU
−0.161<ΔP<0.313 0.88< V <1.1 49.5 < f <50.5 >3
−0.016<ΔQ<0.013 0.88< V <1.1 49.5 < f <50.5 >3

Fig. 9  Effect of load switching on |ΔZ|

Fig. 7  |ΔZ| during islanding with nearly perfect power match for
(a) QF = 0.3, (b) QF = 1.8, (c) QF = 2.5

Table 2 Obtained |ΔZ| during different events

Event type |ΔZ|, PU VPCC, PU
islanding during perfect match (QF = 0.3) 12.30 0.99
islanding during perfect match (QF = 1.8) 12.18 0.99
islanding during perfect match (QF = 2.5) 12.12 0.99
islanding during 22% reactive power mismatch 12.18 1.04
islanding during 25% active power mismatch 9.84 0.88
load switching, increased 50% 1.04 0.99 Fig. 10  |ΔZ| for fault at PCC
three-phase fault at PCC 0.00 0.00 (a) Solid three-phase fault, (b) High resistance fault

DG output variations, reduced 20% 1.00 0.99

UOV- and UOF-based schemes, the active power mismatch limits
(ΔPn and ΔPp) are −0.161 and 0.313 PU, respectively, while the
reactive power mismatch limits (ΔQn and ΔQp) are −0.016 and
0.013 PU, respectively, as depicted in Table 3. On the contrary,
these limits (and NDZ) are very small in the case of the proposed
scheme which is clear from Fig. 8 (all values are in PU). Moreover,
the islanding events can be detected earlier without injecting any
perturbation when the mismatch of power is large enough.

4.1.4 Load switching: The load connected to the distribution

system varies randomly. Therefore, to study the effect of the
sudden load jump, 50% of the rated load is suddenly increased at
0.6 s. The changes are found in ROCONSV and therefore, an AS is
Fig. 8  Comparison of NDZ of proposed scheme with that of UOV/UOF
generated. The |ΔZ| momentarily crosses the threshold but returns
back to 1 PU within 0.02 s, as shown in Fig. 9. Thus, the islanding
4.1.2 Islanding during reactive power mismatch: For an
event is not detected (see Table 2) as per the flowchart given in
islanding event during a mismatch of reactive powers (say ΔQ = 
Fig. 4.
22%), the |ΔZ| and PCC voltage are included in Table 2. The load
requires 0.22 PU reactive power while the DG supplies no reactive
power as it is working at the unity power factor. Therefore, during 4.1.5 Fault at PCC: The ROCONSV crosses the threshold when a
the pre-islanding condition, the reactive power is supplied by the solid three-phase fault is created at PCC at t = 0.6 s and the AS is
grid. After an islanding event, due to reactive power mismatch, the generated. The |ΔZ| is nearly |ΔZthr| momentarily and returns to
ROCONSV crosses the threshold and generates AS. Due to this, almost zero magnitude as depicted in Fig. 10a. Now, as per the
1% perturbation in DG power is injected and ΔZ is calculated. proposed algorithm, this event is not misdetected as an islanding
Subsequently, it is observed that |ΔZ| crosses the threshold and event and can be considered as the other switching event. Further,
remains high (i.e. 12.18 PU) for more than 0.3 s. Therefore, the to investigate the impact of high impedance faults on the proposed
proposed scheme successfully detects the islanding event. scheme, a fault with 10 Ω resistance (more than the load resistance)
is created at PCC. Following the fluctuations, the steady-state value
4.1.3 Islanding during active power mismatch: When islanding of |ΔZ| is nearly 1 PU as shown in Fig. 11. Hence, the proposed
occurs at time 0.6 s with the mismatch of active power (say ΔP =  scheme is unaffected by the value of fault resistance. For the
25%), the AS is generated since ROCONSV has crossed the unsymmetrical fault analysis, an a–g fault is created at PCC and it
threshold and |ΔZ| is obtained by using (3) which is higher than the is found that |ΔZ| for each phase is less than |ΔZthr| and therefore,
threshold (i.e. 9.84 PU) as shown in Table 2; therefore, islanding this event is not misdetected as an islanding event (Fig. 11).
event is detected.
The performance of the proposed scheme is compared with the 4.1.6 DG output variation: As discussed in Section 1, the solar-
UOV- and UOF-based schemes. Corresponding outcome is based DGs are the fastest growing DGs. The output of this type of
included in Table 3 which shows that UOV- and UOF-based DG varies primarily with irradiance and temperature. Therefore,
schemes find limitation during the small power mismatch because the impact of DG output variation on the proposed scheme has
voltage and frequency do not deviate from their respective limits been studied. When the DG output has been step changed from 1 to
while the proposed scheme successfully detects the islanding event 0.8 PU, the proposed scheme is found unaffected since |ΔZ| is less
during all the conditions since the |ΔZ| is found to be >3 PU. For

IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 23, pp. 5365-5374 5369
© The Institution of Engineering and Technology 2019
 Fig. 14  Equivalent diagram for switching of a capacitor bank

Fig. 11  |ΔZ| for asymmetrical (a–g) fault at PCC

(a) Phase-a, (b) Phase-b, (b) Phase-c

Fig. 12  Equivalent diagram for unbalanced load connection

Fig. 15  Switching of a capacitor bank

(a) QL and ΔQ, (b) ROCONSV, (c) AS, (d) Pdg, (e) |ΔZ|

under such conditions. A single-phase load has been increased by

>50%. Fig. 12 includes the equivalent diagram in which the circuit
breaker CBa is switched on to connect to the single-phase load at
0.6 s. The outcome in the form of different plots is presented in
Fig. 13. It can be seen from Figs. 13a‒e that at 0.6 s, the current
increases and disturbs ROCONSV causing the generation of an AS
and enabling |ΔZ| calculation after the perturbation injection. It is
found that the |ΔZ| goes below the threshold |ΔZthr| after being
greater than the threshold for about one cycle i.e. 0.02 s and
therefore, this is considered as a switching event and not an
islanding event.

4.1.8 Switching of a capacitor bank: A capacitor bank of 15 

kVAR (0.3 PU) has been connected to the microgrid at 0.6 s –
equivalent is given in Fig. 14. It can be seen in Fig. 15a that the
reactive power supplied by the bank (QL) is fed to the utility grid
(ΔQ) since the load connected to the system is of unity power
factor and supplied fully by the DG. There is a change in
ROCONSV that leads the generation of an AS, perturbation
injection, and |ΔZ| calculation, as depicted in Figs. 15b–e. Since,
Fig. 13  Connection of unbalanced load after some fluctuations, |ΔZ| falls below the |ΔZthr|, it will be
(a) Load currents of phase-a and -b, (b) ROCONSV, (c) AS, (d) Pdg, (e) |ΔZ| considered as a switching event.

than |ΔZthr| as shown in Table 2. Therefore, it is considered as a 4.1.9 Starting of induction motor: A three-phase squirrel cage
non-islanding switching event. induction motor with the following rating is connected to the
system at 0.6 s, as depicted in Fig. 16.
4.1.7 Unbalanced load: Most of the time, the distribution system
is subjected to unbalanced loading. Therefore, it becomes
important to investigate the performance of the proposed scheme
5370 IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 23, pp. 5365-5374
© The Institution of Engineering and Technology 2019
Fig. 16  Equivalent diagram for the starting of induction motor Fig. 18  Equivalent diagram for the switching of three-phase rectifier

Fig. 17  Starting of induction motor Fig. 19  Switching of three-phase rectifier
(a) Te, (b) Speed, (c) Im, (d) ROCONSV, (e) AS, (f) Pdg, (g) |ΔZ| (a) Rectifier voltages, (b) Rectifier idc, (c) Rectifier iac, (d) ROCONSV, (e) AS, (f)
Pdg, (g) |ΔZ|
Rating: 18.45 kVA (0.369 PU), 440 V, 50 Hz, Rs = 0.5968 Ω, Ls 
= 0.0003495 H, Rr = 0.6258 Ω, Lr = 0.005473 H, inertia = 0.05 kg  can be seen in Fig. 19d. A perturbation in DG power is caused by
the generation of an AS and then |ΔZ| is calculated which is found
m2, friction factor = 0.005879 N m s, pole pair = 8. to be less than |ΔZthr| with some fluctuations, for about 0.5 s, as
When the motor is connected, electromagnetic torque (Te),
depicted in Figs. 19e–g. So the event is assumed to be a switching
motor speed, motor current (im), and ROCONSV are depicted in event.
Figs. 17a–d. The AS is generated since the ROCONSV crosses the
corresponding threshold and |ΔZ| is calculated the following
4.2 Offline verification for modified IEEE 13-bus test feeder
perturbation injection – see Figs. 17e–g. This event is assumed to
be switching event because |ΔZ| does not remain greater than the To test the proposed scheme for a microgrid with different
threshold for >0.1 s. configurations, a modified IEEE 13-bus test feeder [23] is used, as
depicted in Fig. 20a. A 50 kW converter-based DG is placed at bus
4.1.10 Presence of non-linear load: To investigate the 671 and the model does not include the regulating transformer
performance of the proposed scheme under non-linear loading, a between nodes 650 and 632. Both the conditions, i.e. switch ‘SW’
three-phase half-wave rectifier with a series inductive filter is close and open, are considered for testing the proposed scheme.
connected to the microgrid at 0.6 s – circuit shown in Fig. 18. The The value of |ΔZthr|, calculated using (5), is 2.42 PU (base value is |
power drawn by the DC load is 0.145 PU. The rectifier voltage ΔZ|connected). An islanding event is created at 0.6 s and it is found
(DC side and phase-a of AC side) and current (DC side and phase- that the proposed scheme successfully detects this islanding event
a of AC side) are given in Figs. 19a–c. Following the rectifier as |ΔZ| is more than |ΔZthr|, which is clear from Figs. 20b and c.
switching, there will be variations/fluctuations in ROCONSV as

IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 23, pp. 5365-5374 5371
© The Institution of Engineering and Technology 2019
 Fig. 21  Block diagram for experimental setup

Fig. 20  Verification for modified IEEE 13-bus test feeder

(a) Single-line equivalent circuit, (b) |ΔZ| recorded for an islanding event with SW
closed, (c) |ΔZ| with SW open Fig. 22  Experimental results
(a) Grid voltage and current in grid-connected mode, (b) DG voltage and current in
4.3 Experimental validation grid-connected mode, (c) |ΔZ| and DG output power for islanded mode with closely
matched powers, (d) |ΔZ| and DG output power for grid-connected mode
A one-phase laboratory experimental setup has been developed
with a prototype of converter-based DG connected to the load and experimental validation of the proposed scheme, |ΔZ| was
distribution grid. As can be seen in Fig. 21, the hysteresis current measured for two conditions: one, when converter-based DG and
controlled-based firing pulses are generated by the CLP1104 load were not connected to the grid and the powers of DG and load
dSPACE processor board. While generating the firing pulses, the were closely matched (referred as islanded mode with closely
DG current and voltage are sensed using the HTP50 and AD202JN matched powers); and second, when converter-based DG and load
sensors, respectively. These sensors measure the instantaneous were connected to the grid and the powers of DG and load were,
values of the variables. With the help of these sensed voltage and again, closely matched (referred as grid-connected mode – also, a
current, the DG output power, PCC voltage frequency, and phase ‘non-islanding event or mode’). For an islanding condition,
are obtained using the CLP1104 dSPACE processor board. The Fig. 22c depicts the magnitude of measured |ΔZ| and DG power. It
proposed algorithm along with a one-phase PLL block (available in is evident that |ΔZ| is large and nearly equal to the load impedance
MATLAB/Simulink library) and control scheme are loaded on (i.e. 25 Ω). Therefore, the islanding event is successfully detected.
dSPACE processor board to prepare a current template and to On the contrary, |ΔZ| is found to be very small during the
generate the firing pulses based on the current and power errors as connected mode, which is clear from Fig. 22d. So no islanding
depicted in Fig. 5. These pulses are given to a one-phase insulated- event is detected in this situation. Therefore, the results obtained
gate bipolar transistor-based inverter module (Semikron) through from an experimental laboratory setup successfully validate the
an optocoupler (6N136) printed circuit board to provide the proposed scheme.
isolation.
The inverter module consists of an inbuilt driver circuit. The
4.4 Comparison between the proposed scheme and that of
output of the inverter (or DG) is connected to PCC via a series
[16]
inductor (the inverter uses an inbuilt rectified dc supply). The
resistive load (25 Ω) and grid-side supply (50 V peak) are also The scheme proposed in this paper is compared with a passive
connected to the PCC. The results are captured using an Agilent- islanding detection scheme reported in [16], which is based on
built DSO-X 2014A digital storage oscilloscope (DSO). The time ROCOVPA. In the case of the dependent mode of operation (i.e.
step is 50 μs i.e. more than that of the off-line verification (i.e. 10  the converter is using PCC frequency and phase as feedback), both
μs) due to dSPACE processor board limitation. There will only be the schemes detect islanding event as shown in Fig. 23. In this
more ripples in the output of inverter due to increased step size. figure, dPh/dt is the ROCOVPA and ‘th’ is its threshold.
Moreover, nowadays, there are various processors available having There are conditions when the main grid may not be available
high processing speed so the step size will not be a problem while due to load shedding or due to some other reason (normally
practically implementing the proposed scheme. referred as disconnected condition). Moreover, there can be more
For a closely matched power condition, Fig. 22a presents the than one DG connected to the system, as shown in Fig. 24 [24]. Let
grid voltage and current. The grid current should be zero in this us consider a case in which two DGs (DG1 and DG2) are islanded.
case but it is a noisy waveform, which is due to the switching Now, one DG (say DG1) will be maintaining the PCC voltage and
behaviour of the inverter. It is clear from Fig. 22b that the inverter frequency by generating its own reference template. Moreover,
current is sinusoidal and in-phase with voltage. For the other DG (DG2) will be working in a power control mode and not

5372 IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 23, pp. 5365-5374
© The Institution of Engineering and Technology 2019
 Fig. 26  System schematic under disconnected mode with two DGs in the
system

Fig. 23  Nearly perfect power match with dependent mode of operation
(a) ROCOVPA, (b) |ΔZ|

Fig. 27  Nearly perfect power match with independently controlled

converter
(a) ROCOVPA, (b) |ΔZ|

Fig. 24  Modified IEEE-33 bus system with three DGs [24]
Fig. 28  Short-circuit fault at PCC
(a) ROCOVPA, (b) |ΔZ|

version of islanded DG1 and DG2, used to test the proposed

scheme under an independent mode, is shown below in Fig. 26 and
the result is depicted in Fig. 27. After the islanding due to
disconnection of DG2, the |ΔZ| becomes more than the threshold
and detects islanding while the ROCOVPA does not cross the
corresponding threshold fails to detect islanding in this case.
Under fault inception at PCC, the ROCOVPA jumps to a very
high value and remains more than the threshold of 10°/s for >0.15 s
as shown in Fig. 28a. On the other hand, the |ΔZ| remains below
Fig. 25  Switching between dependent to independent mode the threshold value as shown in Fig. 28b.
Therefore, the ROCOVPA-based scheme [16] finds its
in a voltage/frequency control mode. Therefore, DG1 will be in an limitation during the fault while the proposed scheme is unaffected
independent mode and controlling the voltage and frequency of the by a fault at PCC. Overall, it is found that the ROCOVPA-based
system. In the independent mode of operation, DG1 will have a scheme [16] is highly susceptible to disturbances other than
phase θint corresponding to a fixed frequency, ω. The two modes of islanding and fails to detect islanding events if the converter is
operation (i.e. dependent and independent) can be understood with independently controlled with nearly perfect power matching
the help of Fig. 25. During the dependent mode PLL is used to conditions. While the proposed scheme works absolutely fine in all
generate the reference phase for the inverter and during the cases with the use of very small perturbation (i.e. 1%) which
independent mode, fixed frequency reference is used to obtain the barely degrades the power quality. The proposed scheme takes
inverter reference phase [25]. more time to detect the islanding event as compared to the scheme
Now, there can be a disconnection of DG2 due to any known or of [16]. However, it is less than the specified time i.e. 2 s [5].
unknown reason. This disconnection can also be treated as a Furthermore, the proposed scheme is more robust than the scheme
secondary islanding (or islanding) event. The detection of this of [16] and clearly differentiates between the islanding and other
event is also necessary for the implementation of load shedding or switching events.
connection of backup DG (if available) etc. Therefore, the
proposed method has been tested for such scenario. The reduced

IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 23, pp. 5365-5374 5373
© The Institution of Engineering and Technology 2019
5 Conclusion [10] Hashemi, F., Ghadimi, N., Sobhani, B.: ‘Islanding detection for inverter-based
DG coupled with using an adaptive neuro-fuzzy inference system’, Int. J.
This paper presents a new method to detect islanding event in a Electr. Power Energy Syst., 2013, 45, (1), pp. 443–455
[11] Sareen, K., Bhalja, B.R., Maheshwari, R.P.: ‘Islanding detection technique
microgrid with embedded converter-based DGs. To address the based on inverse hyperbolic secant function’, IET Renew. Power Gener.,
limitations of existing schemes, an index i.e. superimposed 2016, 10, (7), pp. 1002–1009(7)
impedance ‘ΔZ’ is used to detect the islanding event. The islanding [12] Bei, T.-Z.: ‘Accurate active islanding detection method for grid-tied inverters
and other switching events are classified based on the variations in in distributed generation’, IET Renew. Power Gener., 2017, 11, (13), pp.
1633–1639
the |ΔZ|. If |ΔZ| is more than a defined threshold for 0.1 s, the [13] Ropp, M., Larson, S., Meendering, D., et al.: ‘Discussion of a power line
islanding event is detected. Else, there can be other type of carrier communications-based anti-islanding scheme using a commercial
switching event such as load switching, fault and capacitor automatic meter reading system’. 2006 IEEE 4th World Conf. on Photovoltaic
switching. Energy Conf., Hilton Waikoloa Village Waikoloa, Hawaii, May 7–12, 2006,
pp. 2351–2354
The performance of the proposed methodology is first verified [14] Menon, V., Nehrir, M.H.: ‘A hybrid islanding detection technique using
using MATLAB simulations for various events such as islanding, voltage unbalance and frequency set point’, IEEE Trans. Power Syst., 2007,
faults, load switching, and DG output variations. Results proved 22, (1), pp. 442–448
the effectiveness of the proposed methodology. Later, the method is [15] Mahat, P., Bak-Jensen, B., Jensen, B.B.: ‘A hybrid islanding detection
technique using average rate of voltage change and real power shift’, IEEE
successfully validated using a one-phase experimental setup of a Trans. Power Deliv., 2009, 24, (2), pp. 764–771
microgrid. The performance of the proposed method is also [16] Ghanbari, T., Samet, H., Hashemi, F.: ‘Islanding detection method for
compared with the recent schemes of [16, 20] in respect of fault inverter-based distributed generation with negligible non-detection zone using
event, the effect of the operational mode of the converter, and energy of rate of change of voltage phase angle’, IET Gener. Transm. Distrib.,
2015, 9, (15), pp. 2337–2350
power quality. The results demonstrate the superiority of the [17] Samui, A., Samantaray, S.R.: ‘New active islanding detection scheme for
proposed methodology in various aspects. constant power and constant current controlled inverter-based distributed
Moreover, the proposed scheme is easy to implement as only generation’, IET Gener. Transm. Distrib., 2013, 7, (7), pp. 779–789
the algorithm is to be updated on the computer and can be [18] Abd-Elkader, A.G., Saleh, S.M., Magdi Eiteba, M.B.: ‘A passive islanding
detection strategy for multi-distributed generations’, Int. J. Electr. Power
implemented at a low cost as it does not require additional Energy Syst., 2018, 99, pp. 146–155
hardware or separate measuring devices. [19] Dubey, R., Popov, M., Samantaray, S.R.: ‘Transient monitoring function-
Additionally, the proposed method is helpful in distinguishing based islanding detection in power distribution network’, IET Gener. Transm.
between fault events and islanding events, so as to enable the Distrib., 2019, 13, (6), pp. 805–813
[20] Yu, B.G., Matsui, M., Yu, G.J.: ‘A correlation-based islanding-detection
protection system to act accordingly and prevent confusion among method using current-magnitude disturbance for PV system’, IEEE Trans.
both these events. The use of the proposed methodology can be Ind. Electron., 2011, 58, (7), pp. 2935–2943
further extended in controlling the mode of operation of the [21] Dash, P.K., Padhee, M., Panigrahi, T.K.: ‘A hybrid time–frequency approach
converter-based DGs when it is required to change from power/ based fuzzy logic system for power island detection in grid connected
distributed generation’, Int. J. Electr. Power Energy Syst., 2012, 42, (1), pp.
current control mode to voltage/frequency control mode in the case 453–464
of microgrid applications. [22] Bahrani, B., Karimi, H., Iravani, R.: ‘Nondetection zone assessment of an
active islanding detection method and its experimental evaluation’, IEEE
Trans. Power Deliv., 2011, 26, (2), pp. 517–525
6 References [23] ‘Distribution Test Feeders’. Available at http://sites.ieee.org/pes-testfeeders/
[1] Kar, S., Samantaray, S.R.: ‘Time-frequency transform-based differential resources/, accessed 25 March 2018
scheme for microgrid protection’, IET Gener. Transm. Distrib., 2014, 8, (2), [24] Rostami, A., Jalilian, A., Hagh, M.T., et al.: ‘Islanding detection of distributed
pp. 310–320 generation based on rate of change of exciter voltage with circuit breaker
[2] Makwana, Y.M., Bhalja, B.R.: ‘Islanding detection technique based on switching strategy’, IEEE Trans. Ind. Appl., 2019, 55, (1), pp. 954–963
superimposed components of voltage’, IET Renew. Power Gener., 2017, 11, [25] Kumar, S., Singh, B.: ‘Seamless transition of three phase microgrid with load
(11), pp. 1371–1381 compensation capabilities’. 2017 IEEE Industry Applications Society Annual
[3] Khamis, A., Xu, Y., Dong, Z.Y., et al.: ‘Faster detection of microgrid Meeting, Cincinnati, OH, USA, 2017, pp. 1–9
islanding events using an adaptive ensemble classifier’, IEEE Trans. Smart [26] Skocil, T., Gomis-Bellmunt, O., Montesinos-Miracle, D., et al.: ‘Passive and
Grid, 2016, 9, (3), pp. 1–11 active methods of islanding for PV systems’. 13th European Conf. on Power
[4] Samantaray, S.R., Samui, A., Babu, B.C.: ‘Time-frequency transform-based Electronics and Applications, 2009, EPE ‘09, Barcelona, Spain, 2009, pp. 1–
islanding detection in distributed generation’, IET Renew. Power Gener., 10
2011, 5, (6), pp. 431–438
[5] I. S. C. C. 21: ‘IEEE standard for interconnection and interoperability of
distributed energy resources with associated electric power systems
7 Appendix
interfaces’, IEEE Std 1547-20018 (Revision IEEE Std 1547-2003), 2018, pp.
1–138
 
[6] Chowdhury, S.P.P., Chowdhury, S.P.P., Crossley, P.A.A.: ‘Islanding protection Microgrid data [26]: PCC rated voltage, VPCC = 0.44 kV and rated
of active distribution networks with renewable distributed generators: a
comprehensive survey’, Electr. Power Syst. Res., 2009, 79, (6), pp. 984–992 frequency, f = 50 Hz, DG rated output power = 50 kW.
[7] Katiraei, F., Iravani, M.R., Lehn, P.W.: ‘Micro-grid autonomous operation Grid data: Rated voltage = 0.44 kV, RIB = 1 mΩ, LIB = 1 mH.
during and subsequent to islanding process’, IEEE Trans. Power Deliv., 2005, Load data: Rated power = 50 kW, R = 3.872 Ω, L = 41.0832 
20, (1), pp. 248–257
[8] Hirodontis, S., Li, H.: ‘An adaptive load shedding method for intentional mH, C = 246.6244 μF, load resonance frequency = 50 Hz.
islanding’. 2009 Int. Conf. on Clean Electrical Power, Italy, 2009, pp. 300– Transformer data: Rated kVA = 100 kVA, voltage = 0.1/0.44 kV
303 (star–star), R1 = R2 = 1 × 10−4 PU, L1 = L2 = 0.04 PU, Rm = 500 PU,
[9] Iravani, M.R.: ‘A control strategy for a distributed generation unit in grid-
connected and autonomous modes of operation’, IEEE Trans. Power Deliv., Lm = 500 PU.
2008, 23, (2), pp. 850–859

5374 IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 23, pp. 5365-5374
© The Institution of Engineering and Technology 2019