You are on page 1of 16

Electrical Power and Energy Systems 143 (2022) 108437

Contents lists available at ScienceDirect

International Journal of Electrical Power and Energy Systems


journal homepage: www.elsevier.com/locate/ijepes

A new hybrid islanding detection method for mini hydro-based microgrids


Alexandre Serrano-Fontova a, *, Reza Bakhshi-Jafarabadi b
a
School of Electrical and Electronic Engineering, The University of Manchester, Manchester M60 1QD, UK
b
Electrical Engineering Department, Engineering Faculty, Ferdowsi University of Mashhad, Mashhad, Iran

A R T I C L E I N F O A B S T R A C T

Keywords: This paper proposes a fast and reliable hybrid islanding detection method (IDM) for mini-hydro-based distributed
Distributed generation (DG) generations (DGs) with zero non-detection zone (NDZ). The proposed IDM aims to tackle the islanding events
Islanding detection method (IDM) caused by a self-excited induction generator (SEIG) when it is driven by a mini-hydro turbine system utilising the
Microgrids
transient dynamic response of the governor for the first time. To achieve such a goal, it takes advantage of a two-
Mini-hydro
Non-detection zone (NDZ)
stage process in which both passive and active techniques are combined. Thus, if the rate of change of frequency
(ROCOF)-based threshold of the first stage is exceeded, the power reference of the mini-hydro unit is modified,
implying a change in the turbine governor gate position. The mechanical torque applied to the prime mover is
accordingly shifted to a new state so that both frequency and its derivative will exceed the established thresholds
in the second stage in the islanding mode. Conversely, the effect of imposed disturbance is eminently negligible
in the grid-connected mode since the frequency is strictly dictated by the main grid. The proposed IDM has been
evaluated through numerous islanding and non-islanding case studies considering both single and multi-DG
scenarios in MATLAB/Simulink. The outcomes highlight the outstanding performance of the proposed algo­
rithm with zero NDZ and 473 ms average detection time, indicating the capability of the governor system as a
reliable tool to identify islanding operations. The proposed technique does not degrade the power quality (PQ) of
the grid, requires a low level of computational complexity and provides a high degree of reliability. Therefore, it
is a robust and cost-effective solution for future microgrids with great penetration of mini-hydro units.
© 2017 Elsevier Inc. All rights reserved.

equipment, and damage to the electrical machines due to the out-of-


1. Introduction phase reclosing [3]. In order to grapple with such islanding effects,
the IEEE Std. 1547–2018 proposes a procedure to be followed in these
During recent years, the growing interest in renewable energy re­ situations and suggests a maximum of 2 s for ceasing/controlling DG
sources (RESs) has increased to mitigate global warming and prevent its generation [4]. The essential aspects of any islanding study are the
hazardous effects. Thus, RESs are being introduced into the electric capability for fast and reliable islanding detection as well as avoiding
power systems at a steady pace, whereas the traditional fossil-fuel-based false tripping during non-islanding events. In this context, various
resources are being dismantled gradually. For instance, the global islanding detection methods (IDMs) have been recently presented. These
installed mini-hydro capacity was around 78 GW in 2019, indicating a IDMs can be generally classified into communication-based (remote),
10% growth with respect to 2013 [1]. passive, active, and hybrid techniques [5].
Given the inherent intermittency of these RESs and the particular­
ities of their energy conversion process, the reliability of the whole 1.1. Literature review
system is being jeopardised. Further, these RESs are widely known as
distributed generations (DGs) when are connected to the distribution The communication-based IDMs commonly use a telecommunication
networks (DNs), thus posing a few technical challenges such as unin­ channel such as optic-fibre and power line communication in an attempt
tentional islanding. This undesired event occurs when a portion of the to connect the circuit breaker (CB) of the substation-feeder with the CB
network that has been isolated from the main grid, remains energised located at the DG point of common coupling (PCC). As an instance, the
[2]. To that end, this situation should be detected timely to avoid power PLC method implemented in [6] detects the islanding condition using
quality (PQ) disturbances, a hazard for the network personnel and the the communication signals through the power distribution lines. In this

* Corresponding author.
E-mail address: alexandre.serrano-fontova@manchester.ac.uk (A. Serrano-Fontova).

https://doi.org/10.1016/j.ijepes.2022.108437
Received 3 March 2022; Received in revised form 22 May 2022; Accepted 15 June 2022
Available online 28 June 2022
0142-0615/© 2022 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
A. Serrano-Fontova and R. Bakhshi-Jafarabadi International Journal of Electrical Power and Energy Systems 143 (2022) 108437

Nomenclature Pm Turbine mechanical power


Tg Governor time constant
Abbreviations Γe Electromagnetic torque
CB Circuit breaker Tsm Servomotor time constant
DG Distributed generation ksm Servomotor gain
DN Distribution network Qt Turbine flow rate
GCPVS Grid-connected photovoltaic system g Governor gate opening
MPPT Maximum power point tracking JT Total moment of inertia
NDZ Non-detection zone Tw Water time constant
IDM Islanding detection method hbase Maximum water head
NDZ Non-detection zone gbase Rated turbine flow
PCC Point of common coupling L Penstock length
PQ Power quality A Penstock area
RES Renewable energy sources Hl Head loss
ROCOP/ROCOQ Rate of change of active/reactive Power ΔPM Change in mechanical power
ROCOV Rate of change of Voltage ΔPDIS Change in active power
SEIG Self-excited induction generator ΔP/ΔQ Active/Reactive power imbalance
THD Total harmonic distortion Δg Variation of governor gate opening
UVP/OVP Under/Over Voltage protection Δf Frequency deviation
UVF/OVF Under/Over Frequency protection df/dt Frequency derivative
HIM SEIG inertia
Variables ωo Natural angular frequency
ΔH Water head Pe Generator electrical power
ga Gravitational acceleration RF Fault resistance
Q Water discharge ωr Rotor angular velocity
ρ Water density ωs Stator angular velocity
ωm Turbine mechanical speed

method, a transmitter device sends a continuous low-voltage signal to Thereby, although the outstanding performance of the presented tech­
the microgrid or DG side using power line carriers equipped with a nique has been indicated for most islanding power mismatches, it has
receiver device. The underlying advantage of all these IDMs is their fast not been investigated for a full-balanced islanding condition. Moreover,
and reliable performance in all scenarios, as they only rely on the reli­ the expressions used for settings implementation have been defined
ability of communication channels disregarding the measured local without being proven by analytical expressions. Bekhradian, et al.
variables. Nevertheless, their implementation requires a high economic defined the rate of change of PCC’s resistance to the angular frequency
cost, making it less cost-effective, especially for small-scale DGs. of the synchronous-based-DG as an islanding indicator in [8]. When the
Passive IDMs are based on local measurements where variables are DG is islanded, this variable drops notably for a given timeframe, e.g.,
compared with predefined threshold setting(s) to identify potential 25 ms. The PSCAD/EMTDC software simulation results remarked that
variations in the state variables. According to the selected variables, the presented IDM detects islanding incidents except for a narrow
these IDMs can be further classified into three main typologies: time- ±0.2% of active power imbalances within 200 ms. However, the
domain, frequency-domain, and pattern recognition. The conventional threshold settings deemed in this work have been defined by conducting
time-domain protective functions are the rate of change of frequency numerous simulations, which makes this method highly dependent on
(ROCOF), the rate of change of voltage (ROCOV), the under/over the studied test system. Consequently, the approach might need to be re-
voltage protection (UVP/OVP), the under/over frequency protection assessed every time the system conditions are modified.
(UFP/OFP), and the rate of change of both active and reactive power Other recent works have taken advantage of the ROCOF as the main
(ROCOP/ROCOQ). In general, the low cost and easy implementation are passive-based indicator to tackle the effects of islanding operations by
the main advantages of these methods. However, as their effectiveness either synchronous or asynchronous generators in DNs [9,10]. Recently,
relies on the power imbalance between generation and load, they suffer several research studies have tried to mitigate the NDZ through
a large non-detection zone (NDZ), i.e., those cases with a small power frequency-based criterion or pattern recognition techniques, e.g., deci­
mismatch where the presented IDM fails to identify islanding. Further­ sion trees [9,10], multi-long short-term memory [11], continuous
more, some methods have set out very low threshold settings to identify wavelet transform and conventional neural network [12], combination
such slight deviations. Nevertheless, it may cause false operation during of various learning and signal processing algorithms [13], grey wolf
some non-islanding events that imply either voltage or frequency de­ optimised neural network [14], and variational mode decomposition-
viations, e.g., short-circuit faults, capacitor and induction motor based mode singular entropy [15]. Cui, et al. adopted the intelligent
switching, and transformer energisation. Therefore, tuning the optimum relay equipped with a decision tree algorithm [9]. The settings of the
threshold settings becomes a major challenge for passive-based IDMs to IDM with several binary conditions have been initially defined by using
avert false tripping in non-islanding incidents and minimum NDZ. The a set of local data in islanding and non-islanding events. The results
ROCOQ criterion has been exploited by Nikolovski, et al. in a 10 MW endorsed that the presented IDM classifies islanding and non-islanding
synchronous-based biomass DG [7]. If the DG injects reactive power into accurately with a small NDZ. The high dependency of the method’s
the grid (either capacitive or inductive) after a sudden disconnection settings on the studied DG/network and the need to perform large sets of
from the main grid, this would lead to reactive power imbalances that both offline and online simulations are known as the main limitation of
imply high ROCOQ values (either positive or negative according to the these IDMs. Hussein, et al. took advantage of several machine learning
sign of the imbalance). On the contrary, if this reactive power mismatch techniques essentially trained with datasets containing the total har­
is negligible and islanding occurs, the latter indicator will be near zero. monic distortion (THD) of both voltage and current signals [13]. The

2
A. Serrano-Fontova and R. Bakhshi-Jafarabadi International Journal of Electrical Power and Energy Systems 143 (2022) 108437

aftermath of this study clearly revealed that the harmonic content accurate and fast IDM with zero-NDZ for mini-hydro-based DGs is
observed during the transient occurred after the switching operation necessary for tackling the problem at hand. The time response of a
suffices to identify islanding in a synchronous-DG-based microgrid. turbine governor system to reach a desired gate opening level may take a
Admasie et al., however, broadened this research by investigating the few seconds, as reported in the available literature [29]. As such, it has
dependability of the optimised artificial neural network technique been neglected in previous islanding detection studies. In this paper
trained with the intrinsic mode function of voltage for islanding detec­ however, the governor has played a pivotal role in the islanding detec­
tion within a multi-DG scenario [14]. Despite the accurate and fast tion as it is used as an active tool in the second stage of the method.
islanding detection exhibited in such a study, the settings of these In this regard, this paper proposes a two-stage methodology for
methodologies highly depend on the studied system, thus having to detecting the islanding condition that may occur when the self-excited
launch additional time-domain simulations to train the AI-based algo­ induction generators (SEIGs) driven by mini-hydro turbines perfectly
rithm every time a new DG/network is to be considered. match local loads in the microgrid and are disconnected from the main
In active IDMs, a disturbance signal is injected into the control loop grid [23,24]. The first stage of the proposed algorithm uses ROCOF as
of the DG forcing the local variables to drift away from the thresholds the main passive-based indicator with minimum voltage supervision as
during islanding operation, i.e., either above or below. Conversely, its explained later. If the measured ROCOF at the DG(s) PCC exceeds such
effect is hardly appreciable in grid-connected mode since the local threshold, the algorithm modifies the power reference of the control
variables follow the voltage and frequency imposed by the grid [16]. loop, which changes the governor opening gate of the hydro-turbine.
Rabuzin, et al. showed that the rate of change of reactive power to the Due to this opening gate variation, the SEIG power output is shifted to
PCC voltage is inversely proportional to the PCC’s reactance which has a new set point. If the system is islanded, this active power change causes
different settings during islanding and grid-connected modes. This var­ an additional power imbalance between the DG(s) and local loads,
iable certainly drops sharply after island formation by injecting periodic wherein the established frequency and ROCOF thresholds of the second
disturbance to the reference set points in the governor control loop. stage are surpassed. On the contrary, these variables barely deviate in
Although the results highlighted the fast detection time and reliability of the grid-tied mode since the main grid strictly dictates voltage and fre­
this method, the thresholds are determined by numerous islanding and quency, and the DG returns to the initial point of operation.
non-islanding simulations. Moreover, if the reactive power mismatch is Even though the cornerstone and principal novelty of this work lie in
near zero due to the DG reactive power contribution (either capacitive or exploring the dynamics of the governor system for islanding detection
inductive), the sensitivity of the mentioned indicator is reduced, and purposes, other advantages of the proposed technique are as follows:
islanding may remain undetected. Albeit active techniques notably • Accurate detection of all possible islanding scenarios, i.e., zero
improve the NDZ, the injected disturbance increases the harmonic and NDZ.
subharmonic current content, implying a PQ degradation [17]. • Fast islanding detection within an average of 473 ms.
Finally, smaller NDZ and lower PQ degradation are achieved when • No false tripping in non-islanding disturbances.
these passive and active IDMs are combined in the so-called hybrid • No adverse effects on PQ as it does not inject any signal.
methods, e.g., a ROCOV setting followed by the connection of a parallel • Low implementation cost with a small degree of computational
inductance was used in [18], whereas a capacitance was deemed in [19] complexity.
if ROCOF is activated. In [19], for example, the reactor is connected at The rest of the paper is organised as follows; Section 2 details the
the main substation when the first stage determines suspicious islanding implemented SEIG and mini-hydro models as well as the DN test system
events. After that, islanding is categorised in case the second-stage under study. Section 3 describes the proposed hybrid IDM, scrutinises
passive criterion is exceeded. In this approach, high performance is the threshold selection process and discusses the zero-NDZ capability.
observed for both inverter-based and synchronous-based DGs. On the The performance of the suggested technique is then assessed and thor­
other hand, a modified active frequency drift has been considered in oughly analysed under numerous islanding and non-islanding events in
[17], where the islanding condition is detected by calculating the Section 4. Section 5 provides an in-depth comparison of the proposed
changing parameter that relates the change of the chopping fraction and method with the existing IDMs developed for synchronous and
the line frequency periodically. In addition, a novel voltage-based cri­ asynchronous-based DGs applications. A discussion regarding the con­
terion and disturbance injection into the maximum power point tracking tributions and the goals achieved in this paper is provided in Section 6.
(MPPT) algorithm has been presented in [20,21]. Having said the above, Finally, Section 7 summarises the main conclusions of this work.
in general, the bottleneck of these hybrid methods lies in setting the
optimal threshold of the passive-based part as well as tuning the optimal 2. Case study system description
size of the disturbance or local load used in the active-based stage.
Crucially, this threshold will determine the NDZ of these hybrid 2.1. Model of mini-hydro unit
methods.
The hydro power plant used in this paper is based on a turbine-
penstock aggregated model, which acts as a prime mover for the SEIG,
1.2. Contribution and paper structure as can be seen in Fig. 1. The control system is aimed at keeping the water
head at the desired level, thus acting as a governor adjusting the wicket
Based on the previous state-of-the-art review, it is seen that an

Fig. 1. Control loop of the SEIG. (a) Governor, (b) servomotor, and (c) set of penstock-hydro turbine.

3
A. Serrano-Fontova and R. Bakhshi-Jafarabadi International Journal of Electrical Power and Energy Systems 143 (2022) 108437

gate opening [25]. The mechanical power developed by the hydro tur­ applications [29]. The control system is composed of the governor and a
bine (Pm) is given by: servo motor. The governor is modelled as a time delay with constant Tg
as displayed in Fig. 1 (a), whilst the transfer function of the servomotor
Pm = ρ ga Q ΔH η (1)
is shown in Fig. 1 (b). The servomotor considers a time constant Tsm, gain
where, ρ, ga, Q, ΔH, η are the water density, gravitational accelera­ ksm, and both gate and speed opening limits. The governor is commonly
tion (9.81 m/s2), initial channel water discharge, water head, and the designed as a droop-based strategy for frequency-speed control in large
turbine efficiency coefficient, respectively [25]. The latter one is a hydropower plants engaged in frequency control commitment [29]. On
function of the turbine water governing system exploited for power the contrary, the grid-connected mini-hydro-based DGs are usually
regulation in most mini-hydro applications with Kaplan turbines [27]. operated in isochronous mode, where the water head level dictates the
Hence, it is supposed in this paper that the power is regulated through wicket gate opening required for the power regulation [25,28]. The
the governor guide vane system, and the angle of the turbine blades is dynamics of the set penstock-turbine can be defined by [27,29]:
kept fixed. The relationship between the guide vane angle and the
dQt (ΔH − Hl ) g A
developed mechanical torque is detailed in [27,28]. = (3)
dt L
The feedback loop designed for the active power control is depicted
in Fig. 1. As can be seen, the active power reference is dictated by the where, Qt is the turbine flow rate (m3/s), g is the governor gate
head water level, which also depends on the channel water discharge. opening, A is penstock area (m2), L is the penstock length (m), and Hl is
This methodology is widely adopted for mini-hydro-based DGs in real the head loss due to the penstock length (m). Equation (3) can be
applications [29]. The control system is composed of the governor and a expressed in per unit (pu) system as follows:
servo motor. The governor is modelled as a time delay with constant Tg dQt (ΔH − Hl )
as displayed in Fig. 1 (a), whilst the transfer function of the servomotor = (4)
dt TW
is shown in Fig. 1 (b). The servomotor considers a time constant Tsm, gain
ksm, and both gate and speed opening limits. The governor is commonly where, TW is the time constant of the water inertia, computed as
designed as a droop-based strategy for frequency-speed control in large follows:
hydropower plants engaged in frequency control commitment [29]. On ( )
L qbase
the contrary, the grid-connected mini-hydro-based DGs are usually TW = (5)
A g × hbase
operated in isochronous mode, where the water head level dictates the
wicket gate opening required for the power regulation [25,28]. The where, qbase and hbase represent the rated turbine flow and the
dynamics of the set penstock-turbine can be defined by [27,29]. By maximum water head level, respectively. If the friction losses along the
considering that both the mechanical speed (ωm) and the water head penstock are neglected, a linear model from Eq. (3) is obtained as in Eq.
(ΔH) remain unchanged, the torque applied to the hydro turbine (Γ m) (6). Hence, small variations in the guide vane opening gate (Δg) are
depends on the channel water discharge as follows: traduced into changes in the mechanical power (ΔPm) as follows:

ρ ga Q ΔH η ΔPm (s) 1 − sTW


Γm = (2) = (6)
ωm Δg (s) 0.5 + sTW
The mini-hydro system is composed of the main chamber, a short Further, the dynamic equation of the SEIG can be given as follows:
penstock, the governor system, the servo motor, and the Kaplan turbine
dωm
[26]. The prime mover output power in Kaplan turbines can be regu­ JT = (Γe − Γm ) (7)
dt
lated via the governor system taking advantage of both the wicket gate
and turbine blade angles. Still, the turbine guide vane is the only gov­ where, JT is the total mass moment of inertia considering the set
erning system exploited for power regulation in most mini-hydro ap­ SEIG-turbine flywheel and Γe is the developed electromagnetic torque
plications with Kaplan turbines [27]. Hence, it is supposed in this paper by the SEIG, expressed in terms of the active power (Pe):
that the power is regulated through the governor guide vane system, and Pe
the angle of the turbine blades is kept fixed. The relationship between Γe = (8)
ωs − ωr
the guide vane angle and the developed mechanical torque is detailed in
[27,28]. where, the angular velocity of both rotor and the stator are denoted
The feedback loop designed for the active power control is depicted by ωr and ωs, respectively. The SEIG has been modelled through the fifth-
in Fig. 1. As can be seen, the active power reference is dictated by the order model for transient studies [30]. The control system parameters of
head water level, which also depends on the channel water discharge. the 160 kVA case study mini-hydro unit are listed in Table 1.
This methodology is widely adopted for mini-hydro-based DGs in real

Table 1
Detailed parameters of the studied system.
SEIG Electrical Data

Electrical specification 160 kVA (215 HP), 2-poles, 0.4 kV rated line to line voltage, and 50 Hz frequency
Inertial constant 2.9 kgm2 (asynchronous inertia of 0.22 s)
Stator impedance 0.077 + j 0.477 (Ω)
Rotor impedance 0.0139 + j 0.477 (Ω)
Magnetizing reactance j 2.481 (Ω)
Hydro DG data
Set Penstock-Turbine L = 10 m, A = 1.5 m2, Qt = 2–5 m3s− 1, ΔH = 4 m, η = 80–92% (depends on Qt), TW = 0.5 s
Control loop of governor and servomotor Tsm = 0.07 s; Ksm = 2 (servomotor); Tg = 0.2 s (governor); gMIN = 0.95 pu; gMAX = 0.05 pu (servomotor position limits); gspeed=± 0.01pu/s
(servomotor speed limits)
Line Parameters and Transformers data
Transmission and distribution lines (TN L1 (TN): Z1 ¼ 0.012 + j0.29, Z0 = 0.38 + j1.17 (Ω/km), C/C0 = 12.7 × 10-9 / 7.7 × 10-9 (F/km)
and DN) L2 (TN): Z1 ¼ 0.687 + j0.416 (Ω/km), Z0 = 3.2 + j1.660 (Ω/km)
Transformers TR1: 10 MVA, 230/120 kV, ε = 10 (%), Wye/Delta; TR2: 10 MVA, 120/25 kV, ε = 10 (%), Wye/Delta;
TR3: 0.4 MVA, 0.26/25 kV, ε = 6 (%), Delta/Wye

4
A. Serrano-Fontova and R. Bakhshi-Jafarabadi International Journal of Electrical Power and Energy Systems 143 (2022) 108437

2.2. Test system under investigation large-scale load-generation imbalances, the DN object of study hangs
from a modified version of the IEEE 9-Bus test system (Fig. 2 (a)). In this
In this subsection, the test system under study is detailed. In order to regard, the 9th bus test system is connected to a 100 km sub-
evaluate the proposed hybrid IDM during frequency events caused by transmission line through a step-down 230:120 kV transformer (TR1).
At the end of the line between buses 10 and 11, the step-down 120:25 kV
transformer (TR2) feeds the medium voltage network at bus 12. As
shown in Fig. 2, the test system includes two parallel medium voltage
feeders named A and B. The single-line diagram of the whole system is
displayed in Fig. 2 (b). The islanding object of study occurs either with
the operation of the CB DG at bus 13 or the CB at the head of the Feeder
A in Fig. 2 (b). The SEIG is a squirrel cage induction generator equipped
with a capacitor bank, a common configuration in mini-hydro units, see
[23,24]. This DG is connected at the medium voltage Bus 13 of the
Feeder A through an RL series impedance and a step-up 0.4:25 kV
transformer (TR3). The parameters of the line impedances and trans­
formers displayed in Fig. 2 (b) are included in Table 1. The sub­
transmission line (named L1) has been modelled as a pi-section, whereas
the medium voltage lines, represented by L2, are modelled as series RL
impedances.

3. Proposed islanding detection method

This section describes the proposed IDM, presents the analytical


expressions used for determining the threshold settings and discusses the
NDZ based on the theoretical analysis.

3.1. Description of the proposed hybrid algorithm

The proposed two-stage IDM takes advantage of a first-stage passive-


based indicator and subsequent DG power curtailment, as illustrated in
Fig. 2. Test system. (a) A modified version of the 9-Bus IEEE test system, (b) the flowchart in Fig. 3. The proposed technique initially focuses on data
DN object of study where islanding occurs. mining to obtain the local measurements at the DG interconnection bus,

Fig. 3. Flowchart of the proposed IDM.

5
A. Serrano-Fontova and R. Bakhshi-Jafarabadi International Journal of Electrical Power and Energy Systems 143 (2022) 108437

i.e., bus 13 of the system displayed in Fig. 2 (a). Then, the algorithm the islanded microgrid. An imbalance between the generation and load
recognises suspicious islanding events whenever the absolute ROCOF (| in grid-connected mode does not cause significant frequency oscilla­
ROCOF|) exceeds the 0.001 Hz/s threshold with a time delay of 50 ms, tions. On the contrary, once the SEIG is islanded, the rotor mechanical
which implies that the first passive-based stage in Fig. 3 outputs a true speed dictates the frequency of the islanded microgrid [33]. Hence, the
signal. The absolute value of the frequency derivative has been adopted speed deviation (Δω) during islanding is derived by:
as variations can be either positive or negative after islanding formation
ΔP ω0
according to the sign of the power mismatch. The PCC voltage super­ Δω = ω − ω0 = t (11)
2HIM
vision is taken into account during this process to avoid false tripping
during non-islanding voltage sags and provide fast tripping during where, ω is the angular frequency during islanding. The frequency
islanding scenarios with large reactive-power imbalances, see the deviation (Δf) can also be given by:
comprehensive analysis about voltage protection in [31]. A 50 ms delay
ΔP f0
is also considered in the under-frequency setting to evade false tripping Δf = t (12)
2HIM
during the non-islanding three-phase faults. A justification of how these
time delays have been tuned is provided in the following subsection. Hence, the time to reach a given Δf threshold during the islanding
According to the flowchart in Fig. 3, if the first stage is activated, the operation can be quantified as follows [32]:
reference of the power control loop is modified in Fig. 1, shifting the 2ΔfHIM
turbine governor gate opening to a new state. Consequently, the me­ t= (13)
ΔP f0
chanical torque applied to the prime mover and the SEIG output power
are modified accordingly. The power disturbance (ΔPDIS) has been set Similarly, the frequency derivative can be computed for a certain
out as 0.1 pu and is introduced into the control loop via the bypass time interval as in Eq. (14):
switch shown in Fig. 1. The next subsection will elucidate an in-depth df ΔPf0
analysis of how this value has been selected and its effects on the = (14)
dt 2HIM
islanding detection.
If islanding occurs when there is a high reactive/active power In view of the interdependency between the threshold settings and
imbalance, voltage changes significantly, and islanding is therefore their effects on the IDM performance observed in the above equations, a
identified. In critical islanding scenarios with small power imbalances, sensitivity analysis has been conducted. The outputs of this analysis are
the power curtailment leads to |ROCOF| greater than 1 Hz/s and abso­ displayed in Fig. 4, where islanding has been simulated for the studied
lute frequency deviation (|Δf|) greater than 0.5 Hz. As illustrated in system under various voltage-dependent load model compositions, i.e.,
Fig. 3, the second stage of the proposed IDM detects such events and constant impedance, current, and power. A dashed line has been added
trips the DG. to highlight the established ROCOF threshold in these figures. The active
The threshold settings of the second stage are likely to be surpassed power imbalance in each scenario has been increased by 0.003% load
in less than 1 s after the introduced disturbance. Contrarily, if the set of steps, choosing the zero-power imbalance as a base case while keeping
thresholds in the second stage is not surpassed within 1 s after the DG the reactive power imbalance fixed to zero, as it is precisely when the
curtailment, the proposed IDM classifies such events as a non-islanding, generator can become self-excited [24]. The provided outputs endorse
returning the bypass switch in Fig. 1 to its initial position. successful islanding classification considering the predefined threshold
The 1 s time delay set out in this algorithm has been selected to setting in all cases regardless of the implemented load type and power
comply with the maximum disconnection time suggested by the IEEE imbalance, e.g., see the third plot of each column in Fig. 4 which mea­
Std. 1547–2018 [4], while allowing sufficient time for the control sys­ sures power at the PCC.
tem to drift away from the ROCOF and Δf threshold settings included in Since the generator inertia is closely intertwined with the islanding
the second stage. detection according to Eqs. (13) and (14), the effectiveness of the pro­
posed technique under three generator sizes has been tested in a set of
simulations whose results are shown in Fig. 5. As expected from such
3.2. Optimum selection of threshold settings equations, the time to reach a particular frequency threshold increases
for a given imbalance as inertia increases. In this analysis, the consid­
According to the flowchart illustrated in Fig. 3, some thresholds are ered SEIG sizes are larger than those in the system object of study. More
included in the proposed two-stage IDM. This section presents the concretely, 1, 5, and 10 MVA-sized SEIGs have been selected with 0.45,
analytical expressions to optimally determine these settings. Addition­ 0.68, and 0.9 s of asynchronous inertia, respectively. It is observed that
ally, several sensitivity analyses have been undertaken to prove the given the thresholds adopted in the second active-based stage of the
optimality of the selected ones and explore the impact of the turbine- proposed IDM, the islanding condition would be appropriately identified
governor dynamics in the performance of the proposed method in Fig. 1. even for larger DGs. However, it is worth pointing out that the DG size is
The dynamic behavior of the SEIG is expressed as follows regarding commonly selected according to the rated active power of the prime
its swing equation: mover, see the detailed survey in [25].
2HIM dω As seen in Eq. (4), the water inertia constant TW included in the
= (Pm − Pe ) (9) turbine model can potentially affect the power developed by the prime
ω0 dt
mover, given a change in the governor system. Therefore, as the pro­
ΔP = Pm − Pe (10) posed method uses the prime mover for islanding detection purposes,
the influence of this parameter must be considered.
where, HIM and ω0 are the SEIG inertia constant, i.e., asynchronous It is seen from Fig. 6 that for higher TW values, large peak power is
inertia [30], and natural angular frequency, respectively [32]. Also, ΔP reached during the transient response following a change of 0.12 pu in
is the power mismatch between the SEIG generation and the demands in

6
A. Serrano-Fontova and R. Bakhshi-Jafarabadi International Journal of Electrical Power and Energy Systems 143 (2022) 108437

Fig. 4. Sensitivity analysis of the ROCOF setting of the first stage.

Fig. 5. Sensitivity analysis of df/dt for several SEIG sizes.

Fig. 6. Influence of the time constant of the water inertia in the active power transient response.

7
A. Serrano-Fontova and R. Bakhshi-Jafarabadi International Journal of Electrical Power and Energy Systems 143 (2022) 108437

Fig. 7. Effects of different power reductions in the control loop.

Table 2 concluded that the effect of this variable on the proposed technique is
Islanding detection times for different DG power curtailment. hardly noticeable.
As mentioned above, if the first stage outputs a true signal, the power
Active Power reduction Load composition tΔf* tROCOF**
(%) (Cnt. Z, Cnt. P, Cnt. I) (ms) (ms) reference is shifted, introducing an additional mismatch if the system is
islanded. Consequently, the value of the total imbalance will depend on
10 (100%,0%,0%) 316 217
20 (100%,0%,0%) 140 179 the initial imbalance plus the added one. Given the cruciality of ΔPDIS in
30 (100%,0%,0%) 95 133 the performance of this IDM, a sensitivity survey is conducted where
* three power curtailments are investigated considering a balanced island.
It represents the time instant when the frequency threshold of the second
The composite load modelling considered throughout this assessment is
stage is exceeded.
**
It represents the time instant when the ROCOF threshold of the second stage a constant impedance as recommended by the IEEE Std. 1547–2018 [4].
is exceeded. The system is islanded at t = 8.5 s for all these cases. As expected from
Eq. (13), given the SEIG size (HIM) and ΔP, the established thresholds in
the wicket gate position. This transient peak power is traduced into the second stage are reached for different instants in time. The time
larger df/dt and Δf according to Eqs. (12)–(14), thus influencing the instants when those thresholds are surpassed have been extracted from
detection time. This value is relatively low in mini-hydro DGs units due the time-domain simulations in Fig. 7 and listed in Table 2, where the
to short penstocks and reduced discharge levels, so it can be safely percentage of these changes in active power is computed with respect to
the pre-islanding point of operation. It is worth mentioning that even

Fig. 8. Influence of the servomotor gain in the dynamic response.

8
A. Serrano-Fontova and R. Bakhshi-Jafarabadi International Journal of Electrical Power and Energy Systems 143 (2022) 108437

though the change in the power reference is subtracted, as seen in Fig. 1, feedback loops, power reductions and DG sizes.
i.e., curtailing DG power, due to the dynamics of the control system and
its parameters, the system experiences a transient power increase. This
3.3. Discussion about the zero-NDZ capability
effect is observed between t = 8.5 s and t = 9.2 s in Fig. 7, where the
torque developed by the SEIG transiently soars. As expected from the
As can be seen in Fig. 3, the first stage of the proposed IDM uses as a
theoretical analysis, the thresholds of the second stage are surpassed in a
passive-based indicator to modify the reference of the turbine control
smaller time interval as ΔPDIS increases. Essentially, the percentage of
feedback loop. Consequently, it forces the state variables to drift away
active power curtailment is defined as a trade-off between the active
from the thresholds established in the second stage. If it can be proven
power degradation and the detection time.
that the algorithm does not miss any event, then the optimality of the
By using the detection times in Table 2, the difference between 30%
thresholds implemented in the first stage is demonstrated, which in turn
and 10% in the ΔPDIS can be readily computed, 218 ms and 138 ms for
implies that the method has no NDZ. To fulfil such achievement, the
the Δf and ROCOF thresholds of the second stage, respectively. Even
constraints in the first stage must be met, and the pre-islanding power
though larger ΔPDIS could, at first sight, seem the most reliable option,
flowing through the PCC be negligible.
this would imply a much larger voltage and frequency during the island,
According to Eqs. (13) and (14), the NDZ of the proposed IDM can be
therefore taking a large time to reach the pre-event point of operation in
theoretically defined by a range of relative active power imbalance (ΔP/
non-islanding events. With this reasoning in mind, ΔPDIS has been set as
PDG), given the SEIG characteristics. For the particular test system under
10%, implying a milder effect in voltage and frequency whilst achieving
investigation with the parameters listed in Table 1, for example, the
reasonable detection times.
theoretical NDZ falls within a narrow ±8.8 × 10-4 % range, which is
The fourth sensitivity analysis focuses on the dynamic response of
traduced into a few hundreds of mW. The simulation outputs of cases 10
the servomotor as it is also a paramount factor in the control loop. The
and 11 with active power imbalance of 0.4 W and − 1.4 W, respectively,
results displayed in Fig. 8 are aligned with those in [26,27], showing
are shown in the third row of plots in Fig. 4. These outputs clearly
that for a step change of 0.1 pu in the opening gate, the system takes
endorsed that those values match the theoretical percentage, but the |df/
approximately 4 s. And yet still, the initial transient response introduces
dt| set out in the first stage is low enough to be surpassed, e.g., the
an additional imbalance in the islanded microgrid regardless of the total
derisory 0.4 W/-1.4 W range in cases 10 and 11, giving imbalances in the
settling time required to reach the desired gate opening value. It is
order of 1 × 10-4 %. To get a better view of these outcomes, Fig. 9 has
precisely because of this transient response observed in the active power
zoomed into these plots. The plotted waveforms in this figure belong to
that the proposed IDM identifies the islanding operation in a relatively
cases 1 through 12 in Table 3, where the load has been increased by
short time. The time constants and gains used in this paper have been
steps of 5 W, which over a 160 kW DG rated active power, it represents a
obtained from [26].
0.003% load increase. For example, the purple line in the third column
Finally, it is seen from the presented analyses that the detection time
in Fig. 9 stands for the power measured at the PCC in case 10 and give a
could slightly vary depending on the system parameters, i.e., gains and
value of 0.45 W. Realistically, this percentage of mismatch is assumed to
time delays in the control loop. However, as mentioned above, regard­
be zero as it has been demonstrated that any other scenario in between
less of the total time required to reach the desired gate opening, the
would surpass the implemented 0.001 Hz/s ROCOF setting. It is worth
proposed IDM takes advantage of the transient response to introduce an
mentioning that the analytical value obtained through Eqs. (13) and
additional imbalance which pushes the state variables away from the
(14) is based on a full constant power load model as the term ΔP does not
established thresholds in the second stage. Hence the presented tech­
consider the voltage dependency, leading then to slightly different
nique classifies all islanding scenarios timely and accurately for a wide
outcomes when such models are considered.
range of scenarios, including several load models, different gains in the
Given that all ranges of power mismatch are being identified, the

Fig. 9. High resolution of the sensitivity analysis shown in Fig. 4.

9
A. Serrano-Fontova and R. Bakhshi-Jafarabadi International Journal of Electrical Power and Energy Systems 143 (2022) 108437

Table 3
Islanding events with the custom-made distribution network test system.
Case Event description ΔP/PDG ΔQ/PDG Load composition* Detection time
no. (%) (%) (% Cnt. Z, % Cnt. P, % Cnt. (ms)
I)

1 Negligible active and reactive power mismatches and various composite load models − 1.3 × 10- − 1.6 × (70%,0%,30%) 729
3
(Opening CB-DG at t = 2 s) 10-2
2 8.5 × 10-4 − 1.6 × (70%,0%,30%) 862
10-2
3 − 2.8 × 10- − 1.6 × (70%,0%,30%) 559
3
10-2
-4
4 − 4 × 10 − 1.6 × (70%,0%,30%) 833
10-2
-
5 − 1.5 × 10 − 6.6 × (100 %,0%,0%) 439
2
10-3
6 − 9.5⋅10-3 − 6.6 × (100 %,0%,0%) 443
10-3
7 − 3.3 × 10- − 6.6 × (100 %,0%,0%) 552
3
10-3
-3
8 2.8 × 10 − 6.6 × (100 %,0%,0%) 451
10-3
9 − 2.1 × 10- − 4 × 10-3 (27%,0%,73%) 584
2

10 − 5.4 × 10- − 4 × 10-3 (27%,0%,73%) 453


3
- -3
11 − 8.3 × 10 − 4 × 10 (27%,0%,73%) 451
3
-4 -3
12 9.6 × 10 − 4 × 10 (27%,0%,73%) 447
13 Various reactive power mismatches with constant impedance and current load model 8.5 × 10-4 2 (70%,0%,30%) 440
14 (Opening CB-DG at t = 2 s) 8.5 × 10-4 − 2 (70%,0%,30%) 472
15 8.5 × 10-4 5 (70%,0%,30%) 423
16 8.5 × 10-4 − 5 (70%,0%,30%) 449
17 8.5 × 10-4 10 (70%,0%,30%) 400
18 8.5 × 10-4 − 10 (70%,0%,30%) 537
19 8.5 × 10-4 20 (70%,0%,30%) 364
20 8.5 × 10-4 − 20 (70%,0%,30%) 504
21 Different active power mismatches and load types 2 1.4 × 10-2 (25%,35%,40%) 472
22 (Opening CB-DG at t = 2 s) − 2 1.4 × 10-2 (25.1%,34.9%,40%) 371
23 5 1.4 × 10-2 (30.9%,20%,49.1%) 585
24 − 5 1.4 × 10-2 (25%,35%,40%) 301
25 10 1.4 × 10-2 (25.1%,34.9%,40%) 792
26 − 10 1.4 × 10-2 (30.9%,20%,49.1%) 255
27 15 1.4 × 10-2 (25%,35%,40%) 203
28 20 1.4 × 10-2 (25.1%,34.9%,40%) 169
29 Islanding after CB fault clearance 8.5 × 10-4 − 1.6 × (70%,0%,30%) 449
(Opening CB-Feeder A at t = 2 s) 10-2
30 8.5 × 10-4 − 1.6 × (70%,0%,30%) 444
10-2
-4
31 8.5 × 10 − 1.6 × (70%,0%,30%) 647
10-2
32 Different active power mismatches and load types with X/R = 5, 7.5, and 10 (Opening 8.5 × 10-4 − 1.6 × (70%,0%,30%) 862
CB-DG at t = 5 s) 10-2
33 8.5 × 10-4 − 1.6 × (70%,0%,30%) 862
10-2
-4
34 8.5 × 10 − 1.6 × (70%,0%,30%) 862
10-2
*
Load composition: It represents the share of each type of load according to the voltage-dependence load modelling [20].

optimality of the deemed threshold settings in the first stage has been 4.1. Various local load composition with negligible power mismatch
endorsed, in turn proving the feasibility of the zero-NDZ capability.
In this part, the capability of the proposed IDM to identify islanding
4. Simulation results events is investigated where solely the SEIG-DG is considered. This set of
simulations and scenarios includes various active/reactive power mis­
In this section, the performance of the proposed hybrid technique has matches (ΔP and ΔQ) and voltage-dependent load models, i.e., constant
been evaluated and thoroughly analysed under several islanding and impedance (Z), constant current (I), and constant power (P).
non-islanding scenarios for two principal case studies. The initial anal­ These case studies and their associated results have been presented in
ysis considers the test system described in Section 2.1 with a single DG Table 3. The initial study entails cases 1 through 12, including islanding
scenario. The studied test system is then developed by adding a grid- scenarios with negligible active and reactive power imbalances,
connected PV system (GCPVS), thus accounting for a multi-DG case. considering different voltage-dependency load models. The time-
The latter scenario explores the interaction between both types of DGs domain simulation results of cases 4 and 5 are depicted in Figs. 10
when the proposed IDM is implemented. and 11, respectively. This figure illustrates a comparison between the
obtained results, including frequency, df/dt, voltage, and the tripping
signal of the CB with and without employing the proposed methodology.
It is seen from both Table 3 and Figs. 10 and 11 that the proposed

10
A. Serrano-Fontova and R. Bakhshi-Jafarabadi International Journal of Electrical Power and Energy Systems 143 (2022) 108437

Fig. 10. Simulation results for case 4 in Table 3.

Fig. 11. Results obtained for case 5 in Table 3 with constant impedance load model.

IDM identifies all islanding cases correctly regardless of the active power 4.2. Different reactive power imbalances
imbalance. The detection time for trained cases with positive power
imbalances, i.e., cases 21, 23, and 25, is larger than those with the same Although the self-excitation condition of the SEIG occurs when the
absolute power mismatch yet negative. The transient peak that occurred capacitors perfectly match its reactive power [24], i.e., thus having a
after the change in the control loop reduces the initial power mismatch zero-reactive power imbalance, some reactive imbalances have also
in these positive imbalance scenarios, increasing their detection time. been considered for testing the reliability and detection times of the
On the contrary, for those cases with an initial negative mismatch, such IDM. To that purpose, cases 13 to 20 summarise the details of the
as cases 22, 24, and 26, the imbalance increases after the change, simulated scenarios with various reactive power imbalances and zero
thereby implying larger Δf and ROCOF and lower detection time. In ΔP. The results of the time-domain simulations obtained in case 13 in
accordance with the presented outputs in Table 3, it has been demon­ Table 3 are displayed in Fig. 12.
strated that the proposed IDM identifies all islanding cases correctly, As expected from these events, the lack of reactive power is directly
even for negligible power mismatches. traduced into a slight voltage deviation, which is adequately identified

11
A. Serrano-Fontova and R. Bakhshi-Jafarabadi International Journal of Electrical Power and Energy Systems 143 (2022) 108437

Fig. 12. Outputs for case 13 in Table 3 with negligible ΔP/PDG and ΔQ/PDG = 2%.

by the proposed IDM in less than 537 ms for the most unfavorable case. 4.4. Multi-DG scenario
Given that the CB feeder A could be operated after a fault occurs
within this feeder, cases 29 to 31 in Table 3 have considered that the The performance of the proposed algorithm is assessed in a multi-DG
whole medium voltage portion of the grid can be islanded. The outcomes system by adding a 100 kW GCPVS at bus 13, where the interaction
of such casuistry reveal the robust performance of the proposed IDM between both DGs is the pivotal point. The GCPVS is a voltage source
algorithm during these events. converter (VSC)-based unit equipped with an incremental conductance
MPPT algorithm with 1.98 kHz frequency in buck-boost converter. The
DC/AC conversion and control is carried out through a 5 kHz three-
4.3. Influence of grid strength-several X/R ratios bridge VSC-based averaged model. In this DC/AC converter, the d-axis

The proposed technique has assessed the strength of the grid with
several X/R impedance ratios. To assess that capability, the impedance Table 4
ratio of the main source has been varied. That is to say, bus 11 has now GCPVS characteristics.
been switched from the IEEE 9-bus test system to a medium voltage Element Description
infeeder. Hence, the X/R ratio of that infeeder does dictate the strength PV Array Parallel Strings: 330; Modules per string: 5; Voc = 64.6 V; Isc
of the grid. From the results illustrated in Fig. 13, it is observed that such (Model SunPower- = 5.96 A; VMPPT = 54.7 V; IMPPT = 5.58 A; PDG = 100 kW;
a feature does not have noticeable consequences in the operation of this 315) QDG = 0 kVAr; DC link voltage = 500 V
IDM. The previously presented results have widely demonstrated that DC/AC converter VDC regulator gains: Kp = 7; Ki = 800; Current regulator
(Controller gains: Kp = 0.3; Ki = 20; Rf = 2 mΩ; Lf = 6 mH (RL filter);
the proposed IDM identifies all islanding scenarios irrespective of where
features) CDC = 0.1 mF; Rated phase-to-phase voltage = 260 V
the DGs are located and the upstream impedances.

Fig. 13. Comparison of cases 32 to 34 in Table 3 with different X/R ratios.

12
A. Serrano-Fontova and R. Bakhshi-Jafarabadi International Journal of Electrical Power and Energy Systems 143 (2022) 108437

Table 5
Islanding detection time for the multi-DG scenario.
Case no. GCPVS Power share ΔP/PDG ΔQ/PDG Load composition Detection time
(%) (%) (%) (Cnt. Z, Cnt. P, Cnt. I) (ms)

1 25 1.5 × 10-2 − 1.6 × 10-2 (87%,13%,0%) 573


2 50 9.5 × 10-3 − 9.1 × 10-3 (86%,14%,0%) 645
3 75 4.2 × 10-2 − 6.5 × 10-3 (93%,7%,0%) 775

current component depends on the DC bus voltage error, and the q-axis 4.5. Non-islanding events
component is set to zero. Therefore, the DG operates at a unity power
factor, a commonly adopted DGs solution in grid-connected applications This section focuses on testing the capabilities of the proposed al­
[2,21]. The details of the GCPVS-based are listed in Table 4, where the gorithm under non-islanding scenarios when voltage and frequency
proportional (P) and integral (I) gains of the PI controllers are shown by oscillations occur, although the mini-hydro DG is not islanded. The
Kp and Ki, respectively [12]. The islanding is simulated by operating the features of the simulated non-islanding events are summarised in
CB at bus 13, where both types of DG units and loads are isolated from Table 6. Besides, the time-domain results of the non-islanding short-
the main grid. In order to properly evaluate the interaction between circuit faults in cases 1–8 are illustrated in Fig. 15. Additionally, the
DGs, any islanding detection functionality has been implemented in the authors have investigated the dependability of the proposed IDM with
GCPVS. The results of this assessment are summarised in Table 5, which three-phase high impedance faults in cases 9 and 10 of Table 6, where
includes three different PV penetration levels. To unnecessarily extend fault resistances (Rf) of 100 Ω and 150 Ω have been simulated,
the length of the manuscript, only the detection time and the time-
domain waveforms of case 3 in Table 5 are displayed in Fig. 14. The
results for SEIG and GCPVS have been illustrated in Fig. 14 (a) and (b), Table 6
respectively. As can be seen from the third plot of both columns in Non-islanding events and results description.
Fig. 14, a PV share of 75% over a total load of 130.385 kW and 45.725 Case no. Event description* Time duration (s)
kVAr is considered, so both the active and reactive power imbalances are 1 LLL Fault with (Rf = 0 Ω) 0.1
4.2 × 10-2 and − 6.5 × 10-3 %, respectively. 2 LLL Fault with (Rf = 10 Ω) 0.1
As is shown in the presented outputs for the modelled mini-hydro 3 SLG Fault with (Rf = 0 Ω) 0.1
4 SLG Fault with (Rf = 10 Ω) 0.1
unit, the df/dt threshold setting deemed in the first stage of the pro­
5 LLG Fault with (Rf = 0 Ω) 0.1
posed IDM is exceeded 170 ms after the islanding occurrence, causing 6 LLG Fault with (Rf = 10 Ω) 0.1
the operation of the passive-based stage in Fig. 1. After this step, the 7 LL Fault with (Rf = 0 Ω) 0.1
active power reference of the control loop is modified accordingly. 8 LL Fault with (Rf = 10 Ω) 0.1
Consequently, the state variables exceed the second stage threshold 9 LLL Fault with (Rf = 100 Ω) 0.1
10 LLL Fault with (Rf = 150 Ω) 0.1
settings 604 ms after the change in the power reference is ordered. The
11 IM starting (250 HP/4 Pole/IM) 0.4
proposed hybrid IDM eventually identifies the islanding condition 12 100 kW load connection with PF = 0.85 0.2
within 774 ms. 13 Frequency event 60
On the other hand, if the GCPVS share is decreased to 50% and 25% 14 Frequency event 60
15 Frequency event 60
and both active and reactive power imbalances in the microgrid are kept
16 Frequency event 60
fixed, the detection time decreases to 645 ms and 573 ms, respectively, 17 Frequency event 60
as the SEIG becomes the predominant source.
* SLG = Single-line to ground fault, LL = Line-to-line fault; LLL = Three-phase
fault; LLG = Two-phase to ground fault; LLLG: Three-phase to ground fault.

Fig. 14. Measurements obtained in case 3 of Table 5 with a multi-machine DG scenario. (a) SEIG, (b) GCPVS.

13
A. Serrano-Fontova and R. Bakhshi-Jafarabadi International Journal of Electrical Power and Energy Systems 143 (2022) 108437

system shown in Fig. 2 (a). The outcomes of the time-domain simula­


tions during events 11 through 15 in Table 6, are depicted in Fig. 16. The
system experiences a certain Δf and ROCOF change according to these
large-scale imbalances between generation and load. A detailed expla­
nation of the features and origin of these events can be found in the
report undertaken by the European association of transmission opera­
tors (TSOs) [34].
If one focuses on the second plot of the first column in Fig. 15, will
note that frequency suddenly drops below 48.5 Hz and takes a few ms to
recover and step up above the established threshold in the second stage,
i.e., 49.5 Hz. Therefore, adding an intentional 50 ms time delay seems
reasonable to prevent maloperation in such cases.
To sum up, this setting is a trade-off between assuring that the al­
Fig. 15. Performance of the proposed IDM during non-islanding short cir­ gorithm does not missoperate during solid three-phase faults, i.e., LLL
cuit faults. Fault with 0 Ω resistance, without unnecessarily extending the islanding
detection process.
respectively.
As a result of the lower voltage and frequency oscillations during 5. Comparison with other existing islanding detection methods
these faults, the algorithm has easily classified such events as non-
islanding. Based on the above, it has been proven that all ranges of This section compares the proposed methodology with several
possible voltage sag magnitudes expected at the PCC between 0 pu (Rf = recently published IDMs used in scenarios with either synchronous,
0 Ω) and 0.90 pu (Rf = 10 Ω) have been covered regardless of the asynchronous-based DGs or a combination of both. In this context, a
detailed comparison considering the most relevant features in IDM
location and/or fault resistance.
In addition, since this method uses both frequency and ROCOF set­ studies is summarised in Table 7, whereas an individualised discussion is
provided as follows:
tings to identify the islanding condition, the non-islanding frequency
events of Table 6 are essential to assess its dependability. These fre­
quency events have been simulated by reducing the mechanical power • Advanced passive-based IDMs in [9,11,31,32,35,36] have been
of the second generator of the modified version of the 9-bus IEEE test presented to mitigate the NDZ of the conventional passive relays by
exploiting frequency and ROCOF threshold settings. However, all
methods have shown some NDZs, as listed in Table 7. On the con­
trary, the proposed technique identifies the islanding condition in
the most challenging scenarios, including cases with zero power
mismatch. To highlight the effectiveness of the implemented
threshold settings, case 6 in Table 3 is used for comparison purposes,
where the df/dt threshold of the proposed method’s first stage and
those used in [36,37] are plotted in Fig. 17.
• Although pattern recognition-based IDMs have shown a robust per­
formance in islanding and non-islanding scenarios, their imple­
mentation is complex and costly [9-12,35-37]. On the contrary, the
structure of the recommended IDM is simple and can therefore be
implemented with minimum cost.
• The insertion of different impedances over suspicious islanding
Fig. 16. Results of under frequency-based non-islanding events. events have been adopted in [18,19]. Even though the capacitor-

Table 7
Comparison of the proposed IDM with other existing methods.
Methodology NDZ Degrade Considers non-islanding/frequency Degree of practical Computational Detection time
[ΔP/PDG]/[ΔQ/PDG] (%) PQ? events? relevance complexity (s)

Ref. [7] NP* No No/No High Low ≈0.15


Ref. [8] [±0.2]/[±0.2] No Yes/No Medium Medium 0.2
Ref. [9] [0.17, − 0.05]/[0.1, No Yes/No High Medium NP*
− 0.04]
Ref. [10] Nearly Zero No Yes/No Medium High 0.02
Ref. [11] [±0.5] No Yes/No High High 0.05
Ref. [12] NP No Yes/No Medium High 0.21
Ref. [13] Almost Zero No Yes/No Medium High 0.436
Ref. [14] NP No Yes/No Medium High 0.083
Ref. [15] Nearly Zero No Yes/No Medium Low 0.01
Ref. [18] NP No Yes/No High Medium 0.36–0.5
Ref. [19] [8,-10]/[22,-10] No Yes/No High Low 0.3
Ref. [22] Zero Yes Yes/No Medium Medium NP
Ref. [31] [10-29] No Yes/No Medium Medium less than2
Ref. [35] [±2.5]/[±2.5] No Yes/No Medium High 0.08–0.2
Ref. [36] [±0.047]/[±0.5] No Yes/No High Medium 0.1
Ref. [37] [±1]/ [±1] No Yes/No Medium Medium 0.1–0.154
Ref. [38] [0.32]/[0.32] No Yes/No High Medium 0.136
Proposed Zero No Yes/Yes High Low 0.473

*NP: Not Provided.

14
A. Serrano-Fontova and R. Bakhshi-Jafarabadi International Journal of Electrical Power and Energy Systems 143 (2022) 108437

Fig. 17. Comparison of ROCOF thresholds with recently published methods [36,37].

based insertion method in [19] proved very effective, it would have analyses as the proposed technique successfully grapples with islanding
mistakenly tripped during the trained cases 13 through 17 of Table 6. detection even with large SEIGs and several mini-hydro features. Since
In [18], however, the insertion of a pure inductance is presented this hybrid methodology entails DG power curtailment and it can affect
whenever the measured ROCOV surpasses 0.1 pu/s, which misses a the normal operation of the plant, setting out an optimal percentage
noticeable amount of islanding scenarios with low power imbalances becomes paramount. To that end, the sensitivity analysis in Fig. 7
below this threshold. highlights that a 10% reduction suffices to identify islanding then
• Although [8-11,13,17,22,32,35-37] have analysed the islanding avoiding a large disturbance.
detection in scenarios with either synchronous or asynchronous DGs Secondly, it is seen that although the response of a turbine governor
and proved remarkably effective with very low NDZs, the governor may require time to reach the desired gate opening, the transient
system has not played a role in any of them, hence being assessed for response is fast enough to be used as a tool for islanding detection
the first time in this article. purposes. As a proof of concept, the sensitivity analysis depicted in Fig. 8
• The listed IDMs in Table 7 have not considered the non-islanding reveals that the governor takes 4 s to reach the desired point of opera­
frequency deviations caused by large-scale generation-load imbal­ tion, but the transient response adds an imbalance large enough to
ances. Indeed, these techniques would have mistakenly tripped identify the islanding condition.
during such events, while the proposed IDM has shown no false
tripping in these challenging cases. 7. Conclusions
• The proposed hybrid technique detects islanding with an average
time of 473 ms, which resembles the fastest IDMs developed for This paper presents a hybrid two-stage IDM for microgrids involving
synchronous-based DGs. Plus, the obtained detection times allow mini-hydro DG units to tackle the harmful effects of the islanding
sufficient time for the transition between islanded and standalone operation of a SEIG. The proposed technique changes the reference of
mode whilst meeting the maximum disconnection time of 2 s pro­ the active power control loop taking advantage of the governor system
vided by the IEEE Std. 1547–2018 in [4]. whenever the ROCOF-based threshold setting of the first stage is sur­
• The suggested technique has also been evaluated with several grid passed. After that, the islanding condition is identified when the
strength levels, standing for different X/R impedance ratios. The thresholds of the second stage are exceeded. The proposed hybrid IDM
bottom line of this sensitivity analysis pinpoints that this technique has been assessed considering the mini hydro as a single DG unit and a
would be equally effective for microgrids embedded in both weak multi-DG scenario in which a PV unit has been added. Crucially, the
and stiff networks. interaction between both types of DGs and their effects on the proposed
IDM have been scrutinised. The accurate, reliable, and fast performance
6. Results discussion and main contributions of the proposed technique has been examined under various islanding
cases, including negligible active and reactive power mismatches. As can
This section is aimed at discussing the results obtained in this article be drawn from Tables 3 and 5 in Section 3, all tested islanding events are
and underscoring the main contributions. far below the 2 s detection time recommended by the IEEE Std.
As can be seen in the flowchart in Fig. 3, the NDZ is uniquely 1547–2018, whilst the average tripping time proved to be 473 ms.
determined by the first passive-based stage as if it misses an event, the Albeit the optimal selection of the threshold settings and their time
power curtailment of the second stage is not initiated. By observing the delays is the bottleneck of most passive and hybrid-based IDMs, this
outcomes of section 4, it can be concluded that the implemented technique proved successful in dealing with such an issue. The analytical
thresholds in the first stage of the method are optimal as are able to expressions and comprehensive sensitivity analyses prove such ability.
identify a balanced island where the power flowing through the PCC, On the other hand, if a non-islanding event occurs, the proposed tech­
either active or reactive, is negligible, i.e., in the order of hundreds of nique avoids false tripping whilst rapidly reaching the DG initial point of
mW/mVAr. Consequently, the NDZ capability offered by the proposed operation following the power curtailment, which is certainly a vital
method has been proven as any islanding event falls below the estab­ aspect in the IDMs requiring such strategies for the islanding detection.
lished threshold regardless of the power imbalance. Moreover, it has The proposed technique overcomes the major drawback of the
also been found that the voltage-dependency load modelling is indeed existing IDMs by eliminating the NDZ with lower detection times while
an influential factor when setting out the thresholds as the term ΔP in considering the dynamics of the governor system in IDM studies for the
Eqs. (13) and (14) does not consider bus voltages. As one can expect, the first time. Additionally, it offers a low implementation cost and a
parameters in the feedback loop can influence the transient behavior of reduced level of computational complexity, making it a promising so­
the governor and in turn the detection time, yet however, according to lution for future microgrids with a high share of mini-hydro-based DGs.
the fifth sensitivity analysis in Section 3.2, this algorithm exhibits high Finally, it can be safely concluded that the two major goals of this
performance. On the other hand, the optimality of the second stage to­ article have been fulfilled as the algorithm exhibits zero-NDZ within
wards islanding detection has also been evidenced by the sensitivity reasonable islanding detection times. Therefore, this novel IDM that

15
A. Serrano-Fontova and R. Bakhshi-Jafarabadi International Journal of Electrical Power and Energy Systems 143 (2022) 108437

exploits the dynamics of the governor system can be widely used for [16] Bakhshi-Jafarabadi R, Sadeh J. New voltage feedback-based islanding detection
method for grid-connected photovoltaic systems of microgrid with zero non-
mini-hydro based microgrids.
detection zone. IET Renew Power Gener Jul. 2020;14(10):1710–9.
[17] A. G. Abo-Khalil, A. M. Eltamaly, W. Alharbi, Abdel-Rahman Al-Qawasmi, M.
Alobaid, and Ibrahem M., “A modified active frequency islanding detection method
CRediT authorship contribution statement based on load frequency and chopping fraction,” Intr. Trans. Electr. Energy. Syst.,
vol. 31, no. 11, 2021.
Alexandre Serrano-Fontova: Methodology, Software, Conceptual­ [18] Rostami A, Jalilian A, Zabihi S, Olamaei J, Pouresmaeil E. Islanding detection of
distributed generation based on parallel inductive impedance switching. IEEE Syst
ization, Investigation, Writing – original draft. Reza Bakhshi-Jafar­
J Mar. 2020;14(1):813–23.
abadi: Conceptualization, Writing – original draft. [19] Bejmert D, Sidhu TS. Investigation into islanding detection with capacitor
insertion-based method. IEEE Trans Power Deliv Dec. 2014;29(6):2485–92.
[20] Bakhshi-Jafarabadi R, Sadeh J, Popov M. Maximum power point tracking injection
Declaration of Competing Interest method for islanding detection of grid-connected photovoltaic systems in
microgrid. IEEE Trans Power Deliv Feb. 2021;36(1):168–79.
[21] Bakhshi-Jafarabadi R, Sadeh J, Chavez J de J, Popov M. Two-level islanding
The authors declare that they have no known competing financial
detection method for grid-connected photovoltaic system-based microgrid with
interests or personal relationships that could have appeared to influence small non-detection zone. IEEE Trans. Smart Grid 2021;12(2):1063–72.
the work reported in this paper. [22] Zamania R, Esmail M, Hamedani Golshana H, Haes Alheloua N. Hatziargyriou A
novel hybrid islanding detection method using dynamic characteristics of
synchronous generator and signal processing technique. Electr. Power Syst Res.
References 2019;175.
[23] Bansal RC, Bhatti TS, Kothari DP. Bibliography on the application of induction
[1] D., Liu, H., Liu, X., Wang, and E., Kremere, Edition 2019. World Small Hydropower generators in non conventional energy systems. IEEE Trans Energy Conversion Sep.
Development Report 2019: Case Studies. United Nations Industrial Development 2003;18(3):433–9.
Organization; International Center on Small Hydro Power, ISSN: 2706-7599 [24] Mahato SN, Singh SP, Sharma MP. Dynamic behavior of a single-phase self-excited
Available at: www. smallhydroworld.org. induction generator using a three-phase machine feeding single-phase dynamic
[2] Serrano-Fontova A, Martinez JA, Casals-Torrens P, Bosch R. A robust islanding load. Int J Electr Power Energy Syst 2013;47:1–12.
detection method with zero-non-detection zone for distribution systems with DG. [25] Laghari JA, Mokhlis H, Bakar AHA, Mohammad H. A comprehensive overview of
Int J Electr Power Energy Syst Dec. 2021;133:107247. new designs in the hydraulic, electrical equipments and controllers of mini hydro
[3] Serrano-Fontova A, Torrens PC, Bosch R. Power quality disturbances assessment power plants making it cost effective technology. Renew Sustain Energy Rev Apr.
during unintentional islanding scenarios. A contribution to voltage sag studies. 2013;20:279–93.
Energies 2019;12(16):3198. [26] Zhang C, Yang M, Li J. Detailed modelling and parameters optimisation analysis on
[4] IEEE Standards Coordinating Committee 21. IEEE Application Guide for IEEE Std governing system of hydro-turbine generator unit. IET Gener Transm Distrib Mar.
1547TM, IEEE Standard for Interconnecting Distributed Resources with Electric 2018;12(5):1045–51.
Power Systems; 2018. [27] Zhao J, Wang L, Liu D, Wang J, Zhao Y, Liu T, et al. Dynamic model of Kaplan
[5] Hussain A, Kim C, Mehdi A. A Comprehensive review of intelligent islanding turbine regulating system suitable for power system analysis. Math Probl Eng 2015;
schemes and feature selection techniques for distributed generation system. IEEE 2015:1–12.
Access Oct. 2021;9:146603–24. [28] Borkowski D, Majdak M. Small hydropower plants with variable speed
[6] Wang W, Kliber J, Zhang G, Xu W, Howell B, Palladino T. A Power Line Signaling operation—an optimal operation curve determination. Energies Nov. 2020;13(23):
Based Scheme for anti-islanding protection of distributed generators—Part II: Field 6230.
test results. IEEE Trans. Power Deliv. 2007;22(3):1767–72. [29] Working Group Prime Mover and Energy Supply, “Hydraulic turbine and control
[7] Nikolovski S, Baghaee HR, Mlakic D. Islanding detection of synchronous generator- models for system dynamic studies,” IEEE Trans. on Power Syst., vol. 7, no. 1, pp.
based DGs using rate of change of reactive power. IEEE Syst J Dec. 2019;13(4): 167–179, 1992.
4344–54. [30] Chen L, Wang X, Min Y, Li G, Wang L, Qi J, et al. Modelling and investigating the
[8] Bekhradian R, Davarpanah M, Sanaye-Pasand M. Novel approach for secure impact of asynchronous inertia of induction motor on power system frequency
islanding detection in synchronous generator based microgrids. IEEE Trans Power response. Int J Electr Power Energy Syst May 2020;117:105708.
Deliv Apr. 2019;34(2):457–66. [31] Motter D, Vieira JCM. Improving the islanding detection performance of passive
[9] Cui Q, El-Arroudi K, Joos G. Islanding detection of hybrid distributed generation protection by using the undervoltage block function. Electr Power Syst Res Jul.
under reduced non-detection zone. IEEE Trans Smart Grid Sep. 2018;9(5):5027–37. 2020;184:106293.
[10] Sawas A, Woon WL, Pandi R, Shaaban M, Zeineldin H. A multistage passive [32] Grebla M, Yellajosula JRAK, Hoidalen HK. Adaptive Frequency Estimation Method
islanding detection method for synchronous-based distributed generation. IEEE for ROCOF Islanding Detection Relay. IEEE Trans Power Deliv Aug. 2020;35(4):
Trans Ind Informatics Mar. 2021;18(3):2078–88. 1867–75.
[11] A. Kaymaz ¨Ozcanlı, and M. Beysol, “A novel Multi-LSTM based deep learning [33] M. A. Ouhrouche, X. D. DO, Q. M Le, and R. Chainé, “EMTP based simulation of a
method for islanding detection in the microgrid,” Electr. Power Syst. Res., vol. 202, self-excited induction generator after its disconnection from the grid,” IEEE Trans.
p. 107574, Sep. 2021. Energy Conversion, vol.3, no. 1, 1998.
[12] Allan O, Morsi Walid G. A new passive islanding detection approach using wavelets [34] ENTSO-e, Dispersed generation impact on CE region security. 2014, p. 11, Available at;
and deep learning for grid-connected photovoltaic systems. Electr. Power Syst. Res. https://www.entsoe.eu.
2021;199. [35] Alam MR, Begum MTA, Mather B. Islanding detection of distributed generation
[13] Hussain A, Kim C, Admasie S. An intelligent islanding detection of distribution using electrical variables in space vector domain. IEEE Trans Power Deliv Apr.
networks with synchronous machine DG using ensemble learning and canonical 2020;35(2):861–70.
methods. IET Gener. Transm. Distrib. 2021;15:3242–55. [36] Xie X, Xu W, Huang C, Fan X. New islanding detection method with adaptively
[14] Admasie S, Ali Bukhari S, Gush T, Haider Raza, Kim C. “Intelligent islanding threshold for microgrid. Electr Power Syst Res 2021;195:107167.
detection of multi-distributed generation using artificial neural network based on [37] Xie X, Huang C, Li D. A new passive islanding detection approach considering the
intrinsic mode function feature”, Journal of Modern Power Systems and clean. dynamic behavior of load in microgrid. Int J Electr Power Energy Syst May 2020;
Energy 2020;8(4):511–20. 117:105619.
[15] Admasie S, Ali Bukhari S, Gush T, Haider R, Kim C. A passive islanding detection [38] Abdi H, Rostami A, Rezaei N. A Novel passive islanding detection scheme for
scheme using variational mode decomposition-based mode singular entropy for synchronous-type DG using load angle and mechanical power parameters. Electr
integrated microgrids. Electr Power Syst Res 2019;vol. 177 p. 105983. Power Syst Res Mar. 2021;192:106968.

16

You might also like