Uploaded by BILAL MUHAMMAD KHAN

Analysis of DERs on technique

advertisement
Analysis of Distributed Energy Resources Impact
on Islanding Detection Techniques
Pedro Inácio Barbalho, Thiago S. Menezes, Denis V. Coury, José Carlos M. Vieira Junior, Mário Oleskovicz
Department of Electrical Engineering and Computation
University of São Paulo - São Carlos School of Engineering
São Carlos, Brazil
pedro.inacio.barbalho@usp.br, thiagosm@usp.br, coury@sc.usp.br, jcarlos@sc.usp.br, olesk@sc.usp.br
Abstract—During an islanding event, the distributed resources
may keep the system loads operating without frequency and
voltage reference from the main grid. Hence, system stability
and some equipment operation can be compromised. Therefore,
there are many studies proposing islanding detection techniques
that allow fast switching of the distributed resources control
mode or its disconnection from the grid. Considering this,
the present paper aimed to evaluate the impact of distributed
energy resources on three islanding detection techniques: one
based on under/overfrequency; another based on the Rate of
Change of Frequency (ROCOF); and the last one based on
the Rate of Change of Power (ROCOP). Thus, it was firstly
simulated fault and islanding events on a modelled microgrid.
The technique’s accuracy was compared by connecting only a
synchronous generator in one scenario and connecting a battery
energy storage system and a synchronous generator in another
one. Additionally, in the second scenario, the energy storage
system output power was changed and it was considered the
injection and the absorption of power. Finally, it could be
concluded that the energy storage system connection caused some
variation on the islanding detection techniques, specially in the
one based on under/overfrequency.
Index Terms—BESS, distributed generation, islanding detection.
I. I NTRODUCTION
The rise of the Distributed Energy Resources (DER) usage
has been motivated mostly by the necessity of reducing greenhouse gas emission, the increased energy efficiency provided
by combining heat and electrical power and the pursuit to
increase the quality and reliability of power suply [1]. The
DERs are comprised of Distributed Generation (DG) and
Distributed Storage (DS) and its increase can be observed in
the Brazilian scenario, where its penetration grows exponentially [2]. Therefore, studying this recent model of generation
and its impact on the distribution grid is important.
Considering the context described, a Microgrid (MG) could
be a good solution to integrate different types of DERs [1]. A
MG is a region comprised of DGs, DSs and loads connected
to low or medium voltage. This system has its own controllers
and can operate connected to the distribution grid or not
(islanded mode) [3]. Furthermore, the MG’s control reduces
the network complexity transforming numerous connections of
DERs and special loads in only one point of common coupling
(PCC) [1].
In a MG, it is important to keep the system frequency
and voltage approximately at their nominal values on both
modes and the operation cost must be minimized. Taking into
account these MG’s features, the role of a islanding detection
system is to guarantee a fast disclosure of the point of common
coupling, making possible a smooth transition between the
MG’s connected to islanded mode [3].
There are many studies proposing different islanding detection techniques and analysing its accuracies. An islanding detection technique based on the extraction of Wavelet transform
coefficients from the voltage and current of a distributed generator was proposed in [4]. With the data acquired, the islanding
event was detected by a decision tree. In addition, in [5], the
performance of this technique was evaluated and compared to
other conventional techniques like under/overvoltage, voltage
vector shift, under/overfrequency and ROCOF. To do so,
a distribution grid with different distributed resources was
modelled and events of faults, load variation and capacitor
bank disconnection were simulated. The proposed technique
presented a good accuracy and fast response compared to the
others. Additionally, a review of different islanding detection
techniques and their recognition rate, including some based on
intelligent systems, was done by [6].
Although there are many studies comparing the existing or
new islanding detection techniques, it is still needed further
investigation of these techniques accuracy variation due to the
increase of distributed energy resources penetration. Thus, in
view of the importance of this techniques to a MG, a system
with high and variable penetration of DERs, this study aimed
to evaluate the impact of the connection of a Battery Energy
Storage System (BESS) and the variation of its power on the
accuracy of three islanding detection techniques encountered
in the literature: the method based on under/overfrequency; the
method based on ROCOF; and the method based on ROCOP.
II. I SLANDING D ETECTION TECHNIQUES
The islanding detection techniques are essential to a microgrid, ensuring smooth transition from connected to islanded
mode in a unexpected event. These techniques can be classified
as remote or locals. The first one can be integrated on a
MG supervisory control and present high performance [6].
However, its implementation is expensive, complex and is
dependent of a communication infrastructure.
978-1-5386-8218-0/19/$31.00 ©2019 IEEE
The local techniques are divided in actives, passives and
hybrids. The active techniques present high accuracy, but their
operation is based on causing a small perturbation in the
system [7]. Hence, these techniques are not commonly used,
because is desired to avoid the occurrence of electrical power
quality problems, system instability and the increase of the
total harmonic distortion [6]. On the other hand, the passive
techniques are faster, cheaper and easy to implement. For
these reasons, this work focused on analysing three passive
techniques. Their description and how they were implemented
are shown in the subsections below.
islanding; G its the machine nominal power; and f it is the
frequency of the generator terminal voltage.
It can be seen in (1), that the frequency presents a low
variation for a small power unbalance. Consequently, this
technique has bad performance in this scenario. The scheme
used to implement the ROCOF is illustrated in Fig. 2.
df (Hz/s)
dt
Low-pass
The technique based in under/overfrequency supervises the
distributed energy resource terminal voltage frequency. So,
since in the loss of mains the voltage frequency deviate from
its nominal value and crosses a threshold (β), it is possible
to detect this event. Moreover, the accuracy of this technique
may be improved by checking if the bias was violated and if
the terminal voltage is superior to a minimum level (Vmin ).
Additionally, to accuse that an islanding is happening, these
two conditions must be met for a specified time. The scheme
used in this study to implement this technique is shown in
Fig. 1.
|x|
>
β
Time delay
VDG
Vmin
>
β
A. Under/overfrequency (81U/O)
f-fref (Hz)
|x|
Filter
Trip
>
Figure 1. Scheme used to implement the under/overfrequency technique.
Adapted from [8].
B. Rate of change of frequency
The technique based in the rate of change of frequency also
supervises the DER terminal voltage just as the previous one,
but, in this case, it is calculated the frequency variation over
time. The value of this parameter is compared to a threshold
and, when this limit is exceeded and the DER terminal voltage
is above the minimum established through a specified time, a
trip signal is sent, accusing an islading event. This technique,
compared to the under/overfrequency, is more sensible to fast
variations of frequency even when they have low amplitudes.
This feature is well observed in a system energized only
by a distributed synchronous generator. The initial frequency
variation of this type of system when an islanding occurs can
be estimated by (1).
df
∆P
=
,
(1)
dt
2HG
where: H its the machine inertia constant; ∆P it is the
power unbalance in the energized portion at the instant of the
Time delay
VDG
Vmin
Trip
>
Figure 2. Detection scheme used to implement ROCOF. Adapted from [8].
The low-pass Butterworth filter, in Fig. 2, is a fourth-order
filter with a cut-off frequency of 5 Hz [9].
C. Rate of change of power
The technique based in the rate of change of power supervises the variation of the DERs terminal power over time.
In a islanding condition, the DG power variation is greater
than in a grid-connected mode [10]. Thus, even if there is a
small mismatch between the distributed generator power and
local loads, any change in load power will reflect in the rate
of change of power. This technique was already implemented
by [11] as a interlock function along to a ROCOF, aiming to
improve ROCOF overall performance.
In this study, the ROCOP technique wasn’t mixed with
another intending to analyse separately its accuracy variation
due the connection of new distributed energy resources. In
Fig. 3 is represented the scheme used to implement the rate
of change of power.
dp (MW/s)
dt
Low-pass
Filter
|x|
>
β
Time delay
VDG
Vmin
Trip
>
Figure 3. Scheme used to implement ROCOP. Adapted from [8].
It is worth pointing out that a second order low-pass filter,
with cut-off frequency of 5 Hz, was added in the power derivative signal to reduce any noise amplified by the derivative
calculation.
III. D ISTRIBUTED E NERGY R ESOURCES
TABLE II
BESS PARAMETERS
In this paper, two different distributed energy resources were
used. One of them is a synchronous generator that, in a MG,
provides frequency and voltage reference when an islanding
event occurs. The other one is a BESS. These two DERs are
better described in the following subsections.
BESS
Battery bank nominal voltage
Battery storage capacity
DC link’s voltage
BESS nominal power
Grid voltage
Nominal active power
A. Synchronous generator
In a microgrid, rotative generators are usually used to
provide frequency and voltage reference because of their high
inertia. When the MG operates connected to the distribution
grid, the synchronous generator usually controls its active and
reactive power. The machine switches to frequency and voltage
control only when a islanding event is detected. Therefore, the
frequency of the system will oscillate until the power injection
control mode is kept enabled. A relay using a islanding
detection technique is responsible to send a trip signal to the
generator, warning it when to switch its control mode.
The generator used in this study is a round rotor synchronous machine of 5 MVA and it was implemented in
PSCAD/EMTDC. The machine control was kept in active and
reactive power control. The parameters of the generator are
presented in Table I.
IV. M ETHODOLOGY
A. Microgrid
The modelled system is a benchmark from the Conseil
International des Grands Réseaux Électriques (CIGRE) in
2014 [14]. It is an European medium-voltage system and
is used for studies involving distributed energy resources
applications. The distribution system is shown in Fig. 5, as
well as the area corresponding to the microgrid.
Distribution grid 110 kV
T1
1
110/20 kV
T2
12
Feeder 2
2 (PCC)
3
Synchronous Generator
Nominal power
Nominal voltage
Inertia constant
Nominal frequency
110/20 kV
Microgrid
TABLE I
S YNCHRONOUS G ENERATOR PARAMETERS [12]
2.13 kV
1.62 kAh
9 kV
2 MVA
4.16 kV
1 MW
5 MVA
6.6 kV
1s
376.99 rad/s
13
4
S3
5
14
11
10
9
8
S1
7
B. Battery energy storage system
S2
6
The battery energy storage system is one of the most used
DS nowadays [13]. The structure of the BESS implemented
in this paper is presented in Fig. 4.
Load
Transformer
Node
Switch
Figure 5. Modelled distribution grid [14].
AC Grid
-P, Q
Node 10
-
IDC
--
VDC
--
-
Figure 4. Battery energy storage system.
This BESS is comprised of a battery bank, a DC-DC
converter and an AC-DC converter. The DC-DC converter is
a buck-boost type and it is capable of controlling the battery
power input and output. The AC-DC converter controls the
DC link’s voltage and the reactive power exchanged between
the storage system and the grid. The BESS was modelled in
PSCAD and its parameters are presented in Table II.
In the MG, there are residential and industrial loads. These
loads are balanced and were modelled as constant impedance
with total nominal power of 4.32 MW. In this study, the BESS
was connected in node 10 and the synchronous generator in
node 5. These nodes are one of the most distant from the PCC
and has one of the biggest loads concentration. In addition,
the switches S1, S2 and S3 were kept opened. Finally, the
parameters for lines, loads and transformers can be found in
the Appendix.
B. Performance tests
To analyse the techniques’ performances, several cases
of short-circuits and disconnection of the PCC for different
load levels were simulated. Firstly, the islanding and fault
scenarios were simulated considering only the connection of
the synchronous generator in node 5. Since the grid simulated
is a balanced system, only four types of faults were considered: phase-ground; phase-phase; phase-phase-ground and
tree-phase-ground. Still, the fault resistance was varied from
0 Ω to 50 Ω with a 12.5 Ω step. Therefore, with 14 nodes, 5
types of fault resistances and 4 fault types, there are a total of
280 fault scenarios. The faults duration was 200 ms and the
DGs power was kept in 1 pu with power factor of 0.99. Along
with that, islanding events were simulated considering only the
disconnection of the PCC for different load profiles and DGs
power injection. The load profile was varied from 0.1 pu to 1.2
pu with a 0.1 pu step. Moreover, the DGs power was varied
from 0.125 pu to 1 pu with a 0.125 pu step and the power
factor always equal to 0.99. Thus, with 12 load levels and 8
DG generation conditions, the islanding scenarios totalise 96
simulations.
In a second analysis, the BESS was connected at node 10
and the fault and islanding scenarios were simulated again the
same way aforementioned. As the goal of this study was to
evaluate the impact of the variation of the DER power output
in the techniques’ performances, its accuracies were calculated
for each BESS power injection considered. The BESS power
output was varied from 0.25 MW to 1 MW with a 0.25 MW
step, when discharging. Also, when the BESS was charging,
its power injection was varied from 0.125 MW to 0.5 MW
with a 0.125 MW step.
The frequency, derivative of frequency and derivative of
power signals were exported from PSCAD to MATLAB,
where the islanding detection techniques were implemented.
The thresholds, β and Vmin and time delay were obtained
empirically.
V. R ESULTS AND D ISCUSSION
The techniques’ performances were evaluated quantitatively
using a different concept of accuracy, the Balanced Accuracy
(BA). This index was chosen given the necessity of avoiding
a biased analysis since the fault scenario dataset is bigger than
the islanding scenario dataset.
The BA is equivalent to the mean between the true positive
rate and the true negative rate. To exemplify, consider a
scenario where there are an islanding event dataset with 10
cases and a non-islanding event dataset with 100 cases. If all
the 110 cases were classified as non-islanding, the accuracy
would be about 91%. Although its a high value, this index
does not reflects the low performance of the technique, since
there was no islanding event detected. On the other hand, the
BA for this scenario would be equal to 50%, which represents
better the situation.
This index, the BA, was already proposed by [16] as
a improvement of the traditional accuracy to avoid biased
evaluation of a binary classifier. In a formal way, given two
sets, one with P positive cases and other with N negatives
cases, the True Positive Rate (TPR) is calculated as the ratio
between the number of cases the classifier indicated correctly
(True Positive cases - TP) and the total number of positive
cases. Also, the True Negative Rate (TNR) is calculated as
the ratio between the number of cases correctly classified as
negative cases (True Negative cases - TN) and the total number
of negative cases. In (2), the traditional accuracy calculation
is shown. The BA is calculated as presented in (3).
Accuracy =
C. Frequency estimation
With the objective to implement the ROCOF and under/overfrequency techniques, it was needed to model a frequency estimator. To do so, methods that track the voltage
frequency, such as the Phase-Locked Loop (PLL) or detects
the signal zero-crossing can be used. In this study, a method
that detects the sign transition (zero-crossing) and estimate the
frequency every half-cycle was used [15]. In this case, there
is a counter that resets every signal transition from negative
to positive and another counter that resets every transition
from positive to negative. Thus, every half-cycle a frequency
estimation is obtained.
The frequency estimation was applied in each phase of the
three-phase voltage and the average of the three results was
calculated. This frequency estimator was implemented in C
language internally in PSCAD. Furthermore, to filter some
noise and harmonics from the voltage measurements, the input
signal of the estimator passed through a second order low-pass
filter with cut-off frequency of 200 Hz. Moreover, the input
of the estimator, after the filtering process, was sampled at
3,840 Hz, which corresponds to 64 samples per cycle. In the
estimator output there is a second order low-pass filter with
cut-off frequency of 30 Hz [9].
BA =
TP + TN
N +P
(2)
TPR + TNR
2
(3)
The thresholds and time delay values for each islanding
detection technique used in this paper were determined in
MATLAB. To do so, different values were tested until the best
BA was found. These parameters were obtained considering
only the synchronous generator connected in the MG and they
are presented in Table III.
TABLE III
T HRESHOLDS AND TIME DELAY FOR THE ISLANDING DETECTION
TECHNIQUES
81U/O
ROCOF
ROCOP
β
Vmin
(pu)
Time delay
(s)
0.50 Hz
0.70 Hz/s
0.08 MW/s
0.41
0.29
0.33
0.110
0.179
0.333
At first, the balanced accuracy of the techniques was calculated for the scenarios were the MG had only the synchronous
generator and, after that, the index was calculated to the
situation where the BESS was included and discharging. In
Fig. 6, it is possible to note the BA variation for this situation.
power, since the higher BA was obtained in this scenario. The
confusion matrix of this last scenario is presented in Fig. 9.
Desired outputs
1
0
76
2
1
20
278
0
Outputs
Figure 8. Confusion matrix for the 81U/O (No BESS).
Desired outputs
Figure 6. Balanced accuracy for different values of power been injected by
the BESS (discharging mode).
It can be seen, in Fig. 6, that the under/overfrequency
technique was the most affected for different BESS power
outputs. This technique’s BA had an standard deviation of
2.36% and the connection of the BESS reflected on a increase
of the index.
Next, it was considered the influence of the BESS charging
and the BA’s variation can be seen in Fig. 7.
1
0
87
2
1
9
278
0
Outputs
Figure 9. Confusion matrix for the 81U/O (BESS discharging with 750 kW).
It is possible to see in Figs. 8 and 9 that the connection
of the BESS discharging 750 kW increased the number of
islanding cases detected by the technique. Still, the number
of non-islanding cases that where falsely detected as islanding
kept the same. This effect reflected in the BA index, increasing
its value.
As can be observed in Figs. 8 and 9, the BA for the 81U/O
that was 89.22% in the situation without BESS, increased to
94.96% when the BESS was connected discharging 750 kW.
Besides that, when the BESS was connected and it was
charging, the BA of all the techniques were less affected if
compared to the connection the BESS discharging.
VI. C ONCLUSION
Figure 7. Balanced accuracy for different values of power been absorbed by
the BESS (charging mode).
In Fig. 7, it is clear that the BESS charging had a smaller
impact on the techniques’ BA. Furthermore, the most influenced technique was the under/overfrequency and its balanced
accuracy had a standard deviation of 0.87%. In general, the
81U/O had the highest BA and it was the most affected
technique by the connection of a BESS.
Additionally, the 81U/O technique was analysed in more
details under two different scenarios. The first one is equivalent
to the situation where only the synchronous generator is
connected to the MG and its confusion matrix is presented
in Fig. 8. The second one is equivalent to the situation where
the BESS was connected discharging with a 750 kW output
The objective of this study was to analyse the impact
of the connection of a distributed energy resource, besides
a rotational machine, in the performance of three islanding
detection techniques. It was possible to notice that the balanced
precision of all of the techniques were influenced by the
connection of a battery energy storage system. Additionally,
the highest variation in the BA with the connection of the
BESS was obtained in the 81U/O with the storage system
discharging. Nevertheless, this technique identified a greater
number of islanding cases without increasing the total number
of false trips.
For future studies, it is suggested to increase the variety
of the DERs and also analyse the impact of other parameters
in BA of the islanding detection techniques, such as sample
frequency, order and cut-off frequency in the filter used in
the measured signals. Moreover, the influence of the DER
controllers in the balanced accuracy can also be considered,
since a smaller variation in the BA of the detection techniques
was noticed for the scenario in which the BESS charges than
when the BESS was discharging.
ACKNOWLEDGMENT
This study was financed in part by the Coordenação
de Aperfeiçoamento de Pessoal de Nı́vel Superior - Brasil
(CAPES) - Finance Code 001 and in part by the São Paulo
Research Foundation (FAPESP) project No 2017/16742-7. The
authors would like to thank the University of São Paulo
(USP), São Carlos School of Engineering (EESC) and the
Electrical Power Systems Laboratory (LSEE) for the support
and infrastructure provided for this work.
A PPENDIX
TABLE IV
T RANSFORMERS ’ PARAMETERS .
Transformer
Configuration
Z2tr
[Ω]
Snominal
[MVA]
1-2
Delta-Star grounded
0.016+j1.92
25
TABLE V
L INE PARAMETERS
Parameter
0
Rph
[Ω/km]
0
Xph
[Ω/km]
0
Bph
[µS/km]
R00
[Ω/km]
X00
[Ω/km]
B00
[ µS/km]
Microgrid
Feeder 2
0.501
0.510
0.716
0.366
47.493
3.172
0.817
0.658
1.598
1.611
47.493
1.280
TABLE VI
L OAD PARAMETERS FOR THE MICROGRID
Node
2
3
4
5
6
7
8
9
10
11
Apparent power [kVA]
Power factor
Residential
Industrial
Residential
Industrial
285
445
750
565
605
490
340
265
90
675
80
-
0.97
0.97
0.97
0.97
0.97
0.97
0.97
0.85
0.85
0.85
0.85
-
R EFERENCES
[1] N. Hatziargyriou, H. Asano, R. Iravani, and C. Marnay, “Microgrids,”
IEEE Power and Energy Magazine, vol. 5, no. 4, pp. 78–94, 2007.
[2] A. N. de Energia Elétrica ANEEL, “Unidades consumidoras com
geração distribuı́da,” Setembro 2018.
[3] F. Katiraei, R. Iravani, N. Hatziargyriou, and A. Dimeas, “Microgrids
management,” IEEE Power and Energy Magazine, vol. 6, no. 3, 2008.
[4] L. W. Arachchige and A. Rajapakse, “A Pattern Recognition Approach
for Detecting Power Islands Using Transient Signals—Part I: Design and
Implementation,” IEEE Transaction on Power Delivery, vol. 25, no. 4,
pp. 3070–3077, 2010.
[5] N. W. A. Lidula and A. D. Rajapakse, “A pattern-recognition approach
for detecting power islands using transient signals-part II: Performance
evaluation,” IEEE Transactions on Power Delivery, vol. 27, no. 3, pp.
1071–1080, 2012.
[6] A. Khamis, H. Shareef, E. Bizkevelci, and T. Khatib, “A review
of islanding detection techniques for renewable distributed generation
systems,” Renewable and Sustainable Energy Reviews, vol. 28, pp. 483–
493, 2013.
[7] R. Nale and M. Biswal, “Comparative assessment of passive islanding
detection techniques for microgrid,” Proceedings of 2017 International
Conference on Innovations in Information, Embedded and Communication Systems, ICIIECS 2017, 2017.
[8] J. C. M. V. Junior, W. Freitas, W. Xu, and A. Morelato, “Performance
of frequency relays for distributed generation protection,” IEEE Transactions on Power Delivery, vol. 21, no. 3, pp. 1120–1127, 2006.
[9] D. Motter, J. C. Vieira, and D. V. Coury, “Development of frequencybased anti-islanding protection models for synchronous distributed generators suitable for real-time simulations,” IET Generation, Transmission
& Distribution, vol. 9, no. 8, pp. 708–718, 2015.
[10] P. Mahat, Z. Chen, and B. Bak-Jensen, “Review of islanding detection
methods for distributed generation,” 3rd International Conference on
Deregulation and Restructuring and Power Technologies, DRPT 2008,
no. April, pp. 2743–2748, 2008.
[11] B. Liu, D. Thomas, K. Jia, and M. Woolfson, “Advanced rocof protection
of synchronous generator,” in Innovative Smart Grid Technologies
(ISGT), 2011 IEEE PES. IEEE, 2011, pp. 1–6.
[12] F. A. M. Moura, Geração distribuı́da-impactos e contribuições para a
qualidade da energia elétrica e dinâmica dos sistemas elétricos-uma
análise através do ATP-EMTP. Universidade Federal de Uberlândia,
2011.
[13] K. Divya and J. Østergaard, “Battery energy storage technology for
power systems—an overview,” Electric Power Systems Research, vol. 79,
no. 4, pp. 511–520, 2009.
[14] CIGRE Task Force C6.04.02, Technical Brochure 575: Benchmark
systems for network integration of renewable and distributed energy
resources. CIGRE, 2014.
[15] Woodward, MRG3 - Generator Protection with Mains Supervision, Time
Overcurrent Protection and Earth Current Supervision - Manual MRG3
(Revision C).
[16] K. H. Brodersen, C. S. Ong, K. E. Stephan, and J. M. Buhmann, “The
balanced accuracy and its posterior distribution,” 2010, pp. 3121–3124.
Download