Design of an Advanced Phasor Data Concentrator for Monitoring of Distributed Energy Resources in Smart Microgrids

1551-3203 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2017.2697438, IEEE
Transactions on Industrial Informatics
1
Design of an Advanced Phasor Data Concentrator
for Monitoring of Distributed Energy Resources in
Smart Microgrids
Reza Pourramezan, Younes Seyedi, Student Member, IEEE, Houshang Karimi, Senior
Member, IEEE, Guchuan Zhu, Senior Member, IEEE and Michel Mont-Briant
Abstract—Synchrophasor networks are subject to commu-
nication delays and packet dropout which can compromise
data integrity and thus jeopardize control and monitoring of
smart microgrids. This paper proposes an advanced phasor data
concentrators (APDC) capable of counteracting the communi-
cation impairments and improving the quality of monitoring of
distributed energy resources (DERs) in microgrids. The proposed
APDC utilizes an adaptive compensation scheme to achieve
an effective estimate of missing data elements. Moreover, a
monitoring unit is proposed to reliably detect frequency ex-
cursions and identify the DERs affected by islanding events.
The performance of the proposed APDC is evaluated based
on realistic phasor measurement unit (PMU) data aligned and
processed with real-time OpenPDC software. The experimental
results confirm a high-level data integrity under both normal and
disturbed conditions of microgrids. Moreover, fast and reliable
detection of islanding events is achieved due to the significant
improvement in detection time even under severe data losses.
Index Terms—Renewable and Distributed Energy Resources,
Synchrophasor, Phasor Measurement Unit, Phasor Data Concen-
trator, Smart Grid, Monitoring and Control
I. INTRODUCTION
Distributed energy resources (DERs) are increasingly de-
ployed in smart utility grids. The integration of DER units into
the power grid has brought about the concept of smart micro-
grids [1]. A microgrid is a local power network consisting of
DER units, loads and energy storage systems and is normally
connected to the main grid at a point of common coupling
(PCC) [2]. In the absence of the main grid, a microgrid
should be able to operate in the islanded mode and supply
power to its dedicated loads. However, DER units based on
renewable energy sources, e.g., wind and solar, give rise to new
challenges regarding power system management, protection,
and control [3].
A microgrid subjected to various types of faults may
partially or completely be islanded from the host grid [4].
Traditionally, some utilities consider unintentional islanding
events as undesirable conditions due to safety, equipment
This work was supported and financed in part by Vizimax Inc., and in
part by the Natural Sciences and Engineering Research Council of Canada
(NSERC) under grant EGP 488749-15.
R. Pourramezan, Y. Seyedi, H. Karimi and G. Zhu are with the Depart-
ment of Electrical Engineering, Polytechnique Montreal, Montreal, QC, H3T
1J4, Canada. (emails: reza.pourramezan@polymtl.ca, s.y.seyedi@ieee.org,
houshang.karimi@polymtl.ca, guchuan.zhu@polymtl.ca)
M. Mont-Briant is with Vizimax Inc., Longueuil, QC, J4G 1G1, Canada.
(email: mmbriant@vizimax.com)
protection, and system stability concerns. Attempts were made
to address these issues through anti-islanding features provided
in DER units and other practices recommended by standards
such as the IEEE 1547 standard [5].
With recent advancements of electronically-coupled DER
units, a more effective approach is to switch to autonomous
operation mode when the microgrid is disconnected from the
main grid, [6,7]. In the autonomous mode, the loads can
be supplied using the available DERs within the islanded
zone, however, it may require load shedding to balance the
power between generations and consumptions [8]. Moreover,
sophisticated robust controllers should be designed and utilized
to guarantee voltage and frequency stability of the islanded
zone despite loads uncertainties and source variations [9,10].
Furthermore, to provide smooth transition from grid-connected
mode to islanded mode and to guarantee continuous supply for
the critical loads, fast, accurate and reliable detection of the
islanding event using a centralized monitoring mechanism is
required.
Centralized control and protection of DER units require a
reliable monitoring mechanism to detect potential islanding
events in a timely manner. It has been recently shown that
synchronized phasor (synchrophasor) data can be effectively
used in centralized monitoring and detection applications,
[11], [12]. A synchrophasor data frame reported by a PMU
is defined as a set of phasor, frequency, rate of change of
frequency (ROCOF), and time tags that are measured at
identical time instants.
Several methods for centralized islanding detection in smart
grids have been proposed in the literature (see, e.g., [13]–[16]).
The majority of the existing methods deal with the detection
of islanding of an entire microgrid based on the observation
of frequency excursion. Phasor measurement units (PMUs)
extract synchrophasor data and transmit them to a phasor
data concentrator (PDC) via communication links. The PDC
collects data and produces synchrophasor datasets which are
then transmitted to centralized monitoring and control systems
at different levels. Note that the imperfect data transmission
may result in higher detection times that can lead to instability
of the islanded microgrid [16], [17].
An interpolative algorithm for phasor data concentration is
proposed in [16] in order to enhance disturbance detection in
smart microgrids. Although this algorithm can significantly
diminish the effect of the noise, it may lead to a slower
detection of disturbances. In particular, if all communica-
Copyright c 2017 IEEE. Personal use of this material is permitted. However, permission to use this material
for any other purposes must be obtained from the IEEE by sending a request to pubs-permissions@ieee.org.

2
tion links incur packet dropout, then with this method the
missing data will be replaced by the most recent samples.
Such circumstances impede fast detection of time-critical
disturbances. Moreover, the method of interpolation is not
applicable to scenarios where only a portion of the microgrid
is subject to severe disturbances. Finally, it should be noted
that, the existing data concentration and detection methods are
not validated based on realistic and real-time synchrophasor
data. The method presented in [17] employs spatio temporal
correlations between synchrophasor measurements of different
time instants to detect the low-quality synchrophasor data.
However, no attempt has been made to compensate for PMU
bad/low-quality data.
In this paper, an advanced PDC (APDC) is designed and
implemented on a real-time software platform, which improves
data integrity and facilitates monitoring of DERs in micro-
grids. The proposed APDC consists of three fundamental units:
a conventional PDC, a compensation unit, and a monitoring
unit. The conventional PDC subsystem collects and aligns
the received synchrophasor data streams. The compensation
unit includes an adaptive estimator in order to compensate
for data losses, PMU failures, and excessive communication
delays. The monitoring unit implements an elaborate detection
algorithm that processes voltage/frequency data in order to dis-
tinguish between possible islanding of DERs and disturbances
of the host grid.
The proposed monitoring unit can be configured in order
to implement several applications, e.g., monitoring, data visu-
alization, data analysis, energy management, event detection,
alarming, control, and protection. Fast and accurate detection
of islanding event is one of the most challenging applications,
particularly, when the data integrity is degraded by communi-
cation impairments. Therefore, in this paper, the performance
of the proposed APDC is evaluated with respect to islanding
event detection. We have used frequency and ROCOF data
for islanding detection since these parameters undergo severe
instability and/or swings subsequent to an islanding event.
The main contribution of this paper is twofold: First, the
missing data elements are predicted by a linear/derivative-
based adaptive estimator depending on the rate of change of
the voltage magnitude/frequency. Therefore, the integrity of
output synchrophasor datasets are preserved under both normal
and disturbed conditions. Second, the monitoring subsystem
of the APDC identifies disturbances in the voltage magni-
tude/frequency of the microgrid and the main grid. Hence, the
APDC can identify the islanded region and its operating DER
units. Such real-time information can significantly improve the
performance of centralized protection and control applications
in smart microgrids. The experimental results based on the
measured PMU data confirm that the proposed APDC enables
fast and reliable event detections in the presence of severe
communication impairments.
II. PROPOSED ADVANCED PDC
Figure 1 illustrates a generic centralized monitoring scheme
employing the PDC for smart microgrids. Several PMUs are
used to measure and report the synchrophasor data at different
Fig. 1: Schematic of monitoring and event detection by ad-
vanced PDC in smart grids
nodes, such as local point of connections (PC) of DERs, the
main PCC, and local loads [3]. The synchrophasor data are
then transmitted to the PDC via communication links that are
generally imperfect and encounter packet dropouts and delay.
The basic functionality of the PDC is to collect synchrophasor
data frames and produce time-aligned datasets that can be
transmitted to other PDCs or high-level control devices [18]. In
a hierarchical control structure, however, more functionalities
can be added to the PDC to enhance monitoring, control, and
protection based on synchrophasor data. The main motiva-
tion behind developing the APDC is to process data frames
transmitted over unreliable communication links, in order to
improve monitoring and event detection within the microgrid
domain.
Suppose that P PMUs report the synchrophasor data from
distributed nodes and the main PCC of the microgrid. The
voltage/current signal xi(t) measured at the ith
node, 0 ≤ i ≤
P − 1, can be expressed as [19]:
xi(t) = Ai(t) cos 2π fi(t)dt + φi , (1)
where Ai(t), fi(t), and φi are the instantaneous magnitude,
frequency, and phase angle of the ith
node, respectively. As
a convention in this paper, the node index i = 0 is associated
with the main PCC.
The main grid can be considered as a large and bulky power
system in which all generators rotate at the same speed. This
implies that a nearly unique frequency is observed at the grid
side and hence, the frequency of the microgrid in the grid-
connected mode is imposed by the main grid. However, the
frequency deviates from its nominal value, denoted by fn (60
Hz in North America). If the total demand exceeds the total
generation, then the frequency reduces until the power balance
is achieved at a new frequency. If the total demand decreases,
then the frequency exceeds the nominal value [20], [21].
Under normal operation, since the load-generation balance
is not always satisfied, power system frequency may slowly
fluctuate around its nominal value. Therefore, the instanta-
neous frequency at each node can be expressed as the sum
of the nominal frequency and a time-varying deviation term

3
Fig. 2: Constituent elements of the proposed APDC
as [19], [22], [23]:
fi(t) = fn + ∆fi(t) (2)
where ∆fi(t) is the frequency deviation at the ith
node.
Hence, Equation (1) can be rewritten as:
xi(t) = Ai(t) cos 2πfnt + [2π ∆fi(t)dt + φi]
= Ai(t) cos 2πfnt + θi(t) (3)
The time domain signal in (3) can be represented as an
instantaneous phasor:
Xi(t) =
Ai(t)
√
2
ejθi(t)
(4)
where θi(t) is the instantaneous phase angle of the ith
node
synchronized to the coordinated universal time (UTC). For the
purpose of comparison, the phasors must be evaluated with
the same time scale and the same nominal frequency. These
synchronized phasors or synchrophasors are defined in (4).
Figure 2 shows the constituent elements of the proposed
APDC. The APDC consists of a conventional PDC along
with a compensation unit and a monitoring unit. Each PMU
transmits the measured data frames to the APDC by means
of communication links and data buses. The conventional
PDC receives data frames and produces a standard output
data stream. Afterward, the compensation unit processes data
streams and, if necessary, compensates for data loss and
delay. The compensated data frames are then delivered to the
monitoring unit that constantly monitors events by detecting
abnormal behavior of voltage/frequency in the microgrid. The
main objective of the compensation unit is to provide the
monitoring unit with consistent and reliable synchrophasor
datasets. Events that are detected by the monitoring unit
are used to trigger applications such as microgrid protection,
control and energy management [26].
III. FREQUENCY AND ROCOF ESTIMATION ERROR
REQUIREMENTS
The IEEE Standard C37.118.1 2011 defines two perfor-
mance classes for PMU, namely class M and class P [24]. The
M class requires filtering to attenuate signals above the Nyquist
frequency significantly and can be used in applications where
high reporting delays are tolerated. The P class can be adopted
in applications where a fast reporting and small measurement
latency are required (e.g. event detection, protection, and
control). However, performance class P may result in higher
measurement errors in the presence of interfering signals
[27]. Network designers should choose a performance class
considering the requirements of the desired application.
The error of frequency and ROCOF measurements are
defined as follows [24]:
FE(t) = ˆf(t) − f(t) (5)
RFE(t) = ˆROCOF(t) − ROCOF(t) (6)
where FE(t) and RFE(t) are measurement errors of frequency
and ROCOF, respectively. f(t) and ROCOF(t) are the actual
frequency and ROCOF, while ˆf(t) and ˆROCOF(t) are the
measured values, respectively. PMUs should comply with
measurement error requirements defined in the synchrophasor
standard [24].
IV. PROPOSED METHODS AND DEVELOPED ALGORITHMS
In this section, the details of the compensation and mon-
itoring units are thoroughly discussed. The compensation
unit employs two different estimation methods to cope with
communication impairments and improve the integrity of
the output data. The monitoring unit implements basically
a conditioning algorithm with the primary goal of detecting
major disturbances and the impacted DERs in the microgrid.
A. Parameter Estimation and Compensation Unit
The main objective of the compensation unit is to com-
pensate for communication losses in order to provide the
monitoring unit with consistent and reliable datasets. Two
methods are employed in the design of the compensation unit:
a linear parameter estimator and a derivative-based estimator.
In case of packet dropouts, PMU failures, or intolerable
communication delays, these estimation methods are used to
predict the value of the measurements based on the previously
received data frames.
The linear estimator is basically a linear minimum mean
square error (LMMSE) estimator [28] that relies on the spatio
temporal correlations between synchrophasor data reported by
different PMUs at different time instants. Let xi[k] denote a
data element (voltage magnitude, frequency, etc.) reported by
the ith
PMU at the kth
instant. The last N measurements
reported by the ith
PMU comprise the data vector xi[k|N]
defined as:
xi[k|N]






xi[k + 1 − N]
...
xi[k − 1]
xi[k]






(7)
where N is the predefined number of data samples that build
the estimation window. Furthermore, let X[k|N] denote the

4
concatenated data vector formed by stacking the data vectors
for all PMUs:
X[k|N]






x0[k|N]
x1[k|N]
...
xP −1[k|N]






(8)
In fact, the concatenated data vector defined in (8) contains
the last N measurements (of the same data type) reported
by all PMUs. Let ˆxi[k + 1|N] be the estimate of data element
xi[k+1] based on the last N data elements reported up to time
instant k. Given X[k|N], ˆxi[k + 1|N] can be computed based
on the one-step-ahead prediction obtained from the LMMSE
estimator:
ˆx[k + 1|N] = E ˆx + CˆxXC−1
XX X[k|N] − E X (9)
where ˆx[k + 1] is the P × 1 predicted data vector:
ˆx[k + 1|N] =






ˆx0[k + 1|N]
ˆx1[k + 1|N]
...
ˆxP −1[k + 1|N]






(10)
E X and E ˆx are the expected values of actual and
predicted data vectors, respectively. CˆxX is the P ×PN cross-
covariance matrix and CXX is the PN ×PN data-covariance
matrix calculated by:
CˆxX = ˆx − E ˆx X − E X
T
(11)
CXX = X − E X X − E X
T
(12)
The expected values of data vectors in (11) and (12) can be
found by averaging over a sliding window of archived data.
The LMMSE method yields the minimum prediction error
variance inasmuch as the variations of input data elements
follow a stationary Gaussian distribution (where the statistics
of data do not change over the time). The assumption of
Gaussian distribution can be justified through the central limit
theorem in non-disturbed conditions provided that a large
number of consumers contribute to the power flow. In fact,
the predicted data vectors computed from (9) have negligible
errors when both the main grid and the microgrid are operating
in normal conditions. However, due to the inherent time
averaging in calculations of (9), the LMMSE method may
exhibit slow response to abrupt changes in the input data.
Specifically, when the microgrid operates under a disturbed
condition, the voltage/frequency data can vary rapidly in a
short time window. Therefore, under disturbed conditions and
during transitions of microgrid between different operation
modes, an alternative approach should be employed.
Note that PMUs are capable of reporting rate of change
of phasor parameters. Hence, a viable alternative is to em-
ploy a derivative-based estimator. Compared to the LMMSE
estimator, a derivative-based estimator has a faster response
with non-stationary data in disturbed conditions. A first-order
Fig. 3: Flowchart of adaptive data compensation by the APDC
derivative-based estimator can predict the data elements at the
next time instant based on their present value and the rate of
change of element as:
ˆxi[k + 1] = xi[k] +
dxi
dt k
×
1
Fs
(13)
where Fs is the reporting rate of PMUs in frames per second
(fps). If the data element to be predicted is the frequency, then
the ROCOF reported by PMUs can be used in (13) and the
derivative-based estimator is called ROCOF-based estimator.
For other types of data, the derivative of the data element can
be approximately calculated as its rate of change during two
consecutive data frames:
dxi
dt k
xi[k] − xi[k − 1] Fs (14)
It should be noted that, the rate of change of phasor param-
eters is highly susceptible to measurement noise. Therefore,
the predictions obtained from the derivative-based estimator
could be relatively inaccurate under stationary (non-disturbed)
conditions. On the contrary, the predicted values based on the
LMMSE estimator are more accurate under stationary condi-
tions. Considering the fact that the operational conditions are
not a priori known, we propose an adaptive data predictor that
switches between the two estimation algorithms depending on
the received data.
Figure 3 shows the functional flowchart of the compensation
unit based on the proposed adaptive data estimation approach.
At the beginning of the procedure, the concatenated data
vectors, X, are stored in the database. When the number of
stored data vectors is equal to N, the estimation is carried
out for the data at the next time instant for each PMU.
Given the kth
data frame, the rate of change of the data
element for the ith
PMU is compared with some threshold
xT . If the rate of change of parameter is lower than the
threshold xT , then the LMMSE estimator is used to predict
the data element for the (k +1)th
time instant. Otherwise, the
derivative-based estimator is used for predictions. Finally, the
estimated values are stored in the same database in a last-in-
first-out (LIFO) manner. At the (k + 1)th
time instant, if the

5
Fig. 4: Flowchart of DER monitoring by the APDC
PMU measurements are not available, then the stored predicted
data elements are retrieved to build the output synchrophasor
dataset.
The derivative of a noise-free synchrophasor parameter un-
der steady-state condition is zero, i.e., dxi/dt = 0. Therefore,
xT should be set to 0 to enable the derivative-based estimator
upon occurrence of any events. However, in practice, syn-
chrophasor measurements are polluted with noise, harmonics,
and disturbances due to the load switching. Therefore, an
essential consideration in designing xT is to minimize the
use of the derivative-based estimator under normal operation
conditions, and hence, xT should be large enough to avoid the
excessive use of the derivative-based estimator. The minimum
value of xT can be determined based on the standard and
maximum deviations of synchrophasor data.
B. Monitoring Unit
After compensation of the missing data elements, the mon-
itoring unit processes the synchrophasor datasets and sends
event indicators as outputs of the APDC. The event indicators
can be used for different purposes such as DER islanding
alarm and protection tasks. The basic premise of the event
detection is to continuously monitor the disturbances in volt-
age and frequency data.
Frequency excursion detection is a common passive method
for islanding detection [15], [29]. In the proposed monitoring
method, first the magnitude of the frequency error ∆fi[k] ,
is calculated from (2), and then it is compared with a positive
threshold, ∆fT . The value of the threshold affects both the
non-detection zone (NDZ) and the probability of false alarms
[30]. Therefore, the monitoring unit must be able to distinguish
between a frequency excursion due to an actual islanding and
those caused by disturbances. Theoretically, ∆fT must be
greater than the frequency band of power system (e.g. ±0.06
Hz in North America [31], [32]). Moreover, in practice, ∆fT
should be greater than the maximum FE of PMUs obtained
from stochastic analysis of frequency data.
Figure 4 shows the flowchart of the proposed DER monitor-
ing algorithm that performs a fast, secure and reliable islanding
detection. The synchronized frequency measurements of all
DERs as well as the main grid are received by the APDC and
further processed by the compensation unit. It is thus cogent to
assume that the estimates of frequency for all nodes are always
available regardless of packet dropouts and communication
delays. Upon arrival of each data frame, the magnitude of
the frequency error of the main grid, |∆f0|, and those of the
DERs, |∆fi|, 1 ≤ i ≤ P −1, are calculated and compared with
∆fT . If the frequency of the ith
node violates the threshold
value, then a counter referred to as the number of consecutive
detections, denoted by NCDi, is incremented by one. Once
the frequency returns to the acceptable range, this counter is
reset to zero. The following expression is thus obtained:
NCDi[k] =
NCDi[k − 1] + 1, |∆fi| > ∆fT
0, otherwise
(15)
If the NCDi becomes equal or greater than a critical
number of consecutive detections, denoted by NCDc, then a
frequency excursion is detected on the ith
node. The trade-off
between the event detection time and the reliability depends
on the choice of NCDc. A small NCDc leads to faster event
detections. However, it increases the likelihood of a false alarm
due to perturbations and noise. The optimal value of NCDc
can be determined based on the requirements of the specific
application. To achieve instantaneous event detections, NCDc
should be set to 1.
As synchrophasor data arrive at the APDC, the parameters
NCDi are updated and the monitoring unit investigates the
following four scenarios:
1) NCDi < NCDc, ∀i = 0, ..., P − 1:
In this case, the number of consecutive detections for
the main grid and all DERs is less than its critical value.
Therefore, there is no significant frequency excursion in
the grid and no islanding event in the microgrid.
2) NCD0 < NCDc and NCDi ≥ NCDc for some i ≥ 1:
This case indicates that the main grid is operating under
normal conditions but the ith
DER is suffering from a
severe frequency excursion. There is a possibility that
the ith
DER is disconnected from the main grid and
thus an islanding alarm is issued.
3) NCDi ≥ NCDc and |∆fi| = |∆f0|, ∀i = 1, ..., P − 1:
In this scenario, the frequency errors for the main grid
and all DERs exceed the normal range, which implies
that there is a global frequency excursion on the grid.
No islanding alarm is issued since the entire microgrid
is connected to the main grid.
4) NCD0 ≥ NCDc and NCDi ≥ NCDc, |∆fi| = |∆f0|
for some i ≥ 1:
This case implies that the frequency errors for some
DERs are not in the same range with that of the main
grid. There is a major frequency disturbance on the main
grid while some of the DERs show frequency errors that
are not equal to that of the main grid. Thus, the islanding
alarm is issued for potentially isolated DERs.

6
Fig. 5: Part of experimental setup: Two PMUs, programmable
AC source, switch, amplifier and controller
V. EXPERIMENTAL RESULTS
A. Experimental Setup
This section presents the implementation of the developed
algorithms and the verification of the performance of the
APDC against the islanding events. The experimental setup
includes three PMUs (that operate according to P class), and
the developed APDC that is responsible for monitoring of
DERs and their connection/disconnection events. All PMUs
are powered by a DC source of 48 VDC with a reporting
rate equal to 60 fps. A PMU is dedicated to reporting the
synchrophasor data of the Hydro-Quebec distribution grid and
sends the data frames to the proposed APDC via an Internet
connection. For the Hydro-Quebec PMU the GPS clock is
used to synchronize and timestamp the data frames. The other
two PMUs, which are shown in Figure 5, are responsible
of measuring the synchrophasors at the two local PCs of
the DERs. The Inter-Range Instrumentation Group time code
(IRIG-B) is used for time synchronization which conforms to
the requirements imposed by the IEEE Std. C37.118.2 standard
[25]. The synchrophasor data obtained by these two PMUs are
transmitted to the APDC via a single Ethernet link.
In this paper, hardware emulation is used for real-time
performance assessment of the proposed APDC. The behavior
of microgrid under disturbed conditions (islanding event) is
emulated by feeding the Electromagnetic Transients Program
(EMTP) simulations of the IEEE 34-bus network to a pro-
grammable AC voltage source [33]. The programmable AC
voltage source is used to generate the voltage signals at the
local PCs of DERs. An electrical switch connects the amplified
output of the programmable voltage source with the input
terminals of the PMUs. In EMTP simulations, the aggregated
power generated by the DER units is close to the total
demanded power by the loads to the extent on which a slow
frequency drift of −0.5 Hz is observed after islanding of the
entire feeder. Unless otherwise stated, the LMMSE estimator
uses N = 10 data frames for prediction and compensation.
TABLE I: Standard and maximum deviation of PMU
measured frequency and ROCOF under normal conditions.
Param. Criteria PMU-0 PMU-1 PMU-2
f
[mHz]
Standard deviation 20.456 20.595 20.598
max |∆f| 96.290 92.980 92.983
ROCOF
[Hz/s]
Standard deviation 0.0929 0.1653 0.1654
max
|ROCOF|
3.6 7.0 7.1
The proposed methods are implemented in the OpenPDC; a
software package comprised of a set of applications for manag-
ing synchrophasor data and real-time processing of time-series
data streams [34]. The reported synchrophasor data from the
PMUs are decimated to 40 fps and then delivered to specific
action adapters. The action adapters are software modules
capable of real-time processing based on the developed com-
pensation and detection algorithms. Moreover, to investigate
the effects of communication impairments an online packet
drop action adapter is utilized inside the OpenPDC platform.
For certain desired time stamps it replaces the data frames
with Not a Number (NaN) indicator and sends them to the
compensation unit.
B. Frequency Estimation/Compensation
The signals at the input terminals of the three PMUs are the
voltages of the distribution network. Therefore, the measured
frequencies include both load-generation imbalance (∆fi(t) in
(2)) and noise. In case of frequency estimation, the threshold
xT is fT , which is a constraint on the ROCOF.
Table I shows the standard and the maximum deviations of
the measured frequency and ROCOF for a period of 30 min-
utes under normal operation. The average standard deviation of
frequency is 20.5 mHz and the maximum frequency deviation
from nominal value is 96.29 mHz (measured by PMU-0 which
is installed at the main PCC). Although the average of ROCOF
is 0 Hz/s over 30 minutes, its average standard deviation is
0.1412 Hz/s with a maximum ROCOF of 7.1 Hz/s (measured
by PMU-2 which is installed at the PC of DER-2).
As discussed in Section IV-A, the threshold value for the
adaptive estimator is designed based on the following criteria.
First, it should be much greater than the average ROCOF
standard deviation under normal condition. Second, it must be
small enough to trigger the ROCOF-based estimator when fre-
quency excursions occur. Therefore, under normal conditions
with the presence of both noise and frequency deviations, the
adaptive estimator mostly employs linear rather than ROCOF-
based estimator. Our experiments show that with the threshold
value fT = 1 Hz/s, the ROCOF-based estimator is activated
at only 0.4 % of the total operation time. In general, the value
of this threshold depends on the performance of PMUs.
Figures 6 (a)-(c) show the actual (measured) and the pre-
dicted frequency data based on the two estimators under the
normal condition, where there are neither frequency excursion
nor communication impairments. The actual frequency on the
grid side, f0, and the actual frequencies at the local PCs,
f1 and f2, exhibit small scale variations mainly due to the
instantaneous load-generation balance in the Hydro-Quebec

7
59.990
59.992
59.994
59.996f0
[Hz]
(a)
Measured Linear Estimator ROCOF-based Estimator
59.470
59.475
59.480
59.485
f1
[Hz]
(b)
151 151.2 151.4 151.6 151.8 152
Time [s]
59.470
59.480
59.490
f2
[Hz]
(c)
Fig. 6: Performance comparison of linear and ROCOF-based
estimators under normal condition: (a) frequencies reported by
PMU of the main PCC, (b) frequencies reported by PMU of
DER 1, (c) frequencies reported by PMU of DER 2
60.002
60.004
60.006
60.008
60.010
f0
[Hz]
(a)
59.600
59.800
60.000
f1
[Hz]
(b)
Measured
ROCOF-based Estimator
Linear Estimator
Adaptive Estimator
178.5 178.6 178.7 178.8 178.9 179
Time [s]
59.600
59.800
60.000
f2
[Hz]
(c)
Fig. 7: Performance comparison of linear, ROCOF-based
and adaptive estimators under the disturbed condition: (a)
frequencies reported by PMU of the main PCC, (b) frequencies
reported by PMU of DER 1, (c) frequencies reported by PMU
of DER 2
network. As expected, both estimation methods follow the
actual frequency smoothly but the LMMSE estimator has a
better accuracy in terms of the variance of the prediction error.
Figures 7 (a)-(c) show the actual and the predicted frequency
data for the PMUs when the entire microgrid becomes islanded
from the Hydro-Quebec network at time instant t = 178.55 s.
It can be seen that the linear estimator has a slow response to
the frequency excursion, whereas the ROCOF-based estimator
shows a faster response. Figure 7 also shows the results of the
proposed adaptive estimator under both normal and disturbed
conditions.
For the main grid’s frequency f0, one can see in Figure
TABLE II: Performance comparison of estimation methods
under normal and disturbed conditions with noise.
Condition
FE Estimator type
[mHz] LMMSE ROCOF Adaptive
Normal
max. 1.8 2.4 1.8
avg. 0.6 1.1 0.6
Disturbed
max. 388.3 161.4 116.1
avg. 214.2 61.6 48.9
7 that the estimated frequencies by both linear and adaptive
estimators are the same during the normal condition. For
the DER frequencies, however, the APDC switches to the
ROCOF-based estimator when frequency excursion occurs. It
can be concluded that the adaptive estimator yields fast and
accurate predictions for missing frequency data.
Table II confirms the results of Figures 6 and 7 by com-
paring the maximum and the average values of FEs for
different estimation methods under both normal and disturbed
conditions. FE is calculated as the absolute value of the
difference between the actual measured frequency by PMU
and the estimated frequency by different estimators. Under the
normal condition, the FE of the LMMSE estimator is less than
that of the ROCOF-based estimator. Moreover, the adaptive
estimator mostly employs the LMMSE method.
The results of frequency estimation due to an islanding event
are also given in Table II. The results imply that the FE of the
LMMSE estimator is higher than that of the ROCOF-based
estimator under disturbed conditions. The adaptive estimator
switches from the LMMSE to the ROCOF-based estimator,
resulting in a significant improvement of frequency estimation.
C. DER Monitoring Results
The critical number of consecutive detections and the
frequency deviation threshold are set to NCDc = 1 and
∆fT = 0.15 Hz, respectively. The selected ∆fT is greater
than both the North American tight frequency band (0.06 Hz)
and the maximum FE given in Table I (0.096 Hz). The selected
values of NCDc and ∆fT facilitate fast islanding detections
with small NDZ while preventing false alarms.
To investigate the performance of the monitoring unit, the
two DERs are simultaneously islanded at t = 334.95 s.
Moreover, four data packets from DER1 are intentionally
discarded at exactly two reporting intervals after islanding
time, i.e., at t = 334.975 s. Figures 8 (a)-(c) show the actual
frequency data (including the communication impairments),
the estimated values, and the compensated frequency data for
the PMU 1. It can be seen that the estimated frequency is
used only when the input data are not available. As soon
as the frequency exceeds the critical value, fc = 59.85 Hz,
and before any data reported by PMUs are available, four
consecutive packet dropouts occur for PMU 1. If there is
no frequency estimation/compensation, then the first possible
islanding detection would be at t = 335.1 s, (i.e., 5 data frames
later). This means a detection delay equal to four reporting
intervals, i.e., 100 ms. However, by using the compensation
unit all the missing data are replaced by the predicted values,

8
60.005
60.006
60.007
60.007
60.008f0
[Hz]
(a)
59.600
59.800
60.000
f1
[Hz]
(b)
Measured
Estimated
Compensated
334.9 334.95 335 335.05 335.1 335.15 335.2 335.25 335.3
Time [s]
59.600
59.800
60.000
f2
[Hz]
(c)
fn
- ∆ fT
= 59.85 Hz
100ms
Fig. 8: Performance of the proposed APDC in terms of island-
ing detection under successive packet dropouts: (a) frequencies
of the main PCC, (b) frequencies of DER 1, (c) frequencies
of DER 2
334.9 334.95 335 335.05 335.1 335.15 335.2 335.25 335.3
Time [s]
0
5
10
15
NCD2
Uncompensated measurements
Compensated measurements
Islanding instant
first packet drop
instant
last packet drop
instant
Fig. 9: The number of consecutive detections for PMU of DER
2 with and without packet dropout compensation
which allows the detection algorithm to process the consistent
frequency data, as shown by the dashed line in Figure 8 (b).
Figure 9 shows the number of consecutive detections over
the time. Without using data compensation, the parameter
NCD2 equals 1 at t = 335.1 s. Whereas by using the compen-
sation unit, the parameter NCD2 reaches 1 at t = 335 s (the
instant of the first packet dropout). These results substantiate
the advantages of using the proposed APDC for time-critical
monitoring applications with communication impairments.
D. Performance Evaluation in the Presence of Noise
In Section V-B, performances of different estimation meth-
ods are evaluated using experimental results of the distribution
network. However, the effect of noise on FE was not explicitly
evaluated due to the fact that PMU inputs were polluted with
unmeasurable noise. In this section, frequency and ROCOF
measurements corrupted by additive white Gaussian noise
(AWGN) are used to evaluate the performance of estimators.
Table III shows the maximum FE values of PMU outputs
and the outputs of estimation methods for different signal-to-
noise ratios (SNRs) of PMU output frequency. The percentage
of usage of ROCOF-based estimator (PRBE) is reported in
Table III to evaluate the robustness of the adaptive estimator
against the ROCOF noise.
TABLE III: Performance comparison of LMMSE,
ROCOF-based, and adaptive estimation methods with
measurable noise in normal condition.
SNR
[dB]
PMU Max.
Errors Max. FE [mHz] PRBE
[%]RFE
[Hz/s]
FE
[mHz] LMMSE ROCOF-based Adaptive
80 1.3 22 13 46 46 0.4
75 2.6 44 27 91 91 14.4
70 4.2 69 43 144 144 36.5
60 11.8 196 120 406 406 74.4
50 46.4 789 473 1624 1624 93.1
TABLE IV: Performance comparison of adaptive and
interpolative estimation methods under steady-state and ramp
tests.
Test conditions Max. FE [mHz]
Steady-
state
f0 = 60 Hz
Test interval:
60 s
f1,2
[Hz]
SNR
[dB]
PMU Interpolative Adaptive
60
85 13.8 6.7 8.5
70 69.3 33.8 144.0
59.8
85 14.5 139.8 7.5
70 72.6 165.7 146.3
Ramp range ±4 Hz
Rf = ±2 Hz/s
85 48.8 3974.8 58.4
70 85.7 3979.7 420.2
It can be seen in Table III that the maximum FE of the
LMMSE estimator is always smaller than that of the PMU
output. On the contrary, the maximum FE of the ROCOF-
based estimator is greater than that of the PMU output. How-
ever, even for smaller SNRs, the ROCOF-based estimations do
not deviate significantly from those of the PMU output. Note
that, the maximum FEs of PMU for SNR value lower than 70
dB, are not compliant with requirements defined in the IEEE
standard for synchrophasor measurements [24].
For SNR values greater than 80 dB, the PRBE is nearly
zero, resulting in highly accurate frequency estimations. How-
ever, for SNR values less than 80 dB, the PRBE increases.
Therefore, the maximum FE of the adaptive estimator exceeds
that of the PMU output due to the greater noise on ROCOF
measurement. Note that, the effectiveness of the adaptive
estimator relies on the trade-off between FE requirements
under normal and disturbed conditions.
Moreover, the noise assessment results show that the de-
signed threshold ∆fT = 0.15 Hz is highly robust against noise
for SNR values greater than 70 dB. It is essential to mention
that, for this SNR value, the PMU itself has a maximum FE
of 0.069 Hz and a maximum RFE of 4.2 Hz/s that are not
compliant with the synchrophasor standard [24].
E. Comparison With Interpolative Method
Table IV compares the maximum FE of the interpolative
estimator proposed in [16] and the adaptive estimator under
normal condition and ramp test. Under normal conditions, with
all the PMUs reporting nominal frequency f0 = f1 = f2 = 60
Hz and SNR= 85 dB, both the adaptive and interpolative
estimators improve the maximum FE of PMUs. However, the
interpolative method has a slightly better performance in case
of nominal frequencies. The key advantage of the adaptive

9
estimator unfolds in predicting off-nominal frequencies (e.g.
f1 = f2 = 59.8 Hz with f0 = 60 Hz). Under such
circumstances, the performance of the adaptive estimator is
much better than that of the interpolative estimator in both
noisy and very noisy situations.
Moreover, frequency ramp tests are performed on f1 and
f2 with a fixed f0 = 60 Hz, under different SNR values. In
these tests, the PMU input signals have the general form of
xi(t) = Ai(t) cos 2πfnt + πRf t2
+ θi(t) (16)
where the Rf is the frequency ramp rate in Hz/s (i.e. the
theoretical value of ROCOF).
The ramp rate is set to Rf = ±2 Hz/s and a frequency range
of ±4 Hz between 58 to 60 Hz is performed. As suggested
by the IEEE standard [24], the first two reporting intervals
(i.e. 2/Fs) before the beginning and after the end of the
ramp, are excluded from FE calculations. Results presented in
Table IV demonstrate that the adaptive estimator successfully
tracks the ramp frequency. Whereas, the interpolative estimator
has a very slow response with poor maximum FE. The ramp
test results further corroborate the superiority of the proposed
adaptive estimator under power system transients.
VI. CONCLUSION
This paper presents an advanced phasor data concentrator
(APDC) that enables the monitoring of DERs in smart micro-
grids. The core of the proposed APDC comprises an adaptive
estimator that can switch between a vector linear minimum
mean square error (LMMSE) estimator and a derivative-based
estimator sensitive to the rate of change of the parameter of
interest. The monitoring algorithm is a centralized frequency
excursion detector that processes the frequency data of DERs
and the main grid in order to detect the islanded resources.
The APDC has been experimentally implemented and its
performance is verified based on real-time synchrophasor
data collected from three PMUs and different communication
media. The frequency excursion detector is tested under severe
communication impairments. The experimental results show
that the proposed APDC has an excellent performance in terms
of data integrity under both normal and disturbed conditions
of microgrids and in the presence of realistic measurement
noise. The numerical results also confirm that the APDC can
reliably monitor DERs and determine the islanded zones under
consecutive packet dropouts.
REFERENCES
[1] X. Fang, S. Misra, G. Xue, and D. Yang, “Smart grid - The new and
improved power grid: A survey,” Commun. Surveys Tuts., vol. 14, no. 4,
pp. 944-980, 2012.
[2] N. Hatziargyriou, H. Asano, R. Iravani and C. Marnay, “Microgrids,”
IEEE Power Energy Mag., vol. 5, no. 4, pp. 78-94, July-Aug. 2007.
[3] D. E. Olivares, et al., “Trends in microgrid control,” IEEE Trans. Smart
Grid, vol. 5, no. 4, pp. 1905-1919, Jul. 2014.
[4] A. H. Kasem Alaboudy, H. H. Zeineldin, and J. Kirtley, “Microgrid sta-
bility characterization subsequent to fault-triggered islanding incidents,”
IEEE Trans. Power Del., vol. 27, no. 2, pp. 658-669, Apr. 2012.
[5] IEEE guide for design, integration, and operation of distributed resource
island systems with electric power systems, Std 1547.4-2011, Jul. 20,
2011.
[6] B. M. Eid, N. Abd Rahim, J. Selvaraj, and A. H. El Khateb, “Control
methods and objectives for electronically coupled distributed energy
resources in microgrids: A review,” IEEE Syst. Journal, vol. 10, no. 2,
pp. 446-458, Jun. 2016.
[7] A. Ovalle, et al.,“Decentralized control of voltage source converters in
microgrids based on the application of instantaneous power theory,” IEEE
Trans. Ind. Electron., vol. 62, no. 2, pp. 1152-1162, Feb. 2015.
[8] I. J. Balaguer, Q. Lei, S. Yang, U. Supatti and F. Z. Peng, “Control for
grid-connected and intentional islanding operations of distributed power
generation,” IEEE Trans. Ind. Electron., vol. 58, no. 1, pp. 147-157, Jan.
2011.
[9] M. Hamzeh, S. Emamian, H. Karimi, and J. Mahseredjian, “Robust
control of an islanded microgrid under unbalanced and nonlinear load
conditions,” IEEE Trans. Emerg. Sel. Topics Power Electron., vol. 4, no.
2, pp. 512-520, Jun. 2016.
[10] M. Babazadeh, and H. Karimi, “A robust two-degree-of-freedom control
strategy for an islanded microgrid,” IEEE Trans. Power Del., vol. 28, no.
3, pp. 1339-1347, Apr. 2013.
[11] Y. Seyedi, H. Karimi, and S. Grijalva, “Distributed generation moni-
toring for hierarchical control applications in smart microgrids,” IEEE
Trans. Power Syst., DOI: 10.1109/TPWRS.2016.2610322
[12] E. Cotilla-Sanchez, P. D. H. Hines, and C. M. Danforth, “Predicting
critical transitions from time series synchrophasor data,” IEEE Trans.
Smart Grid, vol. 3, no. 4, pp. 1832-1840, Sep. 2012.
[13] A. Borghetti, et al., “Synchronized phasors monitoring during the
islanding maneuver of an active distribution network,” IEEE Trans. Smart
Grid, vol. 2, no. 1, pp. 82-91, Mar. 2011.
[14] Y. Guo, K. Li, D. M. Laverty, and Y. Xue, “Synchrophasor-based
islanding detection for distributed generation systems using systematic
principal component analysis approaches,” IEEE Trans. Power Del., vol.
30, no. 6, pp. 2544-2552, May 2015.
[15] Z. Lin et al., “Application of wide area measurement systems to islanding
detection of bulk power systems,” IEEE Trans. Power Syst., vol. 28, no.
2, pp. 2006-2015, May 2013.
[16] Y. Seyedi, H. Karimi, and J. M. Guerrero, “Centralized disturbance
detection in smart microgrids with noisy and intermittent synchrophasor
data,” IEEE Trans. Smart Grid, DOI: 10.1109/TSG.2016.2539947
[17] M. Wu and L. Xie, “Online detection of low-quality synchrophasor
measurements: A data driven approach,” IEEE Trans. Power Syst., vol. PP,
no. 99, pp. 1–1, 2016.
[18] A. Monti, F. Ponci, and C. Muscas, Phasor Measurement Units and
Wide Area Monitoring Systems. Elsevier Science & Technology Books,
2016.
[19] D. R. Gurusinghe, A. D. Rajapakse, and K. Narendra, “Testing and
enhancement of the dynamic performance of a phasor measurement unit,”
IEEE Trans. Power Del., vol. 29, pp. 1551–1560, Aug. 2014.
[20] P. Kundur, N. J. Balu, and M. G. Lauby, Power system stability and
control. New York: McGraw-Hill, 1994. OCLC: 28929603.
[21] L. R. Chang-Chien, Y. S. Wu, and J. S. Cheng, “Online estimation of
system parameters for artificial intelligence applications to load frequency
control,” Transmission Distribution IET Generation, vol. 5, pp. 895–902,
Aug. 2011.
[22] N. Jaleeli and L. VanSlyck, “Control performance standards and proce-
dures for interconnected operation,” EPRI TR-107813, Apr. 1997.
[23] C.-i Chen, “A phasor estimator for synchronization between power grid
and distributed generation system,” IEEE Trans. Ind. Electron., vol. 60,
no. 8, pp. 3248-3255, May 2012.
[24] IEEE standard for synchrophasor measurements for power systems, Std
C37.118.1-2011, Dec. 28, 2011.
[25] IEEE standard for synchrophasor data transfer for power systems, Std
C37.118.2-2011, Dec. 28, 2011.
[26] L. I. Minchala-Avila, L. Garza-Castan, Y. Zhang and H. J. A. Ferrer,
“Optimal energy management for stable operation of an islanded micro-
grid,” IEEE Trans. Ind. Informat., vol. 12, no. 4, pp. 1361-1370, Aug.
2016.
[27] P. Castello, J. Liu, C. Muscas, P. A. Pegoraro, F. Ponci, and A. Monti,
“A fast and accurate PMU algorithm for P+M class measurement of
synchrophasor and frequency,” IEEE Trans. Instrum. Meas., vol. 63,
pp. 2837–2845, Dec 2014.
[28] S. M. Kay, Fundamentals of statistical signal processing, Volume I:
Estimation Theory, Prentice Hall, 1993.
[29] L. A. C. Lopes and H. Sun, “Performance assessment of active frequency
drifting islanding detection methods,” IEEE Trans. Energy Convers.,
vol. 21, pp. 171–180, Mar. 2006.
[30] Z. Ye, A. Kolwalkar, Y. Zhang, P. Du, and R. Walling, “Evaluation of
anti-islanding schemes based on nondetection zone concept,” IEEE Trans.
Power Electron., vol. 19, pp. 1171–1176, Sept. 2004.

10
[31] N. Jaleeli and L. S. VanSlyck, ”NERC’s new control performance
standards,” IEEE Trans. Power Syst., vol. 14, no. 3, pp. 1092-1099, Aug
1999.
[32] Y. G. Rebours, D. S. Kirschen, M. Trotignon and S. Rossignol, ”A survey
of frequency and voltage control ancillary services - Part I: Technical
features,” IEEE Trans. Power Syst., vol. 22, no. 1, pp. 350-357, Feb.
2007.
[33] J. Mahseredjian, S. Dennetire, L. Dub, B. Khodabakhchian and L. Grin-
Lajoie, “On a new approach for the simulation of transients in power
systems,” Electric Power Systems Research, vol. 77, no. 11, pp. 1514-
1520, Sep. 2007.
[34] “The open source phasor data concentrator”, CodePlex, 2016. [Online].
Available: https://openpdc.codeplex.com/. [Accessed: 17- Nov- 2016].
Reza Pourramezan received the B.Sc. degree from
Shahid Beheshti University, Tehran, Iran, in 2003,
and the M.Sc. degree from the University of Tehran,
Tehran, Iran, in 2006, both in electrical engineering.
He is currently working toward the M.Sc. degree in
the Department of Electrical Engineering, Polytech-
nique Montreal, Montreal, QC, Canada. His recent
research fields are power system analysis, monitor-
ing, and control, wide area monitoring systems and
synchrophasor applications in smart grids.
Younes Seyedi (S’12) received the B.Sc. degree
from University of Tehran, Tehran, Iran, in 2010, and
the M.Sc. degree from Tehran Polytechnic, Tehran,
Iran, in 2012, both in electrical engineering. In Sep.
2014, he started his PhD program by joining the
Department of Electrical Engineering at Polytech-
nique Montreal, Montreal, QC, Canada. His recent
research is focused on data analytics for power grid
monitoring and control, intelligent protection appli-
cations for smart grids, and synchrophasor networks.
Houshang Karimi (M’07-SM’12) received the
B.Sc. and M.Sc. degrees from Isfahan University
of Technology, Isfahan, Iran, in 1994 and 2000, re-
spectively, and the Ph.D. degree from the University
of Toronto, Toronto, ON, Canada, in 2007, all in
electrical engineering. He was with the Department
of Electrical Engineering, Sharif University of Tech-
nology, Tehran, Iran, from 2009 to 2012. From June
2012 to January 2013, he was a Visiting Professor
in the ePower lab of the Department of Electri-
cal and Computer Engineering, Queens University,
Kingston, ON, Canada. He joined the Department of Electrical Engineering,
Polytechnique Montreal, QC, Canada, in 2013, where he is currently an
Assistant Professor. His research interests include control systems, microgrid
control, and smart grids.
Guchuan Zhu (M’07-SM’12) received the M.S.
degree in electrical engineering from the Beijing
Institute of Aeronautics and Astronautics, Beijing,
China, in 1982; the Ph.D. degree in mathematics
and control from cole des Mines de Paris, Paris,
France, in 1992; and the graduate Diploma degree
in computer science from Concordia University,
Montreal, QC, Canada, in 1999. Dr. Zhu joined
cole Polytechnique de Montral, Montral, in 2004,
where he is currently an Associate Professor in the
Department of Electrical Engineering. His current
research interests include control of distributed parameter systems, nonlinear
and robust control, and optimization with their applications to microsystems,
aerospace systems, communication networks, and smart grid.
Michel Mont-Briant is a project manager at Viz-
imax Inc. He received his B.Eng from Sherbrooke
University in 1984. He is a registered engineer in
the province of Quebec, and had been working for
Snemo Ltd since 1981 as R&D manager, then R&D
director and finally General Manager. Snemo Ltd
R&D involved in designing, testing and protective
relays, automation platform and control systems
for power utilities, including the flagship products
CBSC and SynchroTeq circuit breaker controlled
switching devices. He owns a Canadian patent on
a device for controlling VHV or UHV circuit breaker. Since the merger of
Snemo Ltd and STR into Vizimax Inc, He is mostly involved in special project
management.

Design of an Advanced Phasor Data Concentrator for Monitoring of Distributed Energy Resources in Smart Microgrids

Recommended

Recommended

More Related Content

What's hot

What's hot (18)

Similar to Design of an Advanced Phasor Data Concentrator for Monitoring of Distributed Energy Resources in Smart Microgrids

Similar to Design of an Advanced Phasor Data Concentrator for Monitoring of Distributed Energy Resources in Smart Microgrids (20)

Recently uploaded

Recently uploaded (20)

Design of an Advanced Phasor Data Concentrator for Monitoring of Distributed Energy Resources in Smart Microgrids