This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON SMART GRID

Wavelet-Based Event Detection Method

Using PMU Data
Do-In Kim, Student Member, IEEE, Tae Yoon Chun, Student Member, IEEE,
Sung-Hwa Yoon, Student Member, IEEE, Gyul Lee, Student Member, IEEE,
and Yong-June Shin, Senior Member, IEEE

AbstractIn order to deal with the nonstationary signatures

of phasor measurement units (PMU) signals, this paper presents
a wavelet-based detection algorithm. Moreover, for an application to PMU for event detection purpose, it is necessary for us
to classify detected events into unexpected real power related
accidents, such as generator trip or automated control, such
as reactive power injection. The proposed normalized wavelet
energy function calculates the root mean square (RMS) of detail
coefficients from time-synchronized voltage and frequency that
reflect nonstationary occurrence of significant changes in signals. For a robust detection, wavelet-based detection parameter
is designed with consideration of nonstationary characteristics of
events. Also, there are distinct transients in voltage and frequency
caused by different event types, and distinct results are key-idea
of event classification. Besides the determination of event occurrences, one can obtain the information of event characteristics
that include event types and zonal information of event from the
proposed method. Moreover, successful results of detection and
classification in real-world cases are presented in this paper.
Index TermsDetection, discrete wavelet transform, generator
trip, phasor measurement units (PMU), reactive power control,
synchrophasor.

I. I NTRODUCTION
ONITORING wide area power systems has become an
important issue for the reliable operation of modern
power systems that have become complicated by developments
of new technology including increase in energy demands,
renewable resource, and high-voltage dc (HVDC) transmission. In order to provide the information to prevent or relieve
the serious malfunctions in power systems, wide area monitoring systems have been implemented to have a capability
of detecting abrupt symptoms in real-time. Implementation of
phasor measurement units (PMUs) is considered one of the
most efficient real-time monitoring devices as it can improve
the ability to monitor wide area of the electric power grid. For

Manuscript received March 18, 2015; revised July 29, 2015 and
August 30, 2015; accepted September 1, 2015. This work was supported
in part by the National Research Foundation (NRF) of Korea through
the Ministry of Science, Information and Communications Technologies,
and Future Planning under Grant NRF-2014R1A2A1A01004780 and
Grant NRF-2012M2A8A4055236, and in part by the Korea Electric
Power Corporation Research Institute through the Korea Electrical
Engineering and Science Research Institute under Grant R13TA20.
Paper no. TSG-00317-2015. (Corresponding author: Yong-June Shin.)
The authors are with the School of Electrical and Electronic Engineering,
Yonsei University, Seoul 120-749, Korea (e-mail: yongjune@yonsei.ac.kr).
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TSG.2015.2478421

example, the root mean square (RMS) data of voltage and current can be obtained at the rate of 30 or 60 samples/s by PMU
system, while they are acquired at the rate of 4 or 5 samples/s
in the traditional supervisory control and data acquisition system [1]. Therefore, signals measured by PMU are expected
to have a precision high enough to represent events that contribute to the serious malfunctions in power systems such as
large area blackout [2], [3].
Nevertheless, nonstationary characteristics in PMU data
make detection task difficult because there is no tendency
or fundamental frequency in PMU signals. There have been
numerous discussions about event detection to determine the
event occurrences that are represented in PMU signals [4], [5].
Examples of event detection techniques include voltage and
frequency stability diversities between each moving average
value [6], linear principal component analysis based approach
to analyze dimensional reduction of synchrophasor data [7],
and geographical visualization of PMU data [8]. This paper
is also motivated by the needs of automated real-time monitoring for synchrophasor data to enable potential applications
of power system monitoring such as [9] and [10], and proposes a PMU-based event detection technique to have robust
automatic detection capability.
The wavelet technique is an efficient signal processing
technique for PMU data that exhibit nonstationary characteristics [11]. While, there is a fair amount of research regarding
wavelet-based detection techniques for sinusoidal signals such
as [12][15], application to PMU signal requires additional
research effort because of differences in signal characteristics. Instead of harmonic analysis, a principle of wavelet-based
detection is based on abrupt changes in wavelet coefficients
caused by events such as generator trip or voltage controls.
Furthermore, in order to improve detection performance by
reducing the false alarm, we propose a normalization of variant signatures caused by instant time [16], [17] and zonal
difference in power system [18], [19].
As an application of event detection technique, detected
events can be recognized into unexpected real power events
such as generator trips, line faults, failure of substation, etc.
or automated power system control includes compensation
from capacitors, reactors, or generators. Classification is also
important with respect to power system operation as event
classification enables system to determine whether system
should take additional actions or not. In other words, real
power events require optimized actions to relieve the serious

c 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
1949-3053 
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
2

IEEE TRANSACTIONS ON SMART GRID

malfunctions of the system, whereas, reactive power events

routinely occur to maintain the voltage stability. The key-idea
of proposed classification method is to observe the difference
between voltage and frequency signal because real power event
and reactive power event have different influences on the voltage and frequency signal, respectively. The entire process of
event detection including recognition of event type is applied
to real-world data in Korea [20].
This paper is organized as follows. Section II presents
the targets of event detection represented in real-world
PMU signals and discusses how different types of events
have effects on voltage and frequency signals. Section III
demonstrates theoretical background of wavelet analysis and
Section IV explains the event detection method using wavelet
analysis demonstrated in Section III. The event detection
technique is applied to real-world PMU data, and detection results are discussed in Section V. Applications of
proposed techniques to real-world PMU data show the
successful application of event detection and classification
algorithm.
II. F ORMULATION OF D ETECTION TARGET
In a multimachine power system, signals measured at nearby
buses will show coherent oscillation characteristics [21].
Therefore, waveforms of transient responses that are caused
by occurrences of power system dynamics exhibit similar transient waveform among the nearby buses. However,
modeling transient phenomena in a real-world power system is complicated, therefore a reduced two-area power
system model is used as an efficient way to interpret transient phenomena. Accordingly, the reduced two-area model
shown in Fig. 1 is composed of the equivalent generators
G1 G2 and the equivalent loads L1 L2 , respectively, in
each electrical area [21]. Assumed equivalent loads are influenced by frequency-independent load variation and frequencydependent load variation [22]. In order to explain the transient
without random variation of the load caused by electrical
demands, only frequency-dependent load variation is considered, and linearized to i fi , where frequency of ith area fi is
multiplied by positive coefficient i .
Theoretical background of event classification using voltage and frequency monitoring is based on principles that the
change in the real power affects the frequency, while the
change in the reactive power affects the voltage. Thus, oscillations in voltage and frequency signals are either oscillations
of frequency depending on the system inertia, or oscillations
of bus voltage magnitude depending on rotor angle deviations
and system organizations [23].
Unexpected generator trip event signals can represent events
caused by real power variation, and is defined as a real-power
event in this paper. In other words, transient response of generator trip can be explained by the general tendency of real
power event. It is reasonably assumed that a generator trip
in area 1 causes the negative variation of mechanical power
output in area 1 Pm,i and positive variation of real power
flow from areas 2 to 1 Ptie as a compensation of generator
trip. Deviation in frequency has negative value according to

Fig. 1.

Two-area with two-generator system.

the application of swing equations to small deviation in ith

area as follows:

di
1 
Pm,i i i + Ptie
(1)
=
dt
2Hi
where total inertia constant of each area H1 , H2 can be theoretically obtained from inertia constant of each generator [23],
and variation of electrical output is i i and (Pm,i +Ptie )
have negative values. Taking the Laplace transform of (1) is
followed as:


1
Pm,i + Ptie .
(2)
i (s) =
2Hi s + i
Finally, we can obtain frequency variation in time domain
representation as follows:

1
1 t
(Pm,1 + Ptie )e 2H1
(3)
2H1

1
2 t
2 (t) =
(Ptie )e 2H2
(4)
2H2
where (3) and (4) represent the frequency variations in
areas 1 and 2, respectively. From the solution of (3) and (4),
variations of frequency 1 and 2 clearly have negative steady state error. On the other hand, the interaction
between voltage and frequency controls is weak and relationship between voltage and frequency has not been thoroughly
analyzed [11], [22]. Instead, by linearizing swing equation and
voltage magnitude relation, we obtain estimation of V that
is defined as the variation of bus voltage caused by real power
event as follows:

1 (t) =

V(x) = f (x)i

(5)

where x indicates the bus locations, V(x) is the voltage variation at the bus indicated by x, i is rotor angle deviation of
ith area bus caused by frequency deviation, and f (x) is location related Jacobian function [23]. From the development of
equations that indicate the indirect relation between the voltage and real power, we can conclude that the frequency signal
functions as an indicator of real power event occurrences.
Unlike real power events, reactive power events are caused
by generators, capacitors, or reactors, and primarily affects the
local voltage magnitudes [25], [26]. Reactive power events are
hard to explain using a reduced model, therefore, we consider
the multiple buses at each area. Reactive power injection in the
ith area is considered as one of the reactive power events, and
total reactive power injections are determined by the weighted
linear combination of generated compensation values in electrical area. The local reactive power compensation at the ith
area can be formulated by [25]
Qi =

b

j=1

PFj D(V(x))

(6)

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
KIM et al.: WAVELET-BASED EVENT DETECTION METHOD USING PMU DATA

(a)

that wavelet has advantages in application to nonstationary signals; due to the fact that wavelet bases are obtained by scaled
and translated versions of a time-localized bases. In wavelet
analysis, expansions of a signal s(t) are represented by coefficients j,k , scale factor j, and translation factor k of designed
basis as follows [32]:

j,k j,k (t)
(7)
s(t) =
k

(b)
Fig. 2. Real-world PMUs data including generator trip event (event 1).
(a) Voltage signal. (b) Frequency signal.

where Qi is total reactive power compensation in the ith area

caused by local bus voltage disturbance, D(V(x)) is a function of voltage variation from the voltage stability range, PFj is
the participation factor, and b is the number of reactive power
source in the ith area [27]. Local bus groups that have a coherent voltage oscillation followed by reactive power variation
are determined by electrical distances. In other words, voltage
variations caused by reactive power events are scarcely shown
at far locations with reactive power sources [26]. On the contrary to real power event, voltage variation will function as an
indicator of reactive power event; due to a weak interaction
between reactive power and frequency [22].
As discussed in Section II, a real power event is characterized by event normalized wavelet energy (NWE) values that
exceed the threshold at both voltage and frequency simultaneously, while a reactive power event is characterized by
event NWE at voltage. Examples of two different type events
measured from real-world PMU are provided in Fig. 2 about
both voltage and frequency, and defined as event 1. From the
observation of the voltage signal in Fig. 2(a) and (b), one
can visually identify two edge points at the voltage signal in
Fig. 2(a) and one edge point at frequency signal in Fig. 2(b),
respectively. In a real-world power system, the process of
maintaining the voltage stability using reactive power events
occurs repeatedly. However, unexpected real power events can
contribute to the serious failure of power system. Therefore,
useful monitoring systems must be capable of distinguishing or
segregating the second edge point (generator trip) from the first
edge point (voltage control). Proposed event detection methods
can distinguish the events by determining event occurrences
in the frequency signal using a wavelet-based detector.

III. WAVELET A NALYSIS

Wavelet analysis is used as an efficient tool to analyze
the transient phenomena. Wavelet expansion transforms the
1-D function into a 2-D sequence of coefficients using timelocalized bases. The important point of wavelet expansion is

where time-localized j,k is defined as wavelet basis. It is a

notable fact that wavelet analysis can provide efficient localization in both time and frequency domains using multiresolution
analysis (MRA) conditions where signal decomposition is in
terms of the different scale bases. In order to construct bases
that are satisfying the multiresolution condition, we will start
from defining the scaling function j,k , then wavelet function
j,k is then determined from the scaling function in order to
satisfy the MRA condition as follows [32]:
(t) =
(t) =

2
2


n=

h0 (n)(2t n)

(8)

h1 (n)(2t n)

(9)

n=

where h0 is low-pass filter coefficient, and h1 is high-pass

filter coefficient, respectively. Thus, (7) can be represented by
signal expansion using wavelet bases in (8) and (9) and result
is represented as
s(t) =


k

cj0 (k)j0 ,k (t) +



dj (k)j,k (t)

(10)

k j=j0

where signal expansion is divided into two components; one

is combination of approximation coefficient cj,k and scaling
function j,k , and the other is combination of detail coefficient dj,k and wavelet function j,k . Coefficients represent a
range from local to global characteristics of the signal, because
their associated functions have different time-frequency scales.
Projection operation is calculated by discrete-time subband
decomposition of input signals using filtering followed by
downsampling [32]. Scaling function and wavelet function
are typically chosen to form orthogonal bases, which have
an advantage in the calculation of coefficients and conservation of the Parseval theorem. If the two basis families are
orthogonal, approximation and detail coefficients can be easily
computed through inner products such as cj,k = s(t), j,k (t)
and dj,k = s(t), j,k (t).
An example of filter bank application to the real-world
PMUs data of event 1 is shown in Fig. 3 using the haar
wavelet function. Fig. 3(a) is the original voltage signal measured at the reference PMU of Korea. Fig. 3(b)(d) represents
calculated detail coefficients that are relevant to d2 , d1 , and d0
in (10). For the signal that has 60 Hz sampling rate, d2 , d1 ,
and d0 represent bandwidth of 3060, 1530, and 7.515 Hz,
respectively. The 2-D representation of real-world PMU data
has some significant benefits compared to the regular 1-D
representation in event detection perspective. From the observation of the signal and coefficients, there is increasing through

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
4

IEEE TRANSACTIONS ON SMART GRID

(a)

(a)

(b)

(c)

(b)

(d)
Fig. 3. Wavelet decomposition of event 1 voltage signal provided in Fig. 2(a).
(a) Voltage signal. (b) d2 coefficients. (c) d1 coefficients. (d) d0 coefficients.

d2 , d1 , and d0 at event points caused by change in the voltage

magnitude. This increase in detail coefficients is the key-idea
of the proposed event detection algorithm.
IV. P ROPOSED WAVELET-BASED E VENT D ETECTION
A. Proposed Event Detection Method
As discussed in Section III, the key-idea of wavelet-based
detection is monitoring the energy of detail coefficients within
the moving window, and determining whether the energy value
exceeds a threshold or not [29]. The reason of designing the
narrow moving window is to retain both reliability and sensitivity of the detection performance simultaneously. Window
length has excessive influencing for results because the energy
range depends on window length. In an effort to prevent the
confusion caused by difference in window length, we choose
a nominal RMS value of coefficients as a criterion that can
reduce effects by window length. Also, designed moving window swin is only applied to past samples that can be used for a
real-time monitoring system, and represented in discrete signal
as follows:
swin [n] = s[p N + n], n = 1, 2, . . . , N

(11)

where N is the number of moving window samples from

current sample p. Detail coefficients are calculated by inner
product between signal and wavelet function to
dj,k [n] = swin [n], j,k [n].

(12)

Then, we define the modified wavelets energy (MWE) function to calculate the RMS value of detail coefficients in moving
window as follows:


Nj
J 
 


1
dj,k [n]2
MWE(n; N, J) = 
(13)
N
j=1 k=1

Fig. 4. Application of real-world PMU data provided in Fig. 2. NWE of

(a) voltage and (b) frequency.

where J is wavelet decomposition level and Nj is the number of

coefficients in each decomposition level j. In order to calculate
the MWE(n; N, J), N, J, and wavelet function j,k defined
in (9) should be determined in advance.
In addition to function deployment, MWE(n; N, J) function
should be normalized to be independent from different characteristics caused by measuring locations and time instance. The
purpose of normalization is to improve the detection reliability by reducing the effects of nonevent disturbance. The PMU
data that have high variability without events can cause the
high values of MWE. For this reason, high-variable signal is
shown as an invalid event when it is compared to low-variable
signal with a typical threshold, These high-value MWE can
contribute to increase of false alarm.
Variability in power system signals shows gradual increase
according to development in power system, and also have a
periodic characteristics according to year, season, week, or
day [8]. One of the efficient methods to define the distortion
characteristics is dividing time instance into units of hours
because a 1 h duration represents both tendency and stationary
characteristics of signals. In a 60 Hz power system, values
of MWE(n; 21 600, J) become a proposed reference value for
normalization where 21 600 samples compose the 1 h duration.
By dividing MWE function into a reference MWE function, a
normalized function can be defined as a designed detector by
MWE(n; N, J)
(14)
MWE(n; 21 600, J)
where the NWE function represents the normalized value of
the MWE function.
Proposed NWE is applied to PMU signals in Fig. 2, and
results are illustrated for both voltage and frequency in Fig. 4.
While, the results nearly have a value of 1 in ambient time,
the value is higher than 1 at the event point excepting the
frequency NWE of the reactive power event. As discussed in
Section II, a notable difference between real power event and
reactive power event is different event occurrence at voltage
NWE(n; N, J) =

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
KIM et al.: WAVELET-BASED EVENT DETECTION METHOD USING PMU DATA

Fig. 6.

Fig. 5.

Diagram of installed PMU in Korea 345 kV substations.

Flowchart of event classification process.

TABLE I
C OMPARISON OF NWE V IA M OTHER WAVELET AND
E ACH D ECOMPOSITION L EVEL

(a)

(b)
Fig. 7. Real-world PMUs data including event 2 case. (a) Voltage signal.
(b) Frequency signal.

and frequency signals. Process of proposed event classification

is illustrated in Fig. 5, and flowchart includes embodiment process of event area. Estimated locations of reactive power event
case are specified into one area, which shows the high value
of NWE. However, estimation of event area in real power event
case has less reliability than reactive power event because real
power event is detected through the whole power system area.
B. Parameter Decision for Event Detection
Selecting the best mother wavelet is based on numerical
experiment [34], and the performance comparison of mother
wavelets is illustrated in this section. In this research, used
whole test cases are 105 PMU data through the five days
that include generator trip event and failure of 154 kV substation, and window length is determined to 32 samples (0.53 s)
to achieve the fast detection and to calculate the four level
wavelet decomposition. As there are variety kinds of designed
mother wavelets, it is not easy to compare the whole kinds of
mother wavelet. Thus, typical wavelets are chosen to targets of

comparison and Daubechies (DB) 1-20 wavelet series results

are compared in Table I.
In the results in Table I, peak NWE values at event time are
calculated to select the best mother wavelet and decomposition
level. Since high value of NWE indicates high performance,
larger than 10 NWE values are high performance indicator in
Table I. From the NWE values through the levels 13 DB 1
wavelet that show larger value than 10, it is reasonable to
select the DB 1 wavelet with three level decomposition as
a best wavelet for both voltage and frequency signal analysis in this research. This phenomenon is caused by the
trade-off relation between time and frequency, and DB 1 is
highest time-localized wavelet among the Daubechies wavelet
series [35].
The threshold () in Fig. 5 is determined from statistical
analysis over 1 h ambient NWE distribution. Theoretically,
since steady state NWE value is 1, values of ambient NWE are
distributed around the 1 for voltage and frequency. However,
in order to maintain the 99.9% confidence level of event detection, the values for voltage and frequency are selected out of
distribution of NWE in ambient to 3.16 and 3.38, respectively.
Note that the value of threshold may be different dependency on statical characteristics of PMU data and confidence
level.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
6

IEEE TRANSACTIONS ON SMART GRID

Fig. 8.

Application of real power event provided in Fig. 7. (a) Voltage MWE. (b) Frequency MWE.

Fig. 9.

Application of real power event provided in Fig. 7. (a) Voltage NWE. (b) Frequency NWE.

V. A PPLICATION TO R EAL -W ORLD PMU DATA

A. Description of Data
The real-world synchrophasor data of Korea are provided
by Korea Electric Power Corporation. This data set includes
information of remote generator trip events for different
days and reactive power events around generator trip events.
Location and connection information of installed PMU in
Korea is mapped in Fig. 6 with PMUs numbered to P1P34.
PMUs collect synchrophasor data of voltage, current, and
frequency with 60 Hz sampling rate. According to known electrical transmission line configuration in Korea, Fig. 6 shows
electrical coherent zones to zones 14. Concepts of electrical distance and zone are developed by coupling between
buses in [27] and [28], and we confirm that buses at same
zone show similar shapes in voltage, current, and frequency
signals.
Data for event 2 are provided in Fig. 7 with real-world
PMUs signals, and show voltage and frequency signals
for 80 min including real and reactive power event. Real
power event causes sharp variations of voltage and frequency
signals at around 60 min, whereas, reactive power event
causes sharp variations of partial voltage signals at around
10 min. By using the proposed wavelet-based algorithm,
event 2 is automatically detected rather than observing the
signals.

B. Detection Results and Discussion

Assuming that the PMU data is being processed in realtime
manner, NWE(n; 32, 3) is calculated instantaneously by shifting the present sample step by step for all the PMUs. Results
of NWE(n; 32, 3) should be represented then by both time
domain and PMU locations instantaneously, in order to detect
and classify events. For this purpose, results are visualized in
the 3-D graph as illustrated in Figs. 810 with 5 min duration.
In each figure, NWE is calculated at 34 PMUs and its value are
plotted over the time (min) and PMU number (PMU #). Note
that the nominal value of the plot is 1 in the ambient case.
Unlike monitoring a single PMU as illustrated in Fig. 4,
Figs. 810 exhibit the result of (13) and (14) at corresponding
PMUs. In the 3-D graph results of voltage NWE [Fig. 9(a)],
It is confirmed that NWE at the generator trip event point
(around 60 min) has values between 2.50 and 12.40 that are
clearly higher than nominal value of 1. At the same time,
NWE about frequency data shown in Fig. 9(b) have values
between 9.06 and 17.40 that are higher than corresponding
voltage results. These results support the fact that frequency
works as the main indicator of real power event caused by
direct relationship with the real power event as described
in (3) and (4).
In order to demonstrate importance of normalization process, we compare the results of MWE [Fig. 8(a)] and NWE

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
KIM et al.: WAVELET-BASED EVENT DETECTION METHOD USING PMU DATA

Fig. 10.

Application of reactive power event provided in Fig. 7. (a) Voltage NWE. (b) Frequency NWE.

(a)

(a)

(b)

(b)

Fig. 11. Peak values of NWE at real power event point. (a) Voltage data.
(b) Frequency data.

Fig. 12. Peak value of NWE at reactive power event point. (a) Voltage data.
(b) Frequency data.

[Fig. 9(a)] for the event 2 case. In the event 2 case, voltage
signals measured at P10P13 PMUs have a high disturbance
at ambient time. For this reason, P10P13 PMUs in Fig. 8(a)
have notably high values for the entire observation time, and
those high value in ambient time will be the cause of invalid
detections. Whereas, there is no notable value in NWE for
ambient time excepting the event at around 60 min as shown
in Fig. 9(a). It means that NWE is more effective detector
to improve detection reliability than MWE considering the
application to real-world data.
Beside the successful detection of the real power event, a
reactive power event is also detected as a high value of NWE
between 4.48 and 29.00 at around 10 min, and plotted over
the time and PMU number in Fig. 10. From the observation
of Fig. 10, the voltage data of zone 2 are detected as an event.
As shown in Figs. 2 and 7, there is no sharp signature in the
frequency signal of the reactive power event case, thus there
is no detected point in Fig. 10(b). It also shares context that
frequency works well as a real power event indicator discussed
in Section II.

Furthermore, localized information of an event can be

obtained from peak NWE values. While frequency signals
have similar transient waveforms for all PMUs, voltage has
high dependency on zonal relations. From the observation of
Fig. 11 which shows the peak NWE values of real power
event, voltage NWE has a notably different value between
zones in Fig. 11(a). Based on the principle that nearby PMUs
with event location tend to be influenced by events, Fig. 11
exhibits the relation between NWE and event locations. If the
PMUs measurement covers the entire area, we could determine
which area has events. In other words, event 2 case exhibits
the high value in zones 3 and 4 depicted in Fig. 6, where
zones 3 and 4 is considered as nearby area with the event
location that is located in downstream of the monitoring area.
Also, each area has a high value at P2, P3, P21, P25, and P26,
respectively, and the distribution is also consistent with event
location.
The notable difference between Figs. 11 and 12 is that
occurrence of reactive power event is limited in the voltage
data of zone 2. It means that we can specify the event location

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
8

IEEE TRANSACTIONS ON SMART GRID

into one zone. For this reason, the estimation of event location is more accurate in reactive power event than real power
events. Nevertheless, the reactive power events case has the
limitation that events can only be detected in the monitoring
area unlike real power event. Regarding the frequency analysis,
Fig. 12(b) shows that reactive power events barely influence
the frequency as discussed in Section II.
The successful detection results of event 2 case are illustrated through Figs. 812. Owing to the limited space, we
cannot present results of other events, but the proposed algorithm can successfully detect and classify the real-world events
similar to the event that we discussed in this paper. As a result
we can prove the efficacy of proposed detection method and
their application to Korea power system.
VI. C ONCLUSION
In this paper, monitoring the time-synchronized voltage and
frequency signal is proposed as a method for determining the
event occurrence and classifying event types. For this purpose, normalized value of the wavelet coefficient energy is
proposed as a criterion for determining event occurrence, and
is defined as an NWE function. The proposed NWE is applied
to real-world PMU event data set that includes generator trip
event cases and reactive power events. Results are presented
then with 3-D graphics to be plotted over the time and PMU
number, and results indicate that the proposed technique can
provide a robust detection of power system events. Besides
the detection of event occurrences, one can obtain the localized information of events that include event types and zonal
information of event occurrences via the proposed method.
Thus, the proposed method can be considered as a successful real-time monitoring tool on PMU, and has the possibility
to become real-time event location estimator and event classifier. In future works, we will demonstrate the more specified
relationship between NWE and event locations, which will be
used to find event locations.
R EFERENCES
[1] S. G. Ghiocel et al., Phasor-measurement-based state estimation for
synchrophasor data quality improvement and power transfer interface
monitoring, IEEE Trans. Power Syst., vol. 29, no. 2, pp. 881888,
Mar. 2014.
[2] J. De La Ree, V. Centeno, J. S. Thorp, and A. G. Phadke, Synchronized
phasor measurement applications in power systems, IEEE Trans.
Smart Grid, vol. 1, no. 1, pp. 2027, Jun. 2010.
[3] Y. V. Makarov et al., PMU-based wide-area security assessment:
Concept, method, and implementation, IEEE Trans. Smart Grid, vol. 3,
no. 3, pp. 13251331, Sep. 2012.
[4] F. Hashiesh, H. E. Mostafa, A. R. Khatib, I. Helal, and M. M. Mansour,
An intelligent wide area synchrophasor based system for predicting and
mitigating transient instabilities, IEEE Trans. Smart Grid, vol. 3, no. 2,
pp. 645652, Jun. 2012.
[5] 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. 18321840, Dec. 2012.
[6] K. Seethalekshmi, S. N. Singh, and S. C. Srivastava, A synchrophasor
assisted frequency and voltage stability based load shedding scheme for
self healing of power system, IEEE Trans. Smart Grid, vol. 2, no. 2,
pp. 221230, Jun. 2011.
[7] L. Xie, Y. Chen, and P. R. Kumar, Dimensionality reduction of synchrophasor data for early event detection: Linearized analysis, IEEE
Trans. Power Syst., vol. 29, no. 6, pp. 27842794, Nov. 2014.

[8] A. Kaci, I. Kamwa, S. A. Dessaint, and S. Guillon, Synchrophasor

data baselining and mining for online monitoring of dynamic security limits, IEEE Trans. Power Syst., vol. 29, no. 6, pp. 26812695,
Nov. 2014.
[9] A. G. Phadke and J. S. Thorp, Synchronized Phasor Measurements and
Their Applications. New York, NY, USA: Springer, 2008.
[10] R. Sodhi, S. C. Srivastava, and S. N. Singh, A simple scheme for wide
area detection of impending voltage instability, IEEE Trans. Smart Grid,
vol. 3, no. 2, pp. 818827, Jun. 2012.
[11] W. Gao and J. Ning, Wavelet-based disturbance analysis for power
system wide-area monitoring, IEEE Trans. Smart Grid, vol. 2, no. 1,
pp. 121130, Mar. 2011.
[12] F. B. Costa, Fault-induced transient detection based on real-time analysis of the wavelet coefficient energy, IEEE Trans. Power Del., vol. 29,
no. 1, pp. 140153, Feb. 2014.
[13] D. C. Robertson, O. I. Camps, J. S. Mayer, and W. B. Gish, Wavelets
and electromagnetic power system transients, IEEE Trans. Power Del.,
vol. 11, no. 2, pp. 10501058, Apr. 1996.
[14] U. D. Dwivedi and S. N. Singh, Enhanced detection of power-quality
events using intra and interscale dependencies of wavelet coefficients,
IEEE Trans. Power Del., vol. 25, no. 1, pp. 358366, Jan. 2010.
[15] A. E. B. Abu-Elanien and M. M. A. Salama, A wavelet-ANN technique
for locating switched capacitors in distribution system, IEEE Trans.
Power Del., vol. 24, no. 1, pp. 400409, Jan. 2009.
[16] I. M. Nejdawi, A. E. Emanuel, D. J. Pileggi, M. J. Corridori, and
R. D. Archambeault, Harmonics trend in NE USA: A preliminary
survey, IEEE Trans. Power Del., vol. 14, no. 4, pp. 14881494,
Oct. 1999.
[17] A. A. Fouad, S. Venkataraman, and J. A. Davis, An expert system for
security trend analysis of a stability-limited power system, IEEE Trans.
Power Syst., vol. 6, no. 3, pp. 10771084, Aug. 1991.
[18] A. C. Rueda-Medina and A. Padilha-Feltrin, Distributed generators as
providers of reactive power supportA market approach, IEEE Trans.
Power Syst., vol. 28, no. 1, pp. 490502, Feb. 2013.
[19] A. Kumar, S. C. Srivastava, and S. N. Singh, A zonal congestion management approach using real and reactive power rescheduling, IEEE
Trans. Power Syst., vol. 19, no. 1, pp. 554562, Feb. 2004.
[20] J. Kim et al., Study of the effectiveness of a Korean smart transmission grid based on synchro-phasor data of K-WAMS, IEEE Trans.
Smart Grid, vol. 4, no. 1, pp. 411418, Mar. 2013.
[21] N. Senroy, Generator coherency using the HilbertHuang transform,
IEEE Trans. Power Syst., vol. 23, no. 4, pp. 17011708, Nov. 2008.
[22] H. Saadat, Power System Analysis. New York, NY, USA: McGraw-Hill,
1999.
[23] L. Fan, Z. Miao, and Y. Wehbe, Application of dynamic state and
parameter estimation techniques on real-world data, IEEE Trans.
Smart Grid, vol. 4, no. 2, pp. 11331141, Jun. 2013.
[24] J. H. Chow, A. Chakrabortty, L. Vanfretti, and M. Arcak, Estimation
of radial power system transfer path dynamic parameters using synchronized phasor data, IEEE Trans. Power Syst., vol. 23, no. 2,
pp. 10821094, May 2008.
[25] A. Keane, L. F. Ochoa, C. J. Dent, and G. P. Harrison, Enhanced utilization of voltage control resources with distributed generation, IEEE
Trans. Power Syst., vol. 26, no. 1, pp. 252260, Feb. 2011.
[26] B. A. Robbins, C. N. Hadjicostis, and A. D. Dominguez-Garcia, A twostage distributed architecture for voltage control in power distribution
systems, IEEE Trans. Power Syst., vol. 28, no. 2, pp. 14701482,
May 2013.
[27] J. Zhong, E. Nobile, A. Bose, and K. Bhattacharya, Localized reactive
power markets using the concept of voltage control areas, IEEE Trans.
Power Syst., vol. 19, no. 3, pp. 15551561, Aug. 2008.
[28] P. Lagonotte, J. C. Sabonnadiere, J. Y. Leost, and J. P. Paul, Structural
analysis of the electrical system: Application to secondary voltage control in France, IEEE Trans. Power Syst., vol. 4, no. 2, pp. 479486,
May 1989.
[29] M. H. J. Bollen and I. Y. H. Gu, Signal Processing of Power Quality
Disturbances. Hoboken, NJ, USA: Wiley, 2003.
[30] C. A. Gross, Power System Analysis. Hoboken, NJ, USA: Wiley, 1986.
[31] P. K. Ray, N. Kishor, and S. R. Mohanty, Islanding and power quality disturbance detection in grid-connected hybrid power system using
wavelet and S-transform, IEEE Trans. Smart Grid, vol. 3, no. 3,
pp. 10821094, Sep. 2012.
[32] C. S. Burrus, R. A. Gopinatha, and H. Guo, Introduction to Wavelets
and Wavelet Transforms: A Primer. Upper Saddle River, NJ, USA:
Prentice-Hall, 1998.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
KIM et al.: WAVELET-BASED EVENT DETECTION METHOD USING PMU DATA

[33] N. W. A. Lidula and A. D. Rajapakse, A pattern recognition approach

for detecting power islands using transient signalsPart I: Design and
implementation, IEEE Trans. Power Del., vol. 25, no. 4, pp. 30703077,
Oct. 2010.
[34] T. Y. Chun et al., Selection of optimal wavelet basis for fault detection of PMU data, in Proc. 20th Int. Conf. Elect. Eng., Jeju, Korea,
Jun. 2014, pp. 406410.
[35] Y.-J. Shin, E. J. Powers, W. M. Grady, and S. C. Bhatt, Optimal
Daubechies wavelet bases for detection of voltage sags in electric power
distribution and transmission systems, Proc. SPIE Conf. Wavelet Appl.
Signal Image Process. VII, vol. 3813, pp. 873883, Jul. 1999.

Sung-Hwa Yoon (S14) received the B.S. degree

in electrical and electronic engineering from Yonsei
University, Seoul, Korea, in 2014, where he is currently pursuing the M.S./Ph.D. degree.
He has been a Research Assistant with the
Power Information Technology Laboratory, Yonsei
University, since 2014. His current research interests include applications of phasor measurement unit
for power systems stability, fault detection/location,
control, power systems, and applications of phasor
measurement unit.

Do-In Kim (S14) received the B.S. degree from

Yonsei University, Seoul, Korea, in 2014, where he
is currently pursuing the M.S./Ph.D. degree with the
Power IT Laboratory, all in electrical and electronic
engineering.
Since 2013, he has been a Research Assistant
with the Power IT Laboratory, Yonsei University. His
current research interests include wide-area power
system monitoring and applications of phasor measurement units.

Tae Yoon Chun (S15) received the B.S. and

M.S. degrees from Yonsei University, Seoul, Korea,
in 2010 and 2012, respectively, where he is currently pursuing the Ph.D. degree with the Control
Engineering Laboratory, all in electrical and electronic engineering.
Since 2012, he has been a Research Assistant
with the Control Engineering Laboratory, Yonsei
University. His current research interests include
approximate dynamic programming/reinforcement
learning, optimal/adaptive control, nonlinear control
theories, synchrophasor, and power systems applications.

Gyul Lee (S15) received the B.S. degree in electrical engineering from Yonsei University, Seoul,
Korea, in 2015, where he is pursuing the M.S./Ph.D.
degree with the Power IT Laboratory, School of
Electrical Engineering.
He is participating in the power system analysis
team. His current research interests include power
system data management algorithm, such as compression of synchrophasor data and big data analysis for data mining from advanced power quality
monitoring devices.

Yong-June Shin (S98SM04) received the B.S.

(Hons.) degree in electrical engineering from Yonsei
University, Seoul, Korea, in 1996; the M.S. degree
in electrical engineering and computer science from
the University of Michigan, Ann Arbor, MI, USA,
in 1997; and the Ph.D. degree in electrical and
computer engineering from the University of Texas
at Austin, Austin, TX, USA, in 2004.
He joined the Department of Electrical
Engineering, University of South Carolina,
Columbia, SC, USA, as an Assistant Professor,
where he was promoted to an Associate Professor with tenure in 2011. He
joined the School of Electrical and Electronic Engineering, Yonsei University,
as an Associate Professor, in 2012. His current research interests include
application of novel digital-signal-processing techniques to a wide variety of
important transient and nonlinear problems in smart electric power grids.
Prof. Shin was a recipient of the United States National Science Foundation
CAREER Award in 2008, and the General Electric Korean-American
Education Commission Scholarship.