A Matlab and PowerFactory-based

WAMS Simulator
R. Cardenas-Javier∗ , V. Gonzalez-Sanchez∗ , M.R.A. Paternina∗ , F.A. Zelaya A.∗ ,
A. Zamora† , V. Torres∗ , and D. Dotta‡
∗ UNAM, Mexico City, Mexico
† U.
Michoacana, Morelia, Mexico
‡ , UNICAMP, Campinas, Brasil

rcardenasj[vgonzalez][fzelaya223]@comunidad.unam.mx,
mra.paternina[vtorres]@fi-b.unam.mx, [email protected],[email protected]

Abstract—Wide-area monitoring systems (WAMS) are spread of them are the power system modal extraction algorithms,
in many power systems around the world, which demonstrate which capture the modal information from raw signals such
their significance and potential for power system operators. Thus, as damping, frequency and amplitude or phase. Two ap-
this paper emphasizes the use of synchronized-measurements-
driven applications in power systems. To this end, two Matlab- proaches of algorithms can be defined according to [5]; the
embedded applications are developed using the graphical user first one is related to the classical linear ringdown analysis,
interface (GUI) environment of Matlab. The former provides such as the widely used Prony method, the Matrix Pencil
synchrophasor estimates that are processed and stored by means and the Eigensystem Realization Algorithm (ERA); and the
of a phasor estimation algorithm and a database, respectively; second one associated with the nonlinear methods such as
meanwhile the latter consists of a power system modal monitor
application that is able to capture the modal information corre- Hilbert-Huang analysis, variational mode decomposition, or
sponding to low frequency oscillations (LFOs) in power systems. the energy tracking operators approach. More recently, some
Also, the DIgSILENT PowerFactory environment is employed for new sophisticated approaches have proposed like the Digital
running EMT simulations, whose results provide instantaneous Taylor Fourier Transform (DTFT) [6] or the Dynamic Mode
signals to the phasor estimation stage in Matlab. Finally, results Decomposition (DMD) [7].
exhibit a WAMS simulator dealing with a LFO captured by
PMUs, showing the performance of three traditional technique The main contribution of the paper is to present a Matlab-
for identifying electromechanical modes in power systems. and PowerFactory-based software platform that can be used as
Index Terms—Wide-area monitoring system, synchrophasor a teaching tool and research framework to explore and develop
estimates, phasor measurement units, low frequency oscillations. applications involved in the wide-area monitoring systems,
such as the Low Frequency Oscillation (LFO) detector, oscilla-
I. I NTRODUCTION
tion source locator, phase angle Difference detector, islanding
Most of the Wide-Area Monitoring Systems (WAMS) archi- detector, and voltage stability detector.
tectures around the world are owned by the System Operators The remainder of the paper is organized as follows. Section
(SO), causing that many researchers and academics at the II presents a brief description of a generic WAMS architecture,
research centers and universities are restricted to work with a comparing the centralized and decentralized approach. Then,
limited amount of PMUs data (provided to them by the SO) or the Matlab and PowerFactory-based WAMS simulator archi-
with a large amount of data coming from offline simulations. tecture is depicted emphasizing its key blocks in Section III,
This is one of the reason that explains why some interesting which are described in detail together with the synchrophasor
PMUs-based projects are deployed in the last years by the processing and the power system modal monitor. Finally,
universities, such as the low voltage WAMS in USA and concluding remarks are given in Section IV.
Brazil [1], [2], together with the construction of Hardware
in the Loop (HIL) simulators [3], [4]. Despite these projects II. W IDE -A REA M ONITORING S YSTEM A RCHITECTURE
show good results for monitoring, researching and developing Figure 1 illustrates a generic WAMS architecture, whose
WAMS applications, they are limited to the complexity and the main elements consist of PMUs or Sensors, PDC, and WAMS
cost of their implementation, which make them a solution not application; in turn they also need a Wide-Area Network
affordable for all the academic institutions. Thus, this paper (WAN) for communication purposes that provides the capa-
proposes one reliable and less expensive alternative which bility of extending over a large geographical distance. There
consists of the construction of a WAMS simulator using the are two commonly types of WAMS architecture: decentralized
data exchange communication between software packages that and centralized in Table I. The former possesses a better
are commonly used in the academic field. performance in comparison with the latter. Some pros and
In the literature, several powerful algorithms are currently cons are exhibited in Table I. The main difference among
available to analyze wide-area monitoring systems. Some both approaches comes from the fact that the decentralized
978-1-7281-0407-2/19/$31.00
c 2019 IEEE architecture has a bigger redundancy of data compared with
 PDC

WAN

Area 1
Data archiving

WAMS WAC
application/ WAP
Energy .
Area k management .
centre .

Area j Supervisory control

and data acquisition
(SCADA)
Sensor/PMU

Fig. 1. Generic WAMS architecture [8].

Fig. 2. WAMS simulator architecture.
TABLE I
C OMPARISONS AMONG WAMS ARCHITECTURES
provides an EMT simulation kernel for solving power system
Centralized WAMS Decentralized WAMS transient problems, so that instantaneous voltage and current
- Global Monitoring - Local monitoring
- Necessity of high-speed signals are created. A communication procedure is used to
- Regular communication links
communication links interface with the Matlab environment, where Graphical User
- Processing of large amount
of data
- Process only local data Interfaces (GUIs) are built for processing synchronized phasor
- Limited to PMU devices - Several IEDs can be used measurements with their respective time-stamps according to
- Despite of the global view it is - Avoid transfer unnecessary data the Std. C37.118 [9]. In this work, voltage channels in the
hard monitoring all the system and time consuming computer
(computer time burden) analysis abc reference frame are processed rendering positive-sequence
- May used as first level of centralized phasor estimates corresponding to the amplitude, phase angle,
mode estimation
frequency at each bus.

A. Electromagnetic Transients Simulations

the centralized architecture. However, the centralized approach
has been already installed in several power systems around The instantaneous simulated data are produced with a step
the world (Brazil, India, USA, Colombia, Mexico, Ecuador, size of 100 us, employing the commercial software called
El Salvador, Russia, Switzerland, Korea, China). P OWER FACTORY 2018 SP1 [10], since this is in charge to
make the EMT simulations via the RMS/EMT tool. Here,
III. M ATLAB AND P OWER FACTORY- BASED S IMULATOR the simulation is set up including a 3-cycle three-phase fault
at Medway substation (bus 16) belonging to the well-known
The Matlab and PowerFactory-based simulator is a suitable
two-area IEEE New England model presented in Fig. 3. This
platform for exploring new synchronized measurements-driven
reduced model contains 10 generators and 39 buses. The DPL
applications, such as: Low Frequency Oscillation Detector,
is used to create a EMT simulation in the PowerFactory
Oscillation Source Locator, Phase Angle Difference Detector,
environment. The simulation runs creating data packages of
Islanding Detector, and Voltage Stability Detector. These are
the instantaneous voltage signals lasting 20 ms and using a
useful for assessing the dynamic performance of power sys-
sampling rate of 10 kHz. Each package is simultaneously pro-
tems. The main software architecture involved in the Matlab-
cessed by the phasor measurement processing stage in Matlab,
and PowerFactory-based WAMS simulator are shown in Fig.
prior a communication procedure between PowerFactory and
2.
Matlab is established in the DPL milieu.
Figure 2 represents a conceptual scheme of the platform’s
implementation. The main blocks become: (i) PMU and Figure 4 presents the input signals of 3 PMUs corresponding
WAMS application processing; (ii) Electromagnetic transient to the instantaneous voltage in the abc reference frame, and
simulations; (iii) Phasor Data Concentrator; (iv) Communi- located at the substations Northfield (1), Pilgrim (22), and
cation protocol; and (v) Graphical User Interfaces for PMUs Maine Yankee (29). Notice that an analysis buffer lasting 20
and Modal Monitor. The algorithms has been implemented in ms is used for the synchrophasor processing.
both DIgSILENT Programming Language (DPL) and Matlab
programming language. B. Synchrophasor Processing
This platform is embedded into the PowerFactory and Mat- This is the first task that is executed in the Matlab envi-
lab environments as displays in Fig. 2. The former is used to ronment, providing the phasor simulated data to the WAMS
perform Electromagnetic Transients (EMT) simulations, which simulator. To process the phasor measurement, the voltage
 400

amplitude (kV)
G8
Area 2 300

Voltage
200
37
G10
25 26 28 29 100
30 0
PMU-1 PMU-6 PMU-7 PMU-16 PMU-21 PMU-22 PMU-29
27 0 0.5 1 1.5 2 2.5 3
2 38 5

phase angle (rad)

18 17 24 G9
1

Voltage
0
3 G6
Area 1 16 35
-5
G1 15 21 22 0 0.5 1 1.5 2 2.5 3
Time (s)
39 4 14

5 Fig. 5. Amplitude and phase angle in the voltage channels for 7 PMUs.
6 12
19 23
7
13 20 By applying the multivariate analysis equation in (2) of
8 11 36
31 10
the DFT for multiple channels, Fourier coefficients (ξ̂) are
34 33 G7
9 obtained for a set of M signals.
G2 32 G5 G4
†
G3 [ξ̂ 1 ξ̂ 2 ··· ξ̂ m ] = WN [s1 s2 ··· sm ], (2)
where ξ̂ 1 = [ξˆ0 ξˆ1 · · · ξˆN −1 ].
Fig. 3. New England Power Grid with 10-generator and 39-bus.
Thus, phasor parameters as amplitude â and angle ϕ̂ are
computed using (3) for the corresponding fundamental fre-
quency that is equal to 60 Hz. Both synchrophasor estimates
are exhibited in Fig. 5.
channels in the abc reference frame are passed by a Non-
recursive DFT, providing phasor estimates with a sampling â = |ξˆ1 |
(3)
rate of 50 Hz for all signals. ϕ̂ = ∠ξˆ1
The Fourier matrix WN in (1) and its harmonic phase
h In order to create the database after the phasor parameters
factors ωN = ej2π/N in each vector h = 0, . . . , N − 1 (N
are processed, it is assigned a time-stamp to each set of
is the number of sample per cycle), are used to accomplish
estimates. The database in Table II is structured using one
the phasor estimation.
csv file, one column per positive sequence phasor, storing
 
1 1 1 ··· 1 50 blocks of estimates per second for each bus with their
2 (N −1)  corresponding time-stamp. For instance, Table II illustrates
1
 ωN ωN · · · ωN 
2 4 2(N −1)  two blocks of phasor parameters for the PMU-1, the first block
1 ωN ωN · · · ωN

WN =   (1)
.
. .. .. .. ..  of data consists of: Time-stamp (UTC Time), amplitude (â)
. . . . .

 and phase angle(ϕ̂). Once the results are stored in a database,
(N −1) 2(N −1) (N −1)2
1 ωN ωN ··· ωN it is possible to extract them in order to show them in the
Graphical-User Interface (GUI) designed in Fig. 6 by using
the GUIDE toolbox from Matlab. The interface is depicted in
600.00
[kV]
300.00
Analysis buffer Fig. 6.
Bus 01: Phase Voltage A
0.00

-300.00
Bus 01: Phase Voltage B
Bus 01: Phase Voltage C
C. WAMS Application: Power System Modal Monitor
-600.00
20 ms
Once the voltage phasor parameters are available, they
-900.00
1.53E-1..

600.00
6.22E-0.. 1.24E-0.. 1.87E-0.. 2.49E-0.. [s] 3.11E-0..
are stored in a database doing the functions as a PDC in
[kV]
300.00
the WAMS architecture. In consequence, any synchronized
0.00
Bus 22: Phase Voltage A
Bus 22: Phase Voltage B measurements-driven application can become built-in as a
Bus 22: Phase Voltage C
-300.00

-600.00
new functionality in the platform. In this investigation, three-
-900.00
1.53E-1.. 6.22E-0.. 1.24E-0.. 1.87E-0.. 2.49E-0.. [s] 3.11E-0..
well known techniques such as Prony method (PM) [11]–
600.00
[kV]
[13], Eigensystem Realization Algorithm (ERA) [14], and
300.00

0.00
Bus 29: Phase Voltage A
Matrix Pencil (MP) [15]–[17] reported in [5] to identify
Bus 29: Phase Voltage B

-300.00
Bus 29: Phase Voltage C electromechanical modes in power systems are incorporated
-600.00
within the Matlab environment. These techniques are exten-
-900.00
1.53E-1.. 6.22E-0.. 1.24E-0.. 1.87E-0.. 2.49E-0..
39 Bus New England System
[s] 3.11E-0..
EMT_WAM Date: 6/12/2019
sively used to accomplish a ringdown analysis, being able to
DIgSILENT Power Flow Annex: /1
extract the linear model of the system that is also used to
Fig. 4. Input signals of 3 PMUs corresponding to the instantaneous voltage in
evaluate the dynamic performance of the power system under
the abc reference frame, and located at the substations Northfield (1), Pilgrim large disturbances on the basis of the rotor angular stability
(22), and Maine Yankee (29). principles.
 TABLE II
DATABASE STRUCTURE FOR THE WAMS SIMULATOR IMPLEMENTATION USING A CSV FILE .

PMU-1 PMU-6 PMU-7 PMU-16 PMU-21 PMU-22 PMU-29

UTC Time
â [kV] ϕ̂ [rad] â [kV] ϕ̂ [rad] â [kV] ϕ̂ [rad] â [kV] ϕ̂ [rad] â [kV] ϕ̂ [rad] â [kV] ϕ̂ [rad] â [kV] ϕ̂ [rad]
00:07:49 296.72 1.34 286.31 1.27 283.00 1.24 288.90 1.25 283.99 1.21 303.94 1.60 298.84 1.37
00:07:49 247.75 1.12 170.75 1.24 174.09 1.19 0.00 -0.01 0.30 3.10 184.40 1.99 204.14 1.43
00:07:49 251.78 2.34 183.22 2.44 185.85 2.39 54.35 2.13 53.04 2.10 186.11 -3.10 212.17 2.63
00:07:49 283.56 -2.61 262.20 -2.55 260.21 -2.59 254.27 -2.48 249.88 -2.52 263.02 -2.00 280.12 -2.39
00:07:49 287.02 -1.36 269.80 -1.31 267.37 -1.35 264.76 -1.24 260.24 -1.28 273.96 -0.77 284.46 -1.13

Fig. 6. Graphical User Interface for the WAMS Simulator developed in Matlab.

1) Prony Analysis: This is used using three steps, assuming in (5) of the form b = T a .
a recorded signal y(t), whose discrete time representation is 
y(n)
  
 a

y(n − 1) y(n − 2) ··· y(0) 1
given by y(tk ), which possesses N samples that are evenly  y(n + 1)   y(n) y(n − 1) ··· y(1) a
 2
 
spaced by an amount ∆t, in terms of the exponential form the

..  =
 
.. ..

 .  (5)
 .. 

.   . .
Prony method is defined in [11] by

y(N − 1) y(N − 2) y(N − 3) · · · y(N − n − 1) a
n

n
X Step 2 The polynomial coefficients a are found by solving
ŷ(t) = Ai eσi t cos(ωi t + θi ), (4)
i=1 a = (T T T )−1 T T b (6)
Step 3 Then, the roots zi are computed from the polynomial
where n ≤ N is the subset of modes to be determined. form in (7).
The Prony analysis is accomplished carrying out the fol-
lowing steps. z n − (a1 z n−1 + a2 z n−2 + · · · + an−1 z 0 ) = 0 (7)
Step 1 The Toeplitz matrix T is assembled by selecting where the coefficients ai are unknown and calculated from the
record elements, then the Least-Squares (LS) problem is stated set of measurement vectors in (6).
 From the roots zi , the frequencies (fi ) and damping λi 3) Matrix Pencil Method: This method uses the singular
for the i − th mode are calculated in (8). Thus, these modal value decomposition of a Hankel matrix in which its roots
parameters are given by provide the modes associated with the output signals [5].
ln(zi ) Thereby, the same Hankel matrices and SVD procedure in
λi = (11)-(16) are employed. Then, the following step consists of
∆t
 
ˆ λi (8) sorting out the right singular vectors in Vn , as defines in (18).
fi = Im
2π V1 = [v1 v2 v3 ... vn−1 ]
σ̂i = Re(λi ) (18)
V2 = [v2 v3 v4 ... vn ]
The damping ratio ζ(t) of the signal is obtained through the
Then, matrices are given by
relation,
σ̂i Y1 = V1T V1
ζ̂i (t) = , (9) (19)
ω̂i
Y2 = V2T V1
where ω̂i = 2π fˆi . Likewise, the modal energy in [5] may be
estimated in terms of the amplitude â and frequency ω̂ by The desired poles zi are found as the generalized eigenvalues
of the Matrix Pencil form (Y2 − λY1 ), which can be computed
1
Êi = ω̂i2 â2i . (10) from:
2
2) Eigensystem Realization Algorithm: This algorithm is zi = eig(Y1† Y2 ) (20)
based on the singular value decomposition of the Hankel
matrix H0 associated with the linear ringdown of the system where † corresponds to the Moore-Penrose pseudoinverse.
[5]. The Hankel matrices are assembled with samples corre- Then, the modal information is extracted as in (8)-(10).
sponding to a recorded signal y(t), such that: D. Identifying Low Frequency Oscillations
 
y0 y1 ··· yr
y1 y2 · · · yr+1  The algorithms PM, ERA, and MP for the identification of
H0 =  .

. .. .. 

(11) electromechanical modes have been implemented and tested
 .. .. . .  according to the presented methodology in the following [5],
yr yr+1 · · · yN −1 [11], [14], [17]. For all the identification methods, an analysis

y1 y2 · · · yr+1
 window of 3.46 s is used to observe at least two cycles of
 y2 y3 · · · yr+2  the low frequency oscillation. The Prony’s order (n) is defined
H1 =  .

. .. .. 

(12) equal to 10 in order to have a suitable performance. Whereas,
 .. .. . .  the singular value decomposition (SVD) thresholds for ERA
yr+1 yr+2 · · · yN and MP are adjusted to 1 × 10−3 , prior to calculate the SVD
and r is N/2 − 1. of the Hankel matrices.
The ERA formulation begins by separating the singular Since the positive sequence synchrophasor information is
value decomposition of H0 into two components according available and accessible from the database, a second graphical
to the relative size of the singular values: user interface is developed and embedded in the Matlab

 Σn 0
 T
Vn environment, which serves as a power system modal monitor.
T

H0 = U ΣV = Un Uz (13) This one depicts three main aspects concerning to the iden-
0 Σz VzT
tification of electromechanical modes in Fig. 7: (i) Voltage
where Σn and Σz are diagonal matrices with their elements magnitude for 7 PMUs; (ii) Fourier spectra for 7 PMUs; and
ordered by magnitude (iii) Four labels that monitor modal parameters as damping and
Σn = diag(σ1 , σ2 , . . . , σn ) (14) frequency in (8), damping ratio in (9), and energy in (10), only
The ratio of the singular values of contained in Σ can using a single channel, corresponding to the voltage magnitude
determine the best approximation of n with p significant at Pilgrim substation (BUS 22)..
decimal digits, as follows. It is observed from results that the MP and ERA methods
σi are very similar, as mentioned, both methods employ the SVD
≈ 10−p . (15) to capture the same effective range of the Hankel matrices,
σmax
unlike the PM that adjusts its results based on the order of
Then, H0 is truncated by
the proposed polynomial; even so, the error using a PM with
H0 ≈ Un Σn VnT . (16) a 10-th order is 6 %.
The matrix of the identified system is derived as
IV. C ONCLUSIONS
A = Σ−1/2
n UnT H1 Vn Σ−1/2
n (17)
The presented proposal in this paper has demonstrated
where zi = eig(A). Then, the modal information is extracted that a WAMS simulator can become to embed in Matlab
as in (8)-(10). and PowerFactory by taking advantage of their potential for
running EMT simulations and for creating new synchrophasor
 and Matrix Pencil. From the results, it is noted that ERA and
Matrix Pencil attain similar behaviors for frequency, damping
and energy. Whereas Prony reports lower values in damping
and energy than the other ones.
ACKNOWLEDGMENT
Authors acknowledge financial support from the Project
Support Program for Research and Technological Innova-
tion of UNAM (DGAPA, PAPIIT-2019) through the project
TA100819.
R EFERENCES
[1] Zhian Zhong, Chunchun Xu, B. J. Billian, Li Zhang, S. . S. Tsai,
R. W. Conners, V. A. Centeno, A. G. Phadke, and Yilu Liu, “Power
system frequency monitoring network (fnet) implementation,” IEEE
Transactions on Power Systems, vol. 20, no. 4, pp. 1914–1921, Nov
2005.
[2] I. C. Decker, D. Dotta, M. N. Agostini, S. L. Zimath, and A. S. de
Silva, “Performance of a synchronized phasor measurements system in
the brazilian power system,” in 2006 IEEE Power Engineering Society
General Meeting, June 2006, pp. 8 pp.–.
[3] A. Chakrabortty and Y. Xin, “Hardware-in-the-loop simulations and
verifications of smart power systems over an exo-geni testbed,” in 2013
Second GENI Research and Educational Experiment Workshop, March
2013, pp. 16–19.
[4] A. Zamora, A. Avalos, J. A. la O de Serna, J. H. Chow, and M. R. A.
Paternina, “Real-time simulation of the prony filter for identifying low
frequency oscillations in short-time,” in 2018 IEEE PES Transmission
Distribution Conference and Exhibition - Latin America (T D-LA), Sep.
2018, pp. 1–5.
[5] J. Sanchez-Gasca and D. Trudnowski, “Identification of electromechan-
ical modes in power system,” IEEE Task Force on Identification of
Electromechanical Modes of the Power System Stability, Power &
Energy Society, Tech. Rep., June 2012.
[6] J. de la O Serna, J. Ramirez, A. Zamora Mendez, and M. Paternina,
“Identification of electromechanical modes based on the digital taylor-
fourier transform,” IEEE Trans. Power Syst., vol. 31, no. 1, pp. 206–215,
Jan 2016.
[7] E. Barocio, B. C. Pal, N. F. Thornhill, and A. R. Messina, “A dynamic
mode decomposition framework for global power system oscillation
analysis,” IEEE Transactions on Power Systems, vol. 30, no. 6, pp.
2902–2912, Nov 2015.
[8] A. R. Messina, Wide Area Monitoring of Interconnected Power Systems,
2015.
[9] “Ieee standard for synchrophasor measurements for power systems,”
IEEE Std C37.118.1-2011 (Revision of IEEE Std C37.118-2005), pp.
1–61, Dec 2011.
[10] P. F. U. M. DIgSILENT, “Powerfactory (2008) version 14.0,” DIgSI-
LENT GmbH, Gomaringen, Germany, 1989.
[11] J. F. Hauer, C. Demeure, and L. Scharf, “Initial results in prony analysis
of power system response signals,” IEEE Transactions on power systems,
vol. 5, no. 1, pp. 80–89, 1990.
Fig. 7. Graphical User Interface developed for monitoring the power system [12] J. Johnson and D. Trudnowski, “Bpa/pnnl dynamic system identification
modal information by using Prony, ERA, and Matrix Pencil. dsitools; ringdown analysis tool,” Columbus, OH, Batelle Memorial
Institute, User?s Manual, 1998.
[13] J. Hauer, L. Scharf, D. Trudnowski, J. Johnson, and N. Zhou, “Dsi
toolbox: Psm tool and ringdown analysis tool,” Available for free at
and WAMS applications as in [13]. The main target of this http://www. naspi. org/resources/pitt/pittresources. stm, 2010.
platform is to facilitate the test of novel algorithms that [14] J.-N. Juang and R. S. Pappa, “An eigensystem realization algorithm
requires instantaneous signals and synchrophasor data, or for for modal parameter identification and model reduction,” Journal of
guidance, control, and dynamics, vol. 8, no. 5, pp. 620–627, 1985.
comparison purposes with the traditional techniques. More- [15] Y. Hua and T. K. Sarkar, “Matrix pencil method for estimating pa-
over, it can be a base for testing new WAMS architectures. rameters of exponentially damped/undamped sinusoids in noise,” IEEE
Transactions on Acoustics, Speech, and Signal Processing, vol. 38, no. 5,
The WAMS simulator is useful for academics and pro- pp. 814–824, 1990.
fessionals, whose interests include to understand the basic [16] L. L. Grant and M. L. Crow, “Comparison of matrix pencil and prony
concepts behind the phasor estimation process carried out by methods for power system modal analysis of noisy signals,” in 2011
North American Power Symposium, Aug 2011, pp. 1–7.
PMUs and any synchronized measurements-driven applica- [17] M. L. Crow and A. Singh, “The matrix pencil for power system modal
tion; specially the Power System Modal Monitor. Finally, the extraction,” IEEE Transactions on Power Systems, vol. 20, no. 1, pp.
WAMS simulator performance is assessed using Prony, ERA, 501–502, 2005.