IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 56, NO.

3, MAY/JUNE 2020 2281

Optimized Frequency Scanning of Nonlinear Devices

Applied to Subsynchronous Resonance Screening
Igor Brandão Machado Matsuo , Member, IEEE, Farshid Salehi , Member, IEEE,
Long Zhao , Student Member, IEEE, Yuhao Zhou , Student Member, IEEE, and Wei-Jen Lee , Fellow, IEEE

Abstract—Frequency scanning is a powerful and versatile ap- devices, and shunt active compensation devices have consider-
proach for subsynchronous resonance (SSR) screening in power ably changed the dynamic behavior of power grids [1]–[10]. The
grids. Among different numerical and analytical frequency scan- complex control systems of these types of equipment have intro-
ning techniques, the harmonic injection method is well suited for
the scan of black-box models with active elements, such as in duced new challenges to the power industry [2]–[4]. One of these
wind farms and solar plants. The results can be used for SSR risk challenges is subsynchronous resonance (SSR), also referred to
assessment, in which accuracy performs a decisive role. The ideal as subsynchronous interaction (SSI) by some authors. SSR/SSI
situation for this method, which leads to the most accurate results, is involves the interaction between two or more elements of power
to perform one simulation per single-frequency injection. However, grids within the subsynchronous frequency range (<60 Hz for
for SSR studies with a wide range of frequencies and simulations
with very small time steps (scale of µs), this comes at the expense a 60 Hz system), leading to subsynchronous oscillations (SSO)
of increased simulation time. This article proposes a technique to [10]. Depending on the equipment involved in the interaction,
improve the accuracy of the harmonic injection method through the phenomenon can be categorized in different groups, which
the optimization of the crest factor while injecting all frequencies are as follows [1]–[4].
at one shot, therefore also reducing simulation time. The proposed 1) Torsional Interaction: The oscillation is a result of an inter-
technique was tested on a wind farm connected to a radial test case
and on a portion of a Texas synthetic grid with multiple active ele- action between the mechanical system and the electrical
ments, including two wind farms and a VSC-based STATCOM. The system [1], [4]. This phenomenon is common for conven-
frequency scan results were benchmarked with both time-domain tional generation with long turbine-generator shafts and
transient simulations and the ideal multiple single-frequency in- the oscillation grows relatively slowly [9]. This interaction
jection case and compared with other techniques. The results show might happen between the turbine-generator shaft and a
that the proposed method is superior in accuracy when compared
to the other techniques and is 11.71 times faster when compared to nearby series-compensated line, such as in the Mohave
multiple single-frequency injections. Power Plant incident [15]. It might also be triggered as a
result of interaction between the turbine-generator shaft
Index Terms—Crest factor, frequency scan, harmonic injection,
impedance estimation, multisine signals, nonlinear systems,
and control systems of power-electronic-based devices,
small-signal analysis, subsynchronous control interaction (SSCI), such as between HVDC lines and mass-drive trains of
subsynchronous resonance (SSR), wind farms. generation systems [1]. The second one is mostly known as
subsynchronous torsional interaction and it was observed
for the first time during field tests of a line-commutated-
I. INTRODUCTION
converter HVDC (LCC-HVDC) line that interacted with a
IGH penetration of renewable generation resources and
H other power electronic-based elements such as high-
voltage direct-current (HVDC) lines, series active compensation
turbine-generator of Square-Butte, in the Milton R. Young
power plant [16].
2) Subsynchronous Control Interaction (SSCI): Interaction
between series-compensated lines and control systems
Manuscript received February 19, 2019; revised September 18, 2019 and
November 29, 2019; accepted January 2, 2020. Date of publication February with power electronic-based devices [4], [9].
4, 2020; date of current version April 24, 2020. Paper 2019-PSEC-0221.R2, 3) Induction Generation Effect (IGE): Situation when an
presented at the 2019 IEEE/IAS 55th Industrial and Commercial Power Sys- induction generator shows negative resistance at a sub-
tems Technical Conference, Calgary, AB, Canada, May 5–8, and approved
for publication in the IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS by synchronous resonant frequency and the magnitude of this
the Power Systems Engineering Committee of the IEEE Industry Applications negative resistance exceeds the total resistance of the grid
Society. The work of I. B. M. Matsuo was supported by CNPq, National Council at the same frequency [2], [4].
for Scientific and Technological Development—Brazil. (Corresponding author:
Igor Brandão Machado Matsuo.) SSCI and IGE are purely electrical incidents with the potential
I. B. M. Matsuo, L. Zhao, Y. Zhou, and W.-J. Lee are with the Department to grow very rapidly [9]. Therefore, they can cause serious
of Electrical Engineering, University of Texas at Arlington, Arlington, TX damage to transmission elements and lead to cascading outage.
76019 USA (e-mail: [email protected]; [email protected]; yuhao.
[email protected]; [email protected]). The first recorded SSCI event took place in Texas, in 2009, where
F. Salehi is with the Department of Power System Advisory, DNV GL Energy, 20 Hz oscillations damaged the crowbar protection of many
Dallas, TX 75207 USA (e-mail: [email protected]). DFIGs in a wind farm [17]. Many other SSR events happened in
Color versions of one or more of the figures in this article are available online
at https://ieeexplore.ieee.org. China between 2012 and 2013, where the frequencies recorded
Digital Object Identifier 10.1109/TIA.2020.2971434 varied in the range of 6–8 Hz [5]. Other SSR events include

0093-9994 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Auckland University of Technology. Downloaded on May 25,2020 at 04:46:37 UTC from IEEE Xplore. Restrictions apply.
2282 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 56, NO. 3, MAY/JUNE 2020

13 Hz oscillations at Oklahoma Gas & Electric, and 9–13 Hz

oscillations in the Buffalo Ridge area of Minnesota [6], [7]. Also,
several SSO were captured by wide-area measurement systems
in north China in 2014 and 2015, where a growing installed
capacity of wind generation is present, as well as LCC-HVDC
Fig. 1. Two sides of a grid connected to a POI.
lines, but no series-compensated lines [18].
Due to the possible severe consequences of SSR events, all
independent system operators require performing SSR screening unchanged. This leads to higher efficiency by a considerable
studies to evaluate the risks for all generation interconnections decrease in simulation time while maintaining good accuracy.
and transmission expansion projects. The results of this screen- The technique was tested on a wind farm connected to a radial
ing are utilized to assess the risks of SSR and the needs for further test case and on a portion of a modified Texas synthetic grid
detailed studies, such as time-domain transient simulations, and with multiple active elements, including two wind farms and a
measures in the grid implementation. If SSR risks are confirmed, VSC-based STATCOM. The frequency scan results were bench-
fast and effective detection and/or mitigation countermeasures marked with both time-domain transient simulations, which
are necessary to prevent the SSO harmful effects [8]–[14]. provide the most accurate results for final SSR analysis, and the
Popular methods for SSR screening consist of frequency ideal multiple single-frequency-injection case, which provides
scanning and eigenvalue analysis [17]–[23]. Eigenvalue analysis the most accurate results for frequency-scan-based SSR screen-
requires a detailed model of all elements of the grid, which is nor- ing, and then compared with other techniques. The results show
mally not accessible in the industry in order to protect intellectual that the proposed method is more accurate than the other studied
properties. However, all industrial manufacturers provide their techniques and is 11.71 times faster when compared to multiple
models as electromagnetic transient (EMT) black-box modules. single-frequency injections.
In this case, the best approach is frequency scanning based on
harmonic injection, which consists of an input–output approach. II. FREQUENCY SCAN
The frequency scan-based screening has been utilized to evaluate
mitigation and detection solutions in [9] and [11]. Although different frequency scan techniques have been pro-
Harmonic injection basically consists of injecting a small sig- posed, the injection method is the only choice for the scan of
nal at frequencies of interest and measuring the voltage/current active black-box devices. In order to protect the intellectual prop-
response, from which the impedance can be calculated. One erties, industrial manufacturers provide their models in the form
may choose to perform one simulation per frequency of interest of EMT black-box modules. Normally, these models include the
or to use a signal that injects each frequency individually and compiled code from the controller of the corresponding plant to
consecutively over time [24]. However, this approach can be represent the most realistic response in the simulation. The har-
time-consuming. Instead, a multisine (multifrequency) signal monic injection approach provides the characteristic impedance
containing all frequencies of interest at the same time may be of the nonlinear elements, such as renewable resources, and of
injected at once to reduce simulation time. This technique is not the transmission grid as a function of frequency over the entire
restricted by frequency range and can be utilized to estimate range of interest. For the SSR screening, independent frequency
the impedance of any active element at any frequency range of scans are performed on the generation and transmission systems
interest. Multifrequency injection is used to frequency-scan a from the point of interconnection (POI), as depicted in Fig. 1.
STATCOM—although it is carried out in the supersynchronous The impedances can be derived in the following format:
range instead of in the subsynchronous range—in [25]. A mul- Zgrid1 (f ) = Rgrid1 (f ) + j ∗ Xgrid1 (f ) (1)
tifrequency injection may lead to harmonic injection distortion
when nonlinear components are present [26]. To alleviate the Zgrid2 (f ) = Rgrid2 (f ) + j ∗ Xgrid2 (f ). (2)
effects of multisine injection, one may inject a smaller number The frequency scan results of both sides of the grid are used to
of frequencies per simulation. In this method, the most accurate calculate the cumulative resistance and reactance over the range
results can be obtained in an ideal situation that involves separate of subsynchronous frequencies. The cumulative resistance at a
frequency scans for every single frequency of interest. However, crossover frequency is an index of SSR risk [9], [10]. To find
two factors add up to the time consumed for SSR screening the crossover frequency, we define
studies. The first one is the wide range of frequencies of interest.
The second one is the small time steps (order of a few μs) RPOI (f ) = Rgrid1 (f ) + Rgrid2 (f ) (3)
required for the simulation of EMT models to correctly simulate XPOI (f ) = Xgrid1 (f ) + Xgrid2 (f ) (4)
the high switching frequencies of power electronic devices (in
the order of a few kHz). where f is considered the crossover frequency (fSSR ) if
To address this issue, this article proposes a technique to d
improve the accuracy of the harmonic injection method through XPOI (f ) = 0 & (XPOI (f )) > 0. (5)
df
the optimization of the crest factor while injecting all frequencies
If the resistance at this frequency, fSSR , is negative, as shown
at one shot. This technique reduces the multisine high-peak
in (6), then undamped oscillations are prone to happen.
effect by scattering the signal energy during the harmonic in-
jection while maintaining the signal root-mean-square (RMS) RPOI (fSSR ) ≤ 0. (6)

Authorized licensed use limited to: Auckland University of Technology. Downloaded on May 25,2020 at 04:46:37 UTC from IEEE Xplore. Restrictions apply.
MATSUO et al.: OPTIMIZED FREQUENCY SCANNING OF NONLINEAR DEVICES APPLIED TO SSR SCREENING 2283

Fig. 2. Scheme of how to apply the frequency scanning tool.

The schematic of the frequency scanning tool through har-

monic injection is depicted in Fig. 2. This method consists of Fig. 3. Typical IGBT current x voltage output characteristic curve.
the following steps.
1) Bring the scanned grid to a steady-state condition (bias
f0 : base frequency in Hz;
point) through an equivalent source.
klo : lowest harmonic of interest;
2) Inject a positive-sequence voltage at a frequency of in-
khi : highest harmonic of interest;
terest, Vinj (f ), between the equivalent source and the
δk : angle at the kth harmonic;
terminals of the grid.
δ: set of angles δk .
3) Measure the voltage, Vmeas , and current, Imeas , at the
terminals of the grid. The frequency scanning tool is used in the same way as
4) Apply the fast-Fourier transform (FFT) to both Vmeas and described in the previous section. Zmeas (f ) is obtained for every
Imeas to get Vmeas (f ) and Imeas (f ) at the same frequency frequency of interest after applying the FFT to the voltage and
of the injected voltage Vinj (f ). current measured signals.
5) Calculate the impedance, Zmeas (f ), as in the following The grid may contain devices that present nonlinear behavior,
equation: as in power electronic-based controllers, such as in wind farms,
Vmeas (f ) LCC-HVDC lines, VSC-HVDC lines, and STATCOM devices,
Zmeas (f ) = . (7) among others. This means that the current–voltage relation-
Imeas (f )
ship is not linear, as seen in Fig. 3 [28] for the example of
In this scheme, Vmeas is a phase-to-ground voltage, while an insulated-gate bipolar transistor (IGBT) characteristic IxV
Imeas is the current measured at the same phase. output curve, which is one of many possible sources of nonlin-
earities. Other possible sources can be found in [29]. IGBTs are
III. MULTIFREQUENCY SIGNAL INJECTION one of the nonlinear devices normally present in wind turbine
The best outcome of frequency scanning using the harmonic controllers. From the figure, we can see that, above a certain
injection technique is through separate frequency scans for each voltage amplitude, the current response is not in the linear region
frequency of interest. However, whenever the goal is to estimate anymore. Therefore, the impedance cannot be simply calculated
the impedance characteristic to active elements over a wide range by the ratio between voltage and current, and (7) is not valid.
of frequencies such as the SSR range, this process can take a great Furthermore, the superposition of frequencies is not valid for
amount of time, especially when dealing with large grids with nonlinear systems, which means that the sum of all individual
power electronic-based devices, in which switching frequencies responses for each frequency is not equal to the overall response
may be high, in the order of a few kHz. In these situations, of the system when injecting all frequencies at the same time, so
simulations require small time steps up to the order of a few F(x) + F(y) = F(x + y) [30]. Thus, (8) is not applicable, as
microseconds, hence increasing simulation time. Doing the same we cannot subsequently apply an FFT to get each frequency’s
process for every frequency of interest can become an inefficient response when injecting many frequencies at the same time.
task. The solution proposed herein involves injecting a signal For (7) and (8) to be valid, a small-signal analysis should be
containing all frequencies of interest at the same time, making it performed, with the voltage high enough to avoid being covered
possible to scan each side of the grid with only one simulation. in noise (high signal-to-noise ratio—SNR), but small enough to
maintain linearity and follow the superposition principle. This
A. Multisine Signal Optimization is possible by performing injections of small magnitude around
a point of steady-state operation, called bias point [30].
The injected voltage signal will be a multisine with period Assuming we have chosen an amplitude that falls within an
T0 , which is a sum of sine waves harmonically related through acceptable range, if we apply the following parameters in (8),
a base frequency f0 = 1/T0 [27], as in the resulted multisine signal will be as in Fig. 4:
khi
 A = 1 V; f0 = 1 Hz; klo = 5; khi = 35; δk = 0.
Vinj (δ) = A ∗ sin (2 ∗ π ∗ fk ∗ t + δk ) (8)
k=klo From Fig. 4, we can see that, for an amplitude of 1 V, and
a range of 5–35 Hz with frequencies 1 Hz apart, if we use
where
δk = 0, large spikes happen every second, reaching a peak of
A: amplitude of each sinusoid; 25.04 V. The period of these spikes is 1/f0 . If instead of setting
fk = k ∗ f0 , where k is an integer number; δk = 0 for every frequency we apply variations in δ for each

Authorized licensed use limited to: Auckland University of Technology. Downloaded on May 25,2020 at 04:46:37 UTC from IEEE Xplore. Restrictions apply.
2284 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 56, NO. 3, MAY/JUNE 2020

However, we need to take into account that the Lagrangian

method is an efficient method under a good initial estimate of
the solution. This estimate can be derived from the Schroeder
equation and is presented in the following equation [25]:
 
fk − flo + 1
δk,0 = rem (fk − flo ) ∗ ∗ 360, 360 (14)
Fig. 4. Multisine signal with frequency range of 5 Hz to 35 Hz and δk = 0. fhi − flo + 1
where
frequency, the energy in the signal will be scattered instead of rem(a, b): remainder operator on a divided by b;
concentrated in a few particular instants of time. This means flo : lowest frequency of interest;
that large spikes can be avoided while the RMS value of the fhi : highest frequency of interest;
signal remains the same, therefore maintaining the small-signal
requirement. A good parameter that depicts this situation is the The frequency scan can then be performed using a low-crest-
crest value, as defined in the following equation: factor multisine containing many frequencies at the same time
while maintaining the small-signal requirements. Hence, we are
|Peak Vinj (δ)|
C (δ) = . (9) able to perform an FFT calculation to get each individual fre-
RMS Vinj (δ) quency response based on the overall response output according
Smaller crest values lead to smaller peak values for the same to the superposition principle.
RMS, helping realize the necessary small signal. With that in
mind, an optimization problem can be defined to minimize B. Minimum Simulation Time
the crest value using the angles δk as variables. The general The minimum total simulation time for multiple single-
formulation is as follows: frequency injections (injecting one frequency per simulation
minimize C (δ) or injecting individual frequencies consecutively) is as follows
[32]:
s.t.
khi 
 
gk (δ) = −δk ≤ 0◦ 1
Tsf = + T stk (15)
fk
hk (δ) = δk − 360◦ ≤ 0. (10) k=klo

where T stk is the settling time at frequency fk .

Using the nonlinear Lagrangian optimization method [31],
Now, assuming that we achieved a low crest value for the
one may achieve first-order optimality through the Karush–
multisine signal (injecting all frequencies at the same time in
Kuhn–Tucker (KKT) conditions. The KKT conditions are analo-
one simulation), the SNR ratio will be high [27]. The minimum
gous to making the gradient equal to zero, that is, to bringing the
total simulation time in this case is as follows:
objective function to a minimum value. The Lagrangian function
is as follows: Tms = T0 + max {T stk } . (16)
khi
 khi
 A good approximation to establish the settling time can be
L (δ, λ) = C (δ) + λg,k gk (δk ) + λh,k hk (δk ) .
achieved by determining the time required for the envelope of
k=klo k=klo
(11) the signal to reach and remain within a certain percentage range
The KKT conditions are shown in (10) above and in the of its final value (usually 2% or 5%) [33].
following equation:
IV. CASE STUDY AND SIMULATION RESULTS
||∇L(δ, λ)|| = 0
The proposed method was implemented in MATLAB, but it
λg,k gk (δk ) = 0 ∀k can also be implemented in different platforms with EMT sim-
λh,k hk (δk ) = 0 ∀k. (12) ulation capabilities, such as PSCAD/EMTDC or ATP/EMTP.
Furthermore, all the case studies include detailed models of wind
The optimality measures associated with (12) are as follows: farms, STATCOM and distributed parameters for transmission

 khi
 lines developed in MATLAB/Simulink.

||∇L(δ, λ)|| = ∇C (δ) + λg,k ∇gk (δk )
 A. Case Study I: Radial Test System
k=klo

khi  In order to evaluate the performance of the proposed opti-

+ λh,k ∇hk (δk ) , mized technique, a case study including a black-box model of

k=klo a 100 MVA wind farm connected to the grid through a series-
−−−−−−−→ compensated line is used. The system is depicted in Fig. 5 and the
 
λg,k gk (δk ) , impedance data of the radial connection are in Table I. As it can
−−−−−−−→ be seen, the wind farm and respective collection system are con-
 
λh,k hk (δk ) . (13) nected to the POI through a station transformer and a gen-tie line.

Authorized licensed use limited to: Auckland University of Technology. Downloaded on May 25,2020 at 04:46:37 UTC from IEEE Xplore. Restrictions apply.
MATSUO et al.: OPTIMIZED FREQUENCY SCANNING OF NONLINEAR DEVICES APPLIED TO SSR SCREENING 2285

TABLE II
METHODS OF DIFFERENT ANGLES AND RESPECTIVE PEAK, RMS,
AND CREST VALUES AFTER NORMALIZATION

Fig. 5. Wind farm connected to the radial test system.

TABLE I
RADIAL TEST SYSTEM PARAMETERS

Moreover, the transmission line has 70% of compensation and

connects the POI to the equivalent grid.
The wind farm uses doubly fed induction generation tech-
nology with a vector-oriented control strategy. The black-box
model includes the detailed representation of the turbine aero-
dynamics, mass-drive train, induction machine, and rotor-side
and grid-side converters with their controllers. The wind turbine Fig. 6. Waveforms of the normalized multisine signals with different angles.
control parameters and induction generator data are based on a
commercial wind turbine [34].
The following sections comprise three steps. First, different disperse the energy of the multisine while maintaining the RMS
methods, including the proposed optimized method, are used to value unchanged. Method 2 utilizes a quadratic relation of the
generate the multifrequency harmonic signals. Second, separate frequency and has been used previously in [26].
frequency scans are performed on the wind farm as well as The methods in Table II are listed in the order of larger to
on the grid side using the different harmonic signals generated smaller crest factors, the latter being the proposed√optimized
in the first step. The frequency scan results are compared and method. A single sine wave has a crest factor of 2, which
benchmarked against ideal single-frequency injections. Also, the is approximately 1.4142. We can then easily reason that, when
cumulative resistance and reactance at the POI are calculated adding more sine waves at other frequencies, the crest factor
to evaluate the risk of SSCI incidents. Finally, a time-domain will only get larger. The proposed optimized method resulted
transient simulation is performed to validate the accuracy of the in a crest factor of 1.49 when generating a 31-frequency signal,
proposed technique. which is an indication that it is close to an optimal value.
1) Generating the Multisine Signal for the Frequency Scan: 2) Frequency Scan Results: At this step, the signals of the
To generate different harmonic signals, different formulas were four methods of the multisine waves were separately injected
applied to calculate the angles of the multisine expressed in (8). into the wind farm black-box model as well as into the grid
The following parameters were used: to perform the frequency scan. The results were benchmarked
against reference results taken from multiple simulations of
A = 345 V; f0 = 1 Hz; klo = 5; khi = 35.
single-frequency injections.
The frequency range was chosen based on the previous SSO In this case study, the grid side consists only of linear compo-
frequencies observed in past events in the USA and China, as nents, such as resistors, capacitors, and inductors; and the wind
shown in the introductory section of this article [5]–[7], [17]. farm consists of active devices with nonlinear behavior, such as
The small-signal voltage amplitude, 345 V, is chosen at 0.1% controlled IGBTs and phase-locked loops.
of the rated voltage, 345 kV. The frequencies injected are sepa- To evaluate the risk of SSCI, the cumulative resistance and
rated by 1 Hz. Different methods of angle calculation and their reactance at the POI were depicted in Fig. 7, which also contains
corresponding crest factors are depicted in Table II. The plots a zoomed-in version around the crossover frequencies. Now,
in Fig. 6 show the normalized waveforms of the multifrequency using (5) we can find the crossover frequency (zero reactance)
signals using different angles in δ. The normalization is done in order to assess the risk of undamped oscillations with (6). The
by dividing the waveforms by the amplitude of each injected frequency of oscillation found for each method and respective
sine wave, in this case, 345 V. In this way, we can easily see the resistance values at these resonant frequencies are shown in
effects of the multisine injection on the peak value for every 1 V Table III along with the single-frequency-injection reference
injected. results.
As it is evident in Fig. 6, simply using zero for all angles If the differences between the resistance/reactance of each
leads to large spikes in the harmonic signal. These can produce method and the reference results are defined as errors, the
errors in the frequency scan. However, more appropriate angles methods with smaller crest factors should lead to smaller errors.

Authorized licensed use limited to: Auckland University of Technology. Downloaded on May 25,2020 at 04:46:37 UTC from IEEE Xplore. Restrictions apply.
2286 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 56, NO. 3, MAY/JUNE 2020

TABLE V
TOTAL REACTANCE ERROR OF METHODS IN RELATION TO THE REFERENCE

Fig. 7. Total resistance and reactance seen at the POI.

TABLE III
FREQUENCY OF OSCILLATION AND RESISTANCE VALUES

Fig. 9. Time-domain simulation results at the POI. (a) Active power.

(b) Voltage. (c) Current.

1) Series capacitor by-passed by closing the bypass switch.

2) The simulation was run until t = 35 s to ensure that steady-
state is achieved.
3) The bypass switch is opened at t = 35 s to create a
radial condition between the wind farm and the series-
compensated line. This creates an SSCI oscillation.
The three-phase active power and instantaneous voltages and
currents at the POI are depicted in Fig. 9.
In line with the frequency scan results, the time-domain
transient simulation shows the SSCI risk. The results from the
FFT analysis performed on the current signal of the time-domain
simulation shows the frequency of oscillation at 14.95 Hz. The
growing behavior of the oscillation indicates an undamped oscil-
lation, which corresponds with the negative sign of the resistance
Fig. 8. Error of each method in relation to the reference. found in the frequency scan.
It can be seen that the captured resonant frequency from the
TABLE IV optimized method is an almost perfect match with the results of
TOTAL RESISTANCE ERROR OF METHODS IN RELATION TO THE REFERENCE the ideal frequency scan and the time-domain simulation.

B. Case Study II: Modified Texas Synthetic Grid

To further corroborate the capability and robustness of the
proposed method, the optimized technique was utilized on a
portion of a modified Texas synthetic grid, as depicted in Fig. 10.
The original model was developed by Texas A&M University
The cumulative resistance and reactance errors are plotted in and is not an actual depiction of the Texas grid, but is based
Fig. 8. The average total error calculated for each method in the on public data to synthetically represent the network [35]. It
frequency range is shown in Tables IV and V. allows researchers to test their algorithms on large-scale cases.
The results indicate that the smaller the crest factor, the In order to create the SSR condition, a series capacitor with a
smaller the error. Among all methods, the proposed opti- compensation level of 70% was added on the line between the
mized injection signal results in a more accurate frequency Miami and Denton stations. Also, three nonlinear elements were
scan. added to the case as follows.
3) Time-Domain Transient Simulation: To validate the re- 1) 210 MW Wind Farm 1: This wind farm involves 140 ×
sults of the frequency scan, a time-domain transient simulation 1.5 MW type-3 turbines which are connected to the Miami
was performed as per the following sequence. 1 161 kV station (POI).

Authorized licensed use limited to: Auckland University of Technology. Downloaded on May 25,2020 at 04:46:37 UTC from IEEE Xplore. Restrictions apply.
MATSUO et al.: OPTIMIZED FREQUENCY SCANNING OF NONLINEAR DEVICES APPLIED TO SSR SCREENING 2287

Fig. 11. Total resistance and reactance seen at the POI.

Fig. 10. Portion of the modified Texas synthetic grid.

2) 100 MVA voltage source converter (VSC)-based STAT-

COM connected to the Miami 1 161 kV station. Fig. 12. Error of each method in relation to the reference.
3) 90 MW Wind Farm 2: This wind farm involves 25 ×
3.6 MW type-3 turbines which are connected to the Pan-
handle 3 (PH3) 161 kV station. 4) Perform a frequency scan on the network side that in-
The remaining grid data can be found in the Appendix (see volves Wind Farm 2 and the rest of the grid under the
Tables X–XII). predefined N–3 contingency. Hereafter, this impedance is
In this case study, the objective is to perform a frequency called network-side impedance (grid side 2).
scan-based screening for Wind Farm 1 to evaluate the SSR risk 5) Calculate the cumulative impedances based on the plant-
of this specific wind farm while the system involves two other side and network-side impedances as per (3) and (4).
active elements (Wind Farm 2 and STATCOM). As a first step, 6) If the cumulative resistance at the crossover frequency
a contingency was defined to create a radial connection between (zero reactance frequency) is negative or zero [(5) and
Wind Farm 1 and the series-compensated line. The radial con- (6)], there is a potential risk of SSCI.
tingency is an N–3 contingency and involves tripping/opening Note that the steps above perform independent frequency
of three transmission lines as follows: scans on Wind Farm 1 and on the STATCOM despite their point
1) 161 kV line between the PH2 1 and PH6 stations; of common connection (at the POI). This requirement reduces
2) 500 kV line between the PH4 and Ralls 1 stations; the error, as the influence of the voltage injections in one element
3) 500 kV line between the Miami and Wichita Falls stations. will not affect the response of the other.
The contingency is depicted with black dashed lines in Fig. 10. 1) Frequency Scan Results: The multisine harmonic signal
If these lines are tripped/opened, the only path to export power angles were generated as in Case Study I, with the difference
from the Panhandle area (within the blue oval shape), which also that the amplitude for each sine wave is now 161 V (0.1% of
includes the wind farms, is through the series-compensated line the POI rated voltage of 161 kV). The cumulative resistance and
between the Miami and Denton stations. reactance for all methods can be found in Fig. 11 and the errors
The screening process for Wind Farm 1 at the 161 kV Miami in Fig. 12.
1 station (POI) is performed as per the following steps. In this case study, several nonlinear components are present
1) Perform an independent frequency scan on Wind Farm 1 on both sides of the frequency scan. The plant side contains not
from the POI. only a wind farm but also a VSC-based STATCOM. As can be
2) Perform an independent frequency scan on the VSC-based observed in Fig. 11, the presence of a second wind farm on the
STATCOM from the POI. network side adds a high-impedance transition from an inductive
3) Calculate the equivalent subsynchronous impedance (re- response to a capacitive response seen at the POI. Fig. 12 shows
sistance and reactance) of Wind Farm 1 and the VSC-based that, during this transition between 15 and 20 Hz, the error
STATCOM based on their parallel connection at the POI. in the frequency scan may increase. However, the optimized
Hereafter, this impedance is called plant-side impedance method (method 4) is still able to maintain very low error values.
(grid side 1). Tables VI and VII show the average total errors seen at the POI

Authorized licensed use limited to: Auckland University of Technology. Downloaded on May 25,2020 at 04:46:37 UTC from IEEE Xplore. Restrictions apply.
2288 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 56, NO. 3, MAY/JUNE 2020

TABLE VI
TOTAL RESISTANCE ERROR OF METHODS IN RELATION TO THE REFERENCE

TABLE VII
TOTAL REACTANCE ERROR OF METHODS IN RELATION TO THE REFERENCE

Fig. 14. Time-domain simulation results at the POI. (a) Active power.
(b) Voltage. (c) Current.

It should be noted that the grid might contain different nonlin-

ear elements and, as a result, different nonlinear characteristics.
Besides, the voltage levels on each device might differ depending
on their location in the grid. One specific signal might maintain
the approximate linearity around the initial steady-state condi-
tion of one device while pushing another device’s response to the
nonlinear boundaries. This means that the small-signal analysis
is maintained for the former but will still result in some error
for the latter. Therefore, there is no ideal choice of amplitude
Fig. 13. Zoomed-in plot of Fig. 11 around the crossover frequency.
and angles for the multisine signal that will not yield errors to
TABLE VIII some extent. The choice of 0.1% amplitude level of the rated
FREQUENCY OF OSCILLATION AND RESISTANCE VALUES voltage used herein has shown to be a good choice for most
cases. The main goal is to reduce the average total error for
both the resistance and reactance values so as to keep the best
possible impedance estimation throughout the entire range of
frequencies.
2) Time-Domain Transient Simulation: A time-domain tran-
sient simulation with the application of a three-phase fault that
will lead to the predefined N–3 contingency was performed per
for the resistance and reactance, respectively, and confirm the the following sequence.
better performance of the optimized method in a test case with 1) The simulation starts with lines PH4-RALLS1 and
a larger grid with more nonlinear components in both sides of MIAMI-WICHITA FALLS out of service, for instance,
the scan. for maintenance purposes.
Fig. 13 shows the zoomed-in plot of the total impedance 2) The simulation is run until t = 19.95 s to ensure that steady-
around the crossover frequency (zoomed-in plot of Fig. 11 state is achieved.
around the crossover frequency). Table VIII shows the frequency 3) At t = 19.95 s, a three-phase fault is applied to line PH2
of oscillation and resistance results for all methods. 1-PH6.
Very important observations can be made from the results. 4) At t = 20 s, a circuit breaker opens line PH2 1-PH6.
First, the optimized method has the closest results to the refer- The three-phase active power and instantaneous voltages and
ence for both the frequency of oscillation and resistance value currents at the POI are depicted in Fig. 14. An FFT analysis is
at the respective frequency. Second, methods 1 and 2 resulted performed on the current signal after the oscillation starts. As
not only in frequencies of oscillation farther from the reference shown in Fig. 15, the SSCI signal has a frequency of 29.34 Hz.
but in positive values of the resistance. This shows that poor Again, the reference and optimized method provide the closest
choices for the angles of the multisine signals injected can even results to the time-domain simulation results.
lead to completely inaccurate conclusions. In this case, methods 3) Frequency Scanning Simulation Time: When we inject a
1 and 2 show that the oscillations would be damped; while signal on top of a steady-state operation point of a system, a
methods 3, 4 and the reference show that the oscillations would transient period precedes the new steady-state. The impedance
be undamped. To further verify the frequency scan results, a estimation can only be calculated after this transient period ends
time-domain transient simulation needs to be performed, as is in order to achieve an accurate estimation through the harmonic
usually mandated by protocols for SSR risk assessment, such as injection technique. The three-phase active power output of the
the ERCOT nodal protocols [36]. grid at the POI is a good index to evaluate the new steady-state

Authorized licensed use limited to: Auckland University of Technology. Downloaded on May 25,2020 at 04:46:37 UTC from IEEE Xplore. Restrictions apply.
MATSUO et al.: OPTIMIZED FREQUENCY SCANNING OF NONLINEAR DEVICES APPLIED TO SSR SCREENING 2289

We can see that performing multiple single-frequency injec-

tions takes at least 6.66 times longer than one multisine injection.
This is the minimum ratio considering the minimum settling
times for each single-frequency injection. In industrial practice,
when performing frequency scanning studies, because we do not
know the exact settling times for each frequency in every grid
Fig. 15. Amplitude of the FFT spectrum of the current signal.
to be scanned, we need to assume a fixed T st with a certain
margin to ensure that the transient response has settled. For
instance, using fixed Tst = 0.5 s in the above case study for
all frequencies, the settling times are as follows:
35 
 
1
Tsf = + 0.5 = 17.563 s
fk
k=5

Tms = T0 + 0.5 = 1 + 0.5 = 1.500 s.

In this case, the proposed method is 11.71 times faster.
The simulations were run in a computer with a processor Intel
Core i7-7700 @ 3.60 GHz with 8 GB of RAM memory. This
computer took approximately 47.5 real-world seconds to run 1
simulation-second using time steps of 5 μs for this case study.
This means that the results of Tsf and Tms take in reality 47.5
Fig. 16. Normalized transient response for an injection of a single sine wave
of 30 Hz. Bottom plot is a zoomed-in version of the top plot. times the simulation time. When performing frequency scanning
studies for SSR risk assessment, a high number of contingencies
TABLE IX are considered. ERCOT, for example, establishes a contingency
SETTLING TIMES FOR EACH INJECTED FREQUENCY
criteria of up to N–14 [36]. It is possible to see that the real-
world time to simulate a large number of cases can be extremely
high. Thus, the proposed method can greatly reduce it while
maintaining good accuracy.

V. CONCLUSION
The harmonic injection method is a strong frequency scanning
technique for black-box models with nonlinear devices, such
as many that are present in active devices. The accuracy of
condition. The system is considered to settle to a steady state if this method is highly dependent on the number of injected
the envelope of the active power reaches and remains within a frequencies in each simulation. Normally, the lesser the number
certain percentage range of its final value (normally 2% or 5%) of multiple frequencies, the higher the accuracy. The ideal case
[33]. This settling time can be utilized to calculate the minimum is performing one simulation per single-frequency injection.
simulation time. However, this accuracy is only achieved at the expense of time.
Using (15) and (16), we can compare the minimum simulation This article proposes a technique to increase the accuracy of the
times between multiple single-frequency injections and the pro- scan results by optimizing the crest factor of a multifrequency
posed multisine injection. Fig. 16 shows the normalized transient injection signal with a small amplitude and applying differ-
response of the network-side active power after injecting a single ent optimized angles for each frequency. This signal enables
sine wave of 30 Hz. The magnitude is normalized as to set the performing only one simulation per side of the grid, therefore
final value of the upper envelope to 1 (by dividing all values by drastically reducing the total simulation time while maintaining
its final value). A ±5% range from the final value is used as good accuracy. The frequency scan cannot be completely error-
the settling limits. We can see that the settling time for 30 Hz is free, as there is no ideal choice of angles and amplitude that
0.197 s. The settling time results of all frequencies are shown in will nullify the errors in all nonlinear components of the grid
Table IX. at the same time, but this is also the case for any frequency
From the values of Table IX, (15) and (16) give the minimum scanning method, including single-frequency injections. The
simulation time for multiple single-frequency injections and for proposed method was utilized to scan two wind farm black-box
one multisine injection with period T0 = 1 s models and a VSC-based STATCOM to evaluate the risk of SSR.
The frequency scan results were benchmarked against the ideal
35  
1 case of multiple single-frequency injections and demonstrated
Tsf = + T stk = 8.010 s high accuracy while being 11.71 times faster. Moreover, when
fk
k=5
compared with other methods, the proposed method led to the
Tms = T0 + max {T stk } = 1 + 0.203 = 1.203 s. smallest errors and closest SSR analysis results to the reference,

Authorized licensed use limited to: Auckland University of Technology. Downloaded on May 25,2020 at 04:46:37 UTC from IEEE Xplore. Restrictions apply.
2290 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 56, NO. 3, MAY/JUNE 2020

thus demonstrating its superiority in terms of accuracy. Finally, [7] K. Narendra et al., “New microprocessor based relay to monitor and protect
time-domain transient simulations were conducted to validate power systems against sub-harmonics,” in Proc. IEEE Elect. Power Energy
Conf., 2011, pp. 438–443.
the results of the frequency scan. [8] L. Zhao, I. B. M. Matsuo, F. Salehi, Y. Zhou, and W. Lee, “Development
of a real-time web-based power monitoring system for the substation of
APPENDIX petrochemical facilities,” IEEE Trans. Ind. Appl., vol. 55, no. 1, pp. 43–50,
Jan./Feb. 2019.
TABLE X [9] F. Salehi, I. Brandao Machado Matsuo, A. Brahman, M. Aghazadeh
PARAMETERS OF THE 161 kV TRANSMISSION LINES ON A 100 MVA BASE Tabrizi, and W. Lee, “Sub-synchronous control interaction detection: A
real-time application,” IEEE Trans. Power Del., vol. 35, no. 1, pp. 106–116,
Feb. 2020.
[10] H. K. Nia, F. Salehi, M. Sahni, N. Karnik, and H. Yin, “A filter-less robust
controller for damping SSCI oscillations in wind power plants,” in Proc.
IEEE Power Energy Soc. General Meeting, 2017, pp. 1–5.
[11] M. A. Chowdhury, M. A. Mahmud, W. Shen, and H. R. Pota, “Nonlinear
controller design for series-compensated DFIG-based wind farms to mit-
igate subsynchronous control interaction,” IEEE Trans. Energy Convers.,
vol. 32, no. 2, pp. 707–719, Jun. 2017.
[12] M. A. Chowdhury and G. M. Shafiullah, “SSR mitigation of series-
compensated DFIG wind farms by a nonlinear damping controller using
partial feedback linearization,” IEEE Trans. Power Syst., vol. 33, no. 3,
pp. 2528–2538, May 2018.
[13] T. Joseph, C. E. Ugalde-Loo, S. Balasubramaniam, and J. Liang, “Real-
time estimation and damping of SSR in a VSC-HVDC connected series-
compensated system,” IEEE Trans. Power Syst., vol. 33, no. 6, pp. 7052–
7063, Nov. 2018.
[14] T. Joseph, C. E. Ugalde Loo, S. Balasubramaniam, J. Liang, and G. Li,
“Experimental validation of an active wideband SSR damping scheme for
series-compensated networks,” IEEE Trans. Power Del., vol. 35, no. 1,
TABLE XI pp. 58–70, Feb. 2020.
PARAMETERS OF THE 500 kV TRANSMISSION LINES ON A 100 MVA BASE [15] D. N. Walker, C. E. J. Bowler, R. L. Jackson, and D. A. Hodges, “Results of
subsynchronous resonance test at Mohave,” IEEE Trans. Power App. Syst.,
vol. PAS-94, no. 5, pp. 1878–1889, Sep. 1975.
[16] M. Bahrman, E. V. Larsen, R. J. Piwko, and H. S. Patel, “Experience with
HVDC - Turbine-generator torsional interaction at square butte,” IEEE
Trans. Power App. Syst., vol. PAS-99, no. 3, pp. 966–975, May 1980.
[17] J. Adams, C. Carter, and S. Huang, “ERCOT experience with sub-
synchronous control interaction and proposed remediation,” in Proc. PES
T&D, 2012, pp. 1–5.
[18] H. Liu et al., “Subsynchronous interaction between direct-drive PMSG
based wind farms and weak AC networks,” IEEE Trans. Power Syst.,
TABLE XII vol. 32, no. 6, pp. 4708–4720, Nov. 2017.
PARAMETERS OF THE TRANSFORMERS [19] I. B. M. Matsuo, F. Salehi, L. Zhao, Y. Zhou, and W. Lee, “An optimized
frequency scanning tool for sub-synchronous interaction analysis of non-
linear devices,” in Proc. IEEE/IAS 55th Ind. Commercial Power Syst. Tech.
Conf., 2019, pp. 1–7.
[20] M. S. Annakkage, C. Karawita, and U. D. Annakkage, “Frequency scan-
based screening method for device dependent sub-synchronous oscilla-
tions,” IEEE Trans. Power Syst., vol. 31, no. 3, pp. 1872–1878, May 2016.
[21] L. Yang, Z. Xu, J. Østergaard, Z. Y. Dong, K. P. Wong, and X. Ma,
“Oscillatory stability and eigenvalue sensitivity analysis of a DFIG wind
turbine system,” IEEE Trans. Energy Convers., vol. 26, no. 1, pp. 328–339,
Mar. 2011.
[22] M. Sahni, D. Muthumuni, B. Badrzadeh, A. Gole, and A. Kulkarni,
REFERENCES “Advanced screening techniques for sub-synchronous interaction in wind
farms,” in Proc. PES T&D, 2012, pp. 1–9.
[1] V. B. Virulkar and G. V. Gotmare, “Sub-synchronous resonance in series- [23] Y. Cheng, S. Huang, J. Rose, V. A. Pappu, and J. Conto, “ERCOT sub-
compensated wind farm: A review,” Renewable Sustain. Energy Rev., synchronous resonance topology and frequency scan tool development,”
vol. 55, pp. 1010–1029, Dec. 2015. in Proc. IEEE Power Energy Soc. General Meeting., 2016, pp. 1–5.
[2] L. Fan, R. Kavasseri, Z. L. Miao, and C. Zhu, “Modeling of DFIG-based [24] A. S. Subburaj et al., “Determination of sub synchronous control interac-
wind farms for SSR analysis,” IEEE Trans. Power Del., vol. 25, no. 4, tion between wind turbines and series compensated transmission lines,”
pp. 2073–2082, Oct. 2010. Int. J. Renewable Energy Res., vol. 6, no. 3, pp. 987–994, 2016.
[3] Y. Song and F. Blaabjerg, “Analysis of middle frequency resonance in [25] A. M. Kulkarni, M. K. Das, and A. M. Gole, “Frequency scanning analysis
DFIG system considering phase-locked loop,” IEEE Trans. Power Elec- of STATCOM - Network interactions,” in Proc. IEEE 6th Int. Conf. Power
tron., vol. 33, no. 1, pp. 343–356, Jan. 2018. Syst., 2016, pp. 1–6.
[4] A. E. Leon and J. A. Solsona, “Sub-synchronous interaction damping [26] X. Jiang and A. M. Gole, “A frequency scanning method for the identi-
control for DFIG wind turbines,” IEEE Trans. Power Syst., vol. 30, no. 1, fication of harmonic instabilities in HVDC systems,” IEEE Trans. Power
pp. 419–428, Jan. 2015. Del., vol. 10, no. 4, pp. 1875–1881, Oct. 1995.
[5] L. Wang, X. Xie, Q. Jiang, H. Liu, Y. Li, and H. Liu, “Investigation of SSR [27] R. Pintelon and J. Schoukens, System Identification: A Frequency Domain
in practical DFIG-based wind farms connected to a series-compensated Approach. New York, NY, USA: IEEE, 2001.
power system,” IEEE Trans. Power Syst., vol. 30, no. 5, pp. 2772–2779, [28] IXYS, IXSH 35N120A High Voltage, High Speed IGBT. [Online]. Avail-
Sep. 2015. able: http://ixapps.ixys.com/datasheet/92774.pdf
[6] L. Vanfretti, M. Baudette, J. L. Domínguez-Garcíaz, A. White, M. S. [29] J. M. Guerrero, M. Leetmaa, F. Briz, A. Zamarron, and R. D. Lorenz,
Almas, and J. O. Gjerdeóy, “A PMU-based fast real-time sub-synchronous “Inverter nonlinearity effects in high-frequency signal-injection-based
oscillation detection application,” in Proc. IEEE 15th Int. Conf. Environ. sensorless control methods,” IEEE Trans. Ind. Appl., vol. 41, no. 2,
Elect. Eng., 2015, pp. 1892–1897. pp. 618–626, Mar./Apr. 2005.

Authorized licensed use limited to: Auckland University of Technology. Downloaded on May 25,2020 at 04:46:37 UTC from IEEE Xplore. Restrictions apply.
MATSUO et al.: OPTIMIZED FREQUENCY SCANNING OF NONLINEAR DEVICES APPLIED TO SSR SCREENING 2291

[30] C. Basso, “Small-signal modeling and analytical analysis of power con- Long Zhao (Student Member, IEEE) received the
verters,” in Proc. Appl. Power Electron. Conf. Expo, 2013, pp. 14–18. B.S. degree from Eastern Washington University, Ch-
[31] S. Boyd and L. Vandenberghe, Convex Optimization. New York, NY, USA: eney, WA, USA, in 2012, and the M.S. degree from
Cambridge Univ. Press, 2004, pp. 215–223. Wichita State University, Wichita, KS, USA, in 2014,
[32] J. Schoukens, R. M. Pintelon, and Y. J. Rolain, “Broadband versus stepped both in electrical engineering. He is currently working
sine FRF measurements,” IEEE Trans. Instrum. Meas., vol. 49, no. 2, toward the Ph.D. degree in power and energy systems
pp. 275–278, Apr. 2000. in the University of Texas at Arlington (UTA), Arling-
[33] K. Ogata, Modern Control Engineering, 5th ed. Englewood Cliffs, NJ, ton, TX, USA.
USA: Prentice-Hall, 2010. He is a member of the Energy Systems Research
[34] N. W. Miller, J. J. Sanchez-Gasca, W. W. Price, and R. W. Delmerico, Center (ESRC), UTA. His research interests include
“Dynamic modeling of GE 1.5 and 3.6 MW wind turbine-generators for power system analysis, demand response, renewable
stability simulations,” in Proc. IEEE Power Eng. Soc. General Meeting, energy, energy storage, power system markets, and arc flash protection.
2003, vol. 3, pp. 1977–1983. Mr. Zhao is currently the IEEE Industry Applications Society Student Chapter
[35] A. B. Birchfield, T. Xu, K. M. Gegner, K. S. Shetye, and T. J. Overbye, Chair at UT Arlington.
“Grid structural characteristics as validation criteria for synthetic net-
works,” IEEE Trans. Power Syst., vol. 32, no. 4, pp. 3258–3265, Jul. 2017.
[36] ERCOT Nodal Protocols: Sub-Synchronous Resonance, ERCOT NPRR
562, 2018.

Igor Brandão Machado Matsuo (Member, IEEE)

received the B.S. and M.S. degrees in electrical engi- Yuhao Zhou (Student Member, IEEE) received the
neering from the University of Sao Paulo, Sao Paulo, B.S. and M.S. degrees from the Huazhong University
Brazil, in 2012 and 2015, respectively, and the Ph.D. of Science and Technology, Wuhan, China, in 2013
degree from the University of Texas at Arlington and 2016, respectively, both in electrical engineering.
(UTA), Arlington, TX, USA, in 2019. He is currently working toward the Ph.D. degree in
He was a member of the Energy Systems Research the University of Texas at Arlington, Arlington, TX,
Center (ESRC), UTA and held full scholarship from USA.
CNPq-Brazil. He has also worked for Techimp; the He is a member of the Energy Systems Research
Foundation for the Technological Development of Center (ESRC). His research interests include dy-
Engineering (FDTE); and Calpine. His experiences namic equivalent modeling of wind farms, power
include renewable energy; subsynchronous resonance; monitoring systems; system analysis, renewable energy, energy storage,
partial discharges; power system planning and analysis; and automation and and arc flash protection.
protection systems in the generation, transmission, distribution, and industrial
levels.

Farshid Salehi (Member, IEEE) received the B.S.

degree from Bu-Ali Sina University, Hamedan, Iran,
in 2007, M.S. degree from the University of Semnan, Wei-Jen Lee (Fellow, IEEE) received the B.S.
Semnan, Iran, in 2011, and the Ph.D. degree from the and M.S. degrees from National Taiwan University,
University of Texas at Arlington (UTA), Arlington, Taipei, Taiwan, in 1978 and 1980, respectively, and
TX, USA, in 2018, all in electrical engineering with the Ph.D. degree from The University of Texas at
focus on power systems. Arlington, Arlington, TX, USA, in 1985, all in elec-
His experiences include advanced power system trical engineering.
modeling and analysis with focus on transient and In 1985, he joined the University of Texas at
stability studies, electromagnetic transient modeling Arlington, where he is currently a Professor with
and simulation of switching over-voltage, temporary the Department of Electrical Engineering and the
overvoltage, harmonics, and subsynchronous interactions. He has worked as Director of the Energy Systems Research Center. He
an Engineer at Monenco consulting engineer, as a Research Assistant with the has been involved in research on power flow, transient
Energy Systems Research Center (ESRC), UTA, and is currently working as a and dynamic stability, voltage stability, short circuit, relay coordination, power
Senior Engineer at DNV GL Energy Advisory. His research interests include quality analysis, renewable energy, and deregulation for utility companies.
power system dynamics and transient stability, arc flash, protection, power Dr. Lee is a registered Professional Engineer in the State of Texas.
system optimization, and grid integration of renewable resources.

Authorized licensed use limited to: Auckland University of Technology. Downloaded on May 25,2020 at 04:46:37 UTC from IEEE Xplore. Restrictions apply.