This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
1

Reduced-Order and Aggregated Modeling of

Large-Signal Synchronization Stability for
Multi-Converter Systems
Mads Graungaard Taul, Student Member, IEEE, Xiongfei Wang, Senior Member, IEEE,
Pooya Davari, Senior Member, IEEE, Frede Blaabjerg, Fellow, IEEE

Abstract—During severe grid faults, grid-following converters activity in the modeling, analysis, and mitigation of loss of
may become unstable and experience loss of synchronization synchronization of converter-based generation is observed [5].
when complying with requirements for low-voltage ride-through As first outlined in [6], grid-synchronization instability or
capability. This phenomenon is well-described, understood, and
modeled for single-converter systems but lacks a modeling loss of synchronization (LOS) may occur during weak-grid
framework when extended to multi-converter systems. To fill this and severe low-voltage conditions. Under such circumstances,
gap, this work presents the necessary stability conditions and the converter operation acts as a positive feedback loop to
aggregated reduced-order models for different multi-converter its own synchronization process, which in turn makes the
configurations, which can be used to assess the transient synchro- PLL unable to remain synchronized with the external grid
nization stability of grid-following converters under symmetrical
grid faults. The necessary conditions for transient stability and voltage [7]. Since this issue happens during large disturbances,
the aggregated models are verified through numerous simulation small-signal linearized models cannot be adopted to repre-
studies, which verify their high accuracy for large-signal synchro- sent the transient instability problem. For that reason, much
nization stability assessment. To that end, the Anholt wind power work has been devoted to model and analyze LOS using the
plant is considered as a case study where the aggregated model is nonlinear dynamics governing the synchronization process. A
compared to the full operation of a wind farm string containing 9
full-order grid-following converter models. High model accuracy necessary condition for transient stability was derived in [6],
is obtained, and the computational burden associated with the which highlights the root-cause of large-signal synchronization
proposed model is reduced with a factor of 100 compared to a instability. This model is based on steady-state conditions
full-order representation on the tested system. Accordingly, the which cannot be used for transient stability analysis when
presented analysis and proposed modeling are attractive as a the necessary condition is met. To address this, the authors
screening tool and a convenient approach for early-stage fault
analysis of a system design. in [7] proposed a quasi-static large-signal model of a PLL-
synchronized converter under grid faults. Building on the
Index Terms—Aggregated Modeling, Grid-Connected Con- foundation of this quasi-static large-signal model, numerous
verters, Grid Faults, Reduced-Order Modeling, Synchronization
Stability, Transient Stability Analysis. work has been done to describe and model LOS [8]–[12]
alongside motivating ideas for LOS mitigation strategies [3],
[13]–[18]. Also, since the nonlinear model has no known
I. I NTRODUCTION analytical solution, nonlinear graphical tools and numerical
approximations have been used to assess the transient stability
ITH the unprecedented integration of renewable energy
W sources to the modern power system, synchronous gen-
erators are being replaced by power-electronics-based genera-
[19]–[22].
Nonetheless, all of the cited publications for modeling
of PLL-synchronized grid-following converters under large-
tion [1]. With this remarkable transition follow requirements signal disturbances are based on a single-converter-infinite-
for the responsible operation of power converters and capabil- bus representation. Therefore, the developed models can only
ity to provide ancillary services and low-voltage ride-through be used to represent one single system and cannot cap-
(LVRT) support [2]. However, during severe grid faults, the ture the behavior of paralleled or more complicated multi-
converter control, in particular the synchronization dynamics, converter systems. Regarding modeling of multi-converter
may become unstable, even when the LVRT requirements are systems with focus on the synchronization dynamics, some
met [3]. Along these lines, it is outlined by the British network previous work has been performed [23]–[28] where [23]–
operator, National Grid, that the risk of synchronization in- [26] focus on the small-signal behavior. A large-signal model
stability of phase-locked loop (PLL)-synchronized converters is presented in [27], but here all converters are assumed to
during grid faults is increasing [4]. Consequently, increased share the same point-of-synchronization (POS) and point-
of-connection (POC), which significantly reduces its usage
Mads Graungaard Taul, Xiongfei Wang, Pooya Davari, and Frede Blaabjerg
are all with the Department of Energy Technology at Aalborg University, Den- for practical applications. Lastly, the authors in [28] present
mark (email: [email protected], [email protected], [email protected], [email protected]). a large-signal model for paralleled converters without any
This work was supported by the Reliable Power Electronics-Based Power assumptions on a shared POS or POC. However, the model
System (REPEPS) project at the Department of Energy Technology, Aalborg
University, as a part of the Villum Investigator Program funded by the Villum in [28] is only defined for two paralleled units, where a
Foundation. generalization for n converters seems too cumbersome since

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
2

the frequency dynamics of all converters are coupled with work. Up until now, the aggregated models are developed
each other. Capturing these couplings for multiple convert- to replicate the dynamics of the PCC [29]. On the contrary,
ers and complex interconnections become impractical and this paper aims to present an aggregated model to describe
time-consuming. Therefore, in regards to the modeling of the internal synchronization stability of the multi-converter
large-signal synchronization stability of grid-following multi- system in an aggregated reduced-order manner, such that fast
converter systems, a generalized modeling framework is still transient instability screening can be performed and a physical
lacking from the existing literature. To that end, an aggregated understanding of the instability phenomenon can be obtained.
method that can accurately capture the collective synchroniza- Besides representing the converter as an ideal current
tion dynamics of multi-converter systems without considering source, oriented by the PLL dynamics, this work also shows
the couplings between all interconnected units is needed. that the outer dc-link voltage control and AC voltage control
can be neglected from the model when focusing on the syn-
A. Model Aggregation chronization stability under grid faults. This is different from
other reduced-order modeling approaches such as singular
In regards to aggregated modeling of multi-converter sys-
perturbation theory, used to neglect the fast dynamics of the
tems and wind farms, this is not a new area of study [29].
model [44]. In such work, using the converter operation under
For analysis of the internal stability of multi-converter sys-
the fault behavior to reduce the outer power loops cannot be
tems, detailed models are considered [30], whereas aggregated
performed since they belong to slow preserved modes.
models are often used to investigate the impact that a wind
Therefore, even though much research has been done on
farm or large-scale system has on the connected power system
aggregation and reduced-order modeling of multi-converter
[31]. The aggregation procedure usually involves develop-
systems, the modeling has not been focused on severe grid
ing an aggregated representation of the converters (either
faults. In this case, the modeling may accurately use the
single-machine or multi-machine representation), an equiv-
fault conditions to employ assumptions, leading to a highly
alent impedance preserving the power flow of the system,
simplified structure, which still preserves the collective internal
and an equivalent representation of the stochastic energy
synchronization dynamics of the system. This is the motivation
yield [32], [33]. For aggregation, single-turbine aggregation
and approach pursued in this work. Finally, in addition to this,
is often considered insufficient in terms of accuracy, which
disclosing how the paralleled converter operation changes the
is why multi-machine aggregation is usually performed for
physical interpretation and necessary stability conditions of the
transient studies [34]. For the multi-machine aggregation,
system has so far not been described.
many methods focus on how to perform proper clustering
Thus, this paper aims to fill these research gaps by consider-
of converters [35]. This includes K-means clustering [36],
ing three system configurations that cover most multi-converter
support vector clustering [37], multi-objective optimization
configurations. For these configurations, the necessary con-
algorithm [35], simple clustering based on similar wind speeds
ditions for transient stability are derived. These conditions
[38], or clustering through coherency equivalence [39], [40].
are beneficial for two reasons. First, the necessary conditions
Thus, much research focuses on the clustering based on the
serve as a reliable tool to understand and assess the transient
incoming wind speed used for small-signal analysis rather than
stability and, secondly, the q-axes voltage components, used
simplified assessment of converter synchronization stability
to derive the necessary conditions, lay the foundation for
during severe grid faults.
developing reduced-order dynamic models for each configu-
To that end, most presented aggregated models employ
ration. In contrary to previous work on LOS, the influence of
the full-order dynamics of the aggregated converters [41],
network capacitance on the necessary stability condition is also
resulting in more complexity compared to reduced-order ag-
revealed. To avoid modeling the coupling dynamics between
gregated models. In [42], a reduced-order aggregated structure
all interconnected units and the time-complexity of doing so,
is presented to characterize the small-signal frequency support
this work presents an aggregated reduced-order large-signal
of aggregated wind turbines. Also, an aggregation model for
model for a daisy-chain configured system, e.g., as used in
a DFIG-based wind farm is proposed in [43] to study the
large offshore wind farms, for transient stability assessment.
low-frequency power oscillations. However, these dynamic Accordingly, the research contributions can be summarized
models are based on transfer functions and only the small- as:
signal angular stability is assessed.
1) Identification of necessary stability conditions for the
three descriptive paralleled converter system configura-
B. Research Gaps and Contributions tions.
Despite a lot of research on aggregated modeling, the 2) Extension of these conditions to reduced-order large-
above works are not explicitly focused on the synchronization signal scalable models, which are designed for a low-
stability during grid faults, for which further simplifying as- order transient stability assessment tool.
sumptions can be made to effectively reduce the computational 3) For daisy-chain collector systems where converters are
burden. During severe fault conditions, all converters can pro- separated by non-negligible impedances, an aggregated
vide 1 pu of reactive current support, despite their initial wind reduced-order large-signal model is proposed. This
speed. Hence, when studying severe faults, a highly simplified model eliminates the need to model all couplings be-
reduced-order aggregated model can be developed to assess tween the interconnected units, brings a high model
the large-signal synchronization stability, as is pursued in this accuracy, and therefore reduces the computational re-

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
3

quirements by a factor of 100 when compared to a

detailed simulation study on a real functioning wind
farm.
4) Showing that the outer power loops of the converters can
be neglected for the stability assessment under severe
grid faults and that the presented method also provides
accurate assessments under heterogeneous converter op-
eration.
The remainder of this paper is structured as follows: Sec-
tion II presents the necessary conditions for transient stability
of three representative multi-converter configurations. Also, Fig. 1. Simplified single-line diagram of paralleled converters injecting
how potential network shunt capacitances influence the static current into an external grid. Here, the parallel converters have a common
stability limits is analyzed. In Section III, the q-axes voltage point of synchronization (POS) and point of connection (POC), here at the
PCC.
components used to derive the necessary stability conditions
are merged with the dynamics of the PLL to establish large-
signal reduced-order models for all system configurations. be equal. However, during severe faults, where the network
Additionally, a proposed aggregated model for analyzing the voltages drop low and a breaking chopper takes care of the
transient stability for e.g., a wind farm collector system is active power from the renewable energy source, all converters
presented. All necessary conditions and large-signal models can supply rated reactive current, which justifies the assump-
are numerically verified in Section IV, whereas a case study tion of equal currents during the fault. To that end, during
on the Anholt offshore wind farm is conducted in Section V severe faults, where the POS voltage is low, converters should
to verify the proposed aggregated model further. Here, the prioritize 1 pu of reactive current injection to support the local
proposed reduced-order aggregated model is also verified network voltage [2]. Considering this, then assumption 2 can
under heterogeneous converter operation. Finally, concluding be justified. Lastly, assumption 3 can be justified for two
remarks are enclosed in Section VI. reasons. First, the dominating dynamics of loss of synchro-
nization lies in the low-frequency range [22], [45]. Secondly,
II. S TATIC S TABILITY L IMITS the bandwidth of the inner current regulator is usually placed
The static limitation or necessary condition for transient much higher than that of the synchronization process, which
stability is evaluated for three representative configurations of facilitates that they can be analyzed individually [7], [14], [15].
paralleled converters: converters with a common point of syn- To that end, only symmetrical faults are considered for this
chronization (POS) and point of connection (POC), converters study. This is done since during a worst-case asymmetrical
with a different POS but common POC, and converters with fault (a double line-to-ground fault [46]), the resulting positive
different POS and POC. These together form a representative and negative sequence voltage magnitudes at the fault location
platform for literally any multi-converter system. Hence, from never drop below 1/3 pu, which implies that the converters is
a system configuration point of view, the models presented in unlikely to operate at an unstable operating point during the
this work will be general and scalable. The static models reveal fault. Thus, unbalanced faults do not pose any threat for the
the fundamental instability phenomenon and are, therefore, rel- instability mechanisms studied in this paper
evant when assessing the large-signal synchronization stability.
To that end, the formulations used to develop the necessary
stability conditions are also the fundamental building blocks
for constructing the reduced-order large-signal models. The A. Parallel Converters with Common Point of Synchronization
assumptions made for the derivation of the static and dynamic and POC
models presented in this work are the following:
The simplest configuration of the paralleled operation is
1) All converters have homogeneous dynamics, i.e., the shown in Fig. 1 where n paralleled converters inject current
converter topology, the control law, the controller pa- into an external grid. The external grid consists of a line
rameters, and the output current are all the same. impedance ZL and an equivalent Thevenin grid (Zth , Vth )
2) Outer power loops are not included and can be neglected with a parallel feeder branch where a symmetrical fault is
during severe faults. considered to occur. Zf e denotes the combined feeder and
3) The converter operation is represented as an ideal con- fault impedance. Such a parallel configuration with a shared
trollable current source, which is oriented using the PLL POS could represent paralleled photovoltaic (PV) inverters
dynamics. and other paralleled systems that are located in an electrical
The first assumption implies that the presented models only vicinity of each other. The voltage at the PCC can from the
applies to multi-converter systems where the converters pos- superposition of the linear network be expressed as
sess a high degree of similarity. Since paralleled converters
used in renewable applications such as wind and solar might vP CC = Kg (ωg )Vth ej(θg +φg ) + Kc (ω1 )I1 ej(θC1 +φc1 )
be exposed to different wind speeds and solar irradiance,
the injected currents from the paralleled converters might not + Kc (ω2 )I2 ej(θC2 +φc2 ) + · · · + Kc (ωn )In ej(θCn +φcn ) (1)

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
4

which succinctly can be written as

n
X
vP CC = Kg (ωg )Vth ej(θg +φg ) + Kc (ωi )Ii ej(θCi +φci ) ,
i=1
(2)
with [5]
 
 Zf e (ωg ) 
Kg (ωg ) =  , (3)
Zf e (ωg ) + Zth (ωg ) 
 
 Zf e (ωi )Zth (ωi ) 
Kc (ωi ) = ZL (ωi ) + , (4)
Zf e (ωi ) + Zth (ωi ) 
 
Zf e (ωg ) Fig. 2. Simplified single-line diagram of paralleled converters with different
φg (ωg ) = ∠ , (5) points of synchronization but a common point of connection (POC), i.e., PCC.
Z (ω ) + Zth (ωg )
 fe g 
Zf e (ωi )Zth (ωi )
φc (ωi ) = ∠ ZL (ωi ) + , (6)
Zf e (ωi ) + Zth (ωi ) nization stability. This is the static stability condition for the
maximum current magnitude that can be injected as
where ωi is the estimated frequency by the ith converter, θg is Vth Kg (ωg )
the angle of Vth , θCi = θP LLi + θI is the angle of the injected IC ≤ . (8)
nKc (ωP LL )| sin(θI + φc (ωP LL ))|
current vector of the ith converter, and θI is the power factor
angle. During severe faults, the converters should inject 1 pu From (8), it can be seen that the issues encountered for weak-
of reactive current to support the local voltages [2]. Hence, grid conditions for a single-converter system are exacerbated
θI can be assumed to be equal for all converters. It should be when multiple parallel converters are considered.
noted that since the paralleled converters in Fig. 1 all share
B. Parallel Converters with Different Point of Synchronization
the same PCC where the synchronization is performed; they
but Common POC
can be assumed to have the same estimated angle from the
PLL which in steady-state satisfies that θP LL = θP CC . This For the second configuration where the parallel converters
also implies that θC1 = θC2 = · · · = θCn , provided that experience different POS, i.e. the place where the voltage mea-
the currents of the paralleled converters are the same. From surement is fed to the PLL, but share a common POC is shown
this, the PCC voltage can be expressed in the shared rotating in Fig. 2. Here, the POS is located on the low-voltage side of
frame (multiplying with e−jθP LL on both sides of (1)). This e.g. the wind turbine step-up transformer. Consequently, the
will cancel θP LLi from θCi . Then, by evaluating the imaginary voltage measurement provided to the synchronization unit will
part, the q-axis component of vP CC can be obtained as be different from the previous case as the injected currents of
each converter have a stronger coupling to the voltage at the
vP CC,q = Kg (ωg )Vth sin(θg + φg (ωg ) − θP LL ) POS and globally a weaker coupling to the external grid. The
| {z } POS voltage of the n paralleled converter in Fig. 2 can be
Grid-Synchronization Term, vq−
expressed as
+ nKc (ωP LL )IC sin(θI + φc (ωP LL )), (7)
| {z } vP OS,1 = vP CC + I1 Ztl,1 ej(θC1 +φtl,1 ) ,
Self- and Cross-Synchronization Terms, vq+
..
where IC is the magnitude of the current vector injected .
by each of the n converters. From (7), it can be seen that vP OS,n = vP CC + In Ztl,n ej(θCn +φtl,n ) . (9)
each converter when measuring the PCC voltage sees the Inserting the expression for the PCC voltage from (2), the POS
contribution from the equivalent grid and the contribution voltage at the pth converter becomes
from all the other converters, i.e., an increased self- and n
cross-synchronization term. This implies that seen from each vP OS,p = Kg (ωg )Vth ej(θg +φg ) +
X
Kc (ωi )Ii ej(θCi +φci )
converter, the equivalent grid impedance seems n times larger i=1
compared to the case where only one converter is considered.
+ Ip Ztl,p ej(θCp +φtl,p ) . (10)
This is consistent with the findings of the linearized systems in
[27], [47], [48]. This is intuitive as each current contribution Assuming equal leakage inductances of all the step-up trans-
from all of the paralleled units generates a voltage drop formers, the q-axis component of each POS voltage becomes
in the line impedance ZL . Accordingly, the synchronization vP OS,q = Kg (ωg )Vth sin(θg + φg − θP LL )
process of each converter becomes harder than that of a single- | {z }
converter system, since the PCC voltage is not only dependent Grid-Synchronization Term, vq−

on the external grid voltage, but also the voltage drops on ZL + IC nKc (ωP LL ) sin(θI + φc (ωP LL ))
generated by the n − 1 neighboring converters.
| {z }
Self- and Cross-Synchronization Term, vq+1
For a stable operating point to exist, a solution for θP LL , + IC Ztl (ωP LL ) sin(θI + φtl (ωP LL ) (11)
which assures that vP CC,q = 0 must exist. Using this, one | {z }
can derive a necessary condition for the large-signal synchro- Additional Self-Synchronization Term, vq+2

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
5

From (11), it can be noticed that the positive feedback term

or self- and cross-synchronization term is further increased
compared to (7) due to the included step-up transformer.
Accordingly, the limit for the injected currents then becomes
constrained to
Vth Kg (ωg )
IC ≤ ,
|nKc (ωP LL ) sin(θI + φc (ωP LL )) + Xtl (ωP LL ) cos(θI )|
(12)
neglecting transformer resistance. During purely reactive cur-
rent injection (θI = −90◦ ) as specified by grid codes during
severe grid faults, the static stability constraint in (12) is iden-
tical to (8). This is because the resistance of the transformer is
neglected and the dynamic behavior of the current controller is
assumed ideal for this static model. For actual implementation,
the system with the transformer will be more prone to instabil-
ity as the dynamics of the system become slower, meaning that
the assumption of an ideal controllable current source for the
converter becomes less valid. Also, during a severe fault event,
the converter has to change its injected currents from active
to capacitive reactive. This transition involves some transient
behavior where the second term in the denominator in (12) will
be non-zero, which increases the value of the denominator in
the transient part. The worst-case is at the fault instant where Fig. 3. Simplified single-line diagram of paralleled converters in wind
cos(θI ) = 1, whereas cos(θI ) → 0 as the current controller farm radial string with different points of synchronization (POS) and point
settles, usually in a few milliseconds. of connections (POC). Each converter is interconnected through collector
impedances represented as Zc,i

C. Parallel Converters with Different Point of Synchronization

and POC impedances usually are much smaller compared to the wind
Parallel converters may topologically be interfaced such farm park transformer, the phase-angle difference between
that they neither share the same POS nor the same POC. An the injected currents from the n converters is also small.
example of this is a wind farm string or collector system as Thus, it can be assumed that the injected currents from the
shown in Fig. 3, where each converter is connected in parallel paralleled converters have the same magnitude and phase (IC
along a string separated by collector impedances, denoted Zc . and θI ). Based on the assumption of small-angle differences
This is also known as a daisy-chain turbine configuration. As along the string and that each converter in steady-state is
depicted in Fig. 3, each of the m paralleled wind farm strings synchronized to its POS voltage, the PLL phase-angles are
consists of n internal converters. It should be mentioned that equal as θP LL1 = θP LL2 = · · · = θP LLn = θP LL . Using the
for offshore wind farm applications, the PCC point is often assumptions described above and multiplying each side with
referred to as the point of the right-hand side of the transport e−jθP LL , θCi reduces to θI . Using the expression for the PCC
line/cable ZL in Fig. 3. However, to be consistent with the voltage in (2) and taking the imaginary part of the POS voltage
previous diagrams, the PCC point is continued to be referred at the pth wind turbine, one can get that
to as the point in the wind farm where all the parallel collector
grids are connected. The POS voltage at the pth wind turbine vP OS,q,p = Kg Vth sin(θg + φg − θP LL )
can be expressed as + mKc Is sin(θI + φc ) + IC Ztl sin(θI + φtl )
  p
p n
X
X
Zc,i ejφcol,i
X + Zc,i IC (n − i + 1) sin(θI + φcol,i ) (14)
vP OS,p = vP CC + Ij ejθCj  i=1
i=1 j=i
where m is the number of parallel strings, Is is the total
+ Ip Ztl,p ej(θCp +φtl,p ) , (13)
string current magnitude of the m paralleled strings, and Ztl
where φcol is the impedance angle of the collector cable, n is considered equal for all converters. To develop a single
is the number of turbines on the string, and n ≥ p ≥ 1 is necessary condition for the whole string, the location in
the POS number of the wind turbine voltage counting from the wind farm string, which is most susceptible to loss of
the PCC towards the end of the string. During a severe grid synchronization is considered. The worst-case location along
fault, all converters in the wind farm string can supply 1 pu the wind farm string will be at the end, i.e., at the nth
of reactive current. This is true even though each converter at converter. This is true since this will create the largest positive
the pre-fault operating point may have significantly different feedback in the PLL synchronization loop as it is evident from
injections of active current based on different wind speeds and (14). Then, if the nth converter can be guaranteed to operate
wake effects in the wind farm. To that end, as the collector at a stable equilibrium point, so can the remaining units in the

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
6

string too. Accordingly for the nth converter, the POS voltage where
is  ZC 2 Zf ec 
   2

Kgc =  , φg = ∠ ZC Zf ec , (20)
K1  K1
vP OS,q,n = Kg Vth sin(θg + φg − θP LL )  
 ZC (ZL Zf e Zth + ZC (ZL Zth + Zf e (ZL + Zth ))) 
| {z } Kcc =  ,
Grid-Synchronization Term, vq− K1 
(21)
+ mKc Is sin(θI + φc ) + IC Ztl sin(θI + φtl )  
| {z } | {z } ZC (ZL Zf e Zth + ZC (ZL Zth + Zf e (ZL + Zth )))
Mutual-String Interaction Term, vq+1 Transformer Leakage Term, vq+2 φcc =∠ ,
K1
Xn (22)
+ Zc,i IC (n − i + 1) sin(θI + φcol,i ) . (15)
i=1
and
| {z }
Self-String Interaction Term, vq+3 K1 = ZC 2 (Zf e + Zth ) + ZC (ZL Zth + Zf e (ZL + 2Zth ))
+ ZL Zf e Zth , (23)
For the q-axis voltage component to be zero, the first term on
the right-hand side must be sufficient to cancel the remaining and ZC = 1/(0.5jωC) since half of the capacitance is
offsetting terms. From this, and assuming that Is = nIC , the distributed at each end of the line. During pure reactive current
necessary stability condition becomes injection (θI = −90◦ ), the static stability margin is determined
by the ratio between Vth Kgc and <{Kcc }. For analytical
Vth Kg
IC ≤ , (16) simplicity, it is assumed that the converter operation does not
|A| influence the voltage at the fault location (Zf e ≈ 0), then
where RL
<{Kcc } = 2 + L2 (ω 2 − ω 2 )) , (24)
1 + (0.5Cω)2 (RL L LC
A = nmKc sin(θI + φc ) + Ztl sin(θI + φtl ) √
Xn where ωLC = 1/ LL C. Since the LC resonance frequency
+ (n − i + 1)Zc,i sin(θI + φcol,i ), (17) is indeed much higher than the network operating frequency,
i=1 (24) can be approximated as

and Kg , Kc , φc , Ztl , φtl , Zc , and φcol are frequency-dependent RL

<{Kcc } = 2 − (L ω . (25)
variables. For the case where the collector grid impedances are 1 + (0.5Cω)2 (RL 2
L LC ) )
identical, A reduces to Hence if RL > LL ωLC , then the real part will be reduced
from RL , which increases the stability margin, whereas if
A = nmKc sin(θI + φc ) + Ztl sin(θI + φtl ) RL < LL ωLC , the real part of Kcc will be increased, which
Zc (n + n2 ) reduces the stability margin. However, it should be noted that
+ sin(θI + φcol ). (18)
2 this term is multiplied with (0.5Cω)2 , which in practice is a
very small number. As an example, using the cable data in
As can be noted from the first term in (17) and (18), each
[50] for the type 2XS(FL)2YRAA 18/30(36) kV, one will find
paralleled string is considered to consist of n converters.
that for the smallest and largest conductor sizes, the change
This is a strong assumption, which may not be applicable in
in <{Kcc } from RL only exceeds 5% when the cable length
practice since different strings may have a different number
is increased above 35-50 km. Hence, for this study involving
of converters. However, the factor nm in the above terms can
interconnecting cables between wind turbines in wind farms,
simply be replaced with ntot , representing the total number of
which are much shorter, the use of the static stability limits
converters in the paralleled strings.
without considering the effect of the cable capacitance is well
justified.
D. Influence of Cable Capacitance
III. AGGREGATED R EDUCED -O RDER L ARGE -S IGNAL
For offshore wind farms and due to the public opposition to M ODELS OF S YSTEM C ONFIGURATIONS
overhead lines on land, submarine and land cables are often The static stability limits derived for the different system
used for the connection and interconnection of wind turbines configurations represent the necessary condition for large-
and wind farms. This implies that the cable capacitance may signal synchronization stability. Besides these models being
not be neglected since it can be 20-50 times higher than for useful in understanding the LOS phenomena and providing a
the capacitance for overhead lines [49]. Therefore, the impact necessary condition for stability, they cannot alone be used for
that this non-negligible capacitance has on the static stability predicting the dynamical response of the frequency estimation
limit must be evaluated. Considering one single converter from of each converter. To capture this, the PLL dynamics need to
Fig. 1 where the line impedance (ZL ) is approximated with a be included to construct a large-signal reduced-order nonlinear
cable Π−model, the static limitation for current transfer can model of the synchronization stability. This is done by taking
be written as the q-axes voltages in (7), (11), (15), used to derive the
Vth Kgc (ωg ) necessary stability conditions, and attaching the synchronizing
IP CC ≤ , (19) PLL dynamics to this simplified q-axis voltage model. This
Kcc (ωP LL )| sin(θI + φcc (ωP LL ))|

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
7

Fig. 4. Reduced-order quasi-static large-signal models for (a): parallel converters with common POS and POC (S open), from the configuration in Fig. 1,
and paralleled converters with different POS but common POC (S closed), from the configuration in Fig. 2, and (b): parallel converters with different POS
and POC, from the configuration in Fig. 3. For subfigure (b), this is shown for the pth PLL-synchronized converter in a wind farm string.

procedure describes the link between the formulation of the

necessary stability conditions and the development of the
reduced-order models.

The large-signal reduced-order nonlinear models of the

three system configurations from Section II are shown in
Fig. 4 where the simplified q-axes voltage equations derived
are combined with the second-order PLL model. The block
calculating the frequency-dependent parameters computes the
actual impedance of a given line segment based on the
static inductance of that segment and the operating frequency
estimated by the PLL, e.g. X = LωP LL . For the first two
system configurations contained in Fig. 4(a), the estimated
converter frequency is either identical or nearly identical for
the n paralleled converters due to the shared POC. Therefore, Fig. 5. Aggregated reduced-order quasi-static large-signal model for parallel
only one frequency estimation needs to be modeled as shown converters with different POS and POC where all converters are represented
as an equivalent frequency estimation.
in Fig. 4(a). However, for the system topology where paralleled
converters are interconnected as for a wind farm string, there
exists a significant impedance between the converters, which fication [52]) from each converter to the fault location is small.
implies that the estimated PLL frequency of the paralleled Hence, the internal converters tend to synchronize coherently
converters will be different during transient conditions. This since they are strongly coupled to one another [53]–[55]. Due
effect is captured using the large-signal reduced-order model to this synchronization consensus, which implies that either
in Fig. 4(b), where the PLL frequency estimated by the pth all converters will experience synchronization instability or
converter is modeled. It should be noted with this reduced- neither will, such converters can be grouped and their transient
order model that the n models for the n paralleled converters stability can be analyzed as a whole [56], [57]. Thus, despite
are not coupled. This means that for a given fault condition, the already simplified structure of the model, the intrinsic
some converters may be unstable, whereas others will remain coherency feature of the string converters can be used to
stable depending on the location in the wind farm string. Such aggregate and reduce the model even further. This is useful
a modeling approach does not represent a real scenario where when performing many consecutive cases to identify stability
the estimated frequencies are coupled as modeled for two boundaries and parameter trends.
paralleled converters in [28]. For power networks involving
potential low-inertia systems, the frequency cannot be viewed
as a global parameter in the initial contingency phase [51]. A. Aggregated Model for Wind Farm String
Therefore, as comprehensively studied in [51], different units To be able to capture the frequency consensus among the
may oscillate at different angular frequencies during a fault n paralleled converters for the system topology where the
depending on their coupling to each other. However, since converters have a different POS and POC, without having
the collector grid impedances are much smaller than the to model the frequency-couplings between all interconnected
park transformer and the transport cables, the difference in converters, an aggregated model is here proposed to assess the
admittance distance (which can be used for coherency identi- large-signal synchronization stability.

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
8

The reduced-order aggregated model in Fig. 5 uses an weighted sum of the methods is due to the overlook of the
equivalent impedance of the wind farm string/collector network capacitance in the equivalent impedance. The systems
system to preserve the collective dynamics of the internal with a higher network capacitance possess a slightly better
frequency by coupling the PLL dynamics with the equivalent result with a higher value of k. Along these lines, a lower k is
impedance. This aggregated reduced-order model only slightly better for networks with a low capacitive effect. This
contains one frequency component, as visualized in Fig. 5 is because the power flow is highly dependent on the network
where Zeq , φeq , and Ieq represent the magnitude and capacitances, due to their provision of reactive power [62].
phase angle of the equivalent impedance and the total injected Attaining better results by fine-tuning k may be achieved using
current of the entire string. To that end, the model aggregation an optimization algorithm where the error between the initial
is developed around the reduced-order quasi-static large-signal frequency drop at the fault instant of the aggregated model and
models from Fig. 4. Two different aggregation methods are the center of frequency drop of the n converters is minimized.
examined, and a combination of them is employed in this Alternatively, the equivalent impedance may take into account
work. These include a method preserving the total apparent the network capacitance, resulting in a more complicated
power loss of the wind farm string (Zeq,S ) [32], [58], [59], equivalent impedance representation and, hence, a higher-order
and a method preserving the average voltage drop along the aggregated model, or by adjusting the equivalent impedance
string (Zeq,∆V ) [60, p. 177]. to match the power flows before and after aggregation, as it is
done in [32]. Neither of these methods will be analyzed in this
1) Preservation of Apparent Power: The total power loss work since a highly simplified structure is desired and, as will
in the wind farm string is be shown later, the selection of k around the value 0.75 has a
n
X small influence on the stability prediction capability, which is
Stot = IC2 (n − i + 1)2 Zc,i + IC2 nZtl = Ieq
2
Zeq,S . (26) the main focus here.
i=1

Since Ieq = nIC is the total current of the collector system, IV. M ODEL VALIDATION
the equivalent impedance can be found as
The developed static and dynamic models are verified
n
1 1 X against a detailed full-order simulation study performed in
Zeq,S = Ztl + 2 (n − i + 1)2 Zc,i . (27)
n n i=1 MATLAB’s Simulink with PLECS blockset. All paralleled
converters under test are operated with the grid-following
2) Preservation of Average Voltage Drop: For this method, control structure shown in Fig. 6 with an averaged represen-
the averaged voltage drop on the impedances in the wind farm tation of the converter switching actions, meaning that the
string can be expressed as converter terminal reference voltage is directly applied to three
dependent voltage sources in the simulation [63]. The full
n
!
1 X
∆Vavg = IC (n − i + 1)Zc,i + IC nZtl = Ieq Zeq,∆V . representation of the converter in Fig. 6 is in the following
n i=1
analysis referred to as the full-order converter model. During
(28)
severe faults, where the instability phenomenon is character-
Again by assuming equal current injections from all convert-
ized by the fundamental frequency component, there is no
ers, the equivalent impedance can be isolated as
loss of accuracy when employing an average model [5]. The
n
1 1X network and controller parameters are as listed in Table I. The
Zeq,∆V = Ztl + (n − i + 1)Zc,i , (29) PLL parameters are designed for a damping ratio of 0.707 and
n n i=1
a bandwidth of 20 Hz. Instead of specifying values for Zth and
3) Weighted-Sum Equivalencing: Since the loss of syn- Zf e , the grid fault is simulated using a voltage source with a
chronization fundamentally is a power transfer issue, the controlled amplitude at the bus to the right of ZL . This implies
total apparent power of the entire wind farm string and the that Vth Kg = VF .
equivalent representation should be preserved. However, as The model validation for the system with a common POS
described in [32], [61], this method may not be accurate, and POC, and different POS but a common POC is shown
and it is observed from simulation studies in this work that in Fig. 7, representing the system cases from Section II-A
the method may underestimate the impedance whereas the and Section II-B, respectively. Based on the nominal current
method on averaged voltage drop is observed to overestimate injection and network parameters from Table I, the critical
the equivalent impedance. Therefore, to avoid using a more fault voltage magnitude from (8) and (12) is VF = 0.123 pu.
complicated modeling framework with shunt impedances con- Hence, if the fault voltage magnitude drops below this value,
sidered, a weighted sum of the two presented equivalencing instability will occur according to the static stability condition.
methods is performed to get a better estimate of the equivalent As seen in Fig. 7(a)-(b), the system clearly becomes unstable
impedance. This is when the static stability limit is not fulfilled. To that end, the
reduced-order large-signal model depicted by dashed red is
Zeq = kZeq,S + (1 − k)Zeq,∆V k ∈ [0, 1]. (30)
capable of capturing the instability and dynamical response.
Based on numerous simulations of the under/over-estimation Fig. 7(c)-(d) show the results for three paralleled converters
using either method, it is found that k = 0.75 is a good with a different POS but common POC, the configuration
compromise between the two. The reason for performing a in Fig. 2, Section II-B. Despite the static stability limit not

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
9

Fig. 6. Grid-following voltage-oriented control structure of each individual

converter connected to the external network used for the full-order simulation
model comparison.
Fig. 8. Aggregated model accuracy shown with the frequency response
of three paralleled full-order string converter models during a stable fault
TABLE I response with VF = 0.08 pu. f1 − f3 denote the estimated frequency of
C ONTROL AND NETWORK PARAMETERS each of the three paralleled full-order converter models where converter 3 is
located at the far end of the string.
Symbol Description Value
Vb Nominal grid voltage (line-line, rms) 400 V
IC Converter phase current magnitude 15 A being different from the previous case, due to the assumption
f0 Rated frequency 50 Hz of perfect tracking in the inner current controller, the added
Lcf Converter-side inductor 0.07 pu
Lgf Grid-side inductor 0.04 pu
transformer leakage inductance adds a destabilizing effect
Cf Filter capacitor 0.07 pu during the transient for the full-order representation. This
Rd Filter damping resistance 0.02 pu destabilizing effect is not captured by the large-signal reduced-
ZL Line impedance 0.04 + 0.1j pu order model in Fig. 4(a) since it is assumed that θI can be
Ztl Transformer leakage reactance 0.05j pu controlled to −π/2, instantaneously, effectively canceling the
Zc Collector impedance 0.0091 + 0.009j pu
Kpc Current controller Kp 20 Ω influence of the leakage inductance. This is a fundamental
Krc Current controller Kr 2000 Ω/s limitation of the reduced-order model, which implies that
Kp∗ Initial PLL Design of Kp 0.3542 [rad/(V s)] some conservatism must be included when using the reduced-
Ki∗ Initial PLL Design of Ki 0.4509 [rad/(V s2 )] order dynamical model for systems with different POS but
common POC. This limitation occurs in cases where the stable
operating points of the converters are very close to violating
the stability border in (12). Under such a scenario, the non-
ideal dynamics of the inner current controller start to have
a decisive effect on the transient stability outcome, which
cannot be predicted using the reduced-order model. However,
it should be mentioned, that when the converter operating point
is not very close to the stability boundary, the bandwidth of
the inner current controller does not have a significant effect
on the stability assessment [64]. Since the static stability limit
is not violated in Fig. 7(c), one is able to stabilize the system
by increasing the damping ratio of the PLL dynamics [13],
[19], as shown in Fig. 7(d).
For the verification of the configuration from Section II-C
and the proposed aggregated model, three paralleled convert-
ers as shown in Fig. 3 are considered where the collector
impedance Zc is as listed in Table I, and the external line
Fig. 7. Frequency responses of the full-order simulation model and reduced- impedance is set to ZL = 0.008 + 0.066j Ω. Both impedances
order large-signal models for different cases and system configurations. The contain shunt capacitance to analyze the findings from Sec-
static stability limit or critical VF is 0.123 pu for all cases. Three paralleled tion II-D. For the simulated results in this section, these are
converters, as shown in Fig. 1 during a grid fault with VF = 0.13 pu and
VF = 0.12 pu are shown in (a) and (b), respectively. (c): Three paralleled 4.3 µF and 336 µF for the collector impedance and external
converters, as shown in Fig. 2 during a grid fault with VF = 0.13 pu. (d): line impedance, respectively. Based on the static stability limit
Same as (c) but with an increased damping ratio of the PLL dynamics (Kp = in (18), the critical fault voltage magnitude considering that
1.2Kp∗ ).
each converter injects 1 pu reactive current is VF = 0.078 pu.

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
10

TABLE II
C ABLE DATA USED FOR STUDYING THE A NHOLT WPP.

Cable Type R [mΩ/km] L [mH/km] C [µF/km]

Collector: 150 mm2 124 0.39 0.19
Collector: 240 mm2 75.4 0.36 0.23
Collector: 500 mm2 36.6 0.32 0.32
Submarine cable 33.75 0.17 0.4
Land cable 33.75 0.56 0.17

stability under severe symmetrical grid faults. To that end,

the accuracy of the proposed simplified aggregated model to
evaluate the transient synchronization stability of a wind farm
string is analyzed. The filter and control parameters used for
the full-order model are as listed in Table I.
The distance between the turbines is approximately 600
meters at the edge and 900-1300 meters inside the farm
Fig. 9. Aggregated model accuracy shown with frequency responses of three [66]. These lengths, including the cable data from [50]
paralleled full-order string converter models during an unstable fault response
with VF = 0.07 pu. f1 − f3 denote the estimated frequency of each of the (2XS(FL)2YRAA 18/30(36) kV), are used to calculate the
three paralleled full-order converter models where converter 3 is located at impedance values for the cable Π-model of the collector
the far end of the string. system in Fig. 10. The data used for the export submarine and
land cable can be found in [67], Table 49, and [68], Table 28,
respectively. Both use aluminum 800 mm2 conductors. A
Therefore, to verify this limit in addition to the reduced-
25 km three-core submarine cable is used offshore, whereas
order aggregated model, tests where VF = 0.07 pu and
three 59.6 km single-core cables laid in flat formation are
VF = 0.08 pu are conducted. The results for these two cases
used on land [49]. The impedance values for the cables are
are shown in Fig. 8 and Fig. 9. At first, it is clear that the static
summarized in Table II. Both transformers are modeled as a
stability limit can be accurately used as a necessary stability
pure reactance with 0.05 pu, whereas all shunt reactors are
condition. This is true despite that the collector impedance and
considered to absorb 120 MVAr. All impedance values are
line impedance contains shunt capacitors, which verifies the
scaled depending on how many wind turbines are considered in
assumption of neglecting the effect of shunt capacitances.
the simulation study. I.e., the impedance of the export system
Furthermore, it is seen that the aggregated model for the
will be scaled much larger when studying only one string
collector system provides an accurate averaged behavior of the
compared to if all strings are operated at the same time. This
three paralleled converters with only a small deviation in the
is done to make sure that the voltage sensitivity at the PCC
frequency response. For the aggregated model, the weighting
caused by each converter remains the same between different
factor k is slightly varied to show the impact on the response.
test cases.
For the stable case in Fig. 8, the difference is unnoticeable.
For the unstable case in Fig. 9, it is seen that the accuracy of
A. Model Validation
the modeled frequency can be improved by selecting k = 0.7
instead of k = 0.75. Even though the estimated frequency For the wind farm string under study as highlighted in
can be improved by selecting a different k, it is observed that Fig. 10, the equivalent collector system impedance is calcu-
the results obtained using k = 0.75 gives an excellent result lated and used for the aggregated model. The critical fault
in terms of stability assessment and transient response in the voltage magnitude considering this system is VF = 0.0788 pu.
first and second swing of the response. A stable and unstable response of the estimated frequencies
of the wind farm string is shown in Fig. 11(a) and Fig. 11(b),
respectively. A zoomed view of the initial transient behavior
V. C ASE S TUDY ON A NHOLT W IND FARM is shown from where all the estimated frequency of all nine
To get a realistic view of the static stability limitation string converters can be seen, and it is observed that the
from Section II-C and the proposed aggregated model, the aggregated model accurately represents the averaged frequency
detailed simulation model and these are compared considering component of the wind farm string.
the Anholt 400 MW offshore wind power plant (WPP) [66]. As previously described, the time-domain transient response
Anholt WPP is located approximately 20 km out of the eastern of the aggregated model can be improved by fine-tuning the
coast in Denmark and consists of 111 3.6 MW wind turbines weighting factor k. However, the initial value of k = 0.75
[65]. The configuration of the wind turbines and the medium- again provides an accurate response capable of assessing
voltage collector cable system can be seen in Fig. 10. Here it the transient stability as desired. Since the results in Fig. 11
is evident that the assumption of a fixed collector impedance contains a higher number of string converters and, hence, a
between all converters in a string is not realistic. This section higher network capacitance, than for the case presented in
aims to analyze a string of the Anholt WPP and the influence Fig. 8 and Fig. 9, a slightly higher value of k provides a better
of different collector cable impedances on the synchronization result due to the increased network capacitance.

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
11

Fig. 10. The physical layout of Anholt 400 MW offshore wind power plant with the electrical export system and connections [49], [65]. One wind farm
string with 9 wind turbines is under study for this work, as highlighted, where the string converter numbers are denoted.

With this, it is seen that the static stability limit, despite TABLE III
its simplicity, is highly accurate in determining the necessary C OMPARISON OF COMPUTATIONAL TIME FOR SOLVING 1 S OF A NHOLT
W IND FARM S TRING USING DIFFERENT MODELS .
condition for transient stability. To that end, this means that
a system planner or designer can quickly assess whether a Switching Model Average Model (full-order) Aggregated Model
given system operation during a severe fault will be physically
642 s 34.8 s 306 ms
possible or not, without having to run a single simulation
of the high-order system. In addition to the frequency plots
in Fig. 12, the voltages and currents at the PCC and all
converter connections points are shown for the same stable needed to compute one-second fault response of the Anholt
and unstable case in Fig. 12. The low-frequency instability WPP string using a full-order switching model, a full-order
oscillations are propagating the network voltage and currents. average model, and the proposed simplified model is shown in
However, since all converters have tightly regulated currents, Table III. The result for the switching model is estimated based
only a small impact is observed here. It should be noted that on the time needed to compute a 40 ms response, whereas
in Fig. 12, the currents deviate a bit from the 1 pu reference the time shown for the averaged and simplified models is an
amplitude. This is because the currents displayed are the average of 100 different simulation runs. As evident, besides
injected currents after the LC-filter and not the controlled correctly predicting the stability outcome, the computational
converter-side current. Hence, the discrepancies between 1 pu requirements can be significantly reduced. It is approximately
and the injected currents originates from the discharge and 100 times faster than the averaged model and more than
charging of the filter capacitors during the fault. 2,000 times faster than the full-order switching model. For
the comparison performed here, the switching frequency was
set to 2 kHz. Accordingly, when the number of desired case
B. Computational Enhancement studies increases, such as for identifying stability boundaries
Using the proposed aggregated model, the 9 full-order paral- and parameter trends, and when the number of paralleled and
leled wind turbine converters in the studied string including 11 interconnected converters increases, the benefit of using the
3rd -order cable models, transformers, and shunt reactors can simplified reduced-order aggregated model rapidly increases.
accurately be represented as one single 2nd -order nonlinear
equation where the equivalent impedance and external line
impedance are updated based on the estimated frequency. C. Evaluation of Assumptions
A comparison of computational requirements for the pro- The presented reduced-order aggregated model is based on
posed aggregated model and the full-order model is performed several assumptions including homogeneous VSC controller
by measuring the time needed to solve one second of the parameters and loading levels. Since these may affect the
fault response when a fault occurs. The computations are effectiveness and applicability of the model, these assumptions
performed on a Lenovo ThinkPad with 8 GB of RAM, a 2.80 are evaluated in the following.
GHz Intel Core i7-7600U processor, and a Windows 10 64-bit 1) Heterogeneous Loading Levels and Current Controller
operating system. The models are implemented in MATLABs Dynamics: First, the 9 converters in the string under study
Simulink version 2017a using PLECS blockset version 4.1.1. are exposed to different loading levels. The loading levels of
The simulation model is solved using the variable step size the converters are linearly distributed between 0.6-1 pu to take
auto-solver in Simulink with a maximum step size and a into account wake effects [69]. To that end, the bandwidth
relative tolerance of 1e-4 and 1e-3, respectively. The time of the inner current controller for each converter is different.

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
12

(a) (b)
Fig. 11. Estimated PLL frequencies during a severe fault condition for full-order models and the proposed aggregated model. (a): Stable operating point with
VF = 0.08 pu. (b): Unstable operating point with VF = 0.07 pu.

(a) Stable operating point with VF = 0.08 pu. (b) Unstable operating point with VF = 0.07 pu.
Fig. 12. The signal amplitudes of all nine POS voltages and the PCC voltage are displayed in the top figures. The amplitudes of all injected converter currents
and the current at the PCC are shown in the bottom figures. All variables are shown in per-unit values during a severe grid fault.

Here, the bandwidths of the current controllers are linearly this is shown in Fig. 14 for a PLL bandwidth of 60 Hz and
distributed between a +20 % and a −20 % in bandwidth a PLL bandwidth of 100 Hz. Despite, some discrepancies in
compared to the initial design. This heterogeneous system is the frequency estimation overshoot for the 100 Hz PLL case,
then compared with the proposed aggregated model under two the aggregated model still persists a good reproduction of the
severe fault conditions. One, where a stable operating point synchronization dynamics of the system.
exists during the fault, and one where the operating point is 3) Impact of Outer Loop Control: As mentioned in Sec-
unstable. The results of this are shown in Fig. 13. Despite tion II, it is assumed that the outer control loops of the
the different loading levels of the converter before the fault converter can be neglected during the fault. This is the case
and different dynamics of the inner current controllers, the since the converter current references are switched to comply
aggregated reduced-order model is well capable of reproducing with the LVRT requirements during a severe fault. Accord-
the synchronization behavior of the string. ingly, the outer direct voltage control (DVC) and alternating
2) Impact of PLL Bandwidth: Since the assumption of voltage control (AVC), controlling d-axis and q-axis current
neglecting the inner current dynamics in the aggregated model references through the dc-link voltage regulation and AC
is justifiable when the PLL dynamics are tuned slow, the voltage magnitude, respectively, have a small impact on the
performance of the aggregated model is here analyzed when synchronization stability during the fault. This assumption is
the PLL bandwidth is increased. At the same time, the different justified in the following. The outer loops of a grid-side wind
loading levels of the converters are considered. The results for turbine converter do usually comprise a DVC for the d-axis

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
13

(a) (b)
Fig. 13. Estimated PLL frequencies during a severe fault condition for full-order models and the proposed aggregated model for different converter parameters.
All string converters operate at different loading levels from 0.6 − 1 pu and different current controller bandwidths in the range +20 % to -20 %. k = 0.75.
(a): Stable operating point with VF = 0.1 pu. (b): Unstable operating point with VF = 0.05 pu.

Fig. 14. Frequency responses of the full-order simulation model and aggre-
gated model for different PLL bandwidths during a fault where VF = 0.1 pu.
k = 0.75. (a): PLL bandwidth of 60 Hz. (b): PLL bandwidth of 100 Hz.

control and an AVC for the q-axis control [70]. When a fault
Fig. 15. Current reference generation during normal operating conditions
occurs, the current references are switched to comply with and during fault conditions. The DVC control i∗d and the AVC control i∗q .
the grid code requirements, and a dc-side breaking chopper During fault conditions, reactive current support is prioritized from the LVRT
with PI control is used to stabilize the dc-link voltage during requirements and the dc-link chopper control is activated to protect the dc-
side from overvoltages. IW T represents the current generated from the wind
the fault and deal with the continuing turbine feed-in power. turbine generator-side converter.
The chopper resistor Rch is sized based on the input nominal
power [70]. A detailed view of the outer loop control and
the current reference generation during normal and faulted
conditions are shown in Fig. 15. The outer DVC and AVC with the activation of the dc-side chopper control, the dc-
are tuned as in [71] with a bandwidth of 30 Hz and 10 Hz, link voltage is restored to its nominal voltage. Also, as can
respectively. The aggregated model is compared with the full- be seen in Fig. 16(b), the ac-side low-frequency oscillations
order string converters with outer loop control as shown in are present in the dc-link voltages as well. Additionally, the
Fig. 15. This is conducted for two different fault conditions as q-axis current component has a non-zero value prior to the
shown in Fig. 16. As can be noticed, the dc-link voltages of the fault due to the inclusion of the AVC. Yet, with the outer
converters rapidly increase when the fault occurs due to the loops considered, the aggregated model well reproduces the
imbalance between dc-side and ac-side active power. However, synchronization dynamics of the faulted system. This is the

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
14

(a) (b)
Fig. 16. Comparison of aggregated model to full-order model with outer control loops (DVC: Direct Voltage Control and AVC: Alternating Voltage Control)
during severe faults. The subfigures contain (a): dq-referenced PCC current, (b): three-phase PCC voltages, (c): Full-order converter PLL frequencies and
estimated frequency of aggregated model, and (d): dc-link voltages of full-order converters with outer loop control. For the aggregated model k = 0.75.
Left-side figure (a): Stable operating point with VF = 0.1 pu. Right-side figure (b): Unstable operating point with VF = 0.05 pu.

case for when a stable operation point exists, as shown in can effectively be applied as a screening tool and a convenient
Fig. 16(a), and when the operating point during the fault is approach for early-stage fault analysis of a system design.
unstable, see Fig. 16(b).
R EFERENCES
VI. C ONCLUSION [1] B. Kroposki, B. Johnson, Y. Zhang, V. Gevorgian, P. Denholm, B. M.
Hodge, and B. Hannegan, “Achieving a 100 % renewable grid: Operating
In this article, the modeling, analysis, and transient stabil- electric power systems with extremely high levels of variable renewable
ity assessment of different paralleled multi-converter systems energy,” IEEE Power Energy Mag., vol. 15, no. 2, pp. 61–73, March
2017.
are addressed. From a modeling framework based on static [2] VDE, “VDE-AR-N 4110: Technical requirements for the connection and
conditions, the necessary conditions for transient stability are operation of customer installations to the medium-voltage network (TCC
derived. The q-axes voltage components adopted to derive the medium-voltage),” 2017.
[3] Ö. Göksu, R. Teodorescu, C. L. Bak, F. Iov, and P. C. Kjær, “Instability
necessary conditions are together with the dynamics of the of wind turbine converters during current injection to low voltage grid
synchronization unit used to develop reduced-order models faults and PLL frequency based stability solution,” IEEE Trans. Power
for different multi-converter systems. Considering a daisy- Syst., vol. 29, no. 4, pp. 1683–1691, July 2014.
[4] NationalGrid, UK, “Performance of phase-locked loop based convert-
chain collector system configuration where a non-negligible ers,” Tech. Rep., 2017.
impedance separates each paralleled converter, an aggregated [5] M. G. Taul, X. Wang, P. Davari, and F. Blaabjerg, “An overview
reduced-order model is proposed to represent the averaged of assessment methods for synchronization stability of grid-connected
converters under severe symmetrical grid faults,” IEEE Trans. Power
frequency response of all paralleled converters. Both the static Electron., vol. 34, no. 10, pp. 9655–9670, Oct 2019.
stability conditions and the aggregated models are verified [6] I. Erlich, F. Shewarega, S. Engelhardt, J. Kretschmann, J. Fortmann,
through numerous simulation studies verifying their high ac- and F. Koch, “Effect of wind turbine output current during faults on
grid voltage and the transient stability of wind parks,” in Proc. IEEE
curacy for large-signal synchronization stability assessment. PESGM, July 2009, pp. 1–8.
Then, the Anholt wind power plant is considered as a case [7] D. Dong, B. Wen, D. Boroyevich, P. Mattavelli, and Y. Xue, “Analysis of
study where the aggregated model is compared to the detailed phase-locked loop low-frequency stability in three-phase grid-connected
power converters considering impedance interactions,” IEEE Trans. Ind.
operation of a wind farm string. High model accuracy is Electron., vol. 62, no. 1, pp. 310–321, Jan 2015.
observed using the aggregated model, and the computational [8] X. He, H. Geng, and G. Yang, “Synchronization stability analysis of
burden required for solving the system is reduced with a factor grid-tied power converters under severe grid voltage sags,” in Proc. IEEE
PEAC, Nov 2018, pp. 1–6.
of 100 compared to the full-order system, which enables ana- [9] Q. Hu, L. Fu, F. Ma, and F. Ji, “Large signal synchronizing instability
lysis of larger-scale systems. Finally, the assumptions used to of PLL-based VSC connected to weak AC grid,” IEEE Trans. Power
derive the aggregated model is evaluated where a comparison Syst., vol. 34, no. 4, pp. 3220–3229, July 2019.
[10] W. Wang, G. M. Huang, P. Kansal, L. E. Anderson, R. J. O’Keefe,
has been made to a heterogeneous converter system operated D. Ramasubramanian, P. Mitra, and E. Farantatos, “Instability of PLL-
at different loading levels with different current controller synchronized converter-based generators in low short-circuit systems and
dynamics. Here, the impact of the PLL bandwidth and the the limitations of positive sequence modeling,” in Proc. IEEE NAPS,
Sep. 2018, pp. 1–6.
introduction of outer control loops have also been analyzed, [11] M. G. Taul, X. Wang, P. Davari, and F. Blaabjerg, “Grid synchronization
showing, still, great applicability of the proposed aggregated of wind turbines during severe symmetrical faults with phase jumps,” in
model to reproduce the synchronization dynamics of the Proc. IEEE ECCE, Sept 2018, pp. 38–45.
[12] Q. Hu, J. Hu, H. Yuan, H. Tang, and Y. Li, “Synchronizing stability of
system. Therefore, with low computational requirements and DFIG-based wind turbines attached to weak AC grid,” in Proc. IEEE
high accuracy, the presented analysis and proposed modeling ICEMS, Oct 2014, pp. 2618–2624.

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
15

[13] H. Wu and X. Wang, “Design-oriented transient stability analysis of pll- system stability studies,” IEEE Trans. Power Syst., vol. 33, no. 6, pp.
synchronized voltage-source converters,” IEEE Trans. Power Electron., 6332–6342, 2018.
vol. 35, no. 4, pp. 3573–3589, April 2020. [36] R. Fang, R. Shang, M. Wu, C. Peng, and X. Guo, “Application of
[14] H. Geng, L. Liu, and R. Li, “Synchronization and reactive current gray relational analysis to k-means clustering for dynamic equivalent
support of PMSG-based wind farm during severe grid fault,” IEEE Trans. modeling of wind farm,” International Journal of Hydrogen Energy,
Sust. Energy, vol. 9, no. 4, pp. 1596–1604, Oct 2018. vol. 42, no. 31, pp. 20 154–20 163, 2017.
[15] S. Ma, H. Geng, L. Liu, G. Yang, and B. C. Pal, “Grid-synchronization [37] W. Teng, X. Wang, Y. Meng, and W. Shi, “An improved support vector
stability improvement of large scale wind farm during severe grid fault,” clustering approach to dynamic aggregation of large wind farms,” CSEE
IEEE Trans. Power Syst., vol. 33, no. 1, pp. 216–226, Jan 2018. Journal of Power and Energy Systems, vol. 5, no. 2, pp. 215–223, 2019.
[16] X. He, H. Geng, R. Li, and B. C. Pal, “Transient stability analysis [38] W. Li, P. Chao, X. Liang, J. Ma, D. Xu, and X. Jin, “A practical
and enhancement of renewable energy conversion system during LVRT,” equivalent method for dfig wind farms,” IEEE Trans. Sust. Energy,
IEEE Trans. Sust. Energy, pp. 1–1, 2019. vol. 9, no. 2, pp. 610–620, 2018.
[17] H. Wu and X. Wang, “An adaptive phase-locked loop for the transient [39] X. Zha, S. Liao, M. Huang, Z. Yang, and J. Sun, “Dynamic aggrega-
stability enhancement of grid-connected voltage source converters,” in tion modeling of grid-connected inverters using hamilton’s-action-based
Proc. IEEE ECCE, Sep. 2018, pp. 5892–5898. coherent equivalence,” IEEE Trans. Ind. Electron., vol. 66, no. 8, pp.
[18] M. G. Taul, X. Wang, P. Davari, and F. Blaabjerg, “Robust fault ride 6437–6448, 2019.
through of converter-based generation during severe faults with phase [40] H. R. Ali, L. P. Kunjumuhammed, B. C. Pal, A. G. Adamczyk, and
jumps,” IEEE Trans. Ind. Appl., vol. 56, no. 1, pp. 570–583, Jan 2020. K. Vershinin, “Model order reduction of wind farms: Linear approach,”
[19] ——, “Systematic approach for transient stability evaluation of grid-tied IEEE Trans. Sust. Energy, vol. 10, no. 3, pp. 1194–1205, 2019.
converters during power system faults,” in Proc. IEEE ECCE, Sep. 2019, [41] S. Vijayshankar, V. Purba, P. J. Seiler, and S. V. Dhople, “Reduced-order
pp. 5191–5198. aggregate dynamical model for wind farms,” in 2019 American Control
[20] H. Wu and X. Wang, “Transient stability impact of the phase-locked loop Conference (ACC), 2019, pp. 5464–5471.
on grid-connected voltage source converters,” in IEEE Proc. ECCE Asia, [42] J. Dai, Y. Tang, and Y. Wang, “Aggregation frequency response modeling
May 2018, pp. 2673–2680. for wind farms with frequency support capabilities,” in 2019 IEEE Power
[21] H. Wu and X. Wang, “Design-oriented transient stability analysis of Energy Society General Meeting (PESGM), 2019, pp. 1–5.
grid-connected converters with power synchronization control,” IEEE [43] J. Bi, W. Du, and H. F. Wang, “An aggregation method of wind farms
Trans. Ind. Electron., vol. 66, no. 8, pp. 6473–6482, Aug 2019. model for studying power system low frequency power oscillation,” in
[22] M. G. Taul, X. Wang, P. Davari, and F. Blaabjerg, “An efficient reduced- 2017 2nd International Conference on Power and Renewable Energy
order model for studying synchronization stability of grid-following (ICPRE), 2017, pp. 422–427.
converters during grid faults,” in 2019 20th Workshop on Control and [44] R. M. G. Castro and J. M. Ferreira de Jesus, “A wind park reduced-order
Modeling for Power Electronics (COMPEL), June 2019, pp. 1–7. model using singular perturbations theory,” IEEE Trans. Energy Conv.,
[23] S. Shah and P. Sensarma, “Auto-synchronization of lc filter based front- vol. 11, no. 4, pp. 735–741, 1996.
end converter with parallel inverters based weak distorted island grid [45] Y. Gu, N. Bottrell, and T. C. Green, “Reduced-order models for
using voltage injection,” in IECON 2012 - 38th Annual Conference on representing converters in power system studies,” IEEE Trans. Power
IEEE Industrial Electronics Society, Oct 2012, pp. 3388–3393. Electron., vol. 33, no. 4, pp. 3644–3654, April 2018.
[24] L. Huan, H. Xin, W. Dong, and F. Dörfler, “Impacts of grid structure [46] M. G. Taul, X. Wang, P. Davari, and F. Blaabjerg, “Current reference
on pll-synchronization stability of converter-integrated power systems,” generation based on next generation grid code requirements of grid-
arXiv:1903.05489v2, Nov 2019. tied converters during asymmetrical faults,” IEEE J. Emerg. Sel. Topics
[25] B. Wen, D. Dong, D. Boroyevich, R. Burgos, P. Mattavelli, and Z. Shen, Power Electron., pp. 1–1, 2019.
“Impedance-based analysis of grid-synchronization stability for three- [47] B. Xie, L. Zhou, C. Zheng, and Q. Zhang, “Stability and resonance
phase paralleled converters,” IEEE Trans. Power Electron., vol. 31, no. 1, analysis and improved design of n-paralleled grid-connected pv inverters
pp. 26–38, Jan 2016. coupled due to grid impedance,” in IEEE Proc. APEC, March 2018, pp.
[26] R. Rosso, M. Andresen, S. Engelken, and M. Liserre, “Analysis of 362–367.
the interaction among power converters through their synchronization [48] J. L. Agorreta, M. Borrega, J. López, and L. Marroyo, “Modeling and
mechanism,” IEEE Trans. Power Electron., vol. 34, no. 12, pp. 12 321– control of n-paralleled grid-connected inverters with lcl filter coupled
12 332, Dec 2019. due to grid impedance in pv plants,” IEEE Trans. Power Electron.,
[27] D. Dong, B. Wen, P. Mattavelli, D. Boroyevich, and Y. Xue, “Grid- vol. 26, no. 3, pp. 770–785, March 2011.
synchronization modeling and its stability analysis for multi-paralleled [49] C. F. Jensen, “Harmonic background amplification in long asymmetrical
three-phase inverter systems,” in IEEE Proc. APEC, March 2013, pp. high voltage cable systems,” Electric Power Systems Research, vol. 160,
439–446. pp. 292 – 299, 2018.
[28] J. Zhao, M. Huang, and X. Zha, “Transient stability analysis of grid- [50] Nexans. (2008) Submarine power cables. [Online]. Available: https:
connected vsis via pll interaction,” in 2018 IEEE International Power //www.nexans.de/Germany/2010/NEX_SubmPowCables_mai08_1.pdf
Electronics and Application Conference and Exposition (PEAC), Nov [51] F. Milano, F. Dörfler, G. Hug, D. J. Hill, and G. Verbič, “Foundations
2018, pp. 1–6. and challenges of low-inertia systems (invited paper),” in 2018 Power
[29] A. M. S. Al-bayati, F. Mancilla-David, and J. L. Domínguez-Garcíal, Systems Computation Conference (PSCC), 2018, pp. 1–25.
“Aggregated models of wind farms: Current methods and future trends,” [52] S. T. Y. Lee and F. C. Schweppe, “Distance measures and coherency
in 2016 North American Power Symposium (NAPS), 2016, pp. 1–6. recognition for transient stability equivalents,” IEEE Trans. Power
[30] W. Li, A. M. Gole, M. K. Das, and I. Kaffashan, “Research on wind Apparatus and Systems, vol. PAS-92, no. 5, pp. 1550–1557, 1973.
farms aggregation method for electromagnetic simulation based on [53] Z. Shuai, Y. Peng, X. Liu, Z. Li, J. M. Guerrero, and Z. J. Shen,
fdne,” in 2019 IEEE PES Innovative Smart Grid Technologies Europe “Dynamic equivalent modeling for multi-microgrid based on structure
(ISGT-Europe), 2019, pp. 1–5. preservation method,” IEEE Trans. Smart Grid, vol. 10, no. 4, pp. 3929–
[31] M. Altin, . Göksu, A. D. Hansen, and P. E. Sørensen, “Aggregated wind 3942, 2019.
power plant models consisting of iec wind turbine models,” in 2015 [54] A. M. Khalil and R. Iravani, “Power system coherency identification
IEEE Eindhoven PowerTech, 2015, pp. 1–5. under high depth of penetration of wind power,” IEEE Trans. Power
[32] J. Martínez-Turégano, S. Añó-Villalba, G. Chaques-Herraiz, S. Bernal- Syst., vol. 33, no. 5, pp. 5401–5409, 2018.
Perez, and R. Blasco-Gimenez, “Model aggregation of large wind farms [55] S. Sastry and P. Varaiya, “Coherency for interconnected power systems,”
for dynamic studies,” in IECON 2017 - 43rd Annual Conference of the IEEE Trans. Automatic Control, vol. 26, no. 1, pp. 218–226, 1981.
IEEE Industrial Electronics Society, Oct 2017, pp. 316–321. [56] M. L. Ourari, L. Dessaint, and V. Do, “Dynamic equivalent modeling
[33] A. P. Gupta, A. Mohapatra, and S. N. Singh, “Apparent power loss based of large power systems using structure preservation technique,” IEEE
equivalent model of wind farm collector system,” in 2018 20th National Trans. Power Syst., vol. 21, no. 3, pp. 1284–1295, 2006.
Power Systems Conference (NPSC), 2018, pp. 1–6. [57] S. Zhao, N. C. Nair, and N. Vong, “Coherency-based equivalencing
[34] W. Du, W. Dong, H. Wang, and J. Cao, “Dynamic aggregation of same method for large wind farms,” in 2009 IEEE Power & Energy Society
wind turbine generators in parallel connection for studying oscillation General Meeting, 2009, pp. 1–8.
stability of a wind farm,” IEEE Trans. Power Syst., vol. 34, no. 6, pp. [58] X. Su, Y. Liu, H. Song, and D. Xu, “Comparison between the two equiv-
4694–4705, 2019. alent methods of collector system for wind farms,” in 2015 International
[35] P. Wang, Z. Zhang, Q. Huang, N. Wang, X. Zhang, and W. Lee, Conference on Estimation, Detection and Information Fusion (ICEDIF),
“Improved wind farm aggregated modeling method for large-scale power Jan 2015, pp. 354–358.

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
16

[59] E. Muljadi, C. P. Butterfield, A. Ellis, J. Mechenbier, J. Hochheimer, Xiongfei Wang (S’10-M’13-SM’17) received the
R. Young, N. Miller, R. Delmerico, R. Zavadil, and J. C. Smith, B.S. degree from Yanshan University, Qinhuangdao,
“Equivalencing the collector system of a large wind power plant,” in China, in 2006, the M.S. degree from Harbin In-
2006 IEEE Power Engineering Society General Meeting, 2006, pp. 9 stitute of Technology, Harbin, China, in 2008, both
pp.–. in electrical engineering, and the Ph.D. degree in
[60] L. H. Kocewiak, “Harmonics in large offshore wind farms,” Ph.D. energy technology from Aalborg University, Aal-
dissertation, Faculty of Engineering and Science at Aalborg University, borg, Denmark, in 2013. Since 2009, he has been
2012. with the Department of Energy Technology, Aalborg
[61] A. P. Gupta, A. Mohapatra, and S. N. Singh, “Apparent power loss based University, where he became an Assistant Professor
equivalent model of wind farm collector system,” in 2018 20th National in 2014, an Associate Professor in 2016, a Professor
Power Systems Conference (NPSC), Dec 2018, pp. 1–6. and Research Program Leader for Electronic Power
[62] J. Ruan, Z. Lu, Y. Qiao, and Y. Min, “Analysis on applicability problems Grid (eGrid) in 2018, and the Director of Aalborg University-Huawei Energy
of the aggregation-based representation of wind farms considering dfigs’ Innovation Center in 2020. He is also a Visiting Professor of power electronics
lvrt behaviors,” IEEE Trans. Power Syst., vol. 31, no. 6, pp. 4953–4965, systems with KTH Royal Institute of Technology, Stockholm, Sweden. His
2016. current research interests include modeling and control of grid-interactive
[63] S. Bacha, I. Munteanu, and A. I. Bratcu, Power Electronic Converters power converters, stability and power quality of power-electronic-based power
Modeling and Control with Case Studies, 1st ed. Springer, 2014, ISBN: systems, active and passive filters. Dr. Wang was selected into Aalborg
978-1-4471-5477-8. University Strategic Talent Management Program in 2016. He has received
[64] X. He, H. Geng, and S. Ma, “Transient stability analysis of grid-tied six Prize Paper Awards at the IEEE Transactions and conferences, the
converters considering PLL’s nonlinearity,” CPSS Transactions on Power 2016 Outstanding Reviewer Award of IEEE T RANSACTIONS ON P OWER
Electronics and Applications, vol. 4, no. 1, pp. 40–49, March 2019. E LECTRONICS, the 2018 IEEE PELS Richard M. Bass Outstanding Young
[65] L. H. Kocewiak, B. L. . Kramer, O. Holmstrøm, K. H. Jensen, and Power Electronics Engineer Award, the 2019 IEEE PELS Sustainable Energy
L. Shuai, “Resonance damping in array cable systems by wind turbine Systems Technical Achievement Award, the 2019 Highly Cited Researcher
active filtering in large offshore wind power plants,” IET Renewable by Clarivate Analytics (former Thomson Reuters), and the 2020 IEEE PES
Power Generation, vol. 11, no. 7, pp. 1069–1077, 2017. Prize Paper Award. He serves as a Member at Large for Administrative
[66] Ørsted, DK. (2019) Anholt offshore wind farm. [Online]. Available: Committee of IEEE Power Electronics Society in 2020-2022, and as an
http://dise.org.pl/dania2019/AnholtOffshoreWindFarm.pdf Associate Editor for the IEEE T RANSACTIONS ON P OWER E LECTRONICS,
[67] ABB, “Xlpe submarine cable systems attachment to xlpe land cable the IEEE T RANSACTIONS ON I NDUSTRY A PPLICATIONS, and the IEEE
systems - user’s guide,” Tech. Rep. J OURNAL OF E MERGING AND S ELECTED T OPICS IN P OWER E LECTRONICS.
[68] ——, “Xlpe land cable systems - user’s guide,” Tech. Rep.
[69] S. Kuenzel, L. P. Kunjumuhammed, B. C. Pal, and I. Erlich, “Impact Pooya Davari (S’11–M’13-SM’19) received the
of wakes on wind farm inertial response,” IEEE Trans. Sust. Energy, B.Sc. and M.Sc. degrees in electronic engineer-
vol. 5, no. 1, pp. 237–245, 2014. ing in 2004 and 2008, respectively, and the Ph.D.
[70] M. H. Qais, H. M. Hasanien, and S. Alghuwainem, “Low voltage ride- degree in power electronics from QUT, Australia,
through capability enhancement of grid-connected permanent magnet in 2013. From 2005 to 2010, he was involved in
synchronous generator driven directly by variable speed wind turbine: a several electronics and power electronics projects as
review,” The Journal of Engineering, vol. 2017, no. 13, pp. 1750–1754, a Development Engineer. From 2013 to 2014, he
2017. was with QUT, as a Lecturer. He joined Aalborg
[71] L. Harnefors, M. Bongiorno, and S. Lundberg, “Input-admittance cal- University, in 2014, as a Postdoc, where he is cur-
culation and shaping for controlled voltage-source converters,” IEEE rently an Associate Professor. He has been focusing
Trans. Ind. Electron., vol. 54, no. 6, pp. 3323–3334, Dec 2007. on EMI, power quality and harmonic mitigation
analysis and control in power electronic systems. He has published more than
140 technical papers. Dr. Davari served as a Guest Associate Editor of IET
journal of Power Electronics, IEEE Access Journal, Journal of Electronics
and Journal of Applied Sciences. He is an Associate Editor of Journal
of Power Electronics, Associate Editor of IET Electronics, Editorial board
member of EPE journal and Journal of Applied Sciences. He is member of
the International Scientific Committee (ISC) of EPE (ECCE Europe) and a
member of Joint Working Group six and Working Group eight at the IEC
standardization TC77A. Dr. Davari is the recipient of a research grant from
the Danish Council of Independent Research (DFF-FTP) in 2016, and 2020
IEEE EMC Society Young Professional Award for his contribution to EMI
and Harmonic Mitigation and Modeling in Power Electronic Applications.

Mads Graungaard Taul (S’17) received the B.Sc.

and M.Sc. degrees in electrical energy engineering
with a specialization in power electronics and drives
in 2016 and 2019, respectively, and the Ph.D. degree
in power electronic systems from Aalborg Univer-
sity, Denmark in 2020. In connection with his M.Sc.
degree, he received the 1st prize master’s thesis
award for excellent and innovative project work by
the Energy Sponsor Programme. Dr. Taul was a
visiting researcher at the University of California,
Berkeley, at the Department of Electrical Engineer-
ing and Computer Science from August 2019 to January 2020. Currently, he is
a Post-doctoral researcher at the Department of Energy Technology at Aalborg
University, Denmark. His main research interests include renewable energy
sources and grid-connected converters with a particular focus on modeling,
control, and stability analysis of power electronics-based power systems.

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2020.3015293, IEEE Journal
of Emerging and Selected Topics in Power Electronics
17

Frede Blaabjerg (S’86–M’88–SM’97–F’03) was

with ABB-Scandia, Randers, Denmark, from 1987
to 1988. From 1988 to 1992, he got the PhD degree
in Electrical Engineering at Aalborg University in
1995. He became an Assistant Professor in 1992, an
Associate Professor in 1996, and a Full Professor of
power electronics and drives in 1998. From 2017
he became a Villum Investigator. He is honoris
causa at University Politehnica Timisoara (UPT),
Romania and Tallinn Technical University (TTU)
in Estonia. His current research interests include
power electronics and its applications such as in wind turbines, PV systems,
reliability, harmonics and adjustable speed drives. He has published more than
600 journal papers in the fields of power electronics and its applications. He is
the co-author of four monographs and editor of ten books in power electronics
and its applications. He has received 32 IEEE Prize Paper Awards, the IEEE
PELS Distinguished Service Award in 2009, the EPE-PEMC Council Award
in 2010, the IEEE William E. Newell Power Electronics Award 2014, the
Villum Kann Rasmussen Research Award 2014, the Global Energy Prize in
2019 and the 2020 IEEE Edison Medal. He was the Editor-in-Chief of the
IEEE T RANSACTIONS ON P OWER E LECTRONICS from 2006 to 2012. He
has been Distinguished Lecturer for the IEEE Power Electronics Society from
2005 to 2007 and for the IEEE Industry Applications Society from 2010 to
2011 as well as 2017 to 2018. In 2019-2020 he serves a President of IEEE
Power Electronics Society. He is Vice-President of the Danish Academy of
Technical Sciences too. He is nominated in 2014-2019 by Thomson Reuters
to be between the most 250 cited researchers in Engineering in the world.

2168-6777 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Monash University. Downloaded on April 06,2021 at 01:58:23 UTC from IEEE Xplore. Restrictions apply.