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/TPWRD.2020.3018149, IEEE
Transactions on Power Delivery
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 1

Impedance Network Model Based Modal

Observability and Controllability Analysis
for Renewable Integrated Power Systems
Ying Zhan, Xiaorong Xie, Senior Member, IEEE, Yang Wang, Member, IEEE

 based devices, such as high-voltage direct current (HVDC)

Abstract—The complex interactions between renewables and systems and flexible AC transmission systems (FACTs). iii)
AC/DC networks have brought new oscillation issues. To effec- The system topology could be complex with meshed networks.
tively monitor and suppress these oscillations, the observability For such systems, the occurred oscillation issues usually in-
and controllability of the oscillatory modes at different locations volve a variety of components. The oscillation energy would
of the system must be investigated. To achieve this, the paper first expand to a large area. To effectively monitor and control the
defines the concept of modal observability and controllability
oscillations, it is desirable to know the distribution of the os-
using the responses of unit-pulse perturbations. Then an eigen-
value-decomposition based method is proposed to calculate these cillatory currents and the role of each component in forming the
two indexes with the impedance network model (INM) as the oscillation. Thus it is of great importance to investigate the
input. The obtained modal observability(controllability) has many observability and controllability of the oscillatory modes at
important applications and one of them is to facilitate the aggre- different locations.
gation of the INM. The paper investigates this application by In view of the above need, some researches have been con-
selecting an optimal aggregation port to ensure that the aggre- ducted on the modal observability and controllability. The work
gated impedance can reflect the complete system dynamics. Fi- in [12]-[14] uses the time-domain modal analysis method,
nally, the proposed method is verified on a simple passive circuit
and a practical wind power system that experienced actual sub- which is based on the traditional state-space model. The modal
synchronous resonance incidents. Theoretical results as well as observability(controllability) in(from) a certain input(output)
EMT simulations demonstrate the effectiveness of the method. can be quantified via the left and right eigenvectors of state
space matrix A. However, the matrix A of a practical system
Index Terms—Impedance network model, Observability, Con- could be non-normal with ill-conditioned eigenvalues, which
trollability, Aggregated impedance, Subsynchronous resonance may affect the accuracy of the calculated eigenvectors. This
issue becomes an obstacle of the time-domain modal analysis
method [14]. The work in [15] gives another solution using the
I. INTRODUCTION resonance mode analysis method. It builds the frequen-

T he recent surge in the number of renewables has dramati-

cally changed the dynamic behaviors of the modern power
system, leading to various oscillation issues. For instance, wind
cy-domain nodal admittance matrix. With the frequency of
system resonance available via frequency scan analysis, the
nodal participation factors (PFs), as a combined measure of
turbine generators may interact with the series compensation or observability and controllability can be computed. However,
weak grid to trigger subsynchronous oscillations [1]-[6], while the method focuses on the steady-state resonance phenomenon,
the interactions between different renewables would induce thus cannot be used for the stability analysis.
high-frequency oscillations [7]-[8]. These emerging oscilla- To address the above challenges, our previous work proposes
tions tend to lower power quality, endanger equipment safety the Laplace-domain modal analysis method to investigate the
and threaten system stability [9], thus has become a major modal observability and controllability [5]. The method is
concern for the accommodation of renewables. based on the impedance network model (INM). Compared with
A practical large-scale renewable-integrated power system the time-domain state-space model, the Laplace-domain INM is
usually has the following features: 1) There are many geo- much easier to build, thus more suitable for renewa-
graphically distributed renewables, of which the controller ble-integrated systems with volatile operation conditions. In [5],
configurations/parameters are different [10]. 2) Besides re- the system modes are obtained and the nodal(loop) PFs are
newables, the system is likely to equip other power-electronic introduced. However, the PFs cannot provide direct insights
into the modal observability and controllability. In addition, the
Manuscript received xxxx, xxxx; revised xxxx, xxxx. This work was sup- physical meaning of loop PFs are also not intuitive as a loop is
ported in part by National Natural Science Foundation of China (51737007, constructed by several branches.
51925701, 51907133). (Corresponding author: Xiaorong Xie). To tackle the above issues, this paper makes a further pro-
Ying Zhan, Xiaorong Xie are with the State Key Lab. of Power System,
Department of Electrical Engineering, Tsinghua University, Beijing, 100084, gress on the direction of the Laplace-domain modal analysis. In
China (e-mail: [email protected]). the paper, the modal observability and controllability at dif-
Yang Wang is with College of Electrical Engineering, Sichuan University, ferent nodes(branches) are rigorously defined using the re-
Chengdu, 610065, China (e-mail: [email protected]).
Digital Object Identifier: xxxxxxx sponses of unit-pulse perturbations applied to the target system.
An analytic method is then proposed to calculate the modal

0885-8977 (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: UNIVERSITY OF ROCHESTER. Downloaded on August 27,2020 at 01:09:46 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/TPWRD.2020.3018149, IEEE
Transactions on Power Delivery
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 2

observability and controllability using the eigenvectors of INM. Y ( s)V N ( s)  J N ( s) (1)

More importantly, the paper discusses one of its applications on
the aggregation of INM. For a large power system, the dimen- Z ( s) J L ( s)  V L ( s) (2)
sion of INM would be high. It is thus more attractive to ag-
gregate the INM into an aggregated impedance [3]. Depending B. Calculation of system modes
on the choice of the port from which an equivalence is made, Mathematically, NAM and LIM represent the system trans-
the aggregated impedance may or may not reflect dynamics of fer function matrices, thus their determinants are characteristic
all system modes. This work adopts the defined modal ob- polynomials of the system [5]. Then the system eigenvalues or
servability to address this issue by selecting an optimal aggre- modes can be obtained by solving the zeros of the determinant
gation port. [20], which is given by
The advantages of the proposed method are demonstrated det[Y ( s)]  0 or det[ Z ( s)]  0 (3)
through case studies on a simple passive circuit and a practical
wind power system that experienced actual subsynchronous where det [] is the determinant of the enclosed matrix.
resonances (SSRs). Time-domain state-space method and The characteristic polynomial of the large-scale renewa-
electromagnetic transient (EMT) simulations are carried out to ble-integrated power system is usually of high-order. QR de-
composition is recognized as one of the efficient methods for
check the consistency.
working out the zeros of high-order polynomials [21]. Gener-
The rest of the paper is organized as follows: Section Ⅱ re-
ally, QR decomposition can be applied to systems with an order
views the concept of INM. In Section Ⅲ, the modal observa-
of several hundred to several thousand. For higher-order sys-
bility and controllability matrix are defined. A practical method
tems, the frequency segmentation technique can be used [5].
to calculate these two matrices is also given. Section Ⅳ ex-
The obtained oscillatory modes are denoted as λm =σm ± jωm
plains one application of this work for the aggregation of an
(m=1, 2, …, M, where M is number of modes), where -σm and
INM. In Section V, case studies are conducted to validate the
ωm denote the damping and angular frequency of the mode,
usefulness of the proposed method. Brief conclusions are
respectively.
drawn in Section Ⅵ.
III. MODAL OBSERVABILITY AND CONTROLLABILITY
II. CONCEPT OF IMPEDANCE NETWORK MODEL
With the mode information available, it is possible to inves-
It is widely recognized that the renewables-related oscilla-
tigate the modal observability and controllability at different
tion issues are caused by interactions between converters and
nodes(branches). This section first defines the concept of modal
networks. These interactions generally belong to small-signal
observability and controllability. Then the procedure to calcu-
dynamics, thus can be analyzed through linearized models
late these two indexes is explained.
[16]-[18]. In our previous work [5], the Laplace-domain im-
pedance network model (INM) was built for investigating such A. Definitions of modal observability and controllability
oscillations, which is reviewed in this section. Supposing that a unit-pulse perturbation current(voltage) is
A. Establishment of impedance network model applied to a node(branch) of a system, the excited nodal volt-
ages(branch currents) can be given by
Each component of the system is built as an impedance
M
model (IM) and INM is constructed by connecting all indi-
vidual ones together according to the system topology. The
unk (t )   U nkm e mt
sin(m t   ) (4)
m 1
impedance modelling has been well illustrated in [16]-[18] and
M
 Ilkm e
will not be repeated here. As an input-output representation, IM
ilk (t )  mt
sin(mt   ) (5)
can be easily identified through experiments or field tests, thus m 1
is suitable for the renewables with the internal parameters and
m
configurations unknown. Since INM is a straightforward in- where U nk and I lkm are initial oscillatory magnitudes of mode
terconnection of IMs, it is easy to reconstruct when the oper- λm in the nodal voltage and branch current, respectively; the
ating condition or system topology changes. first and second letter of the subscript in U nk m
( I lkm ) represent the
The INM can be expressed as either nodal admittance matrix
observation and excitation node(branch), respectively; n=1, …,
(NAM), noted as Y(s) or loop impedance matrix (LIM), noted
N and l=1, …, L, where N and L are the total numbers of nodes
as Z(s) for the stability analysis. Then the small-signal dy-
and branches.
namics of a system can be represented by the nodal voltage
For an oscillatory mode λm, its modal observability in a cer-
equation and the loop current equation, as shown in (1) and (2) N L
[19]. In these two equations, VN(s) is nodal voltage vector; JN(s) tain node(branch), denoted as onm ( olm ), is defined as the rela-
is nodal current source vector indicating the currents injected tive magnitude of the oscillatory component in the nodal volt-
into the nodes; JL(s) is loop current vector; VL(s) is loop voltage age(branch current), i.e.
source vector indicating the voltage sources in the loops. It can m
U nk
be seen that the interactions between different system compo- N
onm  N
(6)
nents are determined by Y(s) or Z(s).
 m
U nk
n 1

0885-8977 (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: UNIVERSITY OF ROCHESTER. Downloaded on August 27,2020 at 01:09:46 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/TPWRD.2020.3018149, IEEE
Transactions on Power Delivery
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 3

I lkm That is to say, the observation node(branch) #k in (8) and (9)

L
olm  L
(7) can be any one in the network.
 Ilkm A diagram of the proposed modal controllability is depicted
l 1 in Fig. 2 for better illustration. Assuming a unit-pulse voltage is
It should be emphasized that the modal observability of all applied to the three branches in turn, the excited oscillatory
nodes(branches) keeps the same no matter which node(branch) currents of branch #1 under the three situations are shown in
the excitation is applied to. In other words, the selection of the Fig. 2(b) and the decomposed currents for each mode are shown
excitation node(branch) #k in (6) and (7) does not affect the in Fig. 2(c) and (d), respectively. Based on the magnitudes of
results. Please refer to Appendix.A for the proof. the excited oscillation, we can determine that branch #2 (z3) has
A diagram of the proposed modal observability is depicted in the largest controllability for mode 1 while branch #3 (z4) has
Fig. 1. The experiment system is a simple circuit containing the largest controllability for mode 2. Further, according to the
three branches as shown in Fig. 1, of which the impedance proof in Appendix.A, we know that the result will be same if
values are arbitrarily selected for demonstration. Assuming a the oscillatory current is measured at branch #2 or #3.
unit-pulse voltage u1 is applied to branch #1, the excited os-
cillatory currents of three branches are shown in Fig. 1(b) and
the decomposed currents for each mode are shown in Fig. 1(c)
and (d), respectively. Based on the magnitudes of the observed
oscillation, we can determine that branch #2 (z3) has the largest
observability for mode 1, while branch #3 (z4) has the largest
observability for mode 2. Further, according to the proof in
Appendix.A, we know that the result will be same if the dis-
turbance is applied to branch #2 or #3.

Fig. 2 The illustration of modal controllability: (a) A conceptual impedance

network model with two modes; (b) Current response in branch #1 under
unit-pulse voltage perturbations;(c) The controllability of mode 1 from three
branches; (d) The controllability of mode 2 from three branches.

Taking all oscillatory modes into account, the modal ob-

servability matrix with regard to nodes(branches), namely ON
(OL), is defined as
O N  {onm
N
}; O L  {olm
L
} (10)
In the same way, the modal controllability matrix CN(CL) is
Fig. 1 The illustration of modal observability: (a) A conceptual impedance
defined as
network model with two modes; (b) Current responses of a unit-pulse voltage
added to branch #1;(c) The observability of mode 1 in three branches; (d) The
C N  {cnm
N
}; C L  {clm
L
} (11)
observability of mode 2 in three branches.
By doing so, the modal observability and controllability at
different nodes(branches) can be fully quantified via the ob-
The excited oscillatory magnitudes by applying a perturba-
served and excited oscillatory magnitudes as (6)-(11).
tion to a node(branch) reflects the ability of the node(branch) to
control the mode. Thus the modal controllability from a B. Calculation of modal observability and controllability
node(branch), denoted as cnmN L
( clm ), is defined as based on NAM
For a given mode λm, the NAM of the network, i.e. Y(λm), is
m
U kn symmetric, which can be diagonalized as [15][22]:
N
cnm  N
(8)
Y ( λm )  LΛR
 m
U kn (12)
n 1
where Λ=diag (μ1, μ2, …, μN) is the eigenvalue matrix; L and R
I klm are left and right eigenvector matrices, respectively.
L
clm  L
(9) According to (1), when a unit-pulse current is injected into
 I klm node #k, the excited nodal voltages VnkN ( s) can be expressed as
l 1

Similarly, the modal controllability of all nodes(branches) VnkN ( s)  [Y -1 ( s )]nk (13)

does not change no matter which node(branch) the oscillation is
observed from. Please also check Appendix.A for the proof.

0885-8977 (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: UNIVERSITY OF ROCHESTER. Downloaded on August 27,2020 at 01:09:46 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/TPWRD.2020.3018149, IEEE
Transactions on Power Delivery
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 4

Then the excited oscillatory magnitudes of mode λm in the with

m
nodal voltage VnkN , or U nk can be obtained as[23] H' =BT H ;W ' =WB (22)
m
U nk =VnkN ( s)(s  m ) |s m  [Y ( s )nk (s  m )] |s m
-1
(14) where a2 is a non-zero constant; B is the loop-branch matrix
that determines the relationship between branch currents and
Based on (12), the elements of the inverse matrix of Y(λm), or loop currents.
Y-1(λm) can be approximately expressed as
Substituting (21) into (7) and (9), the observability and
[Y -1 ( λm )]nk  1-1 Ln1 R1k  ....   N-1 LnN RNk (15) controllability of mode λm at each branch can be computed as
Since the determinant of Y(λm) is zero, the matrix Λ contains H lq'
a zero eigenvalue denoted as μp. Here μp is regarded as a very L
olm  L
(23)
small number close to zero,
 H lq'
thus  p   j ( j  1, 2, ...., N , j  p) [24]. According to (15),
1 1
l 1

it can be obtained as Wql'

[Y ( λm )]nk 
-1
 -1
L
clm  (24)
p Lnp R pk (16) L
 Wql'
m
Taking into account (14) and (16), U nk can be obtained as l 1

L L
U nmk  a1 | Lnp R pk | (17) With olm
and clm
available, the modal observability and
controllability matrix, OL and CL, can be formed according to
where a1 is a non-zero constant. (10) and (11).Similar as ON and CN, OL is numerically equal to
Substituting (17) into (6) and (8), the observability and CL since LIM is a symmetric matrix. Such a result indicates that
controllability of mode λm at each node can be computed as the branch with better observation of the oscillation also has
better capability to control the oscillation.
Lnp
N
onm  N
(18)
IV. OPTIMAL AGGREGATION OF INM
 Lnp This section discusses one of the applications of modal ob-
n 1
servability to facilitate the aggregation of INM. The aggregated
R pn impedance has been proposed in [3][25] to describe the system
N
cnm  N
(19)
rather than the high-order INM for simplicity. However, the
 R pn aggregated impedance may not reflect the complete system
n 1
dynamics if the aggregation is not properly done. This section
N N
With onm and cnm available, the modal observability and addresses this issue using the proposed modal observability.
controllability matrix, ON and CN, can be formed according to The aggregated impedance(admittance) is mathematically
(10) and (11). Note NAM is symmetric, thus L and R are or- the input impedance(admittance) of a certain port. The port can
thogonal and L is the transpose of R. According to (18)-(19), be defined by two nodes with one as the reference node or an
the modal observability at a node equals to its modal control- open branch in the network. When a node(branch) constructs
lability. That is to say, ON and CN are numerically the same. the port, the obtained aggregated impedance(admittance), de-
This finding indicates that the monitoring and controlling of the noted as z nA ( ylA ), can be expressed by
oscillation can be practically achieved at the same bus.
VnN ( s )
C. Calculation of modal observability and controllability znA ( s)  (25)
based on LIM J nN ( s)
For a given mode λm, the LIM of the network, i.e. Z(λm), is J lL ( s)
also symmetric, which can be diagonalized as: ylA ( s)  (26)
Vl L ( s)
Z ( λm )  HΛW (20)
where J nN ( s ) is the current injected at node #n and VnN ( s) is the
where Λ is eigenvalue matrix; H and W are left and right ei-
genvector matrices, respectively. The zero eigenvalue of Z(λm) resulting nodal voltage; Vl L ( s ) is the voltage added in series at
is denoted as μq. branch #l and J lL ( s) is the resulting branch current.
Similarly, when a unit-pulse voltage is added to loop #k, the
According to (25) and (26), an aggregated imped-
excited loop currents can be obtained via (2). Following the
ance(admittance) is equal to the self-impedance(admittance) of
same procedure in the previous subsection, the excited oscil-
a node(branch), which can also be represented by the diagonal
latory magnitudes of mode λm in the branch current can be-
computed as element of inverse matrix of NAM(LIM). Thus z nA ( ylA ) is
practically calculated by the INM as
Ilkm  a2 | H lq' Wqk' | (21)

0885-8977 (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: UNIVERSITY OF ROCHESTER. Downloaded on August 27,2020 at 01:09:46 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/TPWRD.2020.3018149, IEEE
Transactions on Power Delivery
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 5

znA  [Y -1 ( s)]nn (27) demonstration.

ylA ( s)  [ BT Z -1 ( s) B]ll (28)

As seen from (27) and (28), different aggregated impedanc-
es(admittances) can be obtained depending on the selection of
the node(branch). The one used however is desired to contain
the information of all system modes [26]. According to
(25)-(26), this requires the nodal voltage(branch current) of the
node(branch) to contain the oscillatory components of all
modes. Since each row of ON(OL) represents the observability
of all modes in a certain node(branch), the node(branch) cor-
responding to the non-zero row of ON(OL) is the one to pick up. Fig. 3 A simple passive circuit
There usually exists several qualified nodes(branches). In
order to find the optimal aggregation port, the nodal(branch)
Firstly, the NAM and LIM of the circuit are established. By
aggregation index, denoted as AnN ( AlL ), is further proposed as solving zeros of their determinant, the oscillatory modes are
the weighted sum of observed magnitudes of different modes, obtained as
i.e. 1   j 1.5634; 2   j 4.1617; 3   j 0.4860 (31)
M
AnN   km onm
N
(29) Then according to (18)-(19), (23)-(24) and (10)-(11), the
m 1 modal observability and controllability matrices with regard to
M nodes(branches) are obtained, which are given by
AlL   km olmL (30) 32.3% 3.2% 33.7% 
m 1 39.5%
 27.8% 3.9% 
where km is the weight of the mth oscillatory mode, which can be N 88.4% 12.2% 26.0%  L 
O   ; O  14.1% 10.0% 31.2% 
set upon the significance of the mode. The modes with the 11.6% 87.8% 74.0%   
damping close to zero or negative would dominate the system  6.9% 34.5% 1.5% 
 7.2% 24.5% 29.7% 
dynamics following a disturbance. Thus they have greater 
weights than the others. The node(branch) with the largest (32)
aggregation index should be chosen as the input port as it can C O ;
N N
C O
L L
(33)
observe all modes with relatively large magnitudes. By doing
so, the optimal aggregation port is identified. As seen from (32)-(33), the modal observabil-
ity(controllability) of mode λ1 is high in node #1 and branches
V. CASE STUDIES #1, 2. Then it can be indicated that the mode λ1 is mainly caused
by the interactions between L1 (branch #1) and C1 (branch #2).
Case studies on two systems are presented in this section. This is why the frequency of mode λ1 closes to that of the par-
The first system is a simple passive circuit, of which the modal allel resonance constituted by L1 and C1
observability and controllability matrices are calculated. And
( 1  1 L1C1  1.414 ). For modes λ2 and λ3, they are more
based on the modal observability, the appropriate aggregation
port is selected to obtain the aggregated impedance. The per- observable in node #2. The modal observability(controllability)
formance of different aggregation ports is also compared to of mode λ2 is high in branches #2, 4, 5, while the mode λ3 is
illustrate the significance of the aggregation port on the ag- more observable in branches #1, 3, 5. Thus it can be inferred
that the mode λ2 is mainly induced by the interactions between
gregated-impedance based analysis. The results are validated
L3 (branch #5) and the series capacitor of C1, C2 (branch #4), C3
through time-domain state-space method.
(branch #5), while the mode λ3 is mainly determined by the
The second system is a practical wind power system that
interactions between C3 and the series inductors of L1, L2
experienced SSR incidents. With the modal observability and
(branch #3), L3. The finding is further confirmed as the fre-
controllability available, the distribution of the oscillatory
quency of mode λ2 and λ3 closes to that of the corresponding
currents and the origins of the SSR are figured out. Furthermore,
the aggregated impedance of the target system is obtained to resonance frequencies 2  1 L3 (C1 / /C2 / /C3 )  3.61 and
simplify the INM based analysis. The findings are 3  1 C3 ( L1  L2  L3 )  0.5 , respectively.
cross-checked by EMT simulations. Thus it demonstrates the
potential of the proposed method to analyze renewable inte- To demonstrate the accuracy of the method, the circuit is
grated power systems. also analyzed via the time-domain state-space method. The
results are consistent with the proposed method. Please refer to
A. A simple passive circuit Appendix.B for the details.
The passive circuit is shown in Fig. 3. It has 3 nodes (node 3 Finally, the aggregated impedance of above INM is obtained.
is selected as the reference one) and 5 branches. Note the pa- According to (32), ON and OL have no zero elements. This
rameters of each component are arbitrarily selected for means that all nodes or branches in the circuit can be used as the

0885-8977 (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: UNIVERSITY OF ROCHESTER. Downloaded on August 27,2020 at 01:09:46 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/TPWRD.2020.3018149, IEEE
Transactions on Power Delivery
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 6

aggregation port to obtain the aggregated impedance. Assum-

ing the weight of three modes are equal, i.e. [1/3 1/3 1/3], the
aggregation index of each node is computed as
A1N =42.18%, A2N =57.82% (34)
It can be seen node #2 has larger overall observability than
that of node #1, thus is selected as the input port to obtain the
aggregated impedance of the INM. The corresponding aggre-
gated impedance is shown in (35). By solving poles of (35), the
oscillatory modes of the system can be obtained, which are the
same as (31).
6s( s 2  1)(2s 2  5)
z2A ( s)  (35)
s6  20s 4  47 s 2  10
Then we consider another scenario in which L2 changes from
2H to 5H. The resonance frequency between L2 and C2 becomes Fig. 4 The INM of the target system
the same as the one of L1 and C1 (L1C1=L2C2). If the circuit As seen from Fig. 4, the system has 12 nodes (node 12 is
oscillates at this frequency, branches #3, 4 will behave as an selected as the reference one) and 19 branches. They are all
open circuit. As a result, the dynamics of mode λ1 cannot be numbered in the figure. Considering that each impedance con-
observed by node #2. This can be reflected via the zero element structs a branch, the branch number is noted in the brackets near
in the modal observability matrix as the impedance. ZS1 and ZS2 represent Inner Mongolia power
100% 14.3% 14.3%  grid and North China power grid, respectively. ZHH-TP is the tie
ON  
85.7% 85.7%
(36) line between external grids. ZGY-HH and ZGY-TP are two com-
 0 pensated transmission lines. ZTran represents the Guyuan sub-
According to (27), the aggregated impedances with different station. ZCB-GY, ZJLQ-GY, ZHT-GY, ZYY-CB, ZBT-CB, ZBLS-CB, ZLHT-GY
nodes as the aggregation port are obtained as represent the transmission lines in the wind farm systems, while
ZJLQ, ZHT, ZYY, ZLHT, ZBLS and ZBT are the impedances of six
2s( s 4  13s 2  2) 12s( s 2  1) wind farms.
z1A ( s)  ; z2A ( s)  (37)
( s 2  2)( s 4  15s 2  2) s 4  15s 2  2 Considering that each wind farm has multiple identical wind
turbine generators(WTGs) linked to a common bus, the IM of a
It can be seen that z1A ( s) contains the information of three wind farm equals to that of a WTG divided by the number of
oscillatory modes, whereas z2A ( s) ignores the parallel reso- WTGs. The impedance modelling of the WTG takes all its
practical parts into account, including the mechanical compo-
nance mode. The result thus demonstrates that an appropriate nents (shaft & drive train), the electrical components (induction
aggregation port is essential for the aggregated-impedance generator, filter), power electronic circuits (rotor-side &
based analysis. grid-side) and the full-scale control (PLL, inner & outer loops).
B. A practical wind power system The derived IM of a WTG is a 27th -order transfer function and
its details can be found in [4]. Considering that the shunt ca-
The second case involves a practical wind farm system lo- pacitance of transmission lines and the magnetizing reactance
cated at North China. The details of the system and its experi- of transformers has very little impact on SSR issues, they are
enced SSR incidents can be found in [3]. Under the same stable ignored in this study. The parameters of the network compo-
operation point as [5], the INM of the system is established as nents are obtained at the fundamental frequency, thus the model
shown in Fig. 4. established is not suitable for analyzing high frequency oscil-
lations.
1) Calculation of modal observability and controllability
Based on NAM and LIM, the oscillatory modes of the
system can be obtained. Table I lists the modes with negative
and weak damping. The mode λ1 is the unstable subsynchro-
nous mode, whose damping and frequency are consistent with
the actual SSR occurred in this system. As seen from the table,
modes λ2-λ7(λ8-λ13) have similar damping and frequencies,
thus we may expect their observability(controllability) follow
the same rule.
TABLE I
OSCILLATORY MODES
Mode Damp. Freq. Mode Damp. Freq.
λ1 -0.0452 7.45 λ8 0.4577 79.14
λ2 0.0871 33.31 λ9 0.4482 80.46

0885-8977 (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: UNIVERSITY OF ROCHESTER. Downloaded on August 27,2020 at 01:09:46 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/TPWRD.2020.3018149, IEEE
Transactions on Power Delivery
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 7

λ3 0.1309 33.40 λ10 0.4319 80.72 thus its modal observability(controllability) is much larger than
λ4 0.1004 33.46 λ11 0.3909 81.70 the others. 2) The modal observability(controllability) with
λ5 0.1388 33.68 λ12 0.3819 81.97
λ6 0.1022 34.42 λ13 0.3468 82.90 regards to the nodes(branches) of the transmission network are
λ7 0.1646 35.54 numerically similar.

Based on (16)-(17) and (21)-(22), the modal observability 2) Aggregation of the INM
and controllability at different nodes(branches) can be calcu- Once the modal observability is available, the nodal and
lated. Here due to the space limitation, modes λ7 and λ8 are branch aggregation indexes can be computed according to
randomly picked out among λ2-λ7 (λ8-λ13) for further analysis. (29)-(30). The weight of each oscillation mode is determined
The results of modes λ1, λ7 and λ8 are shown in Table Ⅱ and Ⅲ. based on their damping: 1/2 for mode λ1, 1/18 for modes λ2-λ7
The numbers in the tables are expressed as percentages. and modes 1/36 for λ8-λ13. The results are listed in Table Ⅳ and
According to the modal observability(controllability) of Ⅴ, respectively. The numbers in the tables are expressed as
mode λ1 , it can be indicated that the oscillatory current flows percentages. It can be seen that node #11 and branch #6 should
along the series-compensated lines as well as wind farms, and be picked up as the aggregation port to obtain the aggregated
the oscillation is mainly caused by the interactions between impedance(admittance). Node #11 and branch #6 represent the
wind farms and the series-compensated grid (in other words BLS wind farm and Guyuan substation, respectively.
they are the origin of oscillation). Considering that mode λ1 is According to [3]-[4], the oscillatory stability of the system
unstable, it is necessary to monitor its dynamics and take timely can be identified via the frequency characteristics of the ag-
control for the system safety. As λ1 is more observa- gregated impedance. Fig. 5 and Fig. 6 show the frequency
ble(controllable) at nodes #3, 4 and branches #6, 10, the response curves of the reciprocal of aggregated impedances
Guyuan substation is the optimal location to place the moni- with node #11 and node # 2 as the aggregation port. The real
toring and control device [27]. For mode λ7, its modal observ- part and imaginary part of the impedance is called equivalent
ability(controllability) at node #8 and branch #9 is the highest, resistance and reactance, respectively. As shown in Fig. 5, there
while node #9 and branches #10, 11 are the most observa- are two zeros in the resistance curve and one zero in the reac-
ble(controllable) locations for mode λ8. Therefore, it can be tance curve. They each correspond to a mode, where the first
inferred that modes λ7 and λ8 are associated with LHT and YY zero of the resistance curve corresponds to the mode λ1. As
wind farms, respectively. λ2-λ7 and λ8-λ13 are very close to each other, they cannot be
easily distinguished from the frequency response curve. Ac-
TABLE Ⅱ
NODAL OBSERVABILITY(CONTROLLABILITY) INDEX cording to the stability criterion proposed in [3]-[4], the mode
Node λ1 λ7 λ8 whose frequency closes to 7Hz is unstable and the other two
1 0.07 1.68 0.47 modes are stable, which consists with the results in Table I.
2 0.02 0.70 0.20
Considering that the modes are nearly unobservable in node
3 11.73 9.27 1.41
4 11.37 8.68 5.04 #2, the obtained aggregated impedance cannot reflect the dy-
5 10.93 9.54 14.13 namics of the modes as shown in Fig. 6. Such a result demon-
6 11.27 9.00 5.08 strates that the aggregation port should be selected cautiously to
7 11.35 8.87 5.05
ensure that the aggregated impedance can reflect the dynamics
8 11.35 21.75 5.02
9 10.27 11.02 32.25 of the system.
10 10.86 9.70 14.80 TABLE Ⅳ
11 10.78 9.79 16.54 NODAL AGGREGATION INDEX
Node Value Node Value Node Value
1 0.56 5 11.41 9 11.75
TABLE Ⅲ 2 0.23 6 11.63 10 12.66
BRANCH OBSERVABILITY(CONTROLLABILITY) INDEX 3 8.76 7 10.38 11 14.21
Branch λ1 λ7 λ8 4 8.69 8 9.72
1 8.82 8.78 6.95
2 11.75 14.87 12.16
3 2.87 4.48 3.45 TABLE Ⅴ
4 11.32 13.20 10.39 BRANCH AGGREGATION INDEX
5 9.10 10.43 8.72 Branch Value Branch Value Branch Value
6 20.41 23.63 19.11 1 8.05 6 20.06 11, 17 6.92
7, 14 2.18 0.16 0.22 2 12.09 7, 14 3.16 12, 18 4.53
8, 15 1.73 0.16 0.17 3 3.26 8, 15 2.88 13, 19 4.20
9, 16 1.28 22.59 0.15 4 11.11 9, 16 2.63
10 15.26 0.85 19.20 5 8.94 10 12.17
11, 17 8.16 0.50 14.05
12, 18 3.83 0.21 2.29
13, 19 3.29 0.14 3.14

For other modes, the findings are: 1) For each mode among
λ2-λ8(λ9-λ13), one of the six wind farms dominate the oscillation,

0885-8977 (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: UNIVERSITY OF ROCHESTER. Downloaded on August 27,2020 at 01:09:46 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/TPWRD.2020.3018149, IEEE
Transactions on Power Delivery
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 8

subplot for better illustration. Then the modal controllability of

each branch can be calculated via (9). The results of modes λ1,
λ7 and λ8 are listed in Table Ⅶ where the relative errors with
respect to the proposed method are also given. Again, a good
agreement can be observed. Thus, it can be concluded that the
proposed method can accurately calculate the modal observa-
bility and controllability with regard to each node(branch).

Fig.5 Frequency response curve of aggregated impedance with node #11 as the
aggregation port

Fig. 7 Current of the transformer at GY substation and its DFT results

Fig. 6 Frequency response curve of aggregated impedance with node #2 as the

aggregation port
3) Verification using EMT simulations
The detailed electromagnetic transient (EMT) model of the
target system is established in PSCAD/EMTDC. The initial
conditions are set the same as [5]. The total simulation time is 2
s. At 0.5s, a small disturbance is applied to the system. Taking
Fig. 8 The observed oscillatory current of mode λ1 in each branch
branch #4 as an example, its current waveform and DFT-based
spectrum is shown in Fig. 7. It can be seen that the branch
current contains five frequency components corresponding to
the mode λ1, λ7, fundamental frequency, λ8 and the comple-
mentary component of λ1, respectively.
Based on the simulated oscillations, the magnitudes of the
oscillatory components can be obtained. Taking mode λ1 as an
example, its oscillatory currents in each branch are shown in
Fig. 8. The oscillatory currents in the wind farms (branches
#14-19) are shown in the zoomed-in subplot for better illustra-
tion. Then the modal observability of each branch can be cal-
culated via (7). The results of modes λ1, λ7 and λ8 are listed in
Table Ⅵ where the relative errors with respect to the proposed
method are also given. It can be seen that the maximum error is
3.30%, indicating a good match between the analytical value
and the simulations. Fig. 9 The excited oscillatory current of mode λ1 by each branch
By applying a unit-pulse voltage to each branch, the simu-
lated oscillatory magnitudes can be obtained. Taking mode λ1 TABLE Ⅵ
THE MODAL OBSERVABILITY IN EACH BRANCH
as an example, the excited oscillatory currents by each branch Branch λ1 Error(%) λ7 Error(%) λ8 Error(%)
are shown in Fig. 9. The oscillatory currents excited by the 1 9.82 1.00 8.73 0.05 6.26 0.69
wind farms (branches #14-19) are shown in the zoomed-in 2 11.02 0.73 15.53 0.66 13.08 0.92

0885-8977 (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: UNIVERSITY OF ROCHESTER. Downloaded on August 27,2020 at 01:09:46 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/TPWRD.2020.3018149, IEEE
Transactions on Power Delivery
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 9

3 3.63 0.76 5.90 1.42 1.45 2.00

In (38), Y -1 ( s ) can be further expressed as
4 13.13 1.81 13.98 0.78 12.76 2.37
5 8.10 1.00 9.38 1.04 8.04 0.69
Y * ( s) d (s)
6 20.44 0.03 22.34 1.28 19.99 0.88 Y -1 ( s )  = Y * ( s) (39)
7, 14 3.04 0.86 0.51 0.35 0.82 0.60 det(Y ( s )) n( s )
8, 15 2.41 0.68 0.57 0.41 0.57 0.40
9, 16 1.91 0.63 19.29 3.30 0.82 0.67 where Y*(s) is the adjoint matrix of Y(s); det(Y(s)) is the
10 13.21 2.05 3.06 2.21 18.24 0.95 determinant of Y(s); n(s) and d(s) are the numerator and
11, 17 6.29 1.87 0.38 0.12 14.14 0.08 denominator of det(Y(s)).
12, 18 3.76 0.07 0.17 0.04 1.86 0.43
13, 19 3.23 0.06 0.15 0.02 1.97 1.17 The zeros of the determinant of Y(s) are noted as λm (m=1,
TABLE Ⅶ
2, …, M), which means
THE MODAL CONTROLLABILITY IN EACH BRANCH M
Branch λ1 Error(%) λ7 Error(%) λ8 Error(%) n( s)  k1  ( s  m ) (40)
1 9.81 0.99 5.90 2.88 7.53 0.58 m 1
2 11.22 0.53 14.86 0.01 10.97 1.19
3 3.73 0.86 2.62 1.86 2.05 1.40 where k1 is a coefficient.
4 12.84 1.52 13.62 0.42 8.35 2.04 Substituting (39) and (40) into (38), Um can be rewritten as
5 8.38 0.72 11.67 1.24 7.42 1.30
6 20.57 0.16 23.75 0.12 19.57 0.46 U m  k2Y * ( λm ) (41)
7, 14 2.96 0.78 0.44 0.28 0.20 0.02
8, 15 2.34 0.61 0.28 0.12 0.20 0.03 with
9, 16 1.82 0.54 25.36 2.77 0.20 0.05
k1 * d (m )
10 13.24 2.02 0.61 0.24 18.30 0.9
k2  (42)
11, 17 6.24 1.92 0.32 0.18 19.44 5.39 [(m  1 )...(m  m1 )(m  m1 )...(m  M )]
12, 18 3.69 0.14 0.29 0.08 2.30 0.01
13, 19 3.16 0.13 0.27 0.13 3.47 0.33 Since Y(λm) has a zero eigenvalue, the rank of Y(λm) is (n-1)
and the rank of Y*(λm) is 1[28]. Then the rank of Um is 1 ac-
VI. CONCLUSIONS cording to (41). Similarly, the rank of Im can be proved to be 1.
This paper presents a Laplace-domain method to calculate That is to say, the row and column vectors of Um and Im are
the modal observability and controllability for large-scale re- proportional to each other. Thus, it is verified that the ratio of
newable-integrated power systems. The concept of modal ob- the observed(excited) oscillatory magnitudes at different nodes
servability and controllability are established first with rigorous or branches are not affected by the selection of excita-
mathematical proof. It is revealed that two indexes are numer- tion(observation) node or branch.
ically identical and can be obtained through the eigenvalue B. State-space-model analysis for the passive circuit
decomposition of NAM or LIM. Further, one application of the
For the passive circuit in Fig. 3, its state variables, inputs and
proposed concept is presented for determining the optimal
outputs are noted as x, u and y, respectively. They are given by:
aggregated impedance, which is helpful for the simplification
x  iL1 iL 2 uC 3 
T
of the INM based analysis. iL3 uC1 uC 2
The proposed method has been verified using both synthetic T
test circuits and practical wind farms. The results indicate that u  i1n i2n u1l u2l u3l u4l u5l  (43)
the analytical modal observability and controllability obtained T
by the proposed method are consistent with EMT simulations. y  u1n u2n i1l i2l i3l i4l i5l 
Therefore, the method can be served as a useful tool for inves-
tigating the oscillatory characteristics of large systems with where iL1, iL2, iL3 are currents of L1, L2, and L3, respectively; uC1,
high penetration renewables. The information can guide the uC2, uC3 are voltages of C1, C2, and C3, respectively; inn and unn
monitoring and suppression of the oscillations. (n=1, 2) are nodal injected currents and nodal voltages; ull and

APPENDIX ill (l=1, 2,…, 5) are branch added voltages and branch currents.
According to the circuit theory, the state-space model de-
A. Derivation of the rank of oscillatory magnitude matrices
noted as [A, B, C, D] can be built. Due to space limitation, only
The oscillatory magnitude matrices of nodal voltages and the state space matrix A is displayed here as
branch currents are defined as U m  {U nkm
} and I m  {Ilkm } .
0 0 0 1 0 0
m
U nk and I lkm are the oscillatory magnitude of mode λm in the 0 0 0 0 0.5 0 

nodal voltage and branch current, respectively; the first and 0 0 0 1 1 1
m
( I lkm ) represent the obser- A  (44)
 2 0 2
second letter of the subscript in U nk 0 0 0
vation and excitation node(branch), respectively.  0 10 10 0 0 0
According to (14), Um can be obtained as  
 0 0 1 0 0 0 
U m  Y -1 ( s )( s  m )|s m (38)

0885-8977 (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: UNIVERSITY OF ROCHESTER. Downloaded on August 27,2020 at 01:09:46 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/TPWRD.2020.3018149, IEEE
Transactions on Power Delivery
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 10

By calculating the eigenvalues of A, the oscillatory modes of [5] Y. Zhan, X. Xie, H. K. Liu, H. Liu, and Y. Li, “Frequency-domain modal
analysis of the oscillatory stability of power systems with
the system can be obtained, which are the same as (31). Then high-penetration renewables,” IEEE Trans. Sustain. Energy, vol. 10, no. 3,
the matrix B ( C ) are formed via multiplying B(C) by left(right) pp. 1534-1543, July. 2019.
[6] H. K. Liu, X. Xie, J. He, T. Xu, Z. Yu, C. Wang, and C. Zhang, “Sub-
eigenvector matrix of A , which is given by synchronous interaction between direct-drive PMSG based wind farms
T and weak AC networks,” IEEE Trans. Power Syst., vol. 32, no. 6, pp.
0.7259 1.2904 0.1838   0.1510 0.6638 0.1841
5.2394 0.1693 0.5241  1.0899 0.0871 0.5248
4708-4720, Nov. 2017.
  [7] C. Buchhagen, C. Rauscher, A. Menze, et al. “BorWin1 -first experiences
0.1744 0.8254 0.3782   0.0363 0.4246 0.3787  with harmonic interactions in converter dominated grids,” in Proc. In-
    ternational ETG Congress, pp.1-7, 2015.
B  1.5105 1.0087 0.0447  C   0.3142 0.1844 0.0447  [8] X. Wang, F. Blaabjerg, M. Liserre, Z. Chen, J. He, and Y. Li, “An active
 0.5423 0.3586 0.3500   0.1128 0.5189 0.3505 damper for stabilizing power-electronics-based AC system,” IEEE Trans.
    Power Electron., vol. 29, no. 7, pp. 3318–3329, July. 2014.
1.8784 0.1753 0.0165  0.3907 0.0902 0.0166  [9] S. Jan, X. Xie, L. Wang, W. Liu, J. He, and H. Liu, “Overview of
1.3361 0.1833  0.2779 0.3340 
 0.3335   0.0943 emerging subsynchronous oscillations in practical wind power systems,”
Renewable and Sustainable Energy Reviews, vol. 99, pp. 159-168, 2019.
(45) [10] S. Boubzizi, H. Abid, A. El hajjaji and M. Chaabane, “Comparative study
Each row of B reflects the observability of a certain mode in of three types of controllers for DFIG in wind energy conversion system,”
Protection and Control of Modern Power Systems, vol. 3, no. 3, pp.
the outputs, where the first two are nodal voltages of nodes #1, 214-225, 2018.
2 and the rest five are branch currents of branches #1-5. Each [11] B. Appasani and D. K. Mohanta, “A review on synchrophasor commu-
column of C reflects the controllability of a certain mode from nication system: communication technologies, standards and applica-
tions,” Protection and Control of Modern Power Systems, vol. 3, no. 3, pp.
the inputs, where the first two rows are nodal injected currents 383-399, 2018.
of nodes #1, 2 and the rest five are branch added voltages of [12] A. M. A. Hamdan, and A. M. Jaradat, “Modal controllability and ob-
servability of linear models of power systems revisited,” Arabian Journal
branches #1-5. Based on B and C , the modal observability for Science and Engineering, vol. 39, no. 2, pp. 1061-1066, 2014.
and controllability with regard to nodes(branches) can be ob- [13] S. M. Chan, “Modal controllability and observability of power-system
tained as: models,” International Journal of Electrical Power & Energy Systems,
vol. 6, no. 2, pp. 83-88, 1984.
Bmn [14] A. M. A. Hamdan, A M Elabdalla, “Geometric measures of modal con-
N
onm  2
(n  1, 2; m  1, 2,3) (46) trollability and observability of power system models,” Electric Power
 Bmn [15]
Systems Research, vol. 15, no. 2, pp. 147-155, 1988.
W. Xu, Z. Huang, Y. Cui, and H. Wang, “Harmonic resonance mode
n 1
analysis,” IEEE Trans. Power DeL., vol. 20, no. 2, pp. 1182-1190, April.
Cnm 2005.
N
cnm  2
(n  1, 2; m  1, 2,3) (47) [16] I. Vieto and J. Sun, “Small-signal impedance modelling of Type-III wind
 Cnm [17]
turbine,” in Proc. IEEE PES General Meeting, pp.1-5, 2015.
L. Fan, R. Kavasseri, Z. L. Miao, and C. Zhu, “Modeling of DFIG-based
n 1
wind farms for SSR analysis,” IEEE Trans. Power Del., vol. 25, no. 4, pp.
2073–2082, Oct. 2010.
Bml
o(Ll  2) m  7
(l  3,4,5,6,7; m  1,2,3); (48) [18] B. Wen, D. Boroyevich, R. Burgos, P. Mattavelli, and Z. Shen,
“Small-signal stability analysis of three-phase AC systems in the
 Bml presence of constant power loads based on measured d-q frame im-
j 3 pedances,” IEEE Trans. Power Electron., vol. 30, no. 10, pp. 5952-5963,
2015.
Clm
c(Ll  2) m  7
(l  3,4,5,6,7; m  1,2,3) (49) [19] R L. Boylestad, Introductory circuit analysis. Columbus, USA: Pearson
Prentice Hall, 2015.
 Clm [20] W. L. Brogan, Modern control theory. Englewood Cliffs, NJ, USA:
j 3 Prentice-Hall, 1985.
[21] J. Ye, Q. Li, H. Xiong, H. Park, R. Janardan and V. Kumar, “IDR/QR: an
According to (46)-(49), the modal observability matrix ON incremental dimension reduction algorithm via QR decomposition,”
(OL) and the modal controllability matrix CN(CL) are obtained, IEEE Trans. Knowl. Data Eng., vol. 17, no. 9, pp. 1208-1222, Sept. 2005.
[22] Y. Cui, and W. Xu, “Harmonic resonance mode analysis using real
which are the same as (32)-(33). Thus it is verified that the
symmetrical nodal matrices,” IEEE Trans. Power DeL., vol. 22, no. 3, pp.
proposed method is effective and accurate. 1989-1990, July. 2007.
[23] J. Brown, and R. Churchill, Complex variables and applications. New
REFERENCES York, USA: McGraw-Hill, 2008.
[24] Z. Xu, S. Wang, F. Xing and H. Xiao, “Study on the method for analyzing
[1] J. Adams, V. A. Pappu, and A. Dixit, “ERCOT experience screening for electric network Resonance Stability,” Energies, vol. 11, no.3, pp. 1-13,
sub-synchronous control interaction in the vicinity of series capacitor 2018.
banks,” in Proc. IEEE PES General Meeting, pp.1-5, 2012. [25] H. Wang, I. Vieto, and J. Sun, “A method to aggregate turbine and net-
[2] A. Ostadi, A. Yazdani, and R. K. Varma, “Modeling and stability work impedances for wind farm system resonance analysis,” in Proc.
analysis of a DFIG-based wind-power generator interfaced with a series IEEE Workshop on Control and Modeling for Power Electronics, pp. 1-8,
compensated line,” IEEE Trans. Power Del., vol. 24, no. 3, pp. 2018.
1504–1514, Jul. 2009. [26] E. Hendricks, O. Jannerup, and P. H. Sorensen. Linear systems control:
[3] H. Liu, X. Xie, X. Gao, H. Liu, and Y. L, “Stability analysis of SSR in deterministic and stochastic methods. Berlin, Germany: Springer, 2008.
multiple wind farms connected to series-compensated systems using [27] U. S. Park, J. W. Choi, W. S. Yoo, M. H. Lee, and K. Son, “Optimal
impedance network model,” IEEE Trans. Power Syst., vol. 33, no. 3, pp. placement of sensors and actuators using measures of modal controlla-
3118-3128, May. 2018. bility and observability in a balanced coordinate,” J Sci. Technol. vol. 17,
[4] H. Liu, X. Xie, C. Zhang, Y. L, H. L, and Y. H, “Quantitative SSR no. 1, pp. 11–22, 2003.
analysis of series-compensated DFIG-based wind farms using aggre- [28] G. Strang, Introduction to linear algebra. Boston, USA: Wellesley
gated RLC circuit model,” IEEE Trans. Power Syst., vol. 32, no. 1, pp. Cambridge Press, 2009.
474-483, Jan. 2017.

0885-8977 (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: UNIVERSITY OF ROCHESTER. Downloaded on August 27,2020 at 01:09:46 UTC from IEEE Xplore. Restrictions apply.