1

Descriptor State Space Modeling of Power Systems

Yitong Li, Member, IEEE, Timothy C. Green, Fellow, IEEE, Yunjie Gu, Senior Member, IEEE

Abstract—State space is widely used for modeling power that the subsystems in a practical power systems are normally
systems and analyzing their dynamics but it is limited to “terminal-inductive” (e.g., filter inductance of IBRs, stator
representing causal and proper systems in which the number inductance of SGs, winding inductance of transformers, and
of zeros does not exceed the number of poles. In other words,
the system input, output, and state can not be freely selected. line inductance of transmission and distribution networks).
This limits how flexibly models are constructed, and in some This means that the terminal voltage is always the input of
circumstances, can introduce errors because of the addition a subsystem model and the terminal current is always the
arXiv:2303.01701v1 [eess.SY] 3 Mar 2023

of virtual elements in order to connect the mismatched ports state and the output. This can lead to port mismatch when
of subsystem models. An extension known as descriptor state connecting subsystem state space models to construct the
space can model both proper and improper systems and is a
promising candidate for solving the noted problems. It facilitates whole model of a multi-apparatus power system. In order
a modular construction of power system models with flexible to solve this problem, in [4], [6], a sufficiently-large virtual
choice of ports of subsystems. Algorithms for mathematical resistor is added at each bus node, to make the node voltage
manipulation of descriptor state space models are derived such well-defined. Alternatively, in [3], [9], each transmission
as preforming inverse, connection, and transform. Corresponding and distribution line is assumed to be a π-section CLC
physical interpretations are also given. Importantly, the proposed
algorithms preserve the subsystem states in the whole system circuit, which is equivalent to adding a sufficiently-small
model, which therefore enables the analysis of root causes of virtual capacitor at each bus node. Obviously, the additional
instability and mode participation. Theoretical advances are virtual resistors or capacitors change the physical structure
validated by example power systems of varied scales including a of the system, introduce errors (maybe sufficiently-small but
single-inductor or -capacitor system, two-inductor or -capacitor non-zero), and limit the flexibility with which models are
system, and a modified IEEE 14-bus generator-inverter-composite
system. constructed.
An alternative to state space modelling is the transfer
Index Terms—Descriptor state space, power system modeling, function method which has successfully been used for an-
power system stability, dynamics, state participation, eigenvalue,
mode. alyzing dynamics of grid-connected power converters (nor-
mally using impedance or admittance transfer functions in
electrical-centric view) [10]–[14]. This approach has gained
I. I NTRODUCTION increasing interest recently because the electrical impedance
For reliable provision of electricity services, power systems spectrum can be directly measured in practice which enables
must remain stable and be capable of withstanding a wide the black-box analysis when physics-based models are not
range of disturbances [1]. In addition to the traditional available [13], [15], [16]. The transfer function can naturally
stability problems associated with synchronous generators represent both proper and improper systems. However, when
(SGs), the increasing penetration of inverter-based resources numerically modeling a high-order, multi-input, multi-output
(IBRs) lead to more stability threats [2], partly because of system, the transfer function normally has poorer numerical
the increases the system order (number of system states), accuracy and is slower to compute than the equivalent state
and participation analysis (root cause of oscillation modes) space model [17] because of: (a) symbolic calculation with
becomes more difficult. Laplace operator s for transfer function versus linear algebraic
State space, as the most fundamental modeling method, calculation with constant system matrices for state space; and
has successfully been used for SG-based grids [1], and has (b) redundant system states and orders for different input-
more recently also been applied to IBR-dominated grids output combinations of transfer function versus shared state
[3]–[6]. States in a state space model are usually linked to vector x and state matrix A for all inputs and outputs of state
physical variables with clear practical interpretation, in other space. Additionally, the system state information is hidden in
words it is a white-box model, and root causes of instability transfer function model due to the one-to-many mapping from
and oscillations modes can be identified based on modal and it to a state space model, which complicates state participation
participation analysis [1]. This can further guide parameter analysis [15], [18], [19].
tuning and stability enhancement in practice. In this article, descriptor state space (also known as implicit
However, a normally overlooked fact is that the state space state space) is introduced for modeling power systems. It is
method can only model a proper system in which the number a generalized state space representation that can model both
of system zeros does not exceed the number of system poles proper and improper systems. It was originally proposed many
[7], [8]. For example, when modeling the dynamics of an years ago in automatic control theory [20], [21] but has
inductor in the Laplace s-domain, its current is the state and only had very little attention in power engineering community,
therefore has to be the system output, and its voltage has to be perhaps because electrical systems (actually all physically-
the system input. In other words, the system input, output, and realizable systems) can always be represented by proper state
state can not be freely selected. An important observation is space models because they are casual systems in practice [7].
 2

For example, a practical inductor can only be connected to a For representing an improper system by state space method,
voltage source but not a current source, which implies an ad- the descriptor state space (also known as implicit state space)
mittance model of a proper system 1/(sL) (voltage input and can be used, whose format is [8]
current output), and the corresponding state space model can
E ẋ = Ax + Bu
not be inverted. As will be illustrated in this article, descriptor (4)
state space can push these limits which is in fact useful when y = Cx + Du
modeling power systems because it provides a natural solution It is a generalized state space format with an additional matrix
to the noted problems of port mismatch between subsystems E. The corresponding transfer function representation of (4)
with incompatible input-output definitions. To facilitate the is
formation of large power system models in an automated y = C(sE − A)−1 B + D u


(5)
process, algorithms are created to manipulate descriptor state | {z }
space subsystem models (such as inverse, transform, connec- G(s)
tion, etc) and physical interpretations of these are discussed. Obviously, (4) and (5) can also be used for modeling a proper
Importantly, the proposed algorithms preserve the states of system when E is invertable or simply an identity matrix I,
subsystems in the whole-system model, so that analysis of and we can easily convert (4) to (1) as
root causes of instability can indicate specific states and
parameters in the physical system. The theoretical advances ẋ = [E −1 A]x + [E −1 B]u
(6)
are validated through numerical analysis and electromagnetic y = Cx + Du
transient (EMT) simulation of example power systems.
The rest of this article is organized as follows: The funda- By contrast, when (4) is an improper system, E is a singular
mentals of state space and descriptor state space are briefly matrix, i.e., not invertable and not full-rank [8]. In this
reviewed in Section II. In Section III, the calculation and case, (4) cannot be converted to (1). Again, a s-domain
interpretation of descriptor state space for power system appli- inductor is taken as an example here, whose impedance model
cations are elaborated. Section IV gives the case studies with G(s) = sL is improper. Its corresponding descriptor state
numerical results and EMT simulations. Section V concludes space representation is
the article.
       
L 0 i̇ 0 1 i 0
= + i
0 0 v̇ −1 0 v 1
| {z } | {z } |{z}
II. F UNDAMENTALS OF D ESCRIPTOR S TATE S PACE E A
  B (7)
  i  
A standard state space model is [8] v= 0 1 + 0 i
| {z } v |{z}
C D
ẋ = Ax + Bu
(1) where E is not invertable.
y = Cx + Du

where x, u, and y are system vectors of state, input, and output,

III. C ALCULATION AND I NTERPRETATION OF D ESCRIPTOR
respectively; A, B, C, D are system matrices. This state space
S TATE S PACE
system is equivalent to the transfer function system G(s) as
When modularly modeling a large power system, combing
y = C(sI − A)−1 B + D u


(2) subsystems together normally needs the connection of subsys-
| {z } tem models. This further leads to the necessity and advantages
G(s)
of using descriptor state space over standard state space or
where s is the Laplace operator and I is an identity matrix. It transfer function, as discussed in this section. Calculation
is worth highlighting that, the system in (1) and (2) is a proper algorithms of using descriptor state space are also derived next.
system, in which the degree of numerator (order of zeros) does
not exceed the degree of denominator (order of poles) [7], [8].
A. Inverse of Descriptor State Space
This means that the input u and output y can not be freely
selected. For example, for a s-domain inductor, the inductor For a standard state space model in (1), when D is in-
current is the system state which is also the system output, and vertable, its inverse can be easily obtained as [8]
the inductor voltage is the system input. The corresponding ẋ = [A − BD−1 C]x + [D−1 ]y
state space model is (8)
u = [−D−1 C]x + [D−1 ]y
   
i̇ = 0 i + 1/L v
|{z} | {z } However, an invertable D matrix means the system has the
A  B (3) same dimension of input u and output y and the equivalent
i= 1 i+ 0 v system transfer function has the same the degrees of numerator
|{z} |{z} and denominator, i.e., proper but not strictly proper [8].
C D
This is a very strict condition and is always not satisfied in
This corresponds to an admittance model of G(s) = 1/(sL) power systems. Counter examples can be easily found when
with the inductance L, i.e., a proper system. looking at models of inductors and capacitors. Therefore, a
 3

Fig. 1. Descriptor state space model of fundamental electrical circuit elements. (a) Admittance model of an inductor. (b) Impedance model of an inductor.
(c) Impedance model of a capacitor. (d) Admittance model of a capacitor.

generalized inverse of (4) is proposed next. We firstly re-write highlighting that, when Cv and Lv are ideally zero, the systems
the descriptor state space model in (4) as in Fig. 1(b) and (d) can not be represented by conventional
state space anymore because of the non-invertable E matrix.
E ẋ = Ax + Bu + 0y
(9) In conclusion, the physical interpretation of the inverse of
0u̇ = −Cx − Du + y an electrical impedance or admittance in state space is: a new
Then, if regarding this equation as the new state equation (x virtual state of voltage or current is added by adding a virtual
and u are states and y is input), we can get the system model capacitor of Cv = 0 in parallel or a virtual inductor of Lv = 0
as         in series.
E 0 ẋ A B x 0
= + y
0 0 u̇ −C −D u I B. Connection of Descriptor State Space
| {z } |{z}
New A New B (10) When modeling a large power system in a modular way, the
 
  x   models of subsystems are derived first, and then the whole
u= 0 I + 0 y
| {z } u |{z} system model is obtained by connecting subsystem models
New C New D
together. However, as discussed in Section I, subsystems in
which is an inverse system of (4) with new input y and new practical power systems are normally “terminal-inductive”.
output u. Remarkably, u is also added into the state vector This leads to the model port mismatch between subsystems
as a new virtual state, which is the key of achieving the and sufficiently-large virtual resistors [4], [6] or sufficiently-
model inverse. Next, inductors and capacitors will be used small virtual capacitors [3], [9] are added to solve this prob-
to physically interpret this inverse algorithm. lem, which change the system physical structure, introduce
Fig. 1(a) shows the state space representation of an ad- errors, and limit the modeling flexibility.
mittance model of an inductor, i.e., 1/(sL) a strictly proper The underlying reason of the above mentioned port mis-
system. According to (10), the key of obtaining the inverse match problem is essentially because standard state space
model needs to introduce a new virtual state which is exactly cannot represent the connection of improper systems, e.g.,
the system old input, i.e., inductor voltage. Hence, a virtual the series connection of inductors or parallel connection of
capacitor Cv is added in parallel with the inductor L. This capacitors. By contrast, the descriptor state space can naturally
leads to the new model in Fig. 1(b). Remarkably, when Cv solve this problem. In addition to the inverse algorithm in (10),
is ideally zero, we can get iC = 0, iL = i, and a singular other commonly used algorithms (sum, append, feedback, etc)
matrix E = [L, 0; 0, 0]. In this case, the model in Fig. 1(b) of connecting descriptor state space are summarized in Fig. 2.
is essentially the descriptor state space representation of sL, The detailed derivation of these algorithms is similar to that
i.e., an improper system. The model in Fig. 1(b) exactly of standard state space in [22] and is omitted here. Next, an
coincide with (7) in last section and the inverse algorithm in example of modeling the series connection of two inductors
(10). Similarly, for a capacitor, Fig. 1(c) shows the impedance is investigated. The descriptor state space model is illustrated
model 1/(sC), and Fig. 1(d) shows the admittance model in Fig. 3(a), which can be obtained by modeling
sC by adding a virtual inductor of Lv = 0. It is also worth Step 1: 1/(sL1 ) and 1/(sL2 ) via standard state space;
 4

Fig. 2. Algorithms of combining and connecting descriptor state space models. (a) Sum. (b) Append. (c) Matrix append. (d) Feedback.

Fig. 3. Modeling the connection of improper systems. (a) Series connection of inductors. (b) Parallel connection of capacitors.
 5

Step 2: sL1 and sL2 via inverse in (10); for (b). The next key step of achieving the transform is to find
Step 3: (sL1 + sL2 ) via sum in Fig. 2(a); x2 = f (x1 , u1 ) so that we can eliminate the virtual state x2
Step 4: 1/(sL1 + sL2 ) via inverse in (10). in (11). There are two sub-cases, as elaborated next.
which gives the final model. The model correctness can be eas- When A22 is invertable, x2 can be directly solved as
ily validated by deriving the corresponding transfer function
x2 = −A−1 −1
22 A21 x1 − A22 B2 u (12)
according to (5). As summarized in Fig. 3(a), each state equa-
tion in the model has a clear physical interpretation, such as Then using this equation to replace x2 in (11) yields the state
Ohm’s law, Kirchhoff’s voltage law, Kirchhoff’s current law, space representation as
 −1 −1
  −1 −1

etc. For example, three Kirchhoff equations clearly indicate x˙1 = E1 (A11 − A12 A22 A21 ) x1 + E1 (B1 − A12 A22 B2 ) u
the series connection relation of two inductors. Additionally, | {z } | {z }
New A New B
these three Kirchhoff equations are also state equations of v1 ,
C2 A−1 x1 + D − C2 A−1
   
y = C1 − 22 A21 22 B2 u
v2 , and i, which are associated to three virtual circuit elements: | {z } | {z }
v1 is the state of a virtual capacitor in parallel with L1 , given New C New D
(13)
by step 2; v2 is the state of a virtual capacitor in parallel with When A22 is not invertable, we first find the null matrix N of A22
L2 , given by step 2; i is the state of a virtual inductor in series so that [24]
with the whole model, given by step 4. Similarly to modeling
N A22 = 0, Rank(N ) + Rank(A22 ) = Order(A22 ) (14)
inductors, Fig. 3(b) shows the dual example of modeling the
parallel connection of two capacitors. Then, left-multiplying N to the second state equation in (11) yields
In general, according to the calculation algorithms, we can 0 = N A21 x1 + N A22 x2 + N B2 u
get the model of a power system without introducing modeling | {z } (15)
0
errors and restricting the choice of system input or output
The time derivative of this equation is
variables. For instance,
Step1: Similarly to the derivation of nodal admittance matrix 0 = N A21 ẋ1 + N B2 u̇ (16)
[1], [16], [23], the whole-network model can be derived by Replacing ẋ1 in this equation by using the first state equation in (11),
combing the model of each branch, via algorithms of sum and we get
matrix append in Fig. 2(a) and (c). 0 = N A21 E1−1 (A11 x1 + A12 x2 + B1 u) + N B2 u̇
Step 2: Models of all apparatuses (IBRs, SGs, etc) in a power | {z } (17)
grid can also be combined via the append in Fig. 2(b), which N̂

gives the whole-apparatus model. Combining this equation with the second state equation in (11), we
Step 3: The whole-network model (Step 1) and whole- get the augmented equation as
       
apparatus model (Step 2) can be connected via the feedback A21 A22 B2 0
0= x1 + x2 + u+ u̇ (18)
rule in Fig. 2(d), which gives the so-called whole-system N̂ A11 N̂ A22 N̂ B1 N̂ B2
impedance model [18], [19], [23] or port-mapping model | {z } | {z } | {z } | {z }
Â21 Â22 B̂2 B̂d
[16].
The whole-system model can then be used for analyzing where Â21 , Â22 , B̂2 , and B̂d are the augmented matrices. It
system dynamics, tuning parameters, and enhancing stability. is remarkable that Â22 is generalized-invertable (specifically, left-
invertable) because it consists of A22 and its null matrix N [24].
Now, x2 can be solved as
C. Transforming Descriptor State Space to State Space
x2 = −Â−1 −1 −1
22 Â21 x1 − Â22 B̂2 u − Â22 B̂d u̇ (19)
As mentioned in Section II, a physically-realizable system
Using this equation to replace x2 in (11), we can get the state space
must be proper because of causalty [7]. Even though a proper representation as
system can be represented by descriptor state space, it can also    
be transformed back to an equivalent state space model. The ẋ1 = E1−1 (A11 − A12 Â−1 −1 −1
22 Â21 ) x1 + E1 (B1 − A12 Â22 B̂2 ) u
| {z } | {z }
algorithm for that is now derived. New A New B
For a descriptor state space system in (4), if E is invertable,
 
+ −E1−1 A12 Â−1
22 B̂d u̇
then the conversion can be easily done by left-multiplying E −1 | {z }
Bd
to the state equation, as shown in (6). Examples of this case    
can be found in Fig. 1(a) and (c). If E is not invertable, we y = C1 − C2 Â−1 −1
22 Â21 x1 + D − C2 Â22 B̂2 ) u
| {z } | {z }
firstly re-write (4) as New C New D
 
+ −C2 Â−1
22 B̂d u̇
       
E1 0 ẋ1 A11 A12 x1 B1
= + u | {z }
0 0 ẋ2 A21 A22 x2 B2 Dd
  (11) (20)
  x1  
y = C1 C2 + D u We can easily see that (13) is simply a special case of (20) when A22
x2 is full-rank and its null matrix N is empty, and therefore Â21 = A21 ,
where E1 is invertable, x1 consists of practical states, and Â22 = A22 , B̂2 = B2 . It is also worth mentioning that, the additional
matrices Bd and Dd are non-zeros only for improper system because
x2 consists of virtual states. Examples of this case can be of its higher order of zeros than poles (the time derivative of input
found in Fig. 3(a) and (b), where E1 = [L1 , 0; 0, L2 ] and u). As for causal and proper systems in practice, Bd and Dd are
x1 = [i1 ; i2 ] for (a) and E1 = [C1 , 0; 0, C2 ] and x1 = [v1 ; v2 ] zero.
 6

Fig. 5. Whole-system mode map.

Fig. 4. A modified IEEE 14-bus power system.

A simple example is also given here. The admittance model of

two series-connected inductors in Fig. 3(a) is a proper system that
represented by descriptor state space. By using (20), this model can
be converted back to state space representation as
        
i̇1 0 0 i1 1/(L1 + L2 ) 0
= + v+ v̇
i̇2 0 0 i2 1/(L1 + L2 ) 0
Fig. 6. Synchronous mode analysis of lines. (a) Participation on -0.35+j50
Hz mode. (b) Participation on -25.4+j50 Hz mode.
| {z } | {z } |{z}
NewA NewB Bd
  (21)
  i1    
i = 1/2 1/2 + 0 v + 0 v̇
| {z } i2 |{z} |{z} and are integrated into a toolbox [26]. The system dynamics are
NewC NewD Dd analyzed by eigenvalue and state participation (descriptor state space
which can be further simplified to model) and validated by time-domain simulations (EMT model in
    Matlab/Simulink) next. The eigenvalues (also known as modes) of
i̇ = 0 i + 1/(L1 + L2 ) v the whole-system are shown in Fig. 5. Certain modes are highlighted
|{z} | {z } and analyzed in detail next.
New A New B
    (22)
i= 1 i+ 0 v
|{z} |{z} A. Synchronous Mode of Inductance
New C New D The 50 Hz synchronous mode in the mode map of Fig. 5 is
i.e., a standard state space model of 1/[s(L1 + L2 )]. investigated first. Fig. 6 shows the participation analysis on two
modes with different real parts: (a) −0.35 + j50 Hz; and (b)
−25.4 + j50 Hz. The participation pie chart shows that mode (a)
D. Tracked States in Descriptor State Space has a significant participation from branch 7-9 which is a transformer
It is worth highlighting that, the system states are always tracked winding with very large X/R ratio of 20, that is, a component with
in the proposed algorithm. For example, in model inverse algorithm low damping. This is reflected in the mode being close to imaginary
in (10), the new system state consists of the old system state with axis. In contrast, mode (b) is dominated by branch 6-13 which is
an additional virtual state u; in model transform algorithm in (20), a line with X/R ratio of 1.97 = 0.13027/0.06615, i.e., properly
the new system state is the subset of old system state; in model damped. This is reflected in this mode being located far from the
connection algorithm in Fig. 2, the new system state is the union set of imaginary axis.
subsystem states. In other words, in the finally obtained whole system
model, a state can always be mapped to a physical state in certain
subsystem. This enables investigating the participation of a physical
B. Swing mode of SG
state on system oscillation modes, and can be used for analyzing The 2.2 Hz and 3.5 Hz swing modes in Fig. 5 are investigated and
system dynamics and tuning system parameters, as will be illustrated the results are summarized in Fig. 7. As shown in Fig. 7(a), the 2.2
by case studies in next section. The participation analysis can be done Hz mode is dominated by the swing of SG1 (ω1 , θ1 ) and SG3 (ω3 ,
either in the converted state space format in (20), or in the descriptor θ3 ); As displayed in Fig. 7(b), the 3.5 Hz mode is dominated by the
state space format directly as derived in Appendix A. swing of SG6 (ω6 , θ6 ). This coincides with the mode loci in Fig. 7(c)
and (d): The 2.2 Hz mode is damped only when increasing the rotor
damping of SG1 and SG3 in (c); and the 3.5 Hz mode is damped only
IV. C ASE S TUDY when increasing the rotor damping of SG6 in (d). Fig. 7(e) shows
By using descriptor state space method proposed in last section, a the EMT simulation results. After the 1 pu load impulse at 0.5 s,
modified IEEE 14-bus system in Fig. 4 is modeled numerically. The SG1 and SG3 oscillates dominantly at 2.2 Hz, and SG6 oscillates at
system consists of 3 SGs and 3 IBRs (Type-IV wind generation). 3.5 Hz. This coincides with the numerical analysis of the descriptor
The parameters, scripts, and models are open-source online [25] state space model.
 7

model both proper and improper systems. This feature solves the
problem of port mismatches that can arise between subsystems with
incompatible definitions of inputs and outputs. The states are always
tracked during the modeling procedure so that a state of whole-
system model can always be mapped to a physical state in one of
the subsystem. With this preservation of states, participation analysis
can be applied which enables analysis of root causes of instability
and oscillatory modes such that indications for parameter tuning in
the physical plant can be provided. Algorithms for manipulation of
descriptor state space subsystems are derived, including formation
of model inverses, connections of subsystems, and other transforma-
tions. Numerical calculations and EMT simulations of example power
systems were used to validate the theoretical advances.

A PPENDIX A
PARTICIPATION A NALYSIS OF D ESCRIPTOR S TATE S PACE
The participation analysis of descriptor state space can be derived
similarly to that of standard state space in Chapter 12 in [1], as
briefly shown next. For a descriptor state space model in (4), we can
get the generalized right and left eigenvector matrices [φ] and [ψ] as
[8], [27]
ψ1
 
  ψ2 
= φ1 φ2 ... φi ... , [ψ] = 
 ... 
 (23)
ψi
...
where φi and ψi are the ith right and left eigenvectors. These two
eigenvector matrices satisfy
A[φ] = E[φ][λ], [ψ]A = [λ][ψ]E (24)
where [λ] is the eigenvalue matrix as
[λ] = blkdiag(λ1 , λ2 , ..., λi , ...) (25)
Fig. 7. Swing mode analysis of SGs. (a) Participation on 2.2 Hz mode. (b)
If only focusing the ith eigenvalue, we can re-write (24) as
Participation on 3.5 Hz mode. (c) Mode locus with increasing rotor damping
of SG1 and SG3 from 1 pu to 20 pu. (d) Mode locus with increasing rotor Aφi = Eφi λi , ψi A = λi ψi E (26)
damping of SG6 from 1 pu to 20 pu. (e) EMT simulation results with 1 pu
load impulse at all SG buses at 0.5 s. Differentiating the first equation in (26) with respect to akj (i.e., the
element of A in kth row and jth column) yields
∂A ∂φi ∂λi ∂φi
C. Control Interaction Mode of IBR φi + A = Eφi +E λi (27)
∂akj ∂akj ∂akj ∂akj
The 24.7 Hz mode in the mode map Fig. 5 is investigated next.
According to Fig. 8(a), this mode is mainly participated by the ac Left-multiplying by ψi , and noting the second equation in (26), we
current control of IBR8 (id8,ki , iq8,ki , id8 , iq8 ), but also jointly can re-write (27) as
participated by its phase-locked loop (PLL) (θ8 ) and dc-link control ∂λi ∂A
(vdc8 ). In other words, it is an interaction mode of different control ψ Eφ = ψi φi (28)
loops of IBR8. Fig. 8(b), (c), and (d) show the mode loci with | i{z }i ∂akj ∂akj
1 or 0 | {z }
changing corresponding control parameters, which indicates that ψik φji
slower current control bandwidth in (b), or faster PLL in (c), or
faster dc-link control in (d) can lead to more chances of interaction For normalized [φ] and [ψ], the left-hand side equals to 1 (when λi
and instability. Fig. 8(e) shows the EMT simulation of 1 pu load is a practical mode) or 0 (when λi is a virtual mode). The right-
impulse response of IBR8 with rated parameters, as the simulation hand side equals to ψik φji , which coincides with the widely-known
reference. The 24.7 Hz oscillation can be clearly observed. Fig. 8(f), participation factor used in power engineering community [1].
(g), and (h) show the EMT simulations with badly tuning ac current
bandwidth fac8 , PLL bandwidth fpll8 , and dc-link bandwidth fdc8 R EFERENCES
at 0.1 s to de-stabilize the system, and tuning them back at 0.2 s
to re-stabilize the system. Remarkably, (f) shows 14 Hz unstable [1] P. Kundur, Power system stability and control. McGraw-hill New York,
oscillations, (g) shows 29 Hz unstable oscillations, whereas (h) shows 1994, vol. 7.
worse but still stable oscillations compared to the rated case in (e). [2] N. Hatziargyriou, J. Milanovic, C. Rahmann, V. Ajjarapu, C. Canizares,
I. Erlich, D. Hill, I. Hiskens, I. Kamwa, B. Pal et al., “Definition and
All EMT simulation results properly coincide with the mode maps classification of power system stability–revisited & extended,” IEEE
predicted by the descriptor state space model in (a) to (d). Trans. Power Syst., vol. 36, no. 4, pp. 3271–3281, 2020.
[3] U. Markovic, O. Stanojev, E. Vrettos, P. Aristidou, and G. Hug,
V. C ONCLUSIONS “Understanding stability of low-inertia systems,” IEEE Trans. Power
Syst., 2019.
The descriptor state space method has been introduced for mod- [4] Y. Wang, X. Wang, Z. Chen, and F. Blaabjerg, “Small-signal stability
elling multi-bus, multi-apparatus power systems composed of many analysis of inverter-fed power systems using component connection
subsystems. It compliments conventional state space analysis and can method,” IEEE Trans. Smart Grid, vol. 9, no. 5, pp. 5301–5310, 2017.
 8

Fig. 8. Control interaction mode analysis of IBR8. (a) Participation on 24.7 Hz mode. (b) Mode locus with reducing ideal current bandwidth fac8 from 250
Hz to 150 Hz. (c) Mode locus with increasing PLL bandwidth fpll8 from 10 Hz to 30 Hz. (d) Mode locus with increasing dc-link bandwidth fdc8 from 10
Hz to 30 Hz. (e) EMT simulation with rated parameters of IBR8. (f) EMT simulation with reducing fac8 from 250 Hz to 150 Hz at 0.1 s and increasing
it back at 0.2 s. (g) EMT simulation with increasing fpll8 from 10 Hz to 30 Hz at 0.1 s and reducing it back at 0.2 s. (h) EMT simulation with increasing
fdc8 from 10 Hz to 30 Hz at 0.1 s and reducing it back at 0.2 s.

[5] X. Wang and F. Blaabjerg, “Harmonic stability in power electronic-based [16] Y. Li, Y. Gu, and T. C. Green, “Mapping of dynamics between
power systems: Concept, modeling, and analysis,” IEEE Trans. Smart mechanical and electrical ports in SG-IBR composite grids,” IEEE Trans.
Grid, vol. 10, no. 3, pp. 2858–2870, 2018. Power Syst., 2022.
[6] N. Pogaku, M. Prodanovic, and T. C. Green, “Modeling, analysis and [17] “Conversion Between Model Types in Matlab.”
testing of autonomous operation of an inverter-based microgrid,” IEEE [Online]. Available: https://www.mathworks.com/help/control/ug/
Trans. Power Electron., vol. 22, no. 2, pp. 613–625, 2007. conversion-between-model-types.html
[7] A. V. Oppenheim and A. S. Willsky, Signals & systems. Prentice Hall, [18] Y. Zhu, Y. Gu, Y. Li, and T. C. Green, “Participation analysis in
1997. impedance models: The grey-box approach for power system stability,”
[8] S. Skogestad and I. Postlethwaite, Multivariable feedback control: IEEE Trans. Power Syst., 2022.
analysis and design. john Wiley & sons, 2005. [19] ——, “Impedance-based root-cause analysis: Comparative study of
[9] “PSCAD.” [Online]. Available: https://www.pscad.com impedance models and calculation of eigenvalue sensitivity,” IEEE
[10] B. Wen, D. Boroyevich, R. Burgos, P. Mattavelli, and Z. Shen, “Analysis Trans. Power Syst., 2022.
of dq small-signal impedance of grid-tied inverters,” IEEE Transactions [20] D. Luenberger, “Dynamic equations in descriptor form,” IEEE Transac-
on Power Electronics, vol. 31, no. 1, pp. 675–687, 2015. tions on Automatic Control, vol. 22, no. 3, pp. 312–321, 1977.
[11] X. Wang, L. Harnefors, and F. Blaabjerg, “Unified impedance model of [21] P. C. Müller, “Descriptor systems: pros and cons of system modelling
grid-connected voltage-source converters,” IEEE Transactions on Power by differential-algebraic equations,” Mathematics and computers in
Electronics, vol. 33, no. 2, pp. 1775–1787, 2017. simulation, vol. 53, no. 4-6, pp. 273–279, 2000.
[12] Y. Li, Y. Gu, Y. Zhu, A. Junyent-Ferre, X. Xiang, and T. C. Green, [22] E. L. Duke, “Combing and connecting linear multi-input, multi-output
“Impedance circuit model of grid-forming inverter: Visualizing control subsystem models,” NASA Technical Memorandum 85912, 1986.
algorithms as circuit elements,” IEEE Trans. Power Electron., vol. 36, [23] Y. Gu, Y. Li, and T. Green, “Impedance-based whole-system modeling
no. 3, pp. 3377–3395, Mar. 2021. for a composite grid via embedding of frame-dynamics,” IEEE Trans.
[13] L. Fan, Z. Miao, P. Koralewicz, S. Shah, and V. Gevorgian, “Identifying Power Syst., vol. 36, no. 1, pp. 336–345, Jan. 2021.
dq-domain admittance models of a 2.3-mva commercial grid-following [24] S. Axler, Linear algebra done right. Springer Science & Business
inverter via frequency-domain and time-domain data,” IEEE Transac- Media, 1997.
tions on Energy Conversion, vol. 36, no. 3, pp. 2463–2472, 2020. [25] “Future Power Networks.” [Online]. Available: https://github.com/
[14] Y. Li, Y. Gu, and T. Green, “Revisiting grid-forming and grid-following Future-Power-Networks/Publications
inverters: A duality theory,” IEEE Trans. on Power Syst., 2022. [26] “Simplus Grid Tool.” [Online]. Available: https://github.com/
[15] E. O. Kontis, T. A. Papadopoulos, A. I. Chrysochos, and G. K. Future-Power-Networks/Simplus-Grid-Tool
Papagiannis, “Measurement-based dynamic load modeling using the [27] “Eigenvalues and Eigenvectors in Matlab.” [Online]. Available:
vector fitting technique,” IEEE Transactions on Power Systems, vol. 33, https://www.mathworks.com/help/matlab/ref/eig.html
no. 1, pp. 338–351, 2017.