Electrical Power and Energy Systems 93 (2017) 65–74

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Comparison of time-varying phasor and dq0 dynamic models for large

transmission networks
Juri Belikov a,b, Yoash Levron c,⇑
a
Faculty of Mechanical Engineering, Technion—Israel Institute of Technology, Haifa 3200003, Israel
b
Department of Computer Systems, Tallinn University of Technology, Ehitajate tee 5, 19086 Tallinn, Estonia
c
The Andrew and Erna Viterbi Faculty of Electrical Engineering, Technion—Israel Institute of Technology, Haifa 3200003, Israel

a r t i c l e i n f o a b s t r a c t

Article history: In recent years, with increasing penetration of small distributed generators and fast power electronics
Received 19 February 2017 based devices, the assumption of quasi-static phasors is becoming increasingly inaccurate. In order to
Received in revised form 20 April 2017 describe fast dynamic behavior and rapid amplitude and phase variations, more accurate dynamic mod-
Accepted 16 May 2017
els based on the dq0 transformation are used. To better understand the differences between these two
models, in this work we compare their relative accuracy when applied to large-scale transmission net-
works. In this light, the present work describes the two types of models using similar terminology, which
Keywords:
is based on dq0 quantities. Based on this result, we show that quasi-static models may be obtained from
Quasi-static
Time-varying phasors
dq0 models at low frequencies, and that there exists a frequency range in which quasi-static model
Power flow equations approximates the dq0 model well. The obtained results allow to estimate the frequency after which
DQ0 models the quasi-static model cannot accurately describe the system dynamics, and dq0 models should be used
instead.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction In recent years, with increasing penetration of small distributed

Quasi-static models are a well-known approach for modeling
the dynamics of large power systems. A key assumption in such
models is that voltages and currents may be represented by
time-varying phasors, since magnitudes and phases throughout
the network change slowly in comparison to the system frequency
[1,2]. The transmission network is represented in this case by a
constant admittance matrix, and the system dynamics are
described either in terms of phasor components, or using the
power-flow equations. A key advantage of quasi-static models is
relative simplicity, since large transmission networks can be
described by purely algebraic equations [3]. In addition, since
quasi-static models employ phasors instead of sinusoidal AC
signals, the system operating point (or equilibrium point) is
well-defined, a property which enables small-signal and stability
studies. Due to these properties, quasi-static models have been
extensively used in the analysis of dynamic interactions that occur
in time frames of seconds to minutes, and have historically enabled
studies of machine stability, inter-area oscillation, and other slow
dynamic interactions that occur in large power systems [1,4–6].
⇑ Corresponding author.
E-mail addresses: juri.belikov@ttu.ee (J. Belikov), yoashl@ee.technion.ac.il
(Y. Levron).
generators and fast power electronics based devices, the assump-
tion of quasi-static phasors is becoming increasingly inaccurate
[7]. In order to describe fast dynamic behavior and rapid amplitude
and phase variations, more accurate dynamic models are being
used, among them models that are based on the dq0 transforma-
tion [8]. The dq0 transformation maps three-phase signals in the
abc reference frame to new signals in a dq0 reference frame, that
often rotates with the electrical angle of a machine’s rotor. This
approach provides several important advantages when the system
is balanced or symmetrically configured [9], one of them is that
sinusoidal three-phase signals are mapped to constant signals at
steady-state [10,11]. Similarly to quasi-static models, dq0-based
models are often time-invariant, and allow to describe the system
by means of ordinary differential equations, using standard state-
space formalism. Since dq0 signals are constant at steady-state,
the operating point of the system is well-defined, and as a result,
the dynamic equations can be linearized, and a small-signal analy-
sis can be performed [4]. For these reasons, the dq0 transformation
is often used to evaluate the dynamics of symmetric or symmetri-
cally configured units, and provides efficient tools for designing
suitable controllers [12]. A review of simulation techniques based
on dq0 quantities can be found in [10,13,14].
The dq0 transformation is increasingly used today when model-
ing distributed sources, complex loads, renewable generators, and

http://dx.doi.org/10.1016/j.ijepes.2017.05.017
0142-0615/Ó 2017 Elsevier Ltd. All rights reserved.
66 J. Belikov, Y. Levron / Electrical Power and Energy Systems 93 (2017) 65–74

power electronics based devices [15–18]. In addition, widespread dq0 signals and time-varying phasors. Assume a unit with balanced
use of this transformation for modeling large-scale power systems three-phase AC voltages given by
is still a pending issue, that is recently being explored. Toward this
v a ðtÞ ¼ AðtÞ cos ðxs t þ wðtÞÞ;
end, one central challenge is to combine various dq0 models that

use local reference frames to form a complete dynamic model v b ðtÞ ¼ AðtÞ cos xs t þ wðtÞ  23p ; ð3Þ


describing a large-scale power system. Another associated chal- v c ðtÞ ¼ AðtÞ cos xs t þ wðtÞ þ 23p :
lenge is to provide a dq0-based model of general transmission net-
Assuming that variations in the magnitude AðtÞ and phase wðtÞ are
works, that will be of low complexity and easy-to-use. Several
slow in comparison to the frequency xs , these voltages may be rep-
recent works exploring these questions are [19], which shows
resented by the time-varying phasor VðtÞ ¼ AðtÞpffiffi ejwðtÞ . In addition,
dq0 models of elementary passive components, work [9], that pre- 2

sents dq0-based models of three-phase networks with RL elements, based on the dq0 transformation (19), equivalent dq0 voltages with
work [20], that shows a nonminimal state-space model of net- respect to the reference frame xs t are
works with standard branches, and [21], that presents a frequency v d ðtÞ ¼ AðtÞ cosðwðtÞÞ;
domain model of transmission networks using dq0 quantities. v q ðtÞ ¼ AðtÞ sinðwðtÞÞ; ð4Þ
Models based on dq0 quantities (as reviewed above) are known
v 0 ðtÞ ¼ 0:
to extend the classic quasi-static model, and are generally consid-
ered to be more accurate. However, while quasi-static models are It immediately follows that a time-varying phasor relates to dq0
well-known and widely used, dq0 models are only now being quantities as
explored in the context of large-scale power systems. To better AðtÞ AðtÞ 1

VðtÞ ¼ pffiffiffi ejwðtÞ ¼ pffiffiffi ðcosðwðtÞÞ þ j sinðwðtÞÞÞ ¼ pffiffiffi v d ðtÞ þ jv q ðtÞ ;
understand the differences between these two models, in this work 2 2 2
we compare their relative accuracy when applied to large-scale ð5Þ
transmission networks. The two types of models are described using
similar terminology, which is based on dq0 quantities. Based on this or alternatively,
pffiffiffi
result, we show that quasi-static models may be obtained from dq0 v d ðtÞ ¼ 2RefVðtÞg;
models at low frequencies, and that there exists a frequency range pffiffiffi ð6Þ
in which quasi-static model approximates the dq0 model well.
v q ðtÞ ¼ 2ImfVðtÞg;
Results are demonstrated on the basis of several test-cases. and the same relations hold for currents. Let us now examine the
The paper is organized as follows. Section 2 presents the basic active and reactive powers. Assuming a voltage phasor VðtÞ and a
definition of dq0 quantities. Section 3 reviews quasi-static models current phasor IðtÞ, the powers are given by
and links them to dq0 quantities. The dq0 models of large-scale

v d ðtÞid ðtÞ þ v q ðtÞiq ðtÞ ;
PðtÞ ¼ RefVðtÞI ðtÞg ¼ 12
networks are presented in Section 4. Section 5 provides a compar-
 ð7Þ
ison between the two models, followed by simulation results in QðtÞ ¼ ImfVðtÞI ðtÞg ¼ 2 v q ðtÞid ðtÞ  v d ðtÞiq ðtÞ :
1

Section 6. Concluding remarks are presented in Section 7. The first equation in (7) shows the dual meaning of the active power
PðtÞ. In the specific case of a static or quasi-static system, in which
2. Preliminaries transients are slow in comparison to xs , the active power PðtÞ of
each phase is the average power over a line cycle. In addition, the
This section recalls the basic definition of the dq0 transforma- instantaneous sum of powers for the three-phases is
tion. Consider a reference frame rotating with an angle of hðtÞ. P3/ ðtÞ ¼ 3PðtÞ. Note that if the system is not quasi-static (for
For instance, in a synchronous machine, hðtÞ is typically selected instance, during a fast transient) the average powers P and Q are
to be the rotor electrical angle. Let ~f represent the quantity to be not well-defined. However, the expression for P3/ ðtÞ still holds.
transformed (current, voltage, or flux), and use the compact nota- Using time-varying phasors, the network is usually described by
tion fabc ¼ ½fa ; fb ; fc T , fdq0 ¼ ½fd ; fq ; f0 T . The dq0 transformation the nodal admittance matrix Y bus . Assume a network with N buses,
with respect to the reference frame rotating with the angle h can and denote the bus voltages by VðtÞ ¼ ½v 1 ðtÞ; . . . ; v N ðtÞT , and the
be defined as [8, Appendix C] T
bus injected currents by IðtÞ ¼ ½i1 ðtÞ; . . . ; iN ðtÞ . All quantities in
~fdq0 ¼ T h fabc ð1Þ these vectors are time-varying phasors. The network is then
described by the relation
with
2

 3 IðtÞ ¼ Y bus VðtÞ: ð8Þ
cos ðhÞ
cos h  23p cos h þ 23p
26

7 This is a quasi-static model of the transmission network, since it is
T h ¼ 4  sin ðhÞ  sin h  23p  sin h þ 23p 5: ð2Þ
3 1 1 1
based on the assumption that phasors change slowly in comparison
2 2 2 to xs . Due to this assumption, the frequency throughout the net-
work is approximately constant, and as a result, the matrix Y bus is
3. Quasi-static models composed of constant admittances, which are computed at the
fixed frequency xs , such that Y bus ¼ Y bus ðjxs Þ. Eq. (8) can also be
The traditional quasi-static model is based on the assumption written in terms of its real and imaginary parts as
that the frequency of voltage and current signals throughout the
RefIðtÞg ¼ RefY bus gRefVðtÞg  ImfY bus gImfVðtÞg;
network is approximately constant. As a result, AC signals can be ð9Þ
modeled accurately enough by means of time-varying phasors, ImfIðtÞg ¼ ImfY bus gRefVðtÞg þ RefY bus gImfVðtÞg;
and the transmission network can be represented by a constant or alternatively, using the power flow equations
admittance matrix Y bus . The transmission network is then modeled P
N
using the linear relation I ¼ Y bus V, which is equivalent to the well- Pn ðtÞ ¼ jV n ðtÞj jynk jjV k ðtÞj cos ð\ynk þ dk  dn Þ;
k¼1
known power flow equations [3]. ð10Þ
P
N
To present the quasi-static models and dq0 models using simi- Q n ðtÞ ¼ jV n ðtÞj jynk jjV k ðtÞj sin ð\ynk þ dk  dn Þ;
lar terminology, we begin by developing the relationship between k¼1
 J. Belikov, Y. Levron / Electrical Power and Energy Systems 93 (2017) 65–74 67

where for the nth unit, P n ðtÞ, Q n ðtÞ are the powers injected from the
unit into the network, jV n ðtÞj is the voltage amplitude, and dn is the
voltage phase. In addition, the constants ynk are elements of the
admittance matrix Y bus .
Quasi-static models of generators and loads may be of different
types, depending on the device physics and level of complexity.
The following section explains how to derive a simple and popular
quasi-static model of the synchronous generator.
Fig. 2. The resulting signal-flow diagram of a 3-bus system.
4. dq0 signals based models

Transformation from the Laplace domain to the time domain results

Models based on dq0 signals may be presented as signal-flow
in
diagrams, in which each component is modeled by dq0 quantities.
A simple motivating example is illustrated in Figs. 1 and 2. v d ¼ xs Liq þ L didt ; d

In quasi-static models, the transmission network is described

by Eqs. (8)–(10). However, since the model is assumed to be
v q ¼ xs Lid þ L didt ;q
ð15Þ
quasi-static, these equations are only valid at frequencies that v 0 ¼ L didt :
0

are much lower than xs . We now recall a theorem that extends

Similarly, assume a three-phase inductor with currents id , iq , i0 ,
this model to higher frequencies using dq0 quantities. Define the
voltages on one side v d;1 , v q;1 , v 0;1 , and voltages on the other side
dq0 signals v d;n , v q;n , v 0;n to be the voltages on bus n, and id;n , iq;n ,
i0;n to be the injected currents to bus n. In addition, define the vec-
v d;2 , v q;2 , v 0;2 . Following (15), the dynamic model of the inductor
may be written as
tors V d ðtÞ ¼ ½v d;1 ðtÞ; . . . ; v d;N ðtÞT , Id ðtÞ ¼ ½id;1 ðtÞ; . . . ; id;N ðtÞ , etc. The
T
2 3 2 32 3 2 3
Laplace transforms of time domain signals are given as
 T
id 0 xs 0 id v d;1  v d;2
d6 7 6 76 7 1 6 7
V d ðsÞ ¼ ½V d;1 ðsÞ; . . . ; V d;N ðsÞT , Id ðsÞ ¼ Id;1 ðsÞ; . . . ; Id;N ðsÞ , etc. 4 iq 5 ¼ 4 xs 0 0 54 iq 5 þ 4 v q;1  v q;2 5: ð16Þ
dt L
i0 0 0 0 i0 v 0;1  v 0;2
Theorem 1 [21]. In symmetrically configured power networks, a
dynamic model based on dq0 signals can be described as By analogy, the model of a capacitor C is
2 3 2 32 3 2 3 2 32 3 2 3
Id ðsÞ N 1 ðsÞ jN 2 ðsÞ 0 V d ðsÞ v d;1  v d;2 0 xs 0 v d;1  v d;2 id
d6 7 6 76 7 16 7
6 7 6
4 Iq ðsÞ 5 ¼ 4 jN 2 ðsÞ N 1 ðsÞ 0 76 7
54 V q ðsÞ 5; ð11Þ 4 v q;1  v q;2 5 ¼ 4 xs 0 0 54 v q;1  v q;2 5 þ 4 iq 5; ð17Þ
dt C
I0 ðsÞ 0 0 Y bus
ðsÞ V 0 ðsÞ v 0;1  v 0;2 0 0 0 v 0;1  v 0;2 i0

and for a resistor R the model is

where Y bus ðsÞ is the frequency dependent nodal admittance matrix and
2 3 2 32 3
1  bus  v d;1  v d;2 R 0 0 id
N 1 ðsÞ :¼ Y ðs þ jxs Þ þ Y bus ðs  jxs Þ ; 6 7 6 76 7
2 4 v q;1  v q;2 5 ¼ 4 0 R 0 54 iq 5: ð18Þ
ð12Þ
1  bus  v 0;1  v 0;2 0 0 R i0
N 2 ðsÞ :¼ Y ðs þ jxs Þ  Y bus ðs  jxs Þ :
2
Any transmission network can be described in terms of these three
As an example, we will now develop the dq0 dynamic model of
basic models. Additional details may be found in [22].
a simple three-phase inductor, with an admittance matrix of
A major challenge that appears when modeling large systems
Y bus ðsÞ ¼ sL1 . According to (11) and (12),
based on dq0 quantities is to choose the reference frame, or in
2 3 2 s xs 3
0 2 V ðsÞ 3
other words, how to link machines that rotate at different frequen-
Id ðsÞ Lðs2 þx2s Þ Lðs2 þx2s Þ d cies. In this paper we follow an approach presented in [7,12],
6 7 6 76 7
4 Iq ðsÞ 5 ¼ 6 07
xs
4  Lðs2 þx2 Þ
s
Lðs2 þx2s Þ 54 V q ðsÞ 5; ð13Þ which address this problem by modeling the network and its com-
s
I0 ðsÞ 1 V 0 ðsÞ ponents using a dq0 transformation that is based on a unified (glo-
0 0 sL
bal) reference frame, rotating with a fixed frequency xs . The
and following several algebraic manipulations frequency xs is chosen as follows: if there is an infinite bus in
the system, xs is selected as the frequency of the infinite bus. If
V d ðsÞ ¼ xs LIq ðsÞ þ sLId ðsÞ; no generator is large enough to be modeled as an infinite bus, then
V q ðsÞ ¼ xs LId ðsÞ þ sLIq ðsÞ; ð14Þ xs is equal to the steady-state system frequency. In this case dq0
signals will be constant at steady-state, and the system will have
V 0 ðsÞ ¼ sLI0 ðsÞ:
well-defined equilibrium point. Other choices of xs are possible,
but the resulting dq0 models will be time varying, and therefore
will not have an equilibrium point. The dq0 signals associated with
each generator are converted from the local reference frame to the
unified reference frame. In addition, the state-space model describ-
ing the transmission network is also based on this unified refer-
ence frame. Since all models are defined in respect to the same
unified reference frame, a model of the complete power system
may be constructed.
A dq0 transformation based on xs is similar to (2), and is
obtained by substituting h ¼ xs t ¼ 2pf s t:
fdq0 ¼ T xs fabc ð19Þ
Fig. 1. A single-line diagram of a 3-bus system.
68 J. Belikov, Y. Levron / Electrical Power and Energy Systems 93 (2017) 65–74

with which is the classic swing equation. The term J is the rotor constant
2

 3 of inertia, poles is the number of machine poles (must be even), P m is
cos ðxs t Þ cos x st  3
2p
cos x st þ 3
2p
26


2p 7
the mechanical power, and K d is the damping constant. The three-
T xs ¼ 4  sin ðxs t Þ  sin x st  3
2p
 sin x s t þ 3 5: ð20Þ phase power can be computed as
3 1 1 1
2 2 2 3

P3/ ¼ v d id þ v q iq þ 2v 0 i0 : ð24Þ
A formula that allows to convert signals from h reference frame 2
to the unified frame can be derived following [12] as The combination of (23) and (24) with d ¼ /1 results in a state-
2 3 2 32 3
fd sin ðdÞ cos ðdÞ 0 ~fd space model of the simple synchronous machine given as
6 7 6 76 7
4 fq 5 ¼ 4  cos ðdÞ sin ðdÞ 0 54 ~fq 5; ð21Þ /_ 1 ¼ /2 ;


 
f0 0 0 1 f0 /_ 2 ¼ poles
2J xs
Pm  32 V e cosð/1 Þid þ sinð/1 Þiq  K d /2 ; ð25Þ
where dðtÞ ¼ hðtÞ  xs t þ p=2. The variables fd , fq are defined with v d ¼ V e cosð/1 Þ; v q ¼ V e sinð/1 Þ; v 0 ¼ 0:
respect to xs t, and ~fd , ~fq are defined with respect to h. This model has two state variables /1 ; /2 , three inputs id ; iq ; i0 , and
Next, we show two dq0-based models of the synchronous three outputs v d ; v q ; v 0 . Note that the model represents only the
machine. We start by developing the model of a simple syn- machine’s voltage source, and does not include the synchronous
chronous machine connected to an infinite bus. While such a
inductance, which is represented separately in (16) with L ¼ e Ld .
model is popular in many textbooks, we show here how to develop
it based on the unified reference frame xs t, instead of the rotor
Remark 1. The quasi-static model of the simple synchronous
angle h. This is necessary for the model to be compatible with
generator is almost identical to the model in (25), and is obtained
the network model presented in Theorem 1.
by replacing v d with RefVðtÞg and v q with ImfVðtÞg.
Assume a simplified synchronous machine represented as an
For completeness, consider now a more sophisticated (physical)
ideal voltage source behind a synchronous inductance e L d . The model of a synchronous machine, as presented in [8]. This model
machine is connected to an infinite bus, as shown in Fig. 3. Both captures the interaction of the direct- and quadrature-axis mag-
the infinite bus (on the right) and the internal synchronous netic field with the quadrature- and direct-axis mmf, respectively,
machine (on the left) are represented as voltage sources. The syn- as well as the effects of resistances, transformer voltages, field
chronous machine voltage is v ~ d ¼ 0, v~ q ¼ V e , v 0 ¼ 0, with a refer-
winding dynamics, and salient poles. The parameters are explained
ence angle of h. In this example h is the electric angle of the rotor in Table 1.
in respect to the stator. The infinite bus has a constant frequency Using (21) and omitting the algebraic details, the resulting non-
of xs , so its voltage is given by v d ¼ V g , v q ¼ 0, v 0 ¼ 0, with a ref- linear state-space model of the machine with respect to the refer-
erence angle of xs t. ence angle xs t is
Now the goal is to construct a dynamic model of the complete
/_ 1 ¼  La2 ff /1 þ /2 /5 þ La2 af /4 þ sinð/6 Þv d  cosð/6 Þv q ;
2R L 2R L
system based on dq0 signals. However, a potential problem is that
b b
the two voltage sources are defined with respect to two different
reference frames (h and xs t). To solve this, we choose xs t as a uni- /_ 2 ¼  RLqa /2  /1 /5 þ cosð/6 Þv d þ sinð/6 Þv q ;
fied reference frame for both the infinite bus and synchronous
/_ 3 ¼  RL0a /3 þ v 0 ;
machine, and construct a model of the synchronous machine using
signals in this reference frame. The synchronous machine voltage /_ 4 ¼ Lf2 af /1  Lf2 d /4 þ v f ;
3R L 2R L

is now obtained by substituting v ~ d ¼ 0, v ~ q ¼ V e , v 0 ¼ 0 in (21), b b

which leads to 3L2b 6Lff Lq

/_ 5 ¼ poles
3Laf
2 3 2 3 2J
T m þ 2 / /
1 2 þ 2 /2 4 ;
/
vd V e cos ðdÞ
2L L b q L b ð26Þ
6 7 6 7 /_ 6 ¼ /5  xs ;
4 v q 5 ¼ 4 V e sin ðdÞ 5;
ð22Þ
v0 0 id ¼ 
2Lff
sinð/6 Þ/1  L1q cosð/6 Þ/2 þ
2Laf
sinð/6 Þ/4 ;
L2b L2b
d ¼ h  xs t þ p=2:
2Lff 2Laf
iq ¼ cosð/6 Þ/1  L1q sinð/6 Þ/2  cosð/6 Þ/4 ;
In addition, the dynamic behavior of the angle d is described by L2b L2b

  3Laf
€d ¼ poles Pm  K d d_  P3/ ; i0 ¼  L10 /3 ; if ¼  /1 þ 2L d
/4 ;
ð23Þ L2b L2
2J xs b

x ¼ /5 ; d ¼ /6 ;

Table 1
Nomenclature: Synchronous machine.

kd , kq , k0 Flux linkages
kf Field winding flux linkage
v~ d , v~ q , v 0 Stator voltages
vf Field winding voltage
~i , ~iq , i Stator currents
d 0
if Field windings current
Ld , Lq , L0 Synchronous inductances
Laf Mutual inductance between the field winding and phase a
Lff Self-inductance of the field winding
Ra , Rf Armature and field winding resistance
J Rotor moment of inertia
Fig. 3. Single-phase diagram of a simple synchronous machine connected to an Te, Tm Electromagnetic and mechanical torque
infinite bus.
 J. Belikov, Y. Levron / Electrical Power and Energy Systems 93 (2017) 65–74 69

where L2b ¼ 2Ld Lff  3L2af . In this model, the state variables are The same result as above can be obtained by formal redefinition of
/1 ¼ kd , /2 ¼ kq , /3 ¼ k0 , /4 ¼ kf , /5 ¼ x, d ¼ /6 , the inputs are v d , the mapping in (19) as T xs : ½0; 2pÞ ! R23 , which translates abc
v q , v 0 , v f , T m , and the outputs are id , iq , i0 , if , x. Unlike the simple signals to dq quantities as fabc ! fdq . Let s ¼ 0, then definitions of
model presented above, this model includes the inductance terms elements N 1 ðsÞ; N 2 ðsÞ become
Ld , Lq , L0 . Since this model uses the unified reference frame xs t, it
1  bus 

can be connected to an infinite bus directly by substituting N1 ¼ Y þ ðY bus Þ ;
v d ¼ V g , v q ¼ 0, v 0 ¼ 0. 2 ð31Þ
1  bus 

N2 ¼ Y  ðY bus Þ ;
2
5. Comparison of quasi-static and dq0 models 
where Y bus ¼ Y bus ðjxs Þ, ðY bus Þ ¼ Y bus ðjxs Þ, and the last identity
Quasi-static models use time-varying phasors, and are therefore holds due to the Laplace transform conjugation property.
accurate only at frequencies that are much lower than xs . In com- According to (8) and (9), the quasi-static model can be written
parison, dq0 models are accurate over a larger frequency range. In in the dq coordinates as
this section we compare these two models for large-scale trans-   " # 
Id ðsÞ RefY bus g ImfY bus g V d ðsÞ
mission networks, and show that quasi-static models can be ¼ : ð32Þ
obtained from dq0 based models at low frequencies. We also show Iq ðsÞ ImfY bus g RefY bus g V q ðsÞ
that for any given error there exists a frequency range in which the
Comparison of the respective elements in (30) and (32) provides
quasi-static model approximates the dq0 model well.
We open with the simple example of the balanced three-phase  

RefY bus g ¼ 12 Y bus þ ðY bus Þ ;
inductor to highlight the similarities and differences between the
 
 ð33Þ
quasi-static and dq0-based dynamic models. Starting from the ImfY bus g ¼  2j Y bus  ðY bus Þ :
quasi-static model, the admittance matrix describing the inductor
is Y bus ¼ Y bus ðjxs Þ ¼ jx1s L. Using time-varying phasors and (8) and
Now, the left-hand side can be written as RefY bus g þ jImfY bus g ¼
(9), the quasi-static model is given by
Y bus as well as the right-hand side results in
1
I¼ V: ð27Þ
1  bus   j 
jxs L  
Y þ ðY bus Þ þ j  Y bus  ðY bus Þ ¼ Y bus ; ð34Þ
2 2
By taking the real and imaginary parts of this equation, and using
the relations between dq0 signals and phasors provided in (6),
meaning that the dq0 model reduces to the quasi-static model when
equivalent expressions for the quasi-static model are
s ¼ 0. h
v d ¼ xs Liq ;
v q ¼ xs Lid ; ð28Þ The lemma above can be used to realize quasi-static models in
v 0 ¼ 0: state-space form. Assume a state-space realization of a dq0 model
given by [23]
In addition, recall that the dq0 model of the inductor is given by
n_ ¼ An n þ Bn u
v d ¼ xs Liq þ L didt ; d
ð35Þ
y ¼ C n n þ Dn u;
v q ¼ xs Lid þ L didt ; q
ð29Þ
v 0 ¼ L didt : 0 and the matrices are defined as
   
Direct comparison of these equations reveals that both models are Adq 0 Bdq 0
An ¼ ; Bn ¼ ;
similar, except for the additional time derivative terms in the dq0 0 A0 0 B0
    ð36Þ
model, which describe high-frequency effects. Note that at low- C dq 0 Ddq 0
frequencies, where the time derivatives are negligible, the two Cn ¼ ; Dn ¼ ;
0 C0 0 D0
models are equivalent.
This simple example is extended by the following lemma, which where n is the state vector, the input u ¼ ½V d ; V q ; V 0 T is the dq0
shows that quasi-static models can be obtained from dq0 based transformations of the three-phase voltages, and the output
models at low frequencies. y ¼ ½Id ; Iq ; I0 T is the dq0 transformations of the three-phase injected
currents. According to Lemma 2, the dq0 and quasi-static models
Lemma 2. For s ¼ 0, the dynamic model in (11) reduces to the quasi- are equal at s ¼ 0, and therefore a time-domain description of the
static expressions in (9). quasi-static model becomes
      V d 
Id Vd
Proof. The quasi-static model relies on the assumption that the ¼ Dqs ¼ Ddq  C dq A1
dq Bdq : ð37Þ
Iq Vq Vq
system is balanced, so the zero components V 0 ; I0 are equal to
zero. On the other hand, the dq0 model only assumes that the net- Comparison of this equation to (9) reveals that
work is symmetrically configured, meaning that signals may not be " #
RefY bus g ImfY bus g
balanced. Therefore, to provide a meaningful comparison between Dqs ¼ : ð38Þ
both models, the zero component is ignored. ImfY bus g RefY bus g
Recall from Theorem 1 that dq components of the dq0 model are
Next we provide a theorem showing that the quasi-static model
given by
approximates the dq0 model well over a fixed frequency interval.
     Specifically, there always exists a frequency interval within which
Id ðsÞ N 1 ðsÞ jN 2 ðsÞ V d ðsÞ
¼ : ð30Þ the quasi-static model approximates the dq0 model with any given
Iq ðsÞ jN 2 ðsÞ N 1 ðsÞ V q ðsÞ
error.
70 J. Belikov, Y. Levron / Electrical Power and Energy Systems 93 (2017) 65–74

Theorem 3. For any given , there exists a frequency interval ond example considers the IEEE 9-bus power network. The syn-
X ¼ ½0; x  with x > 0 such that kEk22;X 6 , where chronous machines are modeled in the dq0 reference frame, and
" # are controlled using a droop-controller, which adjusts the mechan-
N 1 ðjxÞ  RefY bus g jN 2 ðjxÞ þ ImfY bus g ical power according to the rotor frequency. The third example is
EðjxÞ ¼ ; ð39Þ devoted to the small-signal stability analysis of the 118-bus net-
jN 2 ðjxÞ  ImfY bus g N 1 ðjxÞ  RefY bus g
work. The synchronous machines in this example are represented
and the finite frequency H2 norm is in the dq0 reference frame, but there is no droop control, so the
Z x X
2N X
2N
mechanical power is constant. In the 9- and 118-bus examples
1
kEk22;X ¼ jel;m ðjxÞj2 dx: ð40Þ two simulations are run in parallel – in one simulation the network
2p x l¼1 m¼1 is modeled based on dq0 quantities, and in the second simulation
the network is modeled using time-varying phasors (quasi-static
model).
Proof. Similarly to the proof of Lemma 2, we start by assuming
that signals are balanced, and therefore the zero component can
6.1. 14-bus network: open-loop
be ignored.
Let EðjxÞ ¼ ½ek;l ðjxÞ with k; l ¼ 1; . . . ; 2N denote the matrix of Consider the 14-bus network depicted in Fig. 7 below. In this
transfer functions obtained as the difference between respective figure, values of P in MW and Q in MVar represent the operating
relations of dq0 and quasi-static models point, which is obtained by solving the power flow equations.
  " #
N 1 ðjxÞ jN 2 ðjxÞ RefY bus g ImfY bus g
EðjxÞ ¼  :
jN 2 ðjxÞ N 1 ðjxÞ ImfY bus g RefY bus g
Note that, in general, the H2 norm of EðjxÞ in (41) is infinite, since
transmission power networks often contain elements that can be
modeled only by nonstrictly proper transfer functions. Furthermore,
as discussed in Section 5 the quasi-static model is represented by
the feedthrough matrix Dqs , meaning that Eð1Þ–0. This motivates
us to use the modified definition of the H2 norm restricted to the
finite frequency interval X ¼ ½0; x with x 2 Rþ as [24]
Z xX
2N X
2N
1
kEk22;X ¼ jel;m ðjmÞj2 dm:
2p x l¼1 m¼1
Observe that for stable and strictly proper systems formula (42)
coincides with the standard definition [25]. The interval X can be
S S
further divided into subintervals as X ¼ Kk¼1 Xk ¼ Kk¼1 ½xk ; xkþ1 
with 0 ¼ x1 < x2 <    < xK < xKþ1 < 1. According to definition
(42), the norm kEk22;Xk is finite and positive in each kth subinterval.
Since the intervals can be selected arbitrarily small, from the stan-
dard definition of the Riemann integral and the above remark it
immediately follows that the norm kEk22;X is monotonically increas-
ing over X ¼ ½0; x. We use this fact to show that for any given ,
there exists X ¼ ½0; x ; x > 0 such that EðjxÞ is bounded with
respect to the finite frequency H2 norm (42). This can be done as
follows. Take M large enough such that XM ¼ ½0; xM ; X  XM
and proceed toward the origin x ! 0. Then, at some moment
x ! x corresponding to kEk22;X 6  over X ¼ ½0; x . This is due
to the fact that the sequence of subintervals between 0 and xM is
now considered in the reversed order and therefore monotonically
decreases. Finally, in the special limiting case when  ¼ 0, x con-
verges to 0, corresponding to the degenerate interval X0 ¼ ½0; 0,
which is always possible by Lemma 2. h
Note that idea presented in the proof with small modifica-
tions can be applied to other common H1 and H1 norms. In
addition, for an explicit definition of Y bus ðsÞ, the presented
theorem allows to construct an upper bound for the frequency
interval ½0; x .
6. Numerical examples
This section shows several test-cases that demonstrate similar-
ities and differences between quasi-static and dq0 models. Three
networks with 9, 14, and 118 buses are considered, which param-
eters are taken from [26]. In the first example, the transient
response of the open-loop 14-bus network is addressed. The sec-
Using software available in [23] both quasi-static and dynamic
models are constructed based on dq0 quantities. The dq0 dynamic
ð41Þ
model is constructed in the minimal state-space form (35) with 78
states and 42 inputs/outputs (see Fig. 4(a)), 26 and 14 of which,
respectively, correspond to the zero component. In the considered
case, the zero signals can be safely discarded from comparison,
since the input signal is assumed to be balanced. The quasi-static
model is represented by 28  28 feedthrough matrix D as in (38),
see Fig. 4(b). Frequency of the unified reference frame is selected
to be xs ¼ 2p50 rad/s.
An array of Bode plots representing the dq0 dynamic and
quasi-static models is graphically illustrated in Fig. 5. The plots
ð42Þ depict the magnitude (in dB) and phase (in °) for two input/
output pairs. The step input of 1 kV is applied to buses 1, 2,
and the output is shown for bus 5. Figures reveal that the
frequency responses coincide at low frequencies (x ! 0) and
diverge at high frequencies. Specifically, the quasi-static model
is represented by a constant gain and phase over the whole
range of frequencies. In addition, these figures illustrate the
main concept presented in Theorem 3 that the quasi-static
model can be considered as a good approximation of the dq0
model at low frequencies.
Next, the behavior of the system is analyzed under changing
operating conditions as presented in Fig. 6. The input signals
are subsequently changed from their initial values as follows.
The d components of input voltages v d;1 and v d;2 are stepped
from 0 to 15 kV at t ¼ 0:02 s and from 0 to 10 kV at
t ¼ 0:12 s, respectively. Steps are depicted in the bottom plot
of Fig. 6. The output current dq components are measured on
bus 5, since it is common for both bus 1 and 2. Observe that
both models coincide in the steady-state as expected. In addi-
tion, the dq0 model is able to more accurately describe the
transient behavior.
6.2. 9-bus network: droop control
Consider now the standard 9-bus network. Bus 1 represents
an infinite bus, modeled by voltage source. Synchronous
machines described by more complex Eqs. (26) are connected
to buses 2 and 3. The dq0 model of the transmission network
has 18 states and 9 input/outputs. Frequency of the unified
reference frame is selected to be xs ¼ 2p50 rad/s. Simple droop
controllers based on the frequency variation are used to adjust
the input mechanical power of synchronous machines
according to
1
Pm ¼ ðxs  xr Þ þ 3Dref ; ð43Þ
Dr
 J. Belikov, Y. Levron / Electrical Power and Energy Systems 93 (2017) 65–74
Fig. 4. Sparse system matrices of the 14-bus dq0 and quasi-static models.
Fig. 5. Comparison of dq0 and quasi-static models for the 14-bus network.
Fig. 6. Comparison of transient response (top and middle plots) for current dq
components measured on bus 5. The inputs (bottom plot) are steps in d components
of voltages at buses 1 and 2.
71
where Dr ¼ ½0:2454; 0:4706  106 rad s1 =W represents the droop
control sloop parameter, Dref ¼ ½163; 85 MW is the single-phase
reference power, and xr rad/s denotes the rotor electric frequency.
We now consider two different scenarios reflecting similarities and
differences between quasi-static and dq0 models.
Scenario 1: In the first test-case the complete model of the net-
work was simulated using the nominal parameters of the con-
troller as described above. The reference power of the machine
connected to bus 2 is increased by additional 20 MW at time
t ¼ 10 s. Time domain simulations are shown in Fig. 8. In this figure
the top plot represents the active (single-phase) powers and the
bottom plot depicts the rotor electric frequency measured on bus
2. Observe that in this scenario both quasi-static and dq0 models
yield identical results.
Scenario 2: In the second test-case we change the droop control
sloop parameters Dr ¼ ½0:1227; 0:2353  104 . Time domain sim-
ulations emphasizing differences in dynamic behavior of the quasi-
static and dq0 models are shown in Fig. 9. In this figure the top plot
represents the active (single-phase) power, and the bottom plot
depicts the rotor electric frequency both measured on bus 1.
Observe that the dq0 model predicts that the system is unstable,
but the quasi-static model fails to predict this instability. This is
because the quasi-static model representing the network is inaccu-
rate at high frequencies, and fails to capture high-frequency inter-
actions between the units and the network.
6.3. 118-bus network: small-signal analysis
In this section we perform a small-signal stability analysis of
the interconnected system, which is done in several steps: first
the system operating point is computed by solving the system
power flow equations, and converting the resulting complex volt-
ages and currents to dq0 quantities. Following, unit models are lin-
earized in the neighborhood of this operating point, and the overall
system is described using state equations. The complete dq0 model
of the 118-bus network has 1273 states and 53 external input sig-
nals (mechanical powers). In this example synchronous machines
are described by simplified models (25).
Scenario 1: In the first test-case we show that the dominant
poles are determined by the machines, and coincide for the
quasi-static and dq0 models. The overall real power
consumption of all loads is increased by 50% from 4242 to
6363 MW. Table 2 provides the numeric information, in which
72 J. Belikov, Y. Levron / Electrical Power and Energy Systems 93 (2017) 65–74

Fig. 7. Single-line diagram of the 14-bus system. Values of P and Q denote the operating point (power flow solution).

Table 2
Changes in the largest eigenvalues with respect to the increase in loads.

Model Eig # Initial (4242 MW) Step (50%)

dq0 1 0.1629 0.0848

qs 1 0.1628 0.0848
dq0 2 0.2390 0.1908
qs 2 0.2389 0.1907
dq0 3, 4 0.2941 ± 314.1393j 0.2940 ± 314.1394j
qs 3 0.6200 0.4121

Fig. 8. Comparison of time domain responses. The lines correspond to quasi-static

(‘- -’) and dq0 (‘—’) models.

Fig. 10. Eigenanalysis: root locus of largest eigenvalues when active power
consumption is changed. Diamonds ( ) correspond to quasi-static and crosses
() to dq0 model.

Fig. 9. Comparison of time domain responses. The lines correspond to quasi-static
(‘- -’) and dq0 (‘—’) models.
the first four eigenvalues with the largest real parts are shown
for both quasi-static (qs) and dq0 models. Proceeding further, it
can be verified that for both quasi-static and dq0 models the
maximum allowed increase in power consumption is
 57:879%, after which systems become unstable if no control
is applied. The numeric information, presented in Table 2, is
visualized in Fig. 10. Observe that in this scenario both
quasi-static and dq0 models yield identical 1st and 2nd
dominant poles. Time domain simulations are shown in
Fig. 11 for the nominal case.
 J. Belikov, Y. Levron / Electrical Power and Energy Systems 93 (2017) 65–74 73

Fig. 11. Comparison of time domain responses (nominal case). The lines correspond
to quasi-static (‘ ’) and dq0 (‘—’) models.
Fig. 13. Comparison of time domain responses. The lines correspond to quasi-static
(‘ ’) and dq0 (‘—’) models.

Table 3 7. Conclusion
Largest eigenvalues with respect to changes in damping factor.

Model Kd Eigenvalues Quasi-static models describe dynamic processes using time-

dq0 100% 0.1629 0.2390 29.356 ± 185.681j varying phasors, often by means of the network power flow equa-
qs 0.1628 0.2389 tions. This approach has many benefits, and is used to describe
dq0 25% 0.1630 0.2390 13.916 ± 185.437j slow dynamic interactions that occur in large power systems. A
qs 0.1629 0.2389 developing solution that extends this well-known approach is to
dq0 50% 0.1631 0.2390 2.668 ± 183.11j model the system on the basis of dq0 quantities, where the main
qs 0.1630 0.2389 advantage is that dq0 models are more accurate at high
frequencies.
A current challenge is to understand the difference between
Scenario 2: In the second test-case the damping factor (K d ) of these two models for large power systems, and in the presence
the synchronous machine connected to the bus 56 is changed. of high frequency phenomena that may occur due to the increasing
The dominant poles of the quasi-static and dq0 models are similar integration of fast renewable and distributed power sources. In this
for high damping factors, but differ for a smaller damping factor. In light, the present work describes the two models using similar ter-
this case, the dq0 model predicts that the system is unstable, but minology, and provides means to compare them. The proposed
the quasi-static model fails to predict this instability. The damping lemma and simulation results show that the two models converge
factor is subsequently decreased from the nominal value by 25% at a frequency of zero, or equivalently at steady-state. Extending
and 50%. Table 3 provides the numeric information, in which the this lemma, the proposed theorem shows that for any given error
first several eigenvalues with the largest real part are shown. there always exists a frequency interval in which the quasi-static
Observe that, according to Table 3, unlike the quasi-static model, model approximates the dq0 model well. The obtained results
the dq0 model was able to detect the instability when the damping allow to estimate the frequency after which the quasi-static model
factor was decreased by 50%. This numeric information is visual- cannot accurately describe the system dynamics, and dq0 models
ized in Fig. 12, which reflects the change in location of the most should be used instead.
affected pair of complex conjugate poles. Similarly to Scenario 1 Numeric examples with 9 and 118 buses demonstrate the dif-
mechanical power P m;1 ¼ 10 MW is used to step the generator ference between the dq0 and quasi-static models. Specifically,
located on bus 1, and the current d-axis component is measured there are cases, as demonstrated in Figs. 8 and 11, in which both
on bus 62. Time domain simulations emphasizing differences in models provide identical results. This is expected in case high-
dynamic behavior of the quasi-static and dq0 models are shown frequency phenomena are not dominant, meaning (according to
in Fig. 13. Theorem 3) that the quasi-static model is accurate enough.

Fig. 12. Eigenanalysis: root locus and distribution of largest eigenvalues when damping factor is changed. Diamonds ( ) correspond to quasi-static and crosses () to dq0
model.
74 J. Belikov, Y. Levron / Electrical Power and Energy Systems 93 (2017) 65–74
However, there are other cases, as shown in Figs. 9 and 13, in
which the dq0 and quasi-static models provide different results.
Most notably, in both figures the dq0 model predicts that the
system is unstable, but the less accurate quasi-static model fails
to predict this instability.
Acknowledgment
The work was partly supported by Grand Technion Energy Pro-
gram (GTEP) and a Technion fellowship.
References
[1] Miller L, Cibulka L, Brown M, Meier AV. Electric distribution system models for
renewable integration: status and research gaps analysis [Tech. rep.]. CA
(USA): California Energy Commission; 2013. July.
[2] Stefanov PC, Stanković AM. Modeling of UPFC operation under unbalanced
conditions with dynamic phasors. IEEE Trans Power Syst 2002;17(2):395–403.
[3] Grainger JJ, Stevenson WD. Power system analysis. New York: McGraw-Hill;
1994.
[4] Anderson PM, Fouad AA. Power system control and stability. John Wiley &
Sons; 2008.
[5] Liu Y, Sun K, Liu Y. A measurement-based power system model for dynamic
response estimation and instability warning. Electr Power Syst Res
2015;124:1–9.
[6] Yousefian R, Kamalasadan S. A Lyapunov function based optimal hybrid power
system controller for improved transient stability. Electr Power Syst Res
2016;137:6–15.
[7] Ilić M, Zaborszky J. Dynamics and control of large electric power systems. New
York: Wiley; 2000.
[8] Fitzgerald AE, Kingsley C, Umans SD. Electric machinery. 6th ed. New
York: McGraw-Hill; 2003.
[9] Schiffer J, Zonetti D, Ortega R, Stanković AM, Sezi T, Raisch J. A survey on
modeling of microgrids—from fundamental physics to phasors and voltage
sources. Automatica 2016;74:135–50.
[10] Demiray T, Andersson G. Comparison of the efficiency of dynamic phasor
models derived from ABC and DQ0 reference frame in power system dynamic
simulations. In: The 7th IET international conference on advances in power
system control, operation and management, Hong Kong, China. p. 1–8.
[11] Demiray T, Andersson G, Busarello L. Evaluation study for the simulation of
power system transients using dynamic phasor models. In: Transmission and
distribution conference and exposition, Bogotá, Colombia. p. 1–6.
[12] Krause PC, Wasynczuk O, Sudhoff SD, Pekarek S. Analysis of electric machinery
and drive systems. 3rd ed. Wiley-IEEE Press; 2013.
[13] Demiray TH. Simulation of power system dynamics using dynamic phasor
models, Ph.D. thesis. TU Wien; 2008.
[14] Yang T, Bozhko SV, Asher GM. Modeling of uncontrolled rectifiers using
dynamic phasors. In: Electical system for aircraft, railway and ship propulsion.
p. 1–6.
[15] Eid A. Utility integration of PV-wind-fuel cell hybrid distributed generation
systems under variable load demands. Int J Elec Power 2014;62:689–99.
[16] Huang B, Handschin E. Characteristics of the dynamics of distribution
electrical networks. Int J Elec Power 2008;30(9):547–52.
[17] Schonardie MF, Ruseler A, Coelho RF, Martins DC. Three-phase grid-connected
PV system with active and reactive power control using dq0 transformation.
In: The 9th IEEE/IAS international conference on industry applications, Sao
Paolo, Brazil. p. 1–6.
[18] Teodorescu R, Liserre M, Rodriguez P. Grid converters for photovoltaic and
wind power systems. John Wiley & Sons; 2011.
[19] Szcześniak P, Fedyczak Z, Klytta M. Modelling and analysis of a matrix-
reactance frequency converter based on buck-boost topology by DQ0
transformation. In: The 13th international power electronics and motion
control conference. p. 165–72.
[20] Levron Y, Belikov J. Observable canonical forms of multi-machine power
systems using dq0 signals. In: IEEE international conference on the science of
electrical engineering, Eilat, Israel. p. 1–6.
[21] Levron Y, Belikov J. Modeling power networks using dynamic phasors in the
dq0 reference frame. Electr Power Syst Res 2017;144:233–42.
[22] Levron Y, Belikov J. Open-source software for modeling and analysis of power
networks in the dq0 reference frame. In: The 12th IEEE PES PowerTech
conference, Manchester, UK [in press].
[23] Levron Y, Belikov J. Toolbox for modeling and analysis of power networks in
the dq0 reference frame, MATLAB Central File Exchange, retrieved February 7,
2017; 2016 <www.mathworks.com/matlabcentral/fileexchange/58702>.
[24] Vuillemin P, Poussot-Vassal C, Alazard D. Spectral expression for the
frequency-limited H2-norm of LTI dynamical systems with high order poles.
In: European control conference, Strasbourg, France. p. 55–60.
[25] Zhou K, Doyle JC, Glover K. Robust and optimal control. Englewood Cliffs
(NJ): Prentice Hall; 1996.
[26] Zimmerman RD, Murillo-Sánchez CE, Thomas RJ. MATPOWER: steady-state
operations, planning, and analysis tools for power systems research and
education. IEEE Trans Power Syst 2011;26(1):12–9.