Equal-area criterion in power
rspa.royalsocietypublishing.org
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023
systems revisited
Research
Cite this article: Sun Y, Ma J, Kurths J, Zhan
M. 2018 Equal-area criterion in power systems
revisited. Proc. R. Soc. A 474: 20170733.
http://dx.doi.org/10.1098/rspa.2017.0733
Received: 17 October 2017
Accepted: 9 January 2018
Subject Areas:
complexity, electrical engineering,
applied mathematics
Keywords:
nonlinear dynamics, equal-area criterion,
transient stability, critical cleaning time,
swing equation
Author for correspondence:
Meng Zhan
e-mail: zhanmeng@hust.edu.cn
Electronic supplementary material is available
online at https://dx.doi.org/10.6084/m9.
figshare.c.3985332.
Yong Sun1,2 , Jinpeng Ma1,2 , Jürgen Kurths3,4 and
Meng Zhan5
1 Wuhan Institute of Physics and Mathematics, Chinese Academy of
Sciences, Wuhan 430071, People’s Republic of China
2 University of the Chinese Academy of Sciences, Beijing 100049,
People’s Republic of China
3 Potsdam Institute for Climate Impact Research (PIK),
14473 Potsdam, Germany
4 Department of Physics, Humboldt University, 12489 Berlin,
Germany
5 State Key Laboratory of Advanced Electromagnetic Engineering
and Technology, Hubei Electric Power Security and High Efficiency
Key Laboratory, School of Electrical and Electronic Engineering,
Huazhong University of Science and Technology, Wuhan 430074,
People’s Republic of China
YS, 0000-0002-8847-8337
The classic equal-area criterion (EAC) is of key
importance in power system analysis, and provides
a powerful, pictorial and quantitative means of
analysing transient stability (i.e. the system’s ability
to maintain stable operation when subjected to a
large disturbance). Based on the traditional EAC,
it is common sense in engineering that there is
a critical cleaning time (CCT); namely, a power
system is stable (unstable) if a fault is cleared before
(after) this CCT. We regard this form of CCT as
bipartite. In this paper, we revisit the EAC theory
and, surprisingly, find different kinds of transient
stability behaviour. Based on these analyses, we
discover that the bipartite CCT is only one type
among four major types, and, actually, the forms of
CCT can be diversified. In particular, under some
circumstances, a system may have no CCT or show
a periodic CCT. Our theoretical analysis is verified
by numerical simulations in a single-machine-infinite-
bus system and also in multi-machine systems. Thus,
our study provides a panoramic framework for
diverse transient stability behaviour in power systems
2018 The Author(s) Published by the Royal Society. All rights reserved.
 and also may have a significant impact on applications of multi-stability in various other 2
systems, such as neuroscience, climatology or photonics.

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
1. Introduction
Power systems are among the most complex systems in the world [1], due to their complicated
grid structures and diverse electrical (and/or power-electronic) and control devices for
generating, transferring and using power in the system. These grids include different super-
voltage levels in alternating current (AC) transmission systems, high-voltage direct current
(HVDC) transmission systems and even AC–DC hybrid power systems with the recent fast
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

development of HVDC and flexible HVDC technology. The devices include synchronous
generators, transformers, reactive power compensation and various loads. The general goal of
a power system is to transfer electricity from the generators to the loads with the aid of a power
transmission system. But this is already a difficult task, due to its intrinsic complexity on both the
spatial extension and multi-time scales, and intricate spatio-temporal dynamic interactions.
The stability of a power system is of essential importance for its proper operation [1–3].
Although it is a single problem to determine whether a system is stable or not, stability has
been classified into steady-state stability (or small-signal stability) and transient stability (or
large-signal stability) based on the disturbance magnitude and the corresponding analytical
methods; rotor-angle stability, voltage stability and frequency stability based on key instability
properties; and short-term stability, middle-term stability and long-term stability based on the
time-scale range of the problems concerned. In the dispatch and operation of a power system, its
reliability is usually assessed as often as every 5 min, primarily against three criteria: (i) (N − 1)
feasibility (namely, for any N major components, such as a major generator or transmission line,
tripped from the system, a feasible steady-state power-flow solution should persist), (ii) transient
stability (i.e. the ability of the rotor angles of major synchronous generators to maintain stable
operation when subjected to a large disturbance, such as lightning, loss of loads or generators
and three-phase short-circuit faults), and (iii) voltage stability.
Transient stability analysis for any realistic, multi-machine power system relies heavily
on extensive calculations for different large-disturbance faults, and it is often conducted in
advance and prepared off-line. When a large disturbance happens, the transient process normally
comprises three separate stages, including pre-fault (i.e. before the fault occurs), during fault (i.e.
after the fault occurs and before the protection device is switched on and the fault is cleared), and
post-fault (i.e. after the fault is cleared). Consequently, the system’s structures (or parameters)
change twice—once for the fault’s appearance and once for the fault’s clearing. Usually, the
operation state before a fault occurs is assumed to be stable. The transient stability problem for
whether the post-fault equilibrium point is stable or not is controlled by all three stages, and, from
a dynamics point of view, its stability is determined by whether the state is within or out of the
basin of attraction for the post-fault equilibrium point at the time of clearing the fault. Hence, the
associated global stability analysis is rather difficult.
The basic physical picture for transient stability, however, has been well provided by a nearly
100-year-old theory, the equal-area criterion (EAC) [4–9], in a single-machine-infinite-bus (SMIB)
power system, and it is used in lectures on power system analysis [1–3]. Based on the EAC, if the
accelerating energy (area) during the fault can be balanced by the decelerating energy (area) after
the fault is cleared, the rotor of the synchronous generator could asymptotically approach the
new stable equilibrium point of the post-fault state, and, thus, the system is (transiently) stable.
Generally, we should cut a fault as quickly as possible. The critical time during which the system
can endure a fault is referred to as the critical clearing time (CCT). If the clearing time is shorter
(longer) than the CCT, the system is considered to be stable (unstable). Based on this general
picture, the pattern of the CCT is bipartite. Meanwhile, to enhance transient stability, we should
attempt to decrease the accelerating area during the fault and/or increase the decelerating area
 after clearing the fault. Such a basic strategy has already been extensively used in stability/control
3
and in relay design in power systems [1–3]. On the other hand, as the EAC analyses the transient
stability problem on the basis of energy, but not on the trajectory directly, it has spurred an

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
active development of analytical methods in multi-machine power systems based on the transient
energy function (or direct method) [10–12], such as the potential energy boundary surface (PEBS)
method [13], the boundary of stability region based controlling unstable equilibrium point (BCU)
method [14,15] and the extended equal-area criterion (EEAC) method [16,17].
Recently, with the increasing penetration rate of renewable energy and HVDC around the
world, traditional power systems, which were dominated by synchronous generators, have been
gradually transferring to semi-conductor-dominating power systems [18], and their dynamical
behaviour becomes much more complicated and unpredictable. Several significant accidents with
obscure physical mechanisms have occurred [19,20]. Thus, the stability of power systems poses
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

an interdisciplinary challenge not only in electrical engineering, but also in basic science, and
the increased interest of the physics community in power system stability very recently is to
be expected [21–27]. Under such a circumstance, we will revisit here the traditional EAC and
especially check its applicability. To our surprise, we find that the transient behaviour actually
can be very rich, and the bipartite CCT (referred to as type I in this paper) is only one among
four major types, which also include the system having no CCT (types II and III) and a periodic
CCT (type IV). The numerical results on the SMIB model verify the validity of our theoretical
predictions, and taking into account the effects of damping and multi-machine systems (given in
the electronic supplementary material) helps to uncover some other novel types. Therefore, these
findings are promising for further applications in power electrical engineering and could advance
our understanding of the complexity of power systems.

2. Model and equal-area criterion

(a) Model
We begin with the SMIB power system, where a generator connects an infinitely large bus whose
voltage magnitude V s is constant with its angle being always 0 and unchanged. The scheme is
shown in figure 1a. According to the basic principle of a synchronous generator in power system
analysis, the motion of the rotor angle (power angle) δ of a generator with a constant voltage
magnitude E is determined by the well-known swing equations
EVs
Mδ̈ = Pm − dδ̇ − sin δ, (2.1)
X
where M = 2H/ω0 is a lumped parameter, with H the per unit inertial constant (set as 1) and
ω0 = 2π f0 (f 0 is the constant grid frequency, set at 50 or 60 Hz according to different standards),
Pm is the per unit mechanical power input, d denotes the damping impact and X = XT + XL /2 is
the total reactance consisting of the reactance of the transformer XT and transmission line XL /2,
as each line’s reactance is XL .
We set
EVs
PE (δ) = sin δ = Pe sin δ, (2.2)
X
where PE denotes the electromagnetic power output and Pe is its maximum. Thus, the swing
equation actually characterizes the rotor-angle stability by the balance between the mechanical
input power Pm from the prime mover and the electrical output power PE to the grid. As a
result, such a power system’s electromechanical stability problem has already been changed to
a mechanical one.
A rotor is always assumed to be stable before the occurrence of a fault. A large fault occurs
at t = 0 and this fault is cleared until t = tc . Here, tc denotes the clearing time. Thus, the entire
process is divided into three independent sub-processes by two times, t = 0 and t = tc , and each
sub-process is determined by the same swing equation (2.1), but with different system parameters
 (a) (b) 4

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
PB

power (per unit)

Vs
E A XL
XT Pm
B
B XL A PP
G f
PD
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

0 ds dc dmax dp¢ 180

power angle (°)

(c) (d)

PB PP or PD
power (per unit)

power (per unit)

Pm Pm
B

A
PP
PD

0 ds dcct dp¢ 180 0 180

power angle (°)
power angle (°)

Figure 1. Several schemes used in the paper. (a) Scheme for the SMIB power system, where G represents the synchronous
generator whose internal voltage is E, X T indicates the reactance of the transformer and X L the reactance of the transmission line
A (the same as B) and V s denotes the infinitely large bus voltage. (b,c) Schemes for the traditional description based on the EAC,
where Pd < Pp is usually assumed. Pm denotes the mechanical (input) power. PB , PD and PP are the electromagnetic (output)
powers of the before-fault, during-fault and post-fault states, respectively. Pb , Pd and Pp are their corresponding maximal
values. Here, the accelerating area A within the during-fault state equals the decelerating area B within the post-fault state,
with their separation position corresponding to the fault clearing time t c and the fault clearing angle δ c . When t c increases, a
CCT can be determined, as shown in (c). After that, the system would be unstable, as δ becomes larger than δp . (d) Scheme
for the power-angle relation for both Pd and Pp . The system is likely to be more stable for the post-fault state, and the post-
fault state has more stable redundancy if Pd < Pp ; it is likely to be more stable for the during-fault state otherwise. We mainly
compare the values of Pd and Pp , and accordingly uncover four major types of transient stability behaviour, including types I, II
and III for Pd < Pp , and types II, III and IV for Pd > Pp . For more details, see the text.

Pe (or X); namely,

⎧
⎪
⎨Pm − PB (δ) = Pm − Pb sin δ,
⎪ t ≤ 0,
Mδ̈ = Pm − PD (δ) = Pm − Pd sin δ, 0 < t < tc , (2.3)
⎪
⎪
⎩P − P (δ) = P − P sin δ, t ≥ tc ,
m P m p

where PB , PD and PP are the electromagnetic (output) powers of the before-fault, during-fault
and post-fault states, respectively. Pb , Pd and Pp are their corresponding maximal values. An
unchanged Pm is always assumed and d = 0, based on the fact that the transient is fast, and the
damping and the change of Pm have only limited impact comparatively.
 When a fault happens in a transmission line (t ≥ 0), as shown in figure 1a, its reactance X
5
increases, and correspondingly Pd decreases. For an extreme case, as the three-phase grounding
short-circuit fault, the reactance is infinity and thus Pd is zero. After the fault is cleared, the

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
reactance of the line would recover to a new value depending on the operation status. Usually,
this fault line (say, line B) will be cut off and the total reactance becomes XT + XL , which is
larger than the total reactance of the pre-fault, XT + XL /2. Therefore, we know that generally
Pb > Pd and Pb > Pp . Noting that, as the system is always assumed to be stable before a fault, the
condition Pb > Pm should be maintained. Consequently, Pb is maximal among all four of these key
parameters: Pb , Pm , Pd , Pp . For the usual fault analysis, Pp > Pd is further assumed, as shown by
the order of the three power-angle curves in figure 1b. However, we would like to point out that
Pp < Pd also often happens, due to various forms of faults and protection methods. For example,
in considering a single-phase short-circuit fault, which leads to a three-phase trip, the automatic
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

protection would make the system more unstable [28–30]. Actually, the main motivation of this
paper is just to study all of the different cases including not only Pp > Pd but also Pp < Pd , and
attempt to provide a panoramic framework for different types of transient stability behaviour.
Before introducing the EAC, let us identify the stable and unstable equilibrium points within
all three parameter regimes, for the before-fault (B, t = 0), during-fault (D, tc > t > 0) and post-fault
(P, t ≥ tc ) states, respectively. We use δ (δ  ) for the stable (unstable) equilibrium point. Namely, the
stable and unstable equilibrium points of the before-fault state are
 
Pm
δs = arcsin (2.4)
Pb
and  
Pm
δs = π − arcsin . (2.5)
Pb
The stable and unstable equilibrium points of the during-fault state are
 
Pm
δd = arcsin (2.6)
Pd
and  
Pm
δd = π − arcsin . (2.7)
Pd
The stable and unstable equilibrium points of the post-fault state are
 
Pm
δp = arcsin (2.8)
Pp
and  
Pm
δp = π − arcsin . (2.9)
Pp

As the system does not run on the unstable equilibrium point of the before-fault state, δs is
useless. Except for δs , we will see that all of the other five equilibrium points above play a very
important role in the transient stability analysis.

(b) Equal-area criterion

For the whole physical process of transient stability in figure 1a, before a fault, the rotor’s speed
dδ/dt equals zero and the system is stable, remaining at the equilibrium point δ s . When a
fault appears suddenly at t = 0, the mechanical power Pm exceeds the electrical power PD (δs ), as
Pb > Pd (figure 1b). Now the rotor’s speed dδ/dt > 0 and the resulting accelerating speed causes
the rotor to accelerate from δ s . Note that, because of the inertia of the rotor, the rotor angle cannot
jump abruptly. When t = tc , the fault is cleared at a rotor angle, say δ c in figure 1b, Pm < PP (δc ),
yielding a decelerating speed. But the speed is still positive, which would force the rotor angle to
continue increasing. Until a maximal angle, δ max , is reached, its speed recovers to zero, and then
 (a)
6

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
PB
1.0

power (per unit)

bPp
B
pm
aPd
0.5 A
PP

PD
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

0 ds dc 90 dp¢ 180
power angle (°)
(b) (c)
4
stability index

1 tc = 0.14
tc = 0.13
d 2

0 CCT
0 1 2 3 0 1 2 3
clearing time time

Figure 2. Type I: (a) power-angle relation of PE versus δ, (b) stability index for different clearing times showing a bipartite
behaviour and (c) time-series diagrams. In (a), the red (blue) area A (B) denotes acceleration (deceleration) energy. Based
on the EAC, we get the critical clearing angle δc ≈ 60.25◦ and the corresponding CCT ≈ 0.136, as shown in (b), where if
the fault clearing time is less than the CCT, the system could be stable, indicated by 0, or otherwise, indicated by 1. In (c),
numerical results for two different clearing times (higher or lower than the CCT) prove this point well. The parameters here are:
Pm = 0.65 and Pd = 0.6. The other two key parameters, Pb and Pp , are fixed throughout the paper: Pb = 1.3 and Pp = 0.8.
The critical conditions obtained in the paper for various patterns of CCT should be carefully examined by moving the position of
δ c . In addition, we use solid grey lines in (a) for these critical conditions from analyses, where the definitions of α and β are
in equations (3.4) and (3.8).

its angle decreases under the action of decelerating speed. Consequently, the rotor will oscillate
periodically, surrounding the post-fault equilibrium point δ p . If a positive damping (d > 0) is
considered, the rotor would eventually become stable and damped asymptotically at δ p . Under
this situation, we say that the system is transiently stable. Therefore, the whole physical picture is
quite similar to a mass point under the force of a (nonlinear) spring. The mass point is stable at the
equilibrium point δ s originally. However, under the opposite joint actions of acceleration in the
during-fault state and deceleration in the post-fault state the furthest position, δ max , is reached;
this then decreases and finally stabilizes at the new equilibrium point δ p. For this, the condition
δmax < δp is necessary.
In contrast, considering the unstable transient process, namely if the fault is cleared a little
late, the rotor angle would increase beyond δp ; δmax > δp . Thus, the negative accelerated speed
becomes positive again, the rotor angle increases further after δ = δp , and, finally, it goes to
infinity and never comes back. This situation is quite similar to a spring that has been pulled
beyond breaking point in mechanics. The critical situation for δmax = δp is schematically shown in
figure 1c. Correspondingly, a CCT and a critical clearing angle δCCT exist. Namely, if the clearing
time is less than the CCT, the system is stable; otherwise it is unstable. This bipartite pattern of
CCT is quite usual and well known, and is referred to as type I in the paper, as shown in figure 2.
So far, the physical process of transient stability has become clear. Below let us shortly
introduce the classic EAC method, which deals with the same problem in a more efficient manner
 [1–3]. Multiplying each side of equation (2.1) by dδ/dt and setting d = 0, we obtain
7
2
dδ d δ dδ
M = (Pm − PE (δ)) , (2.10)

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

dt dt2 dt

...................................................
and further
 2 δ2  δ2
1 dδ 
M  = (Pm − PE (δ)) dδ. (2.11)
2 dt  δ1 δ1

Therefore, the motion of the rotor angle is simply determined by the conversion between
kinetic and potential energy, which represents the left and right sides of the above equation,
respectively. At t = 0, δ = δs and the kinetic energy is zero. Owing to the fault, the potential
energy stored in the rotor will be transformed to kinetic energy by acceleration during the
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

fault (tc > t > 0). At t = tc , when the fault is cleared, the kinetic energy at δ c is maximal. It will
further change back to potential energy due to the decelerating action of the post-fault state.
Finally, at δ = δ max , the kinetic energy becomes zero again. Next, the rotor angle will decrease and
return to its original value, if δmax < δp . As a result, we use the integration in equation (2.11) to
denote the transformation between kinetic and potential energy. Here, a positive (negative) value
indicates acceleration (deceleration) energy. In figure 1b, we use the red area A (SA ) to denote the
acceleration energy and blue B (SB ) to denote the deceleration energy. According to the law of
energy conservation (d = 0),
S A = SB , (2.12)

which is exactly where the EAC comes from.

This equality also works for the critical situation in figure 1c. Hence, we have
 δCCT  δ p
(Pm − Pd sin δ) dδ + (Pm − Pp sin δ) dδ = 0, (2.13)
δs δCCT

yielding

[Pm (δ  p − δs ) − Pd cos δs + Pp cos δ  p ]

δCCT = arccos . (2.14)
Pp − Pd
The CCT is exactly the time at which the rotor angle reaches δ CCT during the fault.
Apparently, the EAC provides a powerful, pictorial and quantitative means of analysing the
complicated transient stability problem. Based on this theory, it is easy to obtain δ CCT and further
the corresponding CCT. As an example, a typical parameter set for type I is chosen in figure 2.
We define a useful concept, the stability index SI(t),

0, stable,
SI(t) = (2.15)
1, unstable.

The result as a function of clearing time is shown in figure 2b, where the system is stable when
t < CCT, and unstable otherwise; the bipartite pattern of CCT is obvious. In addition, based on
the EAC, we obtain the critical clearing angle δCCT ≈ 60.25◦ and the corresponding CCT ≈ 0.136.
These predictions have been verified in figure 2c by comparing the stable and unstable behaviour
at t = 0.13 and t = 0.14 for t < CCT and t > CCT, respectively.
However, one may ask what the application conditions for the traditional EAC and the
associated bipartite pattern of CCT are. At least we know that Pp > Pd must be maintained. Under
Pp > Pd , compared with the during-fault state, the post-fault state has a greater stability margin
and is more stable. This implies that a fault should be cleared as quickly as we can. Otherwise,
something unusual may happen. Note that the case for a periodic pattern of CCT (called type IV in
this paper) under Pd > Pp has already been reported [30], where the fault has to be cleared during
the periodic windows to realize system stability. So far, however, we do not know its condition or
any potential connection with the traditional bipartite CCT. Therefore, it becomes very important
to conduct an exhaustive study for all parameters Pb , Pm , Pd , Pp and analyse all possibilities. To
emphasize the importance of the relation between Pd and Pp , we add one illustration in figure 1d.
 (a) III¢
1.2 8

aPd

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
IV

0.8 *

Pd
II I III
0.4 bPp
aPd
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

0 0.4 0.8 1.2

Pm

(b)
III¢(c)
III¢(a)
1.2

f
Pd

aPd
1.0 IV
III

bPd III¢(b)

0.8 *
bPp 0.8 1.0 1.2
Pm

Figure 3. Phase diagram for (a) different types of transient stability behaviour with variation of the mechanical power, Pm ,
and the maximum during-fault electrical power, Pd , and (b) its enlargement to show the details of types III, III and IV. Type III
includes three subtypes: types III (a), III (b) and III (c), as shown in (b). Roughly, the whole phase diagram is divided by two
curves: Pm = αPd and Pm = βPp . The critical parameter set (Pm = βPp ≈ 0.704, Pd = 0.8 = Pp ) serving as the organizing
centre is emphasized by a star, and all parameters chosen in the paper for different typical types are indicated by solid circles.
Except for the curve of f obtained by numerics, all critical curves in the phase diagram are obtained from theoretical analyses
and confirmed by numerical simulations. For more details, see the text.

Here, we intend to revisit transient stability with the aid of the EAC, with all four key
parameters changeable: Pb , Pm , Pd , Pp . As Pb is always maximal, Pb = 1.3 is chosen and fixed
throughout this paper. We further fix Pp = 0.8, without losing generality. Thus, only Pm and Pd
are free; Pm < Pb = 1.3 and Pd < Pb = 1.3. We believe that all phenomena reported in the paper are
very generic and can be observed with other choices of parameter sets as well.
The phase diagram in figure 3a shows the results for all types of transient stability behaviour on
the (Pm – Pd ) parameter plane. It includes four major types of CCT, namely type I for the traditional
bipartite CCT, type II for an always-stable behaviour, type III for an always-unstable behaviour
and type IV for a periodic CCT. Both newly found types II and III have no specific CCT, and they
correspond to a slightly destroyed fault and a deeply destroyed fault, respectively. In addition,
we find a subtype of type III, called type III , which indicates that the system is stable if the fault
is not cleared and, on the contrary, unstable if the fault is cleared. Type III appears for larger
 values of Pm and Pd , and can be further classified into three subtypes: type III (a), III (b) and III (c),
9
shown in figure 3b as a zoom-in picture of figure 3a. We will study them in detail. All division
parameter curves in the phase diagram and the occurrence condition for each type of behaviour

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
have been well studied. Thus, figure 3 provides a panoramic framework for diverse transient
stability behaviour. Based on this observation, we infer that the traditional type I is only typical
when it appears in a relatively narrow parameter region for smaller Pm and Pd , but additionally
more complicated phenomena exist for other system parameters. All of these will form the basis
of our investigations in the following analytical section.

3. Analytical results
(a) Pd < Pp
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

We start from the condition Pd < Pp = 0.8 (the lower part of figure 3a) for the survival of only types
I, II and III.
Type II. Type II stands for an always-stable behaviour. Namely, whether we clear the fault or
not, the system is deemed to be stable. As Pd < Pp , the during-fault state has a smaller stability
margin than the post-fault state. If the system is stable during the fault, it would always be stable.
Thus, we should focus on the during-fault state and its corresponding power-angle curve (PD
versus δ). According to the EAC, the maximal accelerating area is SA,max (red area in figure 4a),
and the minimal decelerating area isSB,min (blue area). Clearly, if

SA,max ≤ SB,min , (3.1)

then the system will always be stable whether the fault is cleared or not. We may move the
clearing angle δ c for different clearing times tc to understand this point well. Inequality (3.1)
further produces
 δd  δ d
(Pm − Pd sin δ) dδ ≤ (Pd sin δ − Pm ) dδ, (3.2)
δs δd

reaching the critical condition for type II:

cos δs − cos δ  d
Pm ≤ Pd = α(Pm , Pb , Pd )Pd , (3.3)
δ  d − δs
with
cos δs − cos δ  d
α(Pm , Pb , Pd ) = . (3.4)
δ  d − δs
Clearly, α is a function of Pd . According to the definition of δs and δd : 0 < δs < π /2 and π/2 <
δd< π , we have 0 < α < 1.
The region for type II and its corresponding critical boundary (Pm = αPd ) under Pd < Pp = 0.8
are shown in figure 3a. As the system is always stable, the stability index is fixed at 0 (figure 4b)
and several stable motions for different clearing times in figure 4c prove this point well.
Type III. Similarly, let us analyse the other extreme case, where the system is always unstable
whether we clear the fault or not. Again the during-fault state has a smaller stability margin than
the post-fault state. But if the system is unstable for the post-fault state, it would never be stable.
Therefore, now we should focus on the post-fault state and its corresponding power-angle curve.
Again according to the EAC, the minimal accelerating area is SA,min (red area in figure 5a), and
the maximal decelerating area is SB,max (blue area). Clearly, if

SA,min ≥ SB,max , (3.5)

we have
 δp  δ p
(Pm − Pp sin δ)dδ ≥ (Pp sin δ − Pm )dδ, (3.6)
δs δp
 (a)
10

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
PB

power (per unit)

1.0

PP
aPd
0.5 B
Pm
A
PD
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

0 ds dd 90 dd¢ 180
power angle (°)
(b) (c)
4
stability index

1
2 tc = 0.5 tc = 1.0 tc = 1.5
d
0
0 1 2 3 0 1 2 3
clearing time time

Figure 4. Similar to figure 2, but type II (Pd < Pp ) is considered instead. In this case, even during the fault the maximal
accelerating area A is less than the minimal decelerating area B. Thus, the system is always stable whether the fault is cleared
or not, and we have no CCT. The parameters here are: Pm = 0.45 and Pd = 0.6.

and further
cos δs − cos δ  p
Pm ≥ Pp = β(Pm , Pb , Pp )Pp , (3.7)
δ  p − δs

with
cos δs − cos δ  p
β(Pm , Pb , Pp ) = . (3.8)
δ  p − δs

The expressions for β and α are similar, except that β is a function of Pp , whereas α is a function
of Pd . Similarly, 0 < β < 1. Again, we may move the clearing angle δ c for various possibilities to
understand this point well. Similarly, as the system is always unstable, the stability index is fixed
as 1 (figure 5b) and several unstable trajectories for different clearing times are given in figure 5c.
The region for type III and its corresponding critical boundary with type I (Pm = βPp ) are shown
in figure 3a. So far, β depends on Pp (Pp = 0.8) and Pm . Solving inequality (3.7), we have the critical
condition Pm = βPp ≈ 0.704, indicating that the critical boundary is a vertical line. In addition,
observing the two critical curves for Pm = αPd and Pm = βPp , we find that they meet at Pd = Pp =
0.8. To emphasize this, we superimpose a star at the crossing point (Pm = 0.704, Pd = 0.8) at the
organizing centre in figure 3a. In addition, the parameters used in this paper for various types are
indicated by solid circles.
Type I. Let us look back at the traditional type I for a bipartite CCT (figure 2). If the system
cannot always be stable during the fault, as in type II, and cannot always be unstable when a fault
is cleared immediately, as in type III, there should be a CCT, indicating that the fault should be
cleared as quickly as possible for the system’s stability. So, types III and II are two limiting cases
giving the upper and lower bounds for type I. Namely, the condition for type I is

αPd < Pm < βPp . (3.9)

 (a)
11

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
PB

power (per unit)

1.0

Pm B
A PP bPp
0.5
PD
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

0 ds dp 90 dp¢ 180
power angle (°)
(b) (c)
4
stability index

1 tc = 0.5, 1.0, 1.5

2
d
0
0 1 2 3 0 1 2 3
clearing time time

Figure 5. Similar to figure 2, but type III (Pd < Pp ) is considered instead. In this case, the fault is cleared immediately at t = 0.
The minimal acceleration energy A is larger than the maximal deceleration energy B for the post-fault state. This indicates
that the system is always unstable once a fault appears. Under this situation, we also have no CCT. The parameters here are:
Pm = 0.75 and Pd = 0.6.

The region for type I, surrounded by regions of types II and III, is shown in figure 3a. The
corresponding illustration is given in figure 2, which again proves our analysis well. It is notable
that, despite the introduction of type I and the associated EAC theory (figures 1 and 2) abounded
in the textbooks [1–3], its appearance condition (inequality (3.9) and Pd < Pp ) has never been
carefully analysed before, to the best of our knowledge.
Now, we know that, under Pd < Pp , only types I, II and III are possible. For Pd > Pp , more
complicated phenomena may happen, as we will see in the sequel to this paper.

(b) Pd > Pp
Under this situation, types II, III and IV happen, accompanied by the deformation of type III, type
III and their subtypes, which will be described in detail next.
Type II. Because now Pd > Pp , the stability of the system for a post-fault state is worse than
that for the during-fault state. The worst situation is that the system runs at the post-fault state
immediately when a fault occurs (tc = 0). Thus, if the system can remain stable during such a
situation, it will always be stable. Different from type II Pd < Pp in figure 4, now the power-angle
curve of the post-fault state should become the focus of attention. Again according to the EAC,
if the maximal acceleration energy SA,max is less than the minimal deceleration energy SB,min in
figure 6a, the system will always be stable, and we obtain its condition
 δp  δ p
(Pm − Pp sin δ) dδ ≤ (Pp sin δ − Pm ) dδ, (3.10)
δs δp

and further
cos δs − cos δ  p
Pm ≤ Pp = β(Pm , Pb , Pp )Pp . (3.11)
δ  p − δs
 (a)
12

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
1.0

power (per unit)

PB
bPp
B
Pm
A PD
0.5
Pp
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

0 ds dp 90 dp¢ 180
power angle (°)
(b) (c)
4
stability index

1
2 tc = 0.5, tc = 1.0
d
0 tc = 1.5
0 1 2 3 0 1 2 3
clearing time time

Figure 6. Similar to figure 2, but type II under Pd > Pp is considered instead. Considering the worst case for an immediate
clearing of a fault, the maximal accelerating area A is always less than the minimal decelerating area B for the post-fault state.
Thus, the system is always stable whether the fault is cleared or not. Different from figure 4 for type II at Pd < Pp , here the
post-fault state should be studied. Again, we have no CCT. The parameters here are: Pm = 0.65 and Pd = 0.9.

Again to understand this condition well, we should move the clearing angle δ c for different
tc values. Clearly, the plot in figure 6a corresponds to the worst case situation under tc = 0 and
δc = δs .
Thus, the critical straight line Pm = βPp ≈ 0.704 is the same as that for types I and III under
Pd < Pp , as shown in figure 3a. Similar to figure 4, the illustrations of the EAC, the stability index
and several trajectories for different clearing times in figure 6 prove the correctness of our analysis.
Type III. Now, type III under Pd > Pp happens when the system cannot be stable all the time.
We have to consider the best situation for the during-fault state. If the system cannot be stable
during the fault, it will be unstable forever. Based on the condition that the minimal accelerating
area is larger than the maximal decelerating area, i.e. SA,min ≥ SB,max in figure 7a, the condition
for type III is
 δd  δ d
(Pm − Pd sin δ) dδ ≥ (Pd sin δ − Pm ) dδ, (3.12)
δs δd

or equivalently
cos δs − cos δ  d
Pm ≥ Pd = α(Pm , Pb , Pd )Pd . (3.13)
δ  d − δs
Clearly, the critical condition Pm = αPd is the same as that for the division of types I and II under
Pd < Pp , as shown in figure 3a, in which we use a dotted line to indicate this condition under
Pd > Pp = 0.8. The corresponding results are shown in figure 7. Similar to figure 5, the stability
index is always 1 in figure 7b, and the trajectories for different clearing times all move outwards
in figure 7c.
Type III . So far, we have obtained the conditions for types II and III under Pd > Pp . For type
III, the system cannot be stable whether we clear the fault or not. Here, we like to introduce a
deformation of type III, type III , which denotes that the system can be stable during the fault,
 (a)
13

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
1.0

power (per unit)

PB
Pm B
A aPd
PD
0.5
Pp
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

0 ds dd 90 dd¢ 180
power angle (°)
(b) (c)
4
stability index

1 tc = 0.5, 1.0, 1.5

2
d
0
0 1 2 3 0 1 2 3
clearing time time

Figure 7. Similar to figure 2, but type III under Pd > Pp is considered instead. Considering the best case for an immediate
clearing of a fault, the minimal accelerating area A is always larger than the maximal decelerating area B for the during-fault
state. Thus, the system is always unstable once a fault appears. Different from figure 5 for type III at Pd < Pp , here the curve of
PD should be studied. The parameters here are: Pm = 0.82 and Pd = 0.85.

but it becomes unstable when the fault is cleared. Moreover, for the remaining parameter region
for larger Pm and Pd (βPp ≤ Pm < αPd ) in figure 3b, type III can be further subdivided into three
subtypes: types III (a), III (b) and III (c).
For type III (a), when Pm ≥ Pp , there is no stable equilibrium point for the post-fault state, i.e.
the horizontal line Pm has no intersection with the power-angle curve of PP , and, thus, the system
cannot be stable once a fault is cleared. The condition is

Pp ≤ Pm < αPd . (3.14)

The region for type III (a) is shown in figure 3b, and the corresponding scheme for this situation
is shown in figure 8a.
For type III (b), the system becomes unstable after the fault is cleared. One possibility is that
the clearing angle is beyond δp , although the system can be stable during the fault. Based on
the condition for SA,min ≥ SB,max for the during-fault state, but for the maximal possible clearing
angle δp instead, namely
 δd  δ p
(Pm − Pd sin δ) dδ ≥ (Pd sin δ − Pm ) dδ, (3.15)
δs δd

we have
cos δs − cos δ  p
Pm ≥ Pd = β(Pm , Pb , Pp )Pd . (3.16)
δ  p − δs

Further combining the conditions for the system being stable during the fault, Pm < αPd , and
having a stable equilibrium point for the post-fault state, Pm < Pp , we get the formal condition
 (a) (b) 14
III¢(a) III¢(b)

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

PB PB

...................................................
power (per unit) 1.0 aPd 1.0
B min (Pp, aPd)
Pm B
A Pp Pm
PD A bPd
PD
0.5 0.5
Pp Pp

0 ds dd dmax dd¢ 180 0 ds dd dp¢ dd¢ 180

Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

power angle (°)

(c) (d)
III¢(c)

PB
power (per unit)

1.0
Pp stability index
Pm 1
f
0.5 PD
Pp 0
0 1 2 3 inf
clearing time
0 dsd1d2 90 dp¢ 180
power angle (°)

Figure 8. Similar to figure 2, but type III including types III (a), III (b), and III (c) under Pd > Pp is considered instead. Here,
the system is stable for the during-fault state, if the fault is never cleared, but it becomes unstable immediately once the fault
is cleared. These special cases have been further classified into (a) type III (a), where there is no stable equilibrium point for the
post-fault state, (b) type III (b), where the system cannot be stable for the post-fault state, and (c) type III (c), where the rotor
angle δ1 cannot reach δ2 . To emphasize the difference from type III in figure 7, a star is superimposed in (d) at an infinitely long
clearing time, to show that the system is stable during the fault, if the fault is never cleared. Correspondingly, the parameters
are: (a) Pm = 0.85 and Pd = 1.0, (b) Pm = 0.79 and Pd = 0.88, and (c) Pm = 0.75 and Pd = 1.2.

for type III (b),

βPd ≤ Pm < min(αPd , Pp ). (3.17)

Note that different from the critical conditions for the usual types II and III for either Pm =
α(Pd )Pd or Pm = β(Pp )Pp , the condition Pm ≥ β(Pp )Pd in inequality (3.16) has combined the
information of both Pd and Pp . The parameter region for type III (b) is shown in figure 3b and
its corresponding scheme is shown in figure 8b.
For type III (c), the final possibility for the system being stable during the fault and unstable
after clearing the fault, the final (stable) during-fault state may be unable to be connected with the
initial (stable) post-fault state. Namely,
 δ1 ⎫
⎪
(Pm − Pd sin δ) dδ ≤ 0, stable from δs to δ1 for the during-fault state, ⎪
⎪
⎪
δs ⎪
⎪
⎪
⎬
 δ p
 (3.18)
(Pm − Pp sin δ) dδ ≤ 0, stable from δ2 to δ p for the post-fault state, ⎪
⎪
δ2 ⎪
⎪
⎪
⎪
⎪
⎭
δ2 > δ1 , i.e. δ1 never reaches δ2 .
 (a)
15

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

PB

...................................................
1.0

power (per unit)

min (Pp, bPd, f)
Pm
PD bPp
0.5
Pp
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

0 ds dc(1,2) dmax dp¢ 180

power angle (°)

(b) dmax ds (c)

4
stability index

1.0 tc = 0.1 tc = 0.6

d 2 tc = 0.9

0 d dc2 tc = 0.3
c1
0 t1 t2 t3 1 2 0 1 2 3
clearing time time

Figure 9. Similar to figure 2, but type IV for a periodic CCT under Pd > Pp is considered instead. In (a), the red (blue) denotes
the accelerating (decelerating) area. The accelerating area and the decelerating area are equal. In (b), a pattern of periodic CCT
appears. δc,1 = δc,2 ≈ 72.51◦ , corresponding to t1 ≈ 0.23 and t2 ≈ 0.46. t3 ≈ 0.69 (δs ≈ 35.23◦ ), δmax ≈ 84.2◦ . Within the
first periodic window of CCT, 0.23 ≈ t1 < t < t2 ≈ 0.46, the system is stable as the fault has been properly cleared, as shown
in (c). The parameters here are: Pm = 0.75 and Pd = 0.9. See the text for more details.

As the above condition depends on all four system parameters, we may denote it by
f (Pm , Pb , Pd , Pp ). Therefore, the condition for type III (c) is expressed as

f ≤ Pm < Pp . (3.19)

A further analytical study for f is unavailable and we have to resort to numerics. Finally, the
region for type III (c) is presented in figure 3b and the corresponding scheme is given in figure 8c.
Clearly, type III is similar to type III, with the only difference being that the system can be stable
during the fault if the fault is never cleared. To emphasize this difference, we superimpose a star
for the stable behaviour at an infinitely long clearing time in figure 8d, to indicate such an unusual
stable behaviour.
Type IV. Now let us move on to the final type of behaviour, type IV. As shown in the phase
diagram in figure 3, type IV exists under the condition

βPp < Pm < min(Pp , βPd , f ). (3.20)

The corresponding results for type IV are shown in figure 9. To our surprise, now the stability
index for CCT is periodic, namely we should properly clear the fault within the periodic windows
of the clearing time. Otherwise, the system would be unstable. The first periodic window appears
within (t1 , t2 ). Specifically, t1 ≈ 0.23 and t2 ≈ 0.46, corresponding to δc,1 = δc,2 ≈ 72.51◦ . The time
for the system returning to δ s after the first periodic window is t3 ≈ 0.69; δs ≈ 35.23◦ . In addition,
the maximal angle without clearing the fault is δmax ≈ 84.20◦ . The parameters are: Pd = 0.9 and
Pm = 0.75, chosen within the region of type IV. Several numerical results in figure 9c confirm this
phenomenon well. Note that type IV has been reported as a peculiar behaviour before [30]. Here,
 we put it in a much broader picture and show its extensive observability for system parameters
16
that have been properly chosen.
As type IV is located outside of type III under Pd > Pp , the system could be stable during the

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
fault if the fault is never cleared. Moreover, the later the fault is cleared, the more decelerating
energy the system can accumulate. The scheme for type IV is given in figure 9a. Different from
the pattern for type III (c) in figure 8c, here the final during-fault state can be connected with the
initial post-fault state. In addition, different from the pattern for type I under Pd < Pp in figure 2,
here we should clear the fault after a certain CCT. Again based on the EAC and the scheme in
figure 9a, we can calculate the critical clearing angle δ c from
 δc  δ p
(Pm − Pd sin δ)dδ + (Pm − Pp sin δ)dδ = 0, (3.21)
δs δc
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

and further

[Pm (δ  p − δs ) − Pd cos δs + Pp cos δ  p ]

δc = arccos . (3.22)
Pp − Pd

This expression is the same as δCCT in equation (2.14) for type I, with the key difference that
now we have to cut the fault after δ c . Hence, in figure 9b, the stability index drops from 1 to 0,
when the first CCT window arrives.
Next let us examine the periodicity of CCT in type IV. The first critical clearing angle δ c is
denoted by δ c,1 and its corresponding CCT as t1 . If we do not clear the fault at δ = δ c , the rotor
angle will increase and move forward to δ max (from left to right in figure 9a) and then move back
(from right to left). During this process, we can clear the fault before δ = δ c,2 to make it stable.
The corresponding CCT for the stability index jumps from 0 to 1 at t = t2 , and the first window of
CCT appears at t1 < t < t2 . Here, δc,1 = δc,2 = δc , but t1 = t2 . For the occurrence of the next periodic
window of CCT, we have to wait for the rotor to move further back and accelerate again for the
next periodic round during the fault. This phenomenon is quite similar to the periodic motion of
a mass point connected with a spring.

4. Numerical results
Above, we have studied various transient stability behaviours and classified them into four major
types: including types I, II, III (where III is regarded as a subtype of III) and IV. Now, we perform
some numerical analyses on the basis of the basin of attraction, i.e. we calculated the basins of
these different states [29]. The numerical technique to evaluate the basins of attraction is quite
standard, and is based on an exhaustive search of all initial conditions which asymptotically
approach the attractor. For example, we choose (Pm = 0.65, Pd = 0.6), (Pm = 0.65, Pd = 0.9),
(Pm = 0.75, Pd = 0.6) and (Pm = 0.75, Pd = 0.9) in figure 10a–d for these four different types,
respectively. The parameters are the same as those used before. Again, the parameters (Pb =
1.3, Pp = 0.8) are unchanged. As the operation state of the post-fault state is only determined
by Pm with Pp fixed, the basins of the post-fault state, as shown by the white areas in figure 10a,b,
are the same, and the basins of the post-fault state in figure 10c,d are the same. In addition, we
plot the transient trajectories on all these subfigures. One can clearly see that in figure 10a a CCT
exists for type I; below the CCT, the system will be within the stable region, and, in contrast,
above the CCT, the system will move out of the basin and become unstable. Thus, a bipartite
CCT is expected. In figure 10b for type II, as the transient trajectory is always within the basin, the
system is stable forever; this represents a slightly destroyed fault, although the nonlinear effect
of transient stability still needs to be considered. On the contrary, in figure 10c for type III, as the
transient trajectory is always outside the basin, the system is unstable forever; this might represent
a deeply destroyed fault, for which any protection devices would not save the system. Figure 10d
for type IV shows an obvious periodic CCT. We can make the system stable within the basin
region (to the right of δ c ) and unstable out of the region (to the left of δ c ); δc,1 = δc,2 = δc ≈ 72.51◦ ,
δmax ≈ 84.20◦ and δs ≈ 35.23◦ . All these results fit with the analysis in figure 9 well.
 10 10
(a) (b) 17

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
w 0 w 0

(Pm = 0.65, Pd = 0.6) (Pm = 0.65, Pd = 0.9)

–10 –10
–2 0 2 –2 0 2
d (p/180°) d (p/180°)
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

10 10
(c) (d )
dc1

w 0 w 0 ds dmax

dc2
(Pm = 0.75, Pd = 0.6) (Pm = 0.75, Pd = 0.9)
–10 –10
–2 0 2 –2 0 dc1 = dc2 2
d (p/180°) d (p/180°)

Figure 10. Basin of attraction analysis for transient stability, with white (blue) for the stable (unstable) domain of the post-fault
operation state. (a) Type I with Pm = 0.65, Pd = 0.6. (b) Type II with Pm = 0.65, Pd = 0.9. The basins of (a) and (b) are the
same, as the value of Pm is unchanged; Pm = 0.65. (c) Type III with Pm = 0.75, Pd = 0.6. (d) Type IV with Pm = 0.75, Pd = 0.9.
Similarly, the basins of (c) and (d) are the same. Clearly, for type I in (a), the critical angle δc = 60.25◦ ≈ 1.05 corresponds to
a single value of CCT. In (b) and (c), for types II and III, there is no CCT; the system is always stable and unstable, respectively. In
(d), a pattern of periodic CCT occurs, based on the periodic motion moving in and out of the basin of attraction.

We summarize all these major types in table 1, in which the patterns of the stability index
and the applicability conditions are listed. All these suggest that we have grasped the prominent
properties of transient behaviours and the key physical and dynamical mechanisms in the
simplest SMIB power system, on the basis of traditional EAC theory. In addition, we present
numerical results for non-zero damping and multi-machine power system effects in the electronic
supplementary material.

5. Discussion
In conclusion, we have revisited the classic EAC theory, which is nearly 100 years old but is a
root for transient stability analysis and relevant protection strategies in power systems. Different
from the traditional picture of a bipartite pattern of CCT, type I, here we discover abundant
further types of CCT. Based on our study, the whole parameter region is decomposed into the
four major different types, including types I, II, III and IV, accompanied by several deformations;
such a classification, summarized in table 1 and figure 3, is complete. Thus, our study provides
a panoramic framework for diverse transient stability behaviour and suggests that, even in the
simplest SMIB power system, the transient behaviours are more complicated than previously
thought. In addition, all four major types have also been extensively observed in multi-machine
power systems, i.e. these are prototypes.
 Table 1. Illustration of various types of transient stability and their corresponding patterns of CCT and occurrence conditions,
18
based on the EAC. Basically, there are four major types of transient stability: types I, II, III (including III ) and IV, in which type I
is typical and generally accepted as common sense. In addition, for some other special cases, such as considering the effects of

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
non-zero damping and multi-machine systems, types I and IV  are possible (see the electronic supplementary material for more
details).
..........................................................................................................................................................................................................

illustration
type of
normal case special case
transient
stability Pd £ Pp Pd > Pp

I¢
stability index

stability index
1 1
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

0 0
clearing time clearing time
aPd < Pm < bPp exist in small damping
stability index

stability index

II 1 1

0 0
clearing time clearing time
Pm £ aPd Pm £ bPp

III¢
stability index

stability index

1
stability index

1 1
III

0 0 0 *
clearing time clearing time clearing time inf
Pm ≥ bPp Pm ≥ aPd min (Pp,bPd, f ) £ Pm < aPd
′
stability index

stability index

IV 1 1

0 0
clearing time clearing time
bPp < Pm
exist in a multi-machine system
< min (Pp, bPd, f )
..........................................................................................................................................................................................................

For the significance of our work, we emphasize the following aspects. (i) The work enhances
our understanding of the physical mechanism of transient stability, still on the basis of the
traditional EAC method. (ii) For the calculation of transient stability in the dispatch and operation
of power systems, because a bipartite pattern of CCT is always assumed a priori and the key
objective is to obtain a specific single value of CCT, some fast algorithms have been proposed.
Here, our novel findings suggest that we have to re-examine these algorithms carefully. (iii)
Similarly, all relay protection methods in power electrical engineering are based on the belief
 that, if a fault happens, we should clear it as fast as we can. Now we know that this works only
19
for the case of type I. We believe that the novel findings, such as types II, III, IV and III , may
have an important impact on possible improving strategies in the operation of power systems.

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
(iv) Finally, as transient stability and the relevant basin of attraction are of general interest, we
expect that our observations could be valuable for global stability problems in diverse multi-
stable systems [31]. As some equations intended to describe the dynamics of electrical systems
are very general and similar to equations used in several other disciplines [32,33], such as those
describing the equilibrium of planar beams under large deformation, a forced pendulum, a phase-
forced loop and a Josephson junction, we expect that these innovative results could indeed have
a multi-disciplinary impact.
Data accessibility. This paper has no associated data.
Authors’ contributions. M.Z. and Y.S. conceived and designed the work. Y.S. conducted the computing method.
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

Y.S., J.M. and M.Z. deduced the theoretical inference. Y.S., M.Z. and J.K. analysed the results. All authors
provided ideas and interpreted the results, and wrote the manuscript.
Competing interests. We have no competing interests.
Funding. M.Z. wishes to acknowledge support given to him from the National Key R&D Program of China and
from the National Nature Science Foundation of China (NSFC) through grant no. 11475253.
Acknowledgements. We thank Li Fan for her encouragement and helpful discussions. We thank the editor and
the two referees very much for reviewing our article.

References
1. Bergen AR, Vittal V. 1986 Power systems analysis. Upper Saddle River, NJ: Prentive Hall.
2. Kundur P. 1994 Power system stability and control. New York, NY: McGraw-Hill.
3. Anderson PM, Fouad AA. 2002 Power system control and stability, 2nd edn. Upper Saddle River,
NJ: IEEE Press.
4. Park RH, Bancker EH. 1929 System stability as a design problem. AIEE Trans. 28, 170–193.
(doi:10.1109/T-AIEE.1929.5055195)
5. Dahl OGC. 1935 Stability of the general 2-machine system. AIEE Trans. 54, 185–188.
(doi:10.1109/T-AIEE.1935.5056954)
6. Skilling HH, Yamakawa MH. 1940 A graphical solution of transient stability. Elec. Eng. 59,
462–465. (doi:10.1109/EE.1940.6435162)
7. Prescott JC. 1952 The transient stability of alternators: a graphical solution of the two-machine
case. Proc. IEE IV: Inst. Monogr. 99, 367–371. (doi:10.1049/pi-4.1952.0037)
8. Rahimi A. 1990 Generalized equal-area criterion: a method for on-line transient stability
analysis. In Proc. of IEEE Int. Conf. on Systems, Man, and Cybernetics, Los Angeles, CA, 4–7
November 1990, pp. 684–688. New York, NY: IEEE.
9. Haque MH. 1994 Equal-area criterion: an extension for multi-machine power systems. IEE
Proc. Gen. Transm. Distrib. 141, 191–197. (doi:10.1049/ip-gtd:19949871)
10. Aylett PD. 1958 The energy-integral criterion of transient stability limits of power systems.
Proc. IEE 105, 527–536. (doi:10.1049/pi-c.1958.0070)
11. Willems J. 1971 Direct method for transient stability studies in power system analysis. IEEE
Trans. Automat. Control. 16, 332–341. (doi:10.1109/TAC.1971.1099743)
12. Pai MA, Sauer PW. 1989 Stability analysis of power systems by Lyapunov’s direct method.
IEEE Control Syst. Mag. 9, 23–27. (doi:10.1109/37.16746)
13. Athay T, Podmore R, Virmani S. 1979 A practical method for the direct analysis of transient
stability. IEEE Trans. Power Appar. Syst. 98, 573–584. (doi:10.1109/TPAS.1979.319407)
14. Chiang HD, Wu FF, Varaiya PP. 1994 A BCU method for direct analysis of power-system
transient stability. IEEE Trans. Power Syst. 9, 1194–1200. (doi:10.1109/59.336079)
15. Chiang HD, Chu CC. 1995 Theoretical foundation of the BCU method for direct stability
analysis of network-reduction power-system models with small transfer-conductances. IEEE
Trans. Circuits Syst. I: Fundam. Theory Appl. 42, 252–265. (doi:10.1109/81.386159)
16. Xue Y, Cutsem TV, Ribbenspavella M. 1988 A simple direct method for fast transient
stability assessment of large power-systems. IEEE Trans. Power Syst. 3, 400–412. (doi:10.1109/
59.192890)
 17. Xue Y, Wehenkel L, Belhomme R, Rousseaux P, Pavella M, Euxibie E, Heilbronn B, Lesigne
20
JF. 1992 Extended equal area criterion revisited (EHV power systems). IEEE Trans. Power Syst.
7, 1012–1022. (doi:10.1109/59.207314)

rspa.royalsocietypublishing.org Proc. R. Soc. A 474: 20170733

...................................................
18. Yuan X, Chen S, Hu J. 2016 Multi-time scale voltage and power angle dynamics in
power electronics dominated large power systems. Proc. CSEE 36, 5145–5154. [In Chinese.]
(doi:10.13334/j.0258-8013.pcsee.161247)
19. Li M, Yu Z, Xu T, He J, Wang C, Xie X, Liu C. 2017 Study of complex oscillation caused
by renewable energy integration and its solution. Power System Technol. 41, 1035–1042. [In
Chinese.] (doi:10.13335/j.1000-3673.pst.2016.3049)
20. Ferreira B. 2016 Understanding the challenges of converter networks and systems: better
opportunities in the future. IEEE Power Electron. Mag. 3, 46–49. (doi:10.1109/MPEL.2016.
2551799)
21. Rohden M, Sorge A, Timme M, Witthaut D. 2012 Self-organized synchronization in
Downloaded from https://royalsocietypublishing.org/ on 25 April 2023

decentralized power grids. Phys. Rev. Lett. 109, 064101. (doi:10.1103/PhysRevLett.109.064101)

22. Motter AE, Myers SA, Anghel M, Nishikawa T. 2013 Spontaneous synchrony in power-grid
networks. Nat. Phys. 9, 191–197. (doi:10.1038/NPHYS2535)
23. Menck PJ, Heitzig J, Marwan N, Kurths J. 2013 How basin stability complements the linear-
stability paradigm. Nat. Phys. 9, 89–92. (doi:10.1038/nphys2516)
24. Menck PJ, Heitzig J, Kurths J, Schellnhuber HJ. 2014 How dead ends undermine power grid
stability. Nat. Commun. 5, 3969. (doi:10.1038/ncomms4969)
25. Gajduk A, Todorovski M, Kocarev L. 2014 Stability of power grids: an overview. Eur. Phys. J.
Special Topics 223, 2387–2409. (doi:10.1140/epjst/e2014-02212-1)
26. Ma J, Sun Y, Yuan X, Kurths J, Zhan M. 2016 Dynamics and collapse in a power
system model with voltage variation: the damping effect. PLoS ONE 11, e0165943. (doi:
10.1371/journal.pone.0165943)
27. Sun Y, Kurths J, Zhan M. 2017 Power-functional network. Chaos 27, 083116. (doi:10.1063/
1.4995361)
28. Xue Y. 1993 Transient energy correction and critical cluster identification. Automat. Elect. Power
Syst. 17, 22–31. [In Chinese.]
29. Xue Y. 1994 The mechanisms of instability modes. Auto. Electric Power Syst. 18, 11–24. [In
Chinese.]
30. Guo Q, Zhao J, Zhang B. 2005 A study on isolated stability domain phenomenon in transient
stability assessment based on OMIB. Auto. Electric Power Syst. China 29, 14–19. [In Chinese.]
31. Nusse HE, Yorke JA. 1996 Basins of attraction. Science 271, 1376–1380. (doi:10.1126/
science.271.5254.1376)
32. Baker GL, Blackburn JA. 2005 The pendulum: a case study in physics. Oxford, UK: Oxford
University Press.
33. Antman SS. 2005 Nonlinear problems of elasticity. New York, NY: Springer.