Electric Power Systems Research 76 (2006) 709715

Power system transient stability assessment based on quadratic

approximation of stability region
Ancheng Xue
a,c,
, Felix F. Wu
b
, Yixin Ni
b
, Shengwei Mei
a
, Qiang Lu
a
a
Department of Electrical Engineering, Tsinghua University, Beijing 100084, PR China
b
Center for Electrical Energy Systems, The University of Hong Kong, Hong Kong, PR China
c
Graduate School at Shenzhen, Tsinghua University, Shenzhen 518055, PR China
Received 2 June 2005; received in revised form 9 October 2005; accepted 10 October 2005
Available online 21 November 2005
Abstract
This paper presents an approach to estimate the critical clearing time (CCT) of the multi-machine power systems based on the quadratic surface
which approximates the boundary of stability region relating to the controlling unstable equilibrium point. While obtaining the coefcient matrix
of the second order term of the quadratic approximation, we partition the coefcient matrix into four blocks to reduce the computation burden. The
CCT is estimated by the crossing point of the quadratic approximation and the continuous faulted trajectory. Simulations in IEEE 9-bus and New
England system show the effectiveness of the proposed approach.
2005 Elsevier B.V. All rights reserved.
Keywords: Transient stability assessment; Stability region; Quadratic approximation; Controlling unstable equilibrium point; Critical clearing time
1. Introduction
In the power system transient stability assessment (TSA),
the concept of stability region (or region of attraction), unsta-
ble equilibriumpoint and controlling unstable equilibriumpoint
(CUEP) has been well recognized [16]. Ref. [7] proved that
the boundary of the stability region for a class of power sys-
tems are composed of the stable manifolds of all the unstable
equilibrium points on the boundary, meanwhile, the CUEP of a
certain fault is the unstable equilibriumpoint whose stable man-
ifold is crossed by the continuous faulted trajectory of the fault
[8]. The CCT of the fault can be estimated by the crossing point
of the stable manifold of the CUEP and the continuous faulted
trajectory. Once we have obtained the description for the stable
manifold of the CUEP, we could determine the CCT. Thus the
description of the stable manifold of the CUEP plays an impor-
tant role in TSA. Various approximations to the stable manifold
have been proposed. One way is to use the equal energy sur-

Corresponding author. Tel.: +86 10 62785490; fax: +86 10 62794778.

E-mail addresses: xueac97@mails.tsinghua.edu.cn (A. Xue),
ffwu@eee.hku.hk (F.F. Wu), yxni@eee.hku.hk (Y. Ni),
meishengwei@mail.tsinghua.edu.cn (S. Mei), luqiang@mail.tsinghua.edu.cn
(Q. Lu).
face [813], which is determined by certain transient Lyapunov
energy function. These approximations which based on tran-
sient Lypunov energy function may always give out good results
but have limitation for the non-existence of Lyapunov transient
energy function for general power system models. Another way
is to obtain truncated approximations by applying the Taylor
series expansion or the normal form method to the partial dif-
ferential equation describing the stable manifold of the CUEP.
With the idea of Taylor series expansion, Ref. [14] obtained the
hyper-plane approximation using the rst order term and Ref.
[15] derived a hyper-surface approximation using the second
order term. Recent work [16] also presented a quadratic approx-
imation, but it lacks theoretical justication and needs the energy
correction procedure. Based on the extension of Poincares clas-
sical result on normal formal theory to approximate the stable
manifold, the early work [17] rst proposed an algorithmto com-
pute the relevant coefcient of stable manifold in power series
presentation. The work [18] originally used the normal form to
compute the boundary of stability region. It got the exact series
representationof the stable manifolds characterizingthe stability
boundary. This series can be computed recursively. However, no
numerical tests have been conducted. With the real normal form,
Ref. [19] obtained the second order approximation for the sta-
bility region boundary, but this method requires the computation
0378-7796/$ see front matter 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.epsr.2005.10.003
710 A. Xue et al. / Electric Power Systems Research 76 (2006) 709715
for all the eigenvectors and eigenvalues of the Jacobian matrix,
which results in burdensome computation. To reduce the com-
putation burden, Ref. [20] presented another method to calculate
the quadratic approximation, based on similarity transformation
which avoiding the computation for all the eigenvectors. How-
ever, the above approximation which is based on normal form
approach are founded on the implicit equation which describ-
ing the stable manifold in a new coordinate. Recently, Cheng
et al. [21,22] discovered the explicit equation which describing
the stable manifold of type-1 equilibrium point in the original
coordinate. Furthermore, they presented a quadratic approxima-
tion for the stable manifold of CUEP, which is the same as the
hyper-surface proposed in [15]. This paper applies the proposed
quadratic approximation to determine the CCT of the multi-
machine power systems with the network-reduced structure.
While obtaining the coefcients matrix of the second order term
of the quadratic approximation, we rst partition the coefcient
matrix into four blocks, then we deduce the relationship among
the matrix blocks and obtain the reduced matrix equation which
is satised by one single block, thus the computation burden is
reduced with the associated reduced matrix equation. The esti-
mated CCT is determined by the crossing point of the quadratic
surface and the continuous faulted trajectory. The simulations
in IEEE 9-bus and New England system show the effectiveness
of the proposed approach.
The remainder of the paper is organized as follows. Section
2 develops an algorithm, which using the matrix partition
approach to calculate the coefcient matrix of the second
order term of the quadratic approximation of the stable man-
ifold of the CUEP, to estimate the CCTs of the multi-
machine power systems which having the network reduction
model and uniform damping. Section 3 applies the pro-
posed approach to different systems. Section 4 gives out the
conclusions.
2. CCT estimation based on quadratic approximation
of stability region
2.1. Preliminaries
Consider a nonlinear autonomous dynamic system described
by the following differential equation dened on a manifold M:
x = f(x) (1)
where f, x R
m
, f C
2
. An equilibrium point is a solution of
the equation f(x) =0. A hyperbolic equilibrium point is an equi-
librium point at which the Jacobian D
x
f has no zero real part
eigenvalue. A type-k equilibrium point is a hyperbolic equilib-
riumpoint at which k eigenvalues of D
x
f are having positive real
parts. A hyperbolic stable equilibrium point x
s
of system (1) is
an equilibrium point at which all the eigenvaules of D
x
f (the
Jacobian of the vector eld) have negative real parts. The ow
(or the solution with an initial state x) of system (1) is expressed
as:

t
(x) = (x, t) (2)
The stable manifold of a hyperbolic equilibrium point x
0
is
the set of all those points from each of which the ow will
converge to x
0
as the time approaches positive innite, i.e.,
W
s
(x
0
) =
_
x| lim
t
(x, t) = x
0
_
(3)
If x
s
is a hyperbolic stable equilibrium point for the system
(1), then V(x
s
) =W
s
(x
s
) is the stability region of the hyperbolic
stable equilibrium point x
s
.
Under certain hyperbolic assumption for the autonomous sys-
tem (1), the boundary of the stability region V(x
s
) is composed
of the stable manifolds of all the unstable equilibrium points on
the boundary of the stability region [5], i.e.:
V(x
s
) =
_
x
e
V
W
s
(x
e
) (4)
Furthermore, the stable manifold of a type-1 equilibrium
point x
e
can locally be described as following set [21,22]:
W
s
(x
e
) =
_
x

h(x) = 0, h(x
e
) = 0
(h/x)
T
f = h, rank(h/x) = 1
_
(5)
where is the unique unstable eigenvalue for the Jacobian
matrix J = D
x
f(x)|
x=x
e
at x
e
.
Expanding the right side of Eq. (1) around the point x
e
, we
obtain
f(x) = f(x
e
) +J (x x
e
)
+

(x x
e
)
T
H
1
(x x
e
)/2
.
.
.
(x x
e
)
T
H
m
(x x
e
)/2

+ (6)
where H
i
is the Hessian matrix of the function f
i
at the point x
e
,
i.e., H
i
= [
2
f
i
/x
k
x
l
]
mm
. Then we can obtain the quadratic
approximation for the stable manifold of the type-1 equilibrium
point x
e
as follows [22]:
h
Q
(x) = [x x
e
]
T
+[x x
e
]
T
Q[x x
e
]/2 (7)
where the coefcient vector =(
1
, . . .,
m
)
T
of the linear term
is the left eigenvector associated with the eigenvalue of of the
Jocobian matrix J, i.e.
J
T
= ,
T
= 1 (8)
and the coefcient matrix Q of the quadratic term is the solution
of the Lyapunov matrix equation:
CQ+QC
T
= H (9)
A. Xue et al. / Electric Power Systems Research 76 (2006) 709715 711
where the matrices C = (I
m
/2 J
T
), H =

m
i=1

i
H
i
, I
m
is
the mm identical matrix.
There are a variety of methods [23] for solving the Lyapunov
matrix equation (9). Two most efcient methods are the Schur
method [24] and Hessenberg-Schur method [25]. The explicit
unique solution of Lyapunov matrix equation (9) is
Q = V
1
c
[(C I
m
+I
m
C)
1
V
c
(H)] (10)
where V
c
is the column stack mapping [26, p. 5], which satises
V
c
(B) = (b
11
, . . . , b
m1
, b
12
, . . . , b
m2
, . . . , b
1m
, . . . , b
mm
)
T
(11)
for any mm square matrix B=(b
ij
)
mm
is the Kronecker
tensor product, which satises
AB =

a
11
B a
1t
B
.
.
.
.
.
.
.
.
.
a
s1
B a
st
B

(12)
for any s t matrix A=(a
ij
)
st
and k l matrix B=(b
ij
)
kl
.
2.2. Quadratic approximation of stability region for
multi-machine systems
We review the classical model for transient stability analysis.
Consider a power system consisting of n generators. Let the
loads be modeled as constant impedances. Then the dynamics
of the kth generator can be written with the usual notation as

k
=
0

k
, 2H
k

k
= P
mk
P
ek
D
k

k
, k = 1, . . . , n
(13)
where D
k
and H
k
are damping ratio and inertia con-
stant of machine k, P
ek
= {E
2
k
G
kk
+E
k
(

n
g
j=k
E
j
(G
kj
cos
kj
+
B
kj
sin
kj
))} is the electrical power at machine k,
kj
=
k

j
,
k
is the rotor angle of machine k, E
k
is the constant voltage behind
direct axis transient reactance, P
mk
is the constant mechanical
power.
0
=2f
B
and Y = (Y
ij
)
n
g
n
g
= (G
ij
+jB
ij
)
n
g
n
g
is the
reduced admittance matrix.
If, furthermore, as usual, uniform damping is assumed, i.e.
d
0
=D
k
/2H
k
(k =1, . . ., n), then using the number n machine as
the reference, Eq. (13) can be transformed into the form of Eq.
(14) as follows:

kn
= f
k
=
0

kn
,

kn
= f
k+n1
= d
0

kn
+g
k
(
1,n
, . . . ,
n1,n
),
k = 1, . . . , n 1 (14)
where g
k
=(P
mk
P
ek
)/2H
k
(P
mn
P
en
)/2H
n
. For an n-
generator system, the number of state variable is 2(n 1).
To any given fault, after obtained the related CUEP of the sys-
tem (14) (e.g. by the BCU method [13]), one important step in
estimating the critical clearing time of the given fault is to obtain
the rst order term coefcient by (8) and second order term
coefcient Q by (9), which determining the quadratic approxi-
mation (9) for the stable manifold of the CUEP.
Let x =(
T
,
T
)
T
=(
1,n
, . . .,
n1,n
,
1,n
, . . .,
n1,n
)
T
,
g =(g
1
, . . ., g
n1
)
T
, and J
0
=D

g, then the Jacobian matrix at the

CUEP (
T
0
, 0
T
)
T
of system (14) is
J =
_
0
(n1)

0
I
n1
J
0
d
0
I
n1
_
(2n2)(2n2)
(15)
where 0
(n1)
is the (n 1) (n 1) matrix of zeros and I
n1
is
the (n 1) (n 1) identity matrix.
Assuming
0
is the unique unstable eigenvalue for matrix J
0
,
and
0
is the corresponding left eigenvector, i.e.,
J
T
0

0
=
0
,
T
0

0
= 1 (16)
The unique unstable eigenvalue of the matrix J and its
corresponding left eigenvector =(
T
,
T
)
T
satises
_
0
(n1)

0
I
n1
J
0
d
0
I
n1
_
T
_

_
=
_

_
(17)
Thus from Eq. (17), we can obtain following Eqs. (18) and (19),
directly.

0
= (d
0
+) (18)
J
T
0
= = ( +d
0
)/
0
(19)
Comparing (19) to (16) we can easily get the left unstable
eigenvalue >0, i.e.:
= d
0
/2 +
_
d
2
0
/4 +
0

0
(20)
and its corresponding left eigenvector as follows:
=
_

_
=
_
1/
_
(d
0
+)
2
+
2
0
_
_
(d
0
+)
0

0
_
(21)
Eqs. (20) and (21) imply that the unstable left eigenvector of
the (2n 2) (2n 2) matrix J can be derived by the unstable
left eigenvector
0
of the (n 1) (n 1) matrix J
0
. Thus the
computation burden is reduced.
At the CUEP, the Hessian matrices of f
k
are
H
k
= 0
(2n2)
, k = 1, . . . , n 1 (22)
H
k+n1
=
_
N
k
0
(n1)
0
(n1)
0
(n1)
_
(2n2)(2n2)
, k = 1, . . . ,
n 1 (23)
712 A. Xue et al. / Electric Power Systems Research 76 (2006) 709715
where N
k
=
2
g
k
/
2
is the Hessian matrix of the function
g
k
.
The coefcient matrix
Q =
_
Q
1
Q
2
Q
T
2
Q
4
_
of the quadratic termsatises the Lyapunov matrix equation (9).
Substituting (9) with the Eqs. (15)(23), we can obtain equation
as follows:
_
I
n1
/2 J
T
0

0
I
n1
(d
0
+/2)I
n1
__
Q
1
Q
2
Q
T
2
Q
4
_
+
_
Q
1
Q
2
Q
T
2
Q
4
__
I
n1
/2
0
I
n1
J
0
(d
0
+/2)I
n1
_
=

n1

i=1

i
N
i
0
(n1)
0
(n1)
0
(n1)

(24)
Thus from Eq. (24), we can obtain the following Eqs. (25)(27),
directly:
Q
4
=
0
/(2d
0
+)(Q
2
+Q
T
2
) (25)
Q
1
= [(d
0
+)Q
2
J
T
0
Q
4
]/
0
(26)
J
T
0
Q
T
2
+Q
2
J
0
Q
1
=
n1

i=1

i
N
i
(27)
Furthermore, sequentially substituting the matrix Q
1
and Q
4
in (27) with (26) and (25), we can obtain the matrix equation of
Q
2
as follows:
J
T
0
Q
T
2
+Q
2
J
0
aQ
2
+b(J
T
0
Q
2
+J
T
0
Q
T
2
)
=
k1

i=1

i
N
i
= N
0
(28)
where a =(d
0
+)/
0
, b =(2d
0
+).
Thus the (2n 2) (2n 2) matrix equation (24) is reduced
to the (n 1) (n 1) matrix equation (28). It implies that the
computation of Q can be replaced by the computation of the
block matrix Q
2
. Once have obtained the Q
2
, we could get the
Q
1
and Q
4
with (25) and (26). Thus the computer burden is
greatly reduced. Meanwhile, we can transform Eq. (28) to the
linear equation [26, p. 50] and obtain the unique solution as
follows:
Q
2
= V
1
c
{[(1 +b)(J
T
0
I
n1
)W
[n1]
+I
n1
J
T
0
aI
(n1)
2 +bJ
T
0
I
n1
]
1
V
c
(N
0
)} (29)
where the matrix W
[n1]
is the swap matrix ([26, p. 5]), whose
elements are zeros expect that the elements at (i +(j 1)(n 1),
j +(i 1)(n 1)) i, j =1, . . ., n 1; are all taking the value 1.
Thus we could obtain the quadratic approximation for the
stability region of multi-machine power system with the CUEP
((
e
)
T
,(
e
)
T
)
T
(
e
=0) as follows:
h
Q
(, ) =
T
(
e
) +
T
(
e
) +(
e
)
T
Q
1
(
e
)/2
+(
e
)
T
Q
2
(
e
) +(
e
)
T
Q
4
(
e
)/2 (30)
2.3. Algorithm for CCT estimation and discussions
Once obtaining the quadratic approximation for the stability
region of the post fault system, we could estimate the CCT with
the continuous faulted trajectory ((t),(t)), i.e., identifying the
time satisfyingh
Q
((t),(t)) =0, whichis the time corresponds to
the continuous trajectory intersecting with the quadratic approx-
imation for the stable manifold. The detailed procedure is as
follows.
(1) Obtain the CUEP for a given fault of the power system
according to certain principle (e.g., by the BCU method
[13]).
(2) Calculate the matrix J
0
=D

g and the submatrix N

k
in (23)
at the CUEP.
(3) Calculate the unstable eigenvalue
0
and the associated
unstable left eigenvector
0
of the matrix J
0
=D

g.
(4) Obtainthe unstable eigenvalue andthe associatedunstable
left eigenvector =(
T
,
T
)
T
according to (20) and (21).
(5) Obtain the coefcient matrix Q
2
of quadratic termaccording
to (28), then obtain Q
4
according to (25), and then obtain
Q
1
according to (26).
(6) Substituting the continuous faulted trajectory, ((t),(t)),
to the quadratic approximation of the stability
region (30) until we obtain some t
1
which satises
h
Q
((t
1
),(t
1
))h
Q
((t
1
+t),(t
1
+t)) 0, where t is
the time step used in the simulation. Thus, the time t
1
is the
estimation for the actual CCT. It can be quickly searched
out by appropriate searching method.
The proposed algorithm explores the special structure of the
Jacobian matrix (15) and Hessian matrix (22)(23) of the power
system with network-reduction model and uniform damping,
thus the computation burden for the coefcient matrix of the
quadratic termof the quadratic approximation is greatly reduced
with (28). In small and median size power systems cases, it
would be effective to use Eq. (29) to obtain the coefcient of
the quadratic term. In the large-scale power systems cases, the
approach, which incorporating the sparse matrix technique, is
to be developed.
In the case of detailed models of power systems (e.g., includ-
ing exciter, AVR, PSS, nonlinear load, motor load, etc.), the
Jacobian matrix and Hessian matrix do not hold the special
A. Xue et al. / Electric Power Systems Research 76 (2006) 709715 713
structural as above. Thus the proposed decomposition approach
could not work well. However, the proposed quadratic approxi-
mation (7)(9) is based on the quadratic approximation for the
stable manifold of the type-1 equilibrium point. Furthermore,
the stable manifolds of type-1 equilibrium point partly char-
acterize the stability region of the power system with detailed
model [27]. Thus, the approximation (7) can be extended to the
detailed model of power systems described by the differential-
algebraic equations, as long as the constraints (e.g. limitations
in exciter, etc.) in the systems are not activated. (Otherwise,
the boundary of the stability region would have different
characterization [27,28].) In a word, the quadratic approxi-
mation (7)(9), which stands for the stable manifold of type-
1 equilibrium point, does not have limitation on the model
which is required by the Lyapunov energy function approach,
so it could be extended to the power systems with detailed
model.
3. Simulation results
This section applies the proposed approach in Section 2.3 to
estimate the CCTs of faults in SMIB, IEEE 9-bus system and
New England system.
3.1. Single machine innite bus system
The approximation method proposed above has been applied
to a SMIB system (Fig. 1). Complete data for this system can be
found in [29, p. 844] except that the damping term with D=0
was added to the generator. The line 2 (X
L2
=0.93) experiences
a solid three-phase fault at the sending terminal of the high volt-
age side (bus 2). The fault is cleared by isolating the faulted
line.
With the simulation time step being 0.001 s, the time domain
numerical integration method shows that the actual CCT is
0.086 s, the quadratic approximation gives out a conservative
estimation 0.084 s, and the linear approximation gives out an
over-estimated estimation 0.128 s. Fig. 2 shows the mecha-
nism of estimating CCT with the approximations for the sta-
bility region. In Fig. 2, the continuous faulted trajectory, which
starting from the pre-fault stable equilibrium point, intersects
the approximated boundary of stability region at the time
corresponding to estimated CCT. For the stable manifold of
the CUEP, the quadratic approximation is slightly conserva-
tive at the exit point, thus, its associated CCT estimation is
slightly conservative, while the linear approximation is seri-
ously over-estimated and its associated CCT estimation is
poor.
Fig. 1. Single machine innite bus system (SMIB).
Fig. 2. Mechanism of CCT estimation by stability region approximations.
3.2. The 3-generator 9-bus system
The simulation results in this subsection are based on a 3-
generator, 9-bus power system, which is an IEEE test system
[30, p. 38]. The system is showed in Fig. 3. The types of faults
are three-phase faults. The system are modeled by the network-
reduction model with classical generators having the uniform
damping d
0
=0.1661. The time step adopted in the simulation is
0.001 s.
Table 1 displays the estimated CCTs of six faults using
ve different methods: the time-domain numerical integration
method (T-D), the quadratic and linear approximation of stabil-
ity region, the BCU method [13] and the PEBS method [10].
The results from time-domain numerical integration method are
used as a benchmark. For example, the last row of Table 1 states
that a three phase fault occurs at bus 7 and the post fault system
is the system with the transmission line between buses 7 and 5
tripped, due to the openings of circuit breakers at both ends of
the line. The CCT estimated by the time-domain numerical inte-
gration method is 0.202 s, the CCT estimated by the quadratic
approximation method is 0.199 s, the CCT estimated by the lin-
ear approximation is 0.259 s, and the CCTestimated by the BCU
and PEBS method is 0.204 and 0.214 s, respectively.
Fig. 3. IEEE 3-generator and 9-bus system.
714 A. Xue et al. / Electric Power Systems Research 76 (2006) 709715
Table 1
CCTs for the faults of IEEE-9 bus system (d
0
=0.1661)
Fault bus Line tripped CCTs (unit second) and deviations
T-D Quadratic Linear BCU PEBS
CCT(s) Error (%) CCT(s) Error (%) CCT(s) Error (%) CCT(s) Error (%)
4 46 0.385 0.389 1.04 0.393 2.08 0.381 1.04 0.384 0.26
4 45 0.386 0.397 2.85 0.402 4.15 0.385 0.26 0.386 0.00
9 96 0.254 0.267 5.12 0.302 18.90 0.231 9.06 0.269 5.91
9 98 0.253 0.259 2.37 0.265 4.74 0.230 9.09 0.258 1.98
7 78 0.209 0.215 2.87 0.231 10.52 0.183 12.44 0.223 6.70
7 75 0.202 0.199 1.49 0.259 28.22 0.204 0.99 0.214 5.94
Table 2
CCTs for the generator bus faults of New England system
Fault bus Line tripped CCTs (unit second) and deviations
T-D Quadratic Linear BCU PEBS
CCT Error (%) CCT(s) Error (%) CCT(s) Error (%) CCT(s) Error (%)
25 2526 0.204 0.190 6.86 0.256 25.49 0.188 7.84 0.210 2.94
25 252 0.135 0.115 12.21 0.196 49.62 0.102 22.14 0.141 7.63
23 2322 0.223 0.198 11.21 0.273 22.42 0.174 21.97 0.199 10.76
10 1013 0.242 0.249 2.89 0.324 33.88 0.258 6.61 0.215 11.15
10 1011 0.245 0.256 4.49 0.343 40.00 0.251 2.45 0.224 8.57
21 2122 0.167 0.148 11.38 0.308 84.43 0.224 34.13 0.139. 16.77
Table 1 demonstrates that the proposed quadratic approxi-
mation method offers fairly accurate direct analysis of transient
stability for these faults. The quadratic approximation gives
out slightly over-estimated or conservative results in estimat-
ing CCTs, which is agree with the engineer requirements as
BCU and PEBS methods, while the linear approximation gives
out over-estimated and impractical results (which may exceed
25%). This suggests that the quadratic approximation is also a
good alternative to estimate the CCT.
3.3. New England system
The simulation results presented in this subsection are based
on a 10-generator, 39-bus NewEngland system[1]. The types of
faults are three-phase faults with fault locations both at generator
and load buses. The systemis modeled by the network-reduction
model with classical generators having the uniform damping
d
0
=0.2. The time step adopted in the simulation is 0.001 s.
Tables 2 and 3 display the estimated CCTs of the different
faulted systems using ve different methods: the time-domain
numerical integration method (T-D), the quadratic and linear
approximation of stability region, the BCU method [13] and the
PEBS method [10]. The results fromthe time-domain numerical
integration method are used as a benchmark.
Table 2 demonstrates that the proposed quadratic approxima-
tion method offers not quite accurate direct analysis of transient
stability for some faults at the generation buses. The deviations
of the CCTs obtained by the quadratic approximation method
may be 15% or so, and the deviations of the CCTs obtained
by the linear approximation method are too serious to use. This
suggests that the quadratic approximation may be not a good
alternative to estimate the CCTs of the faults located at genera-
tor buses.
Table 3 demonstrates that the proposed quadratic approxi-
mation method offers fairly accurate direct analysis of transient
stability for faults located at non-generator buses. The quadratic
Table 3
CCTs for the non-generator bus faults of New England system
Fault bus Line tripped CCTs (unit second) and deviations
T-D Quadratic Linear BCU PEBS
CCT Error (%) CCT(s) Error (%) CCT(s) Error (%) CCT(s) Error (%)
3 34 0.266 0.280 5.26 0.320 20.30 0.261 1.88 0.258 3.01
5 56 0.243 0.245 0.82 0.310 27.57 0.241 0.82 0.222 8.64
3 318 0.258 0.268 3.88 0.322 24.81 0.269 4.26 0.259 0.39
18 183 0.247 0.261 5.67 0.334 35.22 0.274 10.93 0.248 0.40
26 2629 0.113 0.106 6.19 0.156 38.05 0.083 26.55 0.123 8.85
28 2829 0.075 0.070 6.67 0.131 74.67 0.062 17.33 0.066 12.00
A. Xue et al. / Electric Power Systems Research 76 (2006) 709715 715
approximation gives out slightly conservative or over-estimated
results in estimating CCTs, the deviations are about 5%, while
the linear approximation gives out much more over-estimated
results (deviations range from 20 to 80%), and the BCU and
PEBSmethods maygive out results withgreat deviationonocca-
sion. This suggests that the quadratic approximation method is
a good alternative to estimate the CCTs of the faults located at
non-generator buses.
4. Conclusions
This paper obtains the quadratic approximation for the sta-
bility region of the multi-machine power systems with network-
reduced model, based on the quadratic approximation for the
stable manifold of the type-1 equilibriumpoint. While obtaining
the coefcient matrix of the second order term of the quadratic
approximation, we explore the special structure of the model.
We partition the coefcient matrix into four blocks to reduce the
computation burden, by only solving the matrix equation for a
single matrix block. Furthermore, we propose an algorithm to
calculate the CCT. The CCTis estimated by the time correspond-
ing to the continuous trajectory intersecting with the quadratic
approximation of stability region (i.e. the stable manifold of the
CUEP). The simulations in the IEEE 9-bus system and New
England system show that the proposed approach could give
out fairly good estimation for the CCTs of median-size power
systems, while the simulation results which incorporating the
sparse matrix technique in large-scale systems are still need to
be developed.
Acknowledgements
This project is supported by National Key Basic Research
Special Fund of China (G2004CB217900, G1998020305,
G1998020309), National Science Foundation of China (No.
50377018, No. 50595411) and Research Grant Council of
HKSAR(HKU7176/03E). The authors also gratefully acknowl-
edge the helpful comments and suggestions of the reviews.
References
[1] M.A. Pai, Energy Function Analysis for Power System Stability, Kluwer
Academic Publishers, Boston, 1989.
[2] A.A. Fouad, V. Vittal, Power System Transient Stability Analysis using
the Transient Energy Function Method, Prentice-Hall, Englewood Cliffs,
NJ, 1992.
[3] M. Ribbens-Pavella, P.G. Murthy, Transient Stability of Power Systems:
Theory and Practice, Wiley, New York, 1994.
[4] H.D. Chiang, C.C. Chu, G. Cauley, Direct stability analysis of electric
power systems using energy functions: theory, applications and perspec-
tive, Proc. IEEE 83 (11) (1995) 14971529.
[5] S. Liu, J. Wang, Energy Function Analysis for Power System Tran-
sient Stability, Shanghai Jiaotong University Press, Shanghai, 1996 (in
Chinese).
[6] Y.X. Yu, C.S. Wang, Theories and Methods of Electrical Power System
Stability, Science Press, Beijing, 1999 (in Chinese).
[7] H.D. Chiang, M.W. Hirsch, F.F. Wu, Stability regions of nonlinear
autonomous dynamical systems, IEEE Trans. Autom. Contr. 33 (1)
(1988) 1627.
[8] H.D. Chiang, F.F. Wu, P.P. Varaiya, Foundation of direct methods for
power systems transient stability analysis, IEEE Trans. Circuits Syst.
CAS-34 (2) (1987) 160173.
[9] H.D. Chiang, J.S. Thorp, The closest unstable equilibrium point method
for power system dynamic security assessment, IEEE Trans. Circuits
Syst. 36 (9) (1989) 11871200.
[10] N. Kakimoto, Y. Ohsawa, M. Hayashi, Transient stability analysis of
electric-power system via LureLypunov function, Part I and II, Trans.
IEE Jpn. 98 (5/6) (1978) 6379.
[11] T. Athay, R. Podmore, S. Virmani, Practical method for directed anal-
ysis of transient stability, IEEE Trans. Appar. Syst. PAS-98 (2) (1979)
573584.
[12] H.D. Chiang, F.F. Wu, P.P. Varaiya, Foundations of the potential energy
boundary surface method for power system transient stability analysis,
IEEE Trans. Circuits Syst. 35 (6) (1988) 712728.
[13] H.D. Chiang, C.C. Chu, Theoretical foundation of the BCU method
for direct stability analysis of network-reduction power system models
with small transfer conductance, IEEE Trans. Circuits Syst. I: Fundam.
Theory Appl. 42 (5) (1995) 252265.
[14] H. Yee, B.D. Spading, Transient stability analysis of multi-machine sys-
tems by the method of hyperplances, IEEE Trans. PAS PAS-96 (1977)
276284.
[15] P.A. Cook, A.M. Eskicioglu, Transient stability analysis of electric power
systems by the method of tangent hypersurface, IEE Proc., Part C 130
(1983) 183193.
[16] Y. Min, H. Li, Y.J. Liu, Critical fault clearing time of multi machine sys-
tem with the quadratic surface method, J. Tsinghua Univ. (Sci. Technol.)
44 (4) (2004) 454457.
[17] S. Ushiki, Analytic expression of the unstable manifolds, Proc. Jpn.
Acad. Ser. A 56 (1980) 239244.
[18] F.M.S. Faithi, A. Arapostathis, P.P. Varaiya, Analytic expressions for
the unstable manifold at equilibrium point in dynamical systems of dif-
ferential equations, in: Proceedings of the 22nd IEEE Conference on
Decision and Control, vol. 3, San Antonio, TX, USA, 1983, pp. 1389
1392.
[19] S. Saha, A.A. Fouad, W.H. Kliemann, V. Vittal, Stability boundary
approximation of a power system using the real normal form of vector
elds, IEEE Trans. Power Syst. 12 (2) (1997) 797802.
[20] V. Venkatasubramanian, J.I. Weijun, Numerical approximation of (n 1)
dimension stable manifold in large systems such as power systems,
Automatica 33 (10) (1997) 18771883.
[21] D.Z. Cheng, J. Ma, Calculation of stability region, in: Proceedings of the
2003 (42nd) IEEE Conference on Decision and Control, vol. 6, Maui,
HI, USA, 2003, pp. 56155620.
[22] D.Z. Cheng, J. Ma, Q. Lu, S.W. Mei, Quadratic form of stable sub-
manifold for power systems, Int. J. Robust Nonlin. Contr. 14 (9/10)
(2004) 773788.
[23] B.N. Datta, Numerical Linear Algebra and Applications, Brooks/Cole
Publishing Co., Pacic Grove, California, 1995.
[24] R.H. Bartels, G.W. Stewart, Algorithm 432: solution of the
matrix equation AX+XB=C, Commun. ACM 15 (9) (1972) 820
826.
[25] G.H. Golub, S. Hash, C. Van Loan, A HessenbergSchur methods for
the problem AX+XB=C, IEEE Trans. Autom. Contr. 24 (6) (1979)
909913.
[26] D.Z. Cheng, Matrix and Polynomial Approach to Dynamic Control Sys-
tem, Science Press, Beijing, 2002.
[27] V. Venkatasubramanian, H. Schattler, J. Zaborszky, Dynamics of large
constrained nonlinear systems: a taxonomy theory, Proc. IEEE 83 (11)
(1995) 15301561.
[28] V. Venkatasubramanian, Stability boundary analysis of nonlinear dynam-
ics subject to state limit, in: Proceedings of the 34th Hawaii International
Conference on System Science, 2001.
[29] P. Kundur, Power System Stability and Control, McGraw-Hill, New
York, 1992.
[30] P.M. Anderson, A.A. Fouad, Power System Control and Stability, The
Iowa State University Press, Ames, IA, USA, 1977 (revisited printing,
IEEE Press, 1994, p. 375).