g
=
g
n
G
(2.1)
g
=
1
M
g
(P
0
m
P
gt
D
g
g
)
V
g
=
X
d
X
d
T
do
_
E
f
V
g
X
d
X
d
. .
1
V
g
Q
g
(V
g
)
V
g
X
d
+
V
t
X
d
cos(
t
g
)
_
. .
+
1
V
g
Q
tg
P
gt
=
V
g
V
t
X
d
sin(
g
t
)
Q
gt
=
V
2
t
X
d
+
V
g
V
t
X
d
cos(
g
t
)
Here P
gt
and Q
gt
are the active and reactive powers injected by the generator at
its terminal bus t. The term Q
tg
could be regarded as injected reactive power
at the internal generator bus g. V
g
g
and V
t
t
are the voltage phasors at the
internal and terminal buses. M
g
represents the machine inertia in seconds
2
, and
D
g
is the generators shaft damping constant in seconds. The variable
g
stands
for the generator frequency deviation from the nominal frequency in rad/sec. The
10
mechanical power applied to the generator shaft, P
0
m
, is assumed constant, though
suitable governor models can be added to represent dynamic behavior of P
m
. E
f
represents a constant generator eld voltage, X
d
is the daxis synchronous reactance,
and X
d
is the daxis transient reactance. T
do
is the daxis opencircuit transient time
constant in seconds.
For transient simulations, a constant voltage source with swing dynamics model
can also be used to represent the generator. This model captures some of the termi
nal voltage control loop characteristics of the exciter, at the expense of eliminating
some internal transient dynamics in the generator. In this case, exciter saturation
can be represented by limits on the reactive power delivered by the synchronous
machine [16].
2.2.2 Transmission System
A equivalent circuit model is employed. Transformers are included as part of the
transmission network.
P
sr
= g
sr
V
2
s
g
sr
V
s
V
r
cos(
s
r
) + b
sr
V
s
V
r
sin(
s
r
) (2.2)
P
rs
= g
sr
V
2
r
g
sr
V
s
V
r
cos(
r
s
) + b
sr
V
s
V
r
sin(
r
s
)
Q
sr
= (b
s
+ b
sr
)V
2
s
g
sr
V
s
V
r
sin(
s
r
) b
sr
V
s
V
r
cos(
s
r
)
Q
rs
= (b
r
+ b
sr
)V
2
r
g
sr
V
s
V
r
sin(
r
s
) b
sr
V
s
V
r
cos(
r
s
)
Here P
sr
, P
rs
, Q
sr
, and Q
rs
are the transmitted powers between buses s and r. V
r
r
and V
s
s
are the voltage phasors at the respective buses. The transfer admittance
is represented by g
sr
jb
sr
; jb
r
and jb
s
denote the shunt bus admittances at
both ends of the line/transformer.
11
2.2.3 Loads
Voltage and frequency dependent load models are used.
P
l
= P
l
1
_
V
l
V
0
l
_
2
P
l
2
_
V
l
V
0
l
_
P
l
3
(2.3)
D
l
(
n
G
) P
l
Q
l
= Q
l
1
_
V
l
V
0
l
_
2
Q
l
2
_
V
l
V
0
l
_
Q
l
3
Q
l
where l = 1 + n
G
, 2 + n
G
, . . . , n
G
+ n
L
. Here P
l
and Q
l
are the powers injected
by the load, and V
l
l
is the load phasor voltage at bus l. P
l
1
, P
l
2
, P
l
3
, P
l
, Q
l
1
,
Q
l
2
, Q
l
3
,and Q
l
are constant weighting factors. D
l
represents the time constant of
the frequency dependent term in seconds, and is the parameter that displays the
functional dependence of slow time scale load increase. For most studies of voltage
collapse, it is assumed that the pattern of load change can be represented with one
degree of freedom as shown above, and that this evolution of load drives the system
to a saddlenode bifurcation.
2.3 HVDC Line Model
The dc line is simulated using an RL circuit, and the HVDC controllers are modeled
to replicate the simple control scheme shown in gure 2.1. Its key characteristics
can be reproduced by the control circuit depicted in gure 2.2 [23]. Although these
circuits are approximations to the more complicated HVDC control structures, they
recreate several of the main properties of actual systems. VDCOL can be introduced
into this model by representing the controller current order as a nonlinear function
of the ac converter voltages [21, 22]. This will be represented by assuming voltage
dependence of the current order settings in the converter current controllers depicted
below.
12
I
d
V
d
Equilb. Point
r
max
VDCOL
r
min
i
min
i
max
I
o
r
I
o
i
Figure 2.1: HVDC control criteria. The rectier is allowed to go into inverter
operation for faster recovery after fault conditions.
K
P
K
I
S
I
d
V
r
I
o
r
+
cos
r
min
cos
r
max
+
+
cos
r
min
cos
r
max
cos
r
x
r
Figure 2.2: PI rectier current controller. The inverter side has a similar control
circuit.
13
Assuming ideal harmonic ltering, equations (2.4) below are typically used to
simulate the behavior of the HVDC system under balanced 3phase operation, and
include the control mode switching depicted in gure 2.1 [21, 22, 23, 24]. The
convention used throughout this thesis will apply the subscripts r and i to
rectier and inverter quantities, respectively.
I
d
=
1
L
d
(V
dr
V
d
i
)
R
d
L
d
I
d
(2.4)
x
r
= l
1
(K
I
[I
or
(V
r
) I
d
], y
r
)
x
i
= l
1
(K
I
[I
d
I
o
i
(V
i
)], y
i
)
cos
r
= l
2
(x
r
+ K
P
[I
or
(V
r
) I
d
])
V
dr
=
3
a
r
V
r
cos
r
3
X
cr
I
d
S
r
= k
r
V
n
I
n
S
n
3
a
r
V
r
I
d
P
r
= k
r
V
n
I
n
S
n
V
dr
I
d
Q
r
=
_
S
2
r
P
2
r
k
r
=
_
[cos 2
r
cos 2(
r
+
r
)]
2
+ [2
r
+ sin 2
r
sin 2(
r
+
r
)]
2
_
1/2
4[cos
r
cos(
r
+
r
)]
1
cos
i
= l
2
(x
i
+ K
P
[I
d
I
o
i
(V
i
)])
V
d
i
=
3
a
i
V
i
cos
i
X
c
i
I
d
S
i
= k
i
V
n
I
n
S
n
3
a
i
V
i
I
d
P
i
= k
i
V
n
I
n
S
n
V
d
i
I
d
Q
i
=
_
S
2
i
P
2
i
14
k
i
=
_
[cos 2
i
cos 2(
i
+
i
)]
2
+ [2
i
+ sin 2
i
sin 2(
i
+
i
)]
2
_
1/2
4[cos
i
cos(
i
+
i
)]
1
Here V
dr
and V
d
i
are the per unit dc terminal voltages at the rectier and inverter
ends. Per unit normalization is employed with I
n
and V
n
as the base quantities for
the dc system, while S
n
is the base power for the ac side. X
cr
and X
c
i
are the per
unit commutation reactances, and R
d
and L
d
are the per unit dc line parameters.
The products a
r
V
r
and a
i
V
i
are the per unit ac bus voltages at the secondary ac side
of the transformers. S
r
and S
i
are the magnitudes of the HVDC complex powers
at the ac side, and P
r
, P
i
, Q
r
, and Q
i
are the active and reactive powers absorbed
by the dc system. Variables
r
and
i
represent the overlap angles at each of the
dc converters. The variables y
r
and y
i
are feedback measurements used to enforce
ring and extinction angle limits in the current controllers; e.g., y
r
= x
r
(t + t)
and y
i
= x
i
(t + t) are approximations that avert saturation problems during the
numeric integration process.
The controllers are designed so that their actions typically do not overlap, i.e.,
either the inverter or rectier controls the dc current, but (normally) not both. For
large dc transient currents produced by voltage changes at the converter buses, and
under normal operating conditions the rectier side controls the current while the
inverter current controller is saturated at is minimum value
i
min
. When the tran
sient current is small the opposite applies. However, during recovery from abnormal
operating conditions one can expect to have both converters controlling the current
for small periods. Under fault conditions the current controllers can also be forced to
their maximum limits, denoted by
rmax
and
imax
. This behavior is approximated
15
here by the limit functions l
1
() and l
2
(), which are dened as follows:
l
1
(x, y) =
_
_
x if y
min
y y
max
0 otherwise
l
2
(x) =
_
_
x
min
if x x
min
x if x
min
< x < x
max
x
max
if x x
max
In equations (2.4), the evaluation of the limit function l
1
() with y = x(t + t)
depends on the time step t and the numerical method used on the integration of the
dynamical system. Its implementation requires an additional variable representing
the system state at t + t.
Dierential equations (2.4) are good approximations to true system behavior
when overlap angles
r
and
i
are less than 60
0
. These equations cannot be used to
estimate the behavior of the dc system when converters change commutation mode,
since under these conditions the converter must be represented by a dierent set of
equations [38]. Figure 2.3 depicts the possible range of values that variable k
r
can
take for rectier ring angle values between 0
0
and 120
0
, and overlap angles from
0
0
to 60
0
. Assuming
k
r
1 and k
i
1
introduces a maximum error of about 4%, with the advantage of signicantly reduc
ing the analytic complexity of the model.
16
0.955
0.96
0.965
0.97
0.975
0.98
0.985
0.99
0.995
1
0 10 20 30 40 50 60
k
r
(
p
.
u
.
)
overlap angle (deg.)
Figure 2.3: Variable k
r
as a function of the rectier overlap angle, for ring angles
between 0
0
and 120
0
.
17
2.4 Matrix Equations
The equations above can be rewritten in vector dierential equation form to simplify
the notation and analysis. For the k
th
HVDC link, the following vectors are dened:
x
dc
k
= [x
r
k
I
d
k
x
i
k
]
T
k = 1 +n
G
+ n
L
, . . . , n
G
+ n
L
+ n
dc
y
dc
k
= [cos
r
k
V
dr
k
S
r
k
cos
i
k
V
d
i
k
S
i
k
]
T
V
dc
k
= [V
r
k
V
i
k
]
T
P
dc
k
= [P
r
k
P
i
k
]
T
Q
dc
k
= [Q
r
k
Q
i
k
]
T
Thus, equations (2.4) can be rewritten in vector form as follows:
x
dc
k
= h
dc
k
(V
dc
k
, x
dc
k
, y
dc
k
) (2.5)
0 = w
dc
k
(V
dc
k
, x
dc
k
, y
dc
k
)
P
dc
k
= f
k
(V
dc
k
, x
dc
k
, y
dc
k
)
Q
dc
k
= g
k
(V
dc
k
, x
dc
k
, y
dc
k
)
Equations (2.1), (2.2), (2.3), and (2.4) can then be arranged into vector dier
ential equations (2.6) below for an n bus ac/dc system, where n = n
G
+ n
L
+ 2n
dc
.
x
dc
= h
dc
(V
dc
, x
dc
, y
dc
) (2.6)
G
=
G
M
G
G
= f
G
(, V) D
G
G
D
x
V
G
= g
G
(, V)
= U
1
G
g
G
(, V)
D
L
L
= f
L
(, V, )
0 =
_
_
g
L
(, V, )
f
dc
(, V, x
dc
, y
dc
)
g
dc
(, V, x
dc
, y
dc
)
_
_
= fg(, V, x
dc
, y
dc
, )
18
0 = w
dc
(V
dc
, x
dc
, y
dc
)
Vectors f() and g() stand for the ac active and reactive power mismatch equations.
In particular, for k = 1, 2, . . . , n
f
k
(, V)
= P
0
k
n
j=1
[G
kj
V
k
V
j
cos(
k
j
) + B
kj
V
k
V
j
sin(
k
j
)] (2.7)
g
k
(, V)
= Q
k
(V
k
)
n
j=1
[G
kj
V
k
V
j
sin(
k
j
) B
kj
V
k
V
j
cos(
k
j
)]
where G
kj
+ jB
kj
is the kj term of the bus admittance matrix, P
0
k
and Q
k
(V
k
)
are the active and reactive powers injected at bus k (note that reactive power is
allowed to depend on bus voltage magnitude), and V
k
k
is the corresponding bus
voltage phasor. Vectors V and are dened as [V
T
G
V
T
L
V
T
dc
]
T
and [
T
G
T
L
T
dc
]
T
,
respectively. The reference generator n
G
is assumed to be an innite bus so that all
system equilibria are guaranteed to have
G
= 0, otherwise the transfer conductance
losses may produce a slight shift in generation frequency at the equilibria. Matrices
M
G
, D
G
, and D
L
are all positive denite diagonal matrices representing the gen
erators inertia and damping, and the loads frequency dependent time constants,
respectively. Matrix D
x
is also a positive denite diagonal matrix where
D
x
ii
=
X
d
i
X
d
i
T
d
0
i
i = 1, 2, . . . , n
G
1
Finally, U
G
is a diagonal positive denite matrix with diagonal entries equal to the
generator voltage magnitudes.
Chapter 3
Saddlenode Bifurcations
3.1 Introduction
The term bifurcation in nonlinear dynamic systems is used to describe changes in the
qualitative structure of the phase portrait when certain system parameters change
[5, 6]. Local bifurcations, as opposed to global bifurcations, can be studied by
analyzing the vector dierential equations (vector eld) near the bifurcating equi
librium points. For the simplest bifurcations of equilibria, known as the saddlenode,
transcritical, and pitchfork, a suitable change of coordinates allows the relevant be
havior to be locally reduced to a onedimensional dierential equation with a single
parameter . Another simple bifurcation is the Hopf, which can be locally reduced
to a twodimensional vector eld with a single parameter.
Considerable evidence in the literature suggest that the onset of voltage collapse
phenomena in power systems can, under certain assumptions regarding system mod
elling, be accurately predicted by identifying parameter values that lead to saddle
node bifurcations [3, 7, 8]. This chapter concentrates on this particular class of local
bifurcations.
19
20
To introduce the basic characteristics of saddlenode bifurcations, consider a
general oneparameter nonlinear system represented by the vector eld
z = s(z, ) (3.1)
where z R
n
and R. Assume that the vector eld has an equilibrium point
(z
0
,
0
) with a zero eigenvalue for the system linearization (Jacobian). Of all the
possible types of local bifurcations characterized by a singular Jacobian at the equi
librium, i.e., saddlenode, transcritical, and pitchfork, only the saddlenode bifur
cation occurs generically so that it is expected to be typical in practice
1
[5, 6, 8].
Then, under certain transversality conditions discussed in detail later, the behavior
of equations (3.1) about the degenerate equilibrium point can be reduced to the fol
lowing single variable oneparameter dierential equation using a LyapunovSchmidt
reduction [7, 40]:
x = x
2
= f(x, ) (3.2)
Figure 3.1 shows a diagram of the equilibria for f(x, ) as the parameter
changes. For < 0 equation (3.2) presents no equilibria, whereas for > 0 the vec
tor eld has stable and unstable equilibrium points. For the special case when = 0,
the linearization of f(x, ) at the equilibrium (0,0) is singular, i.e., f/x[
0
= 0.
Hence, the saddlenode is characterized by two equilibria coalescing and then disap
pearing as the parameter changes. Notice also the quadratic shape of the bifur
cation diagram close to the degenerate equilibrium point, which is a characteristic
of saddlenodes that will be displayed in the behavior of actual ac/dc system bus
voltage magnitudes when plotted versus a varying parameter representing load
change.
1
More formally: a general oneparameter family of dierential equations such as (3.1)
with singular Jacobian at an equilibrium can be perturbed to produce saddlenode bifur
cations satisfying certain transversality conditions, and these saddlenode bifurcations are
robust to perturbations in the model.
21
x
t
bifurc.
stable
unstable
f ()
=
_
_
h
dc
()
G
f
G
() D
G
G
g
G
()
f
L
()
_
_
and g()
=
_
_
fg()
w
dc
()
_
_
Assume that in the neighborhood of the equilibria to be studied, the HVDC link
is subject to either the rectier or the inverter controlling the dc current, so that
in this range the derivatives of the limit functions l
1
() and l
2
() are well dened.
Furthermore, if no limits are considered in the ac equations of the system model
represented by vector equations (2.6), functions
f () and g() can be assumed to be
C
r
functions, r 2, so that their derivatives D and D
2
are also well dened.
When the algebraic constraints g() have an invertible Jacobian D
u
g() along the
system trajectories of interest, the algebraic variables can be eliminated (Implicit
Function Theorem [41]), and (2.6) can be reduced to:
M z =
f (z, u, )
0 = g(z, u, )
_
M z =
f (z,
h(z, ), )
= s(z, )
(3.3)
23
where M is a positive denite diagonal matrix dened by
M
=
_
_
I
2n
dc
0 0 0 0
0 I
n
G
1
0 0 0
0 0 M
G
0 0
0 0 0 D
x
0
0 0 0 0 D
L
_
_
Here I
n
represents an n by n identity matrix. Notice that matrix M is just dening
positive scaling factors for the state variables z.
The condition of nonsingularity of the Jacobian D
u
g() guarantees a well posed
dynamic system as explained in [42]. If matrix D
u
g() becomes singular then the
model represented by (3.3) breaks down and singular perturbation techniques or
load models with dynamic voltage dependance [43] should be used. For example,
a singular D
u
g[
0
at a particular equilibrium point (z
0
, u
0
,
0
) of a simple power
system model consisting of one generator, a transmission line and a load, constitutes
an impasse point that divides two stable equilibria in a system with no unstable
equilibrium points; this type of behavior is nonphysical in a real power network.
The same phenomenon was observed in a 173 bus ac/dc system, where two stable
equilibria coalesce at a singular Jacobian. In this case, assuming a dynamic voltage
dependence of some of the load buses removed the impasse point forcing one of the
stable equilibria to become unstable, so that saddlenode theory can be applied to
explain the bifurcation.
A bifurcation, or structural instability, occurs when the Jacobian D
z
s() of (3.3)
becomes singular at the equilibrium (z
0
,
0
). Several types of bifurcation are possible
in this situation, but of these only the saddlenode occurs generically. Moreover,
the following conditions apply at the saddlenode (z
0
,
0
) [5]:
1. D
z
s(z
0
,
0
) has a simple and unique zero eigenvalue, with normalized right
24
eigenvector v and left eigenvector w, i.e.,
D
z
s(z
0
,
0
) v = 0 (3.4)
w
T
D
z
s(z
0
,
0
) = 0
T
2. w
T s
(z
0
,
0
)
,= 0. (3.5)
3. w
T
[D
2
z
s(z
0
,
0
) v] v ,= 0. (3.6)
In equation (3.6), the square bracketed product of the 3tensor D
2
z
s[
0
and the eigen
vector v yields a matrix D
2
z
s[
0
v R
mm
(m = 3n
G
+n
L
+ 3n
dc
3). Conditions 1
through 3 guarantee the generic quadratic behavior near the bifurcation point, and
also prevent singular augmented Jacobians of the NewtonRaphson based methods
used to determine bifurcation points as described below.
Use of the vector eld s(z, ) in voltage collapse analysis of power systems
presents several problems, such as the diculty of nding the explicit function
u =
h(z, ). Typically function
h(z, ) will not be expressible in closed form, though
its derivatives are available. While ultimately only the derivatives of
h() are needed
for computation of the saddlenode bifurcation point, use of the reduced equations
sacrices sparsity, signicantly increasing the computational costs. Therefore, it is
useful to relate known methods for detecting bifurcations based on saddlenode con
ditions for the reduced system (3.3) to conditions expressed in terms of the complete
set of system equations (2.6). This will be the focus of section 3.3, to follow below.
25
3.3 Equivalence of SaddleNode Conditions
An important computational issue in bifurcation studies for power systems is the
relationship between eigenvalues of the power ow
2
Jacobian, J
PF
, and those of the
system dynamics linearized at the equilibrium point, denoted by J
TS
= M
1
D
z
s(z
0
,
0
).
This subject has been studied for a number of simple ac only system models in
[7, 45, 46]. The following discussion demonstrates that singularity of J
PF
implies a
zero eigenvalue for the linearized system dynamics.
Linearizing
f() and g() at the equilibrium yields a block Jacobian structure
satisfying:
D
(z,u)
(
f, g) [
0
=
_
_
D
z
f[
0
D
u
f[
0
D
z
g[
0
D
u
g[
0
_
_
By eliminating the algebraic variables, one obtains the Jacobian for the reduced
system, i.e.,
MJ
TS
= D
z
f [
0
D
u
f[
0
D
u
g[
1
0
D
z
g[
0
which can be shown to have the following structure:
MJ
TS
=
x
dc
G
G
V
G
L
x
dc
P
1
P
2
0 P
3
P
4
G
0 0 I
n
G
1
0 0
M
G
G
P
5
P
6
D
G
P
7
P
8
Dx
V
G
P
9
P
10
0 P
11
P
12
D
L
L
P
13
P
14
0 P
15
P
16
(3.7)
Using standard block determinant formulas, the determinant of J
TS
can be expressed
2
The power ow equations in this chapter dier slightly from the typical set of
equations in the power system literature, inasmuch as the internal generator variables
are included in these equations. This approach is more consistent with nonlinear system
theory for nding equilibria of a general set of dierential equations.
26
in the form given by (3.8) below, where k is a positive integer.
det J
TS
= (1)
k
det M
1
det
P
1
P
2
P
3
P
4
P
5
P
6
P
7
P
8
P
9
P
10
P
11
P
12
P
13
P
14
P
15
P
16
= (1)
k
det P
det M
(3.8)
On the other hand, the power ow Jacobian can be obtained from (2.6) with
G
= 0. Hence,
J
PF
= (3.9)
_
_
D
x
dc
h
dc
[
0
0 0 0 D
u
h
dc
[
0
D
y
dc
h
dc
[
0
0 D
G
f
G
[
0
D
V
G
f
G
[
0
D
L
f
G
[
0
D
u
f
G
[
0
0
0 D
G
g
G
[
0
D
V
G
g
G
[
0
D
L
g
G
[
0
D
u
g
G
[
0
0
0 D
G
f
L
[
0
D
V
G
f
L
[
0
D
L
f
L
[
0
D
u
f
L
[
0
0
D
x
dc
fg[
0
D
G
fg[
0
D
V
G
fg[
0
D
L
fg[
0
D
u
fg[
0
D
y
dc
fg[
0
D
x
dc
w
dc
[
0
0 0 0 D
u
w
dc
[
0
D
y
dc
w
dc
[
0
_
_
J
PF
=
_
_
D
z
f[
0
D
u
f[
0
D
z
g[
0
D
u
g[
0
_
_
Here
f()
= [h
T
dc
() f
T
G
() g
T
G
() f
T
L
()]
T
, and z
= [x
T
dc
T
G
V
T
G
T
L
]
T
. Thus, the determi
nant of J
PF
can be calculated using:
det J
PF
= det (D
u
g[
0
) det (D
z
f [
0
D
u
f [
0
D
u
g[
1
0
D
z
g[
0
) (3.10)
Manipulating the matrix of (3.9), it can be readily shown that matrix P in (3.8)
is equal to
P = D
z
f[
0
D
u
f[
0
D
u
g[
1
0
D
z
g[
0
(3.11)
27
Hence, from (3.8) and (3.10) we have that
det J
TS
= (1)
k
det J
PF
detM det (D
u
g[
0
)
(3.12)
Theorem 3.1 Let D
u
g[
0
be nonsingular at the equilibrium point
(z
0
, u
0
,
0
) R
N+1
(N = 3n
G
+ 2n
L
+ 9n
dc
3). Then, the matrix resulting from
linearizing the reduced nonlinear system dynamics at the equilibrium, D
z
s[
0
, has a
zero eigenvalue if and only if the corresponding power ow Jacobian is singular.
Proof: Equation (3.12) above shows that a zero eigenvalue of J
PF
is equivalent to
having a zero eigenvalue of D
z
s[
0
, since
det (D
z
s[
0
) = (1)
k
det J
PF
det (D
u
g[
0
)
Hence, by assuming a well posed dynamic system at the equilibrium (z
0
, u
0
,
0
)
(i.e., D
u
g[
0
nonsingular), singularity of the power ow equations necessarily implies
having a zero eigenvalue for the linearized system dynamics, and vice versa. For
completeness, notice that det D
u
g[
0
cannot become innity since all terms in matrix
D
u
g[
0
are bounded above due to the structure of g() (sines and cosines functions).
Theorem 3.1 shows that the rst saddlenode condition is equivalent for both
the transient and the power ow equations. Moreover, theorems 3.2 and 3.3 below
prove that saddlenode bifurcation conditions 2 and 3 for the reduced dynamic model
(3.3), represented by equations (3.5) and (3.6), have an analogous representation in
the power ow equations. In order to simplify the notation, the vector elds
f()
and g(), which represent the power ow equations of the proposed ac/dc system
model, are grouped in the vector eld
F( z, u, )
=
_
f( z, u, )
g
T
( z, u, )
_
_
28
where z is a subset of the state variables z (
G
are not included in this vector), and
u are the variables implicitly dened by the constraint manifold g(z, u, ) = 0
Theorem 3.2 Let D
u
g[
0
be nonsingular at the saddlenode bifurcation point (z
0
, u
0
,
0
)
R
N+1
satisfying transversality conditions (3.4), (3.5) and (3.6). Let F : R
Nn
G
+1
R R
Nn
G
+1
be as dened above. Hence, for properly normalized left eigenvec
tors w R
m
and w R
Nn
G
+1
corresponding to zero eigenvalues of D
z
s[
0
and
D
( z,u)
F[
0
, respectively, i.e., w
T
D
z
s[
0
= 0
T
, and w
T
D
( z,u)
F[
0
= 0
T
, it follows that
w
T
s
0
= w
T
F
0
(3.13)
Furthermore, uniqueness of the zero eigenvalue of D
z
s[
0
guarantees uniqueness of
the zero eigenvalue of matrix D
( z,u)
F[
0
.
Proof: Since D
u
g[
0
is nonsingular at the equilibrium, there exists a smooth local
function
h() around (z
0
, u
0
,
0
) such that u =
h(z, ) (Implicit Function Theorem
[41]). Then,
g(z, u, ) = g(z,
h(z, ), ) = 0
0
= D
u
g[
1
0
g
0
On the other hand, since s(z, ) =
f(z,
h(z, ), ), we have that
s
0
=
0
+ D
u
f [
0
0
=
0
D
u
f[
0
D
u
g[
1
0
g
0
(3.14)
Now, from the ac/dc reduced dierential equations (3.3) and considering that
[ w
T
z
w
T
u
] D
(z,u)
(
f, g)[
0
= 0
T
29
it follows that
w
T
u
= w
T
z
D
u
f[
0
D
u
g[
1
0
(3.15)
0
T
= w
T
z
_
D
z
f [
0
D
u
f[
0
D
u
g[
1
0
D
z
g[
0
_
. .
D
z
s[
0
(3.16)
Note that equation (3.16) implies that w
z
is a null left eigenvector for D
z
s[
0
. But
by denition of (z
0
, u
0
,
0
), the zero eigenvalue is unique with a uniquely dened
left eigenvector w, hence
w = w
z
(3.17)
Equations (3.15) and (3.16) guarantee uniqueness of the zero eigenvalue of D
(z,u)
(
f, g)[
0
.
This follows rst from the assumption that (z
0
, u
0
,
0
) is a saddlenode bifurcation
satisfying condition (3.4); this guarantees uniqueness of the zero eigenvalue of D
z
s[
0
and its corresponding eigenvector w = w
z
. Secondly, since w
z
is unique and D
u
g[
0
is invertible, vector w
u
is also unique. Thus, D
(z,u)
(
f , g)[
0
has a unique zero eigen
value due to uniqueness of its associated eigenvector [ w
T
z
w
T
u
]
T
. Furthermore, from
equations (3.14), (3.15) and (3.17)
w
T
s
0
= w
z
_
0
D
u
f [
0
D
u
g[
1
0
g
0
_
= w
T
z
0
w
T
z
D
u
f [
0
D
u
g[
1
0
. .
w
u
= w
T
z
0
+ w
T
u
g
0
Based on (3.15), the matrix structure of D
z
s[
0
shown in (3.7), and the denition of
30
matrix P in (3.11), one can readily show that
w =
_
_
w
x
dc
w
G
w
G
w
V
G
w
L
_
_
w
G
= D
1
G
w
G
w
z
=
_
_
w
x
dc
w
G
w
V
G
w
L
_
_
(3.18)
Therefore, since
G
/ = 0, we have that
w
T
z
0
= w
T
z
0
w
T
s
0
= w
T
z
0
+ w
T
u
g
0
= w
T
F
0
Notice that (3.18) guarantees uniqueness of the zero eigenvalue of the power ow
Jacobian D
( z,u)
F[
0
.
Theorem 3.3 Let D
u
g[
0
be nonsingular at the saddlenode bifurcation point (z
0
, u
0
,
0
)
R
N+1
satisfying transversality conditions (3.4), (3.5) and (3.6). Let F(), w, and
w be dened as in theorem 3.2. Then, for the properly normalized right eigenvectors
v R
m
and v R
Nn
G
+1
corresponding to unique zero eigenvalues of D
z
s[
0
and
D
( z,u)
F[
0
, respectively, i.e., D
z
s[
0
v = 0, and D
( z,u)
F[
0
v = 0, it follows that
w
T
_
D
2
z
s[
0
v
_
v = w
T
_
D
2
( z,u)
F[
0
v
_
v (3.19)
Proof: First, dene y
= [z
T
u
T
]
T
,and
F()
= [
f
T
() g
T
()]
T
, where
F : R
N
R
R
N
. Note that vector y stands for the set of system variables of the dynamic
31
equations represented by the vector eld
F(). Then, since g(z, u, ) = 0 and
D
y
F[
T
0
w = D
y
F[
0
v = 0 (where w = [ w
T
z
w
T
u
]
T
, v = [ v
T
z
v
T
u
]
T
R
N
), following the
arguments employed in the proof of theorem 3.2 we have that
w
T
u
= w
T
z
D
u
f [
0
D
u
g[
1
0
and w
z
= w
(3.20)
v
u
= D
u
g[
1
0
D
z
g[
0
v
z
and v
z
= v
(3.21)
Furthermore, the product of the tensor D
2
y
F[
0
and the vector v yields the matrix
D
2
y
F[
0
v =
_
_
D
2
z
f[
0
v
z
+ D
2
zu
f[
0
v
u
D
2
uz
f[
0
v
z
+ D
2
u
f [
0
v
u
D
2
z
g[
0
v
z
+ D
2
zu
g[
0
v
u
D
2
uz
g[
0
v
z
+ D
2
u
g[
0
v
u
_
_
=
_
_
D
2
z
f[
0
v D
2
zu
f [
0
D
u
g[
1
0
D
z
g[
0
v D
2
uz
f[
0
v D
2
u
f[
0
D
u
g[
1
0
D
z
g[
0
v
D
2
z
g[
0
v D
2
zu
g[
0
D
u
g[
1
0
D
z
g[
0
v D
2
uz
g[
0
v D
2
u
g[
0
D
u
g[
1
0
D
z
g[
0
v
_
_
Hence, from (3.20) and (3.21) it can be shown that
w
T
[D
2
y
F[
0
v] v =
_
w
T
w
T
D
u
f[
0
D
u
g[
1
0
_
T
[D
2
y
F[
0
v]
_
_
v
D
u
g[
1
0
D
z
g[
0
v
_
_
= w
T
_
D
2
z
f [
0
v D
2
zu
f[
0
D
u
g[
1
0
D
z
g[
0
v (3.22)
D
2
uz
f [
0
vD
u
g[
1
0
D
z
g[
0
+
D
2
u
f[
0
D
u
g[
1
0
D
z
g[
0
vD
u
g[
1
0
D
z
g[
0
D
u
f[
0
D
u
g[
1
0
D
2
z
g[
0
v +
D
u
f[
0
D
u
g[
1
0
D
2
zu
g[
0
D
u
g[
1
0
D
z
g[
0
v +
D
u
f[
0
D
u
g[
1
0
D
2
uz
g[
0
vD
u
g[
1
0
D
z
g[
0
D
u
f[
0
D
u
g[
1
0
D
2
u
g[
0
D
u
g[
1
0
D
z
g[
0
vD
u
g[
1
0
D
z
g[
0
_
v
On the other hand, since in a neighborhood of (z
0
, u
0
,
0
) the Implicit Function
theorem implies that u =
h(z, ), and from the denition of the vector elds s()
32
and g() it follows that
D
z
s = D
z
f + D
u
f D
z
h (3.23)
0 = D
z
g + D
u
gD
z
h D
z
h[
0
= D
u
g[
1
0
D
z
g[
0
(3.24)
Let p = D
z
f v + D
u
f D
z
h v)
D
2
z
s[
0
v = D
2
z
f [
0
v + D
2
uz
f [
0
v D
z
h[
0
+ D
2
zu
f[
0
D
z
h[
0
v +
D
2
u
f[
0
D
z
h[
0
v D
z
h[
0
+ D
u
f[
0
D
z
p[
0
(3.25)
Now, from (3.24) it follows that
0 = D
2
z
g v + D
2
uz
g v D
z
h + D
2
zu
g p +
D
2
u
g p D
z
h + D
u
gD
z
p
D
z
p[
0
= D
u
g[
1
0
_
D
2
z
g[
0
v + D
2
uz
g[
0
v D
z
h[
0
+
D
2
zu
g[
0
D
z
h[
0
v + D
2
u
g[
0
D
z
h[
0
v D
z
h[
0
_
(3.26)
From equations (3.24), (3.25), and (3.26)
D
2
z
s[
0
v = D
2
z
f [
0
v D
2
zu
f[
0
D
u
g[
1
0
D
z
g[
0
v
D
2
uz
f[
0
vD
u
g[
1
0
D
z
g[
0
+
D
2
u
f [
0
D
u
g[
1
0
D
z
g[
0
vD
u
g[
1
0
D
z
g[
0
D
u
f [
0
D
u
g[
1
0
D
2
z
g[
0
v +
D
u
f [
0
D
u
g[
1
0
D
2
zu
g[
0
D
u
g[
1
0
D
z
g[
0
v +
D
u
f [
0
D
u
g[
1
0
D
2
uz
g[
0
vD
u
g[
1
0
D
z
g[
0
D
u
f [
0
D
u
g[
1
0
D
2
u
g[
0
D
u
g[
1
0
D
z
g[
0
vD
u
g[
1
0
D
z
g[
0
Therefore, from (3.22) we have that
w
T
[D
2
y
F[
0
v] v = w
T
[D
2
z
s[
0
v]v
33
Finally, based on (3.18) and since the right eigenvectors v and v are the same up
to the zero component corresponding to the generator frequency variables
G
, i.e.,
v =
_
_
v
x
dc
v
G
0
v
V
G
v
L
v
u
_
_
and v =
_
_
v
x
dc
v
G
v
V
G
v
L
v
u
_
_
we conclude that
w
T
_
D
2
y
F[
0
v
_
v = w
T
_
D
2
y
F[
0
v
_
v = w
T
_
D
2
z
s[
0
v
_
v
where y
= [ z
T
u
T
]
T
represents the set of variables in the power ow equations
F(y, ) = 0 (
G
is not included).
Theorems 3.1, 3.2, and 3.3 are used in section 3.6 to prove nonsingularity of
the Jacobians for the NewtonRaphson based methods used in this thesis to detect
bifurcation points. Also, as a corollary, these theorems yield the relationship between
the reduced vector eld s() eigenvectors at the saddlenode bifurcation and the
eigenvectors corresponding to the singular power ow jacobian D
y
F[
0
. This is useful
information since it can be employed to analyze the local dynamic behavior of the
power system at the bifurcation point as shown in [8].
3.4 The Point of Collapse Method
This technique is better known in the literature as the direct or extended system
method, and it was initially proposed by Seydel for nding saddlenode bifurcations
of nonlinear systems without constraint manifolds [6]. An extended version of this
34
method is applied in [9, 10, 13] to the complete set of power ow equations, including
algebraic constraints, to detect voltage stability problems in ac power networks. The
approach consists of augmenting the equations for equilibria with constraints (3.4)
ensuring a zero eigenvalue at the point of interest. Hence, the equations for z, v,
and take the following form:
s(z, ) = 0 (3.27)
D
z
s(z, ) v = 0
v ,= 0
Solving (3.27) is equivalent to solving equations (3.28) below for the power low
variables z and u, the parameter , and the right eigenvector v, since under the
assumption of D
u
g[
0
invertible, singularity of the power ow Jacobian is a necessary
and sucient condition for singularity of D
z
s[
0
. Furthermore, a nonsingular D
u
g[
0
also guarantees uniqueness of the zero eigenvalue of D
( z,u)
(
f, g)[
0
for a unique zero
eigenvalue of matrix D
z
s[
0
.
_
f ( z, u, )
g( z, u, )
_
_ = 0 (3.28)
_
_
D
z
f D
u
f
D
z
g D
u
g
_
_
_
_
v
z
v
u
_
_ = 0
_
_
v
z
v
u
_
_ ,= 0
The nonzero condition for the eigenvector can be guaranteed by requiring a
particular component of v
z
to be equal to one ( v
i
= 1). (Replacing v ,= 0 by v
i
= 1
can fail in the unlikely event of equations (3.28) yielding a nonzero eigenvector such
that v
i
= 0.) This technique also yields the eigenvector at the bifurcation point,
35
oering insight into the local dynamic behavior of the system close to bifurcation
[8, 53].
Based on theorems 3.1, 3.2, and 3.3, an alternative method for nding the bifur
cation point is to impose saddlenode conditions 1 and 2 on the composite function
[
f
T
( z, u, ) g
T
( z, u, )]
T
representing the power ow equations. This is equivalent to
the approach proposed by Van Cutsem for an ac system in [11], where a Lagrangian
of the reactive power ow equations is used to nd the bifurcation. Equation (3.29)
below shows the constraints used in this approach. Note that (3.29) are essen
tially the same as (3.28), except that the left zero eigenvector w is used to ensure
singularity.
_
f ( z, u, )
g( z, u, )
_
_ = 0 (3.29)
_
w
T
z
w
T
u
_
_
_
D
z
f D
u
f
D
z
g D
u
g
_
_ = 0
T
_
w
T
z
w
T
u
_
_
f /
g/
_
_ = K
Here K is an arbitrary nonzero scalar (such as K = 1), and
f
i
=
_
_
P
l
for load buses
0 otherwise
g
i
=
_
_
Q
l
for load buses
0 otherwise
The equations dened above may have several solutions, indicating that the
system model could have dierent parameter values for several bifurcation points.
However, since a NewtonRaphson method is employed for nding the bifurcations,
36
a good starting guess is needed to obtain the desired solution . A technique for
obtaining these initial guesses, and the computational issues involved in applying
the Point of Collapse method to large ac/dc systems are discussed in section 3.6.
Because the dc system model used here can represent hard limits (and, hence,
has discontinuous derivatives), it is necessary to check whether controller limits
have been reached during the solution process. If so, the equation structure can be
changed to reect this fact and allow for further analyses of the stability of the new
equilibria. The same approach can be used to handle any kind of system limits like
transformer tap limits, and generator voltage and reactive power limits (Qlimits).
3.5 Parameterized Continuation Methods
Continuation methods have been used since the 60s in a variety of engineering
elds, as described by Seydel in [6]. In the power systems literature, for example,
one can nd references to these methods when utilized for solving optimization
problems of power networks as shown in [44]. In [6], continuation methods are
employed for detecting proximity to saddlenode bifurcations in dynamic systems.
These methods are also applied to the analysis of voltage collapse for ac only systems
in [12, 13, 14]. Reference [14] shows a direct application of Seydels parameterized
continuation methods to the ac power ow equations. In [12], the authors do not
use parameterization techniques, but the idea of a perpendicular intersection to
improve the convergence characteristics of these methods is used. Finally, paper
[13] presents and alternative parameterized continuation method to nd the point
of voltage collapse in ac systems. The method presented in this thesis uses both
parameterization and perpendicular intersection techniques, to trace the branch of
equilibria (voltage proles) associated to saddlenode bifurcations in ac/dc networks.
37
1
2
PoC
1
z
z
o
90
z
1
()
Figure 3.2: Continuation method geometry in R
m
R. The state space variables
are z R
m
, and R is the varying parameter.
The continuation method consists of a three step approach to tracing the equi
librium points as one parameter in the system changes, i.e., nd the solutions to
the steady state equations s(z, ) = 0 in (3.3), for a given set of parameter values.
Typically the loading factor is the varying parameter; however, as the system gets
closer to bifurcation the Jacobian D
z
s[
0
becomes illconditioned. This can create
illconditioning in the numerical process used to trace the equilibria, but a change of
parameter, such as switching from to a state variable z
i
z, makes the Jacobian
nonsingular, as it is demonstrated later.
Figure 3.2 shows graphically how the proposed method works. The boldface
curve represents a branch of system equilibria as the system parameter changes.
Assuming that the system is initially at the state (z
1
,
1
), one can move to a neigh
borhood of the new equilibria (z
2
,
2
) by using and the scaled tangent vector
38
z
1
, since for a small step z
2
z
1
+ z
1
and
2
1
+ . To obtain the actual
values of z
2
and
2
, one can use a perpendicular plane to the tangent vector to nd
the desired point in the manifold.
For the power system dynamic model used throughout this thesis, where the
power ow equations are represented by the vector eld F() = [
f
T
() g
T
()]
T
and
the corresponding variables y = [ z
T
u
T
]
T
, and for a nonsingular Jacobian of g(),
there is a one to one correspondence between the equilibria of s() and the solutions
of F(y, ) = 0, i.e., the manifold depicted in gure 3.2 has a corresponding smooth
curve in R
Nn
G
+1
when no limits are reached in a neighborhood of the region of
interest, since F() is C
r
, r 2, as discussed in section 3.2. Furthermore, based
on theorems 3.1, 3.2 and 3.3, a saddlenode bifurcation in the vector eld s() can
be detected using F(). Therefore, one can trace a bifurcation branch utilizing the
power ow equations.
The proposed continuation method algorithm can be summarized in the following
three steps:
1. Predictor: From F(y, ) = F( y, p) = 0, where initially y = y and p = , it
follows that
D
y
F( y
1
, p
1
)
d y
dp
+
F
p
1
= 0
where p
1
and y
1
come from the previous iteration of the method. The direction
y to move in state space can be found by solving equations (3.30) below for
d y/dp.
D
y
F[
1
d y
dp
=
F
p
1
(3.30)
Hence,
y
= p
d y
dp
39
where the parameter increment p can be dened as a function of a scaling
constant k to vary the speed at which the equilibrium branch is traced, i.e.,
p
=
k
 d y/dp 
(3.31)
As the process approaches the bifurcation, p is likely to change to one of the ac
bus voltages, according to step 3, with the loading factor becoming part of y.
The scaling constant k allows for varying the speed at which the equilibrium
branch is traced.
2. Corrector: Find the intersection between the perpendicular hyperplane to the
tangent vector and the equilibrium branch, i.e., solve equations
F( y, p) = 0 (3.32)
p (p p
1
p) + y
T
( y y
1
y) = 0
By initially setting y to y
1
+ y and p to p
1
+ p, solving this set of equa
tions usually takes one or two iterations. Nevertheless, if the process fails to
converge, the steps y and p are cut in half until convergence is attained.
Enforcing ac and dc limits is done with ease, since the initial guesses are good
estimates of the actual solution. If the initial values violate ac/dc limits, the
steps are cut so that all limits are met, and then the continuation equations
(3.32) are solved. When [p[ is cut to less than certain tolerance (typically
10
4
), the corresponding ac equations, or dc state variables, are changed ac
cording to the limits that have been reached.
3. Parameterization: Check the relative change in all the system variables, and
trade p with the variable that presents the largest change. In other words,
p max
_
y
1
y
1
y
2
y
2
, ,
y
n
y
n
p
p
_
40
Experience with the method has demonstrated that as the process approaches
the bifurcation, p changes from to the system bus voltage that is varying the
most, and after a few iterations of the method it returns back to . However, in
all the cases tested, good results were obtained without switching parameters.
By changing the parameter p from to a state variable y
i
y, one guarantees
that the Jacobian of equations (3.30) is nonsingular at the bifurcation point. Fur
thermore, it can be shown that the Jacobian of equations (3.32) is also nonsingular
at the bifurcation point, even for p = (singular power ow Jacobian D
y
F[
0
). These
statements are formally proved in the next section.
The method naturally goes around the collapse point, allowing the user to
trace the unstable side of the equilibrium branch. However, this turning point
must be detected in order to change the sign of p in equations (3.30). For p = ,
the bifurcation point can be spotted by looking for a sign change in the determinant
of the power ow Jacobian. On the other hand, for p = y
i
, the turning point can be
detected by looking for a sign change in .
3.6 Computational Issues
One of the main concerns when applying the Point of Collapse and continuation
methods is the singularity of the equations Jacobians at the bifurcation point,
which could cause numerical problems when a NewtonRaphson solution algorithm
is used. In this section, the formal proofs of nonsingularity of the Jacobians for
both methods are presented. Also, practical considerations are discussed, so that
the methods can be used to reliably and consistently solve large ac/dc systems with
a variety of ac and dc operational limits.
41
3.6.1 PoC Methods
Using F() and y as dened above, the left eigenvector PoC equations (3.29) can be
rewritten as
D
y
F(y, )
T
w = 0 (3.33)
F(y, ) = 0
F
T
w = K
Since F(y, ) is linear with respect to , the Jacobian for (3.33) at the bifurcation
point is
J
PoC
=
_
_
D
2
y
F[
T
0
w D
y
F[
T
0
0
D
y
F[
0
0
F
0
0
F
T
0
0
_
_
(3.34)
In spite of the individual block D
y
F(y
0
,
0
) being singular, conditions (3.5) and
(3.6) guarantee that J
PoC
is nonsingular as shown in theorem 3.4. (Reference [47]
presents a similar proof.)
Theorem 3.4 Let F : R
Nn
G
+1
R R
Nn
G
+1
, F(y, ), vectors y, v, w
R
Nn
G
+1
, and scalar R be dened as theorem 3.3. Then, if vector eld s() in
(3.3) has a saddlenode bifurcation satisfying transversality conditions (3.4), (3.5)
and (3.6) at the equilibrium (y
0
,
0
), and matrix D
u
g[
0
is nonsingular, then matrix
J
PoC
is also nonsingular.
Proof: In order to prove J
PoC
is nonsingular, one has to prove that the only vector
that satises equation (3.35) below is the zero vector, i.e., [p
T
q
T
r]
T
= 0, where p,
42
q R
Nn
G
+1
, and r R.
_
_
D
2
y
F[
T
0
w D
y
F[
T
0
0
D
y
F[
0
0
F
0
0
F
T
0
0
_
_
_
_
p
q
r
_
_
=
_
_
0
0
0
_
_
(3.35)
From (3.35) one has that
D
y
F[
0
p +
F
0
r = 0
T
F
0
r =
T
D
y
F[
0
p R
Nn
G
+1
In particular, if = w
w
T
F
0
r = w
T
D
y
F[
0
p = 0
Hence, from (3.5) and (3.13) we conclude that r = 0, and p = k v, k R, or p = 0.
If p = k v, from (3.35) it follows that
k[D
2
y
F[
T
0
w] v + D
y
F[
T
0
q = 0
k w
T
[D
2
y
F[
0
v] = q
T
D
y
F[
0
R
Nn
G
+1
If = v, then
k w
T
[D
2
y
F[
0
v] v = q
T
D
y
F[
0
v = 0
But from (3.6) and (3.19), this is a contradiction; therefore, p = 0 and
D
y
F[
T
0
q = 0
q =
_
_
k w for k R
0
If q = k w, from (3.35) it follows that
k
F
T
0
w = 0
43
which is also a contradiction, then k = 0. Hence, q = p = 0 and r = 0.
The same arguments can be used to formally prove nonsingularity of the Jaco
bians corresponding to the PoC equations (3.36) and (3.37) below, where  
is
the innity norm the respective eigenvector.
D
y
F(y, ) v = 0 (3.36)
F(y, ) = 0
 v 
= 1
D
y
F(y, )
T
w = 0 (3.37)
F(y, ) = 0
 w 
= 1
Experimentation with these equations on large ac/dc systems has demonstrated that
the left eigenvector PoC equations (3.37), yield better computational performance.
Notice that the Jacobians for all PoC equations are symmetric for an ac lossless
system [11], and topologysymmetric for a general ac system. However, the equations
for the HVDC links cause the power ow Jacobian to become unsymmetric. This
results in a computational penalty, from both performance and programming points
of view.
An important issue in solving the Point of Collapse equations by Newtons
method is the choice of initial guess for eigenvectors v or w. The approach used
here is to apply several iterations of the Inverse Power Method [48] to the power
ow Jacobian, i.e., solve the linear set of equations
D
y
F(y
)
T
w
k+1
= w
k
k = 0, 1, 2, . . .
at the initial power ow solution y
, and for
any initial vector w
0
R
Nn
G
+1
(usually w
0
= [1 1 1]
T
). A normalization of w
k
44
at each iteration k must be introduced to avoid numerical problems. This typically
yields a vector close to the span of the real and imaginary parts of the eigenvec
tor(s) corresponding to the eigenvalue(s) with the smallest magnitude. Although
this is not guaranteed to be the eigenvalue that eventually goes to zero at bifurca
tion, the method has proven eective in a number of sample systems. Notice that
close to the bifurcation this method would yield a vector very close to the desired
eigenvector, since simple eigenvectors at equilibria vary smoothly with system pa
rameter changes. Experience with the PoC technique in large ac/dc systems has
demonstrated that when ac and dc limits are reached, new values for the eigenvec
tors should be recalculated to improve convergence of this method. This is done by
applying a few iterations of the inverse power method to the Jacobian of the new
set of equations at the switching point.
In large power networks, ac control limits can be represented in the steady state
power ow model by voltage regulating transformer tap limits, phase shifter angle
limits, and voltage and reactive power limits at generator buses. For the HVDC
system some further limits must also be considered (in addition to the ring angle
and extinction angle limits represented in the proposed converter dynamic model).
Introducing commutation transformer tap limits, and current and power limits yields
a better representation of the actual control behavior of the dc system.
In the case of the ac system, enforcing limits does not present major diculties
if active and reactive power mismatches at all generator buses, including slack and
PV buses, are explicitly represented in the power ow problem. However, when a
variable reaches an operational limit in the HVDC link, the system changes control
mode. This can be modelled by a change in the state variables used to simulate
the dc steady state behavior. Therefore, a scheme has to be devised in order to
accurately represent the dynamic control logic at both converter stations. Figures
45
3.3 and 3.4 depict the transition logic used in this thesis to simulate the switching
between dierent control modes for rectiers and inverters, when the ac bus voltage
at either side of the dc line changes. In these gures, the boxes represent the four
dc variables that must be kept constant in order to obtain a solution to the ac/dc
power ow; N stands for nominal values, M represents maximum values, and m
corresponds to minimum values. For constant power control, one can substitute
active power for dc current in both gures.
For systems relatively far from the bifurcation point, the initial guesses described
above are not sucient to obtain consistent results. By initially loading the system,
i.e., moving the system variables closer to the collapse point, one can resolve this
diculty. This initial load can be calculated using the predictor equations (3.30)
of the continuation method. These yield a direction in which to move so that the
system gets closer to the bifurcation. Once the step is calculated, the power ow
equations are solved again at the new load level. Several tests have demonstrated
that an accurate solution to this new power ow is not required to obtain good
initial values for the PoC variables.
The scaling constant k in equations (3.30) is used to simulate dierent load
levels. Experimenting with several networks and a variety of values for k, one can
observe that the bifurcation point is closely related to the number of generators in
the system, especially when reactive power limits are enforced. In particular, the
heuristic
k =
n
G
has proven to provide a good value for k. This k is used unless more than 10 %
of the generators reach limits when the system is initially loaded. If this limit is
exceeded, the value of k is reduced by half until the criterion is met. Although this
heuristic worked in most of the cases tested, one cannot guarantee that this will
46
r
N
, I
d
N
i
m
, V
d
i
N
a
r
M
, I
d
N
i
m
, V
d
i
N
a
r
m
, I
d
N
i
m
, V
d
i
N
a
r
m
,
r
m
I
d
N
, a
i
SHUT DOWN
Figure 3.3: Control modes of HVDC link for power ow solution, when the ac bus
voltage at the rectier side changes. Margin switching is assumed.
r
N
, I
d
N
i
m
, V
d
i
N
r
N
, I
d
N
i
m
, a
i
M
r
N
, I
d
N
i
m
, a
i
m
a
r
m
,
r
m
i
m
, a
i
M
SHUT DOWN
a
r
M
, I
d
N
i
m
, a
i
m
a
r
M
,
r
m
I
d
N
, a
i
m
Figure 3.4: Control modes of HVDC link for power ow solution, when the ac bus
voltage at the inverter side changes. Margin switching is assumed.
47
always yield the desired results. Since this process is used only to obtain an initial
guess for the PoC equations, the value used for k can be changed by the user.
With a proper initial guess, the computational costs of solving the PoC equations
are somewhat higher than those for the original power ow equations. The number
of equations has doubled, but the Jacobian J
PoC
retains its sparsity. Furthermore,
close to the bifurcation point the power ow Jacobian J
PF
becomes illconditioned,
causing slow convergence of power ow methods, whereas the Jacobian for the pro
posed Newton solution is typically far from singularity. Chapter 5 depicts compu
tation times of the PoC method for several networks, and compares its performance
against continuation methods.
3.6.2 Continuation Methods
Techniques similar to the ones used in the proof of theorem 3.4 can be used to show
that the Jacobians corresponding to equations (3.30) and (3.32) of the continuation
method are also nonsingular at the bifurcation point. Although parameterization
is necessary to guarantee nonsingularity of the power ow Jacobian in the predic
tor step of the method, the continuation equations of the corrector step can be
shown nonsingular at the collapse point without this assumption. This is the reason
why the parameterization technique described in step 3 is not necessary in practice
for obtaining voltage proles, as long as automatic step reduction is utilized (this
technique assures convergence of equations (3.32)). Theorems 3.5 and 3.6 below es
tablish nonsingularity of the continuation equation Jacobians at the point of voltage
collapse.
Theorem 3.5 Let F : R
Nn
G
+1
R R
Nn
G
+1
, F(y, ), y R
Nn
G
+1
, and
R be dened as in theorem 3.3. At the bifurcation point (y
0
,
0
) satisfying
48
transversality conditions (3.4), (3.5) and (3.6), where F(y
0
,
0
) = 0 and D
y
F[
0
v =
0, let the parameter p
0
R and the vector y
0
R
Nn
G
+1
be dened as
p
0
= y
0
M
where [ v
M
[ [ v
i
[ i ,= M, v = [ v
1
v
Nn
G
+1
]
T
y
0
=
_
y
0
1
y
0
M1
0
y
0
M+1
y
0
Nn
G
+1
_
T
Hence, F( y
0
, p
0
) = 0, and D
y
F[
0
is nonsingular if vectors F/ = F/ y
M
[
0
and
F/y
M
[
0
= F/p[
0
in R
Nn
G
+1
are not collinear.
Proof: From theorem 3.2, matrix D
y
F[
0
has a unique zero eigenvalue at the
saddlenode bifurcation (y
0
,
0
), with a unique right eigenvector v. Then, equa
tion D
y
F[
0
v = 0 can be rewritten as
Nn
G
+1
j=1
v
j
F
y
j
0
= 0
F
y
M
0
=
Nn
G
+1
i=1,i=M
i
F
y
i
0
where all
i
= v
i
/ v
M
are unique, and v
M
,= 0 by denition. Therefore, replacing
column M in D
y
F[
0
by vector F/ makes matrix D
y
F[
0
nonsingular at the saddle
node bifurcation point, since F/ is not collinear with F/y
M
[
0
by assumption.
The assumption of lack of collinearity between vectors F/ and F/y
M
[
0
is
generic, since it is likely that for some index j, 1 j N n
G
+ 1, F
j
/ ,= 0
when F
j
/y
M
= 0 (F = [F
1
F
2
. . . F
Nn
G
+1
]
T
).
At the saddlenode bifurcation (y
0
,
0
) satisfying the transversality conditions
(3.4), (3.5) and (3.6), predictor equations (3.30) can be rewritten, for y = y and
p = , as
D
y
F[
0
dy =
F
d
49
where [t
T
y
t
]
T
= [dy
T
d]
T
is the tangent vector to the manifold F(y, ) = 0 at
the bifurcation point in R
Nn
G
+1
R. Then, from the transversality condition
w
T
F/ ,= 0, and since w
T
D
y
F[
0
= 0
T
, it follows that
w
T
D
y
F[
0
t
y
= w
T
F
D
y
F[
0
t
y
= c D
y
F[
0
v = 0 c R, c ,= 0 (3.38)
Therefore, at the bifurcation, the tangent vector dy/d of the corrector step is
collinear with the right eigenvector v. This implies, from theorem 3.5, that a proper
parameterization, i.e., changing p = to p = y
M
, should make the Jacobian of pre
dictor equations (3.30) generically nonsingular along the bifurcation branch, since
singularity points represented by saddlenodes can be parameterized to avoid nu
merical problems.
Theorem 3.6 Let F(), y and be dened as in theorem 3.5. Let the bifurcation
branch of equilibria be represented by the smooth manifold F(y, ) = 0 in R
Nn
G
+1
) t
y
=
F
where F(y
) = 0. Assume that t
y
is nonperpendicular to the properly normal
ized unique right eigenvector v dened in theorem 3.3. Then, the Jacobian of the
corrector equations (3.32), i.e.,
J
C
=
_
_
D
y
F[
0
F
y
T
_
50
is nonsingular at the saddlenode bifurcation point. Here
= k/  t
y
, and y
= t
y
. The scalar k R is chosen so that the saddle
node bifurcation (y
0
,
0
) is the solution to predictor equations (3.32).
Proof: Once again, to prove that J
C
is nonsingular, one has to prove that vector
a R
Nn
G
+1
and scalar b R are identical to zero in (3.39).
_
_
D
y
F[
0
F
y
T
_
_
_
a
b
_
_ = 0 (3.39)
From equation (3.39), it follows that
D
y
F[
0
a =
F
b
T
D
y
F[
0
a =
T
F
b R
Nn
G
+1
(3.40)
If = w ( w
T
D
y
F[
0
= 0
T
), then equation (3.40) implies b = 0, since w
T
(F/) ,=
0. Therefore, a must be either c v, c R, or 0. However, if a = c v, it follows from
(3.39) that
y
T
v = t
T
y
v = 0
but this contradicts the assumption of vectors t
y
and v being nonperpendicular.
Consequently, a = 0 and b = 0.
The assumption of vectors t
y
and v being nonperpendicular can be justied
based on t
y
= c v at the bifurcation, as shown above, and the smoothness of the
bifurcation branch of equilibria. The branch of equilibria near a saddlenode bifur
cation satisfying the transversality conditions of section 3.2 can be proved smooth
using the LyapunovSchmidt reduction [40]. Hence, generically, one can expect to
fulll this assumption in a neighborhood of the bifurcation point.
Theorem 3.6 shows that predictor equations (3.32) have a nonsingular Jacobian
at the bifurcation point for parameter p = . This property yields a numerical well
51
Element G B B
s
Line 12 3.68 54.13 4.68
Line 23 3.68 54.13 4.68
Transf. G
1
166.67
Transf. G
2
100.00
Transf. 100.00
Capac. 1 13.00
Capac. 3 13.68
Table 3.1: AC transmission system data (ESCR=6.2). All quantities are in p.u.
for a 550 kV and 100 MVA base.
behaved predictor step of the continuation method without the need of parameter
ization.
A computer implementation of the PoC and continuation methods, including
all practical considerations regarding handling of limits in large ac/dc systems, is
briey described in appendix B.
3.7 Voltage Proles and Bifurcations in a Sample
System
Figure 3.5 shows the sample ac/dc system used throughout this thesis to simulate
several ac balanced faults and voltage collapse scenarios, and to illustrate the use
of the proposed stability analysis techniques. This system is a reduction of the 14
bus system depicted in chapter 5 and described in appendix A, designed to roughly
approximate some of the characteristics of power systems in the western part of the
United States. The ac and dc data for the reduced network are shown in tables 3.1
and 3.2, respectively.
52
G
1
G
2
1
2
3
P
L
+jQ
L
R
I
Figure 3.5: Sample ac/dc system
53
Variable Rectier Inverter
a 1.7634 1.7678
X
c
0.1345 0.1257
min
5
0
120
0
max
120
0
142
0
min
40
0
18
0
max
155
0
40
0
I
o
1.0 0.9
I
o
min
0.1 0.0
V
ac
max
0.95 0.95
V
ac
min
0.5 0.5
R
d
0.0624
Assuming 20
0
VDCOL
Table 3.2: DC system data. All quantities are in p.u. for a 550 kV and 2.5 kA
base.
In this system, generator G
2
supports the voltage at the intermediate load bus,
and bus 1 has relatively strong voltage support from generator G
1
. Two dierent
eective short circuit ratios (ESCR) at bus 1, 21.4 and 6.2, are used to simulate
dierent system strengths, whereas the ESCR at the inverter side is kept constant
at about 4. The dc line is designed to supply approximately two thirds of the power
needed at bus 3. The active power for the load at bus 2 is assumed frequency
dependent, whereas the reactive power is modelled as a constant reactance. The
generators are simulated as constant voltage sources with classical second order
swing dynamics. The initial load at bus 2 is P
0
L
= 475 MW, and Q
0
L
= 156.1
MVAR.
For variations in the load at bus 2, a computer assisted symbolic algebra package
[50, 51] was used to solve the continuation method equations, and obtain proles
54
for the three bus voltage magnitudes shown in gures 3.6 through 3.11. Figures
3.6, 3.7, 3.8, and 3.9 show the nose curves (bifurcation diagrams) for changes in
the power demand at bus 2, when VDCOL is not considered. Eigenvalue analysis
of system equilibria demonstrates that voltage magnitudes higher than those at
the point of collapse (saddlenode bifurcation) are stable, whereas equilibria with
voltages below this point are unstable. Figures 3.6 and 3.8 show several bifurcation
points in the unstable region, and also sharp turning points for the reduced ESCR
due to changes in the control modes of the HVDC link. This behavior seems to
be dierent from the way an ac only system behaves (at least for the type of ac
model employed here), since in such systems the shape of the nose curves appears
to be globally quadratic, at least for the numerical examples depicted in chapter
5. (Notice that all bifurcation diagrams, regardless of the vector eld, are quadratic
about the saddlenode point since the phenomenon can be locally represented by
equation (3.2) using a LyapunovSchmidt reduction [7, 40].) The unique character
of the combined ac and HVDC system is more evident when VDCOL is included, as
depicted in Figures 3.10 and 3.11. Here the ac/dc system has a completely dierent
set of unstable equilibria, as well as dierent points of collapse, i.e., the loadability
margin of the system has changed with the inclusion of VDCOL.
Table 3.3 shows the values of system variables at the point of collapse, obtained
by solving equations (3.29). The voltage magnitude at bus 1 is relatively high due
to the strong voltage support from generator G
1
and the HVDC ltering system, es
pecially for a large rectier ESCR. Notice also that the loadability margin increases
with the ESCR, which is to be expected. For the case where VDCOL is not con
sidered, the voltage magnitudes at bifurcation are higher when collapse is reached
via active load increase as compared to reactive increase, which is also reasonable
due to the high reactive demands in the latter case. Although the points of collapse
55
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
PoC
Bif. Bif.
V1
V2
V3
Loading Factor (MW)
V
o
l
t
a
g
e
(
p
.
u
.
)
Figure 3.6: AC voltage proles for active power changes at load bus 2 (P
2
= 1,
Q
2
= 0). ESCR=21.4, and no VDCOL.
56
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
1.05
0 500 1000 1500 2000 2500 3000
PoC
V1
V3
V2
Loading Factor (MVAR)
V
o
l
t
a
g
e
(
p
.
u
.
)
Figure 3.7: AC voltage proles for active power changes at load bus 2 (P
2
= 0,
Q
2
= 1). ESCR=21.4, and no VDCOL.
57
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
PoC
Bif.
Bif.
Bif.
V1
V2
V3
Loading Factor (MW)
V
o
l
t
a
g
e
(
p
.
u
.
)
Figure 3.8: AC voltage proles for active power changes at load bus 2 (P
2
= 1,
Q
2
= 0). ESCR=6.2, and no VDCOL.
58
0.5
0.6
0.7
0.8
0.9
1
1.1
0 500 1000 1500 2000 2500
PoC
V1
V2
V3
Loading Factor (MVAR)
V
o
l
t
a
g
e
(
p
.
u
.
)
Figure 3.9: AC voltage proles for reactive power changes at load bus 2 (P
2
= 0,
Q
2
= 1). ESCR=6.2, and no VDCOL.
59
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
VDCOL
PoC
Bif. V1
V2
V3
Loading Factor (MW)
V
o
l
t
a
g
e
(
p
.
u
.
)
Figure 3.10: AC voltage proles for active power changes at load bus 2 (P
2
= 1,
Q
2
= 0). ESCR=6.2, and VDCOL included.
60
0.5
0.6
0.7
0.8
0.9
1
1.1
0 500 1000 1500 2000 2500
PoC
VDCOL
V1
V2
V3
Loading Factor (MVAR)
V
o
l
t
a
g
e
(
p
.
u
.
)
Figure 3.11: AC voltage proles for reactive power changes at load bus 2 (P
2
= 0,
Q
2
= 1). ESCR=6.2, and VDCOL included.
61
do not change signicantly when VDCOL is included, tracing the unstable equilib
rium points with the continuation method becomes more dicult due to the series
of dc limits that are encountered during the searching process. However, nding
the bifurcation points with the PoC method is straightforward in all cases. This
makes the PoC method very appealing when compared to other methods that need
unstable equilibrium information (e.g., [15, 52]), since the degree of diculty in the
PoC method does not change with the mode of operation of the HVDC system.
This is corroborated in larger systems in chapter 5, where the PoC method con
sistently shows better time performance than the continuation method, especially
when ac/dc limits are considered.
Based on the bifurcation values for active power load, the system has a large
maximum loadability due in part to the innite bus and the transmission line
design. The active load changes are supplied mainly by the innite bus, whereas
the reactive increase is supported by the generators at buses 1 and 2; this in part
explains the smaller reactive bifurcation values. Although the loadability margin
changes when VDCOL is included, one cannot draw a denite conclusion regarding
the advantages or disadvantages on voltage stability of this mode of operation.
Another interesting characteristic of the sample ac/dc system relates to the be
havior of the eigenvalues before bifurcation when VDCOL is included. As shown
in gure 3.12, a complex conjugate pair crosses the imaginary axis from left to
right as the system load increases (at about 4680 MW of additional load), this
happens slightly before the system becomes unstable due to voltage collapse. This
phenomenon is known as Hopf bifurcation, which is another mechanism of transi
tion to instability in dynamical systems characterized by complex eigenvalues of the
linearized system dynamics crossing the imaginary axis as the system parameters
change [5, 6]. This type of bifurcation has been proven not to exist for simple ac
62
Rectier ESCR=21.4 Rectier ESCR=6.2
No VDCOL No VDCOL VDCOL
Var. P
L
Q
L
P
L
Q
L
P
L
Q
L
45.58 26.37 44.66 21.96 46.84 20.05
1
48.80
0
85.33
0
43.6
0
84.7
0
44.5
0
84.2
0
2
69.30
0
61.15
0
66.1
0
58.7
0
75.0
0
58.7
0
3
6.88
0
29.57
0
6.0
0
29.1
0
8.1
0
29.1
0
g
1
45.03
0
89.12
0
28.2
0
100.
0
28.9
0
99.7
0
g
2
66.00
0
64.86
0
62.8
0
62.3
0
71.4
0
61.1
0
V
1
0.978 0.969 0.904 0.882 0.889 0.902
V
2
0.826 0.734 0.812 0.749 0.766 0.778
V
3
0.792 0.767 0.802 0.782 0.773 0.812
I
d
1.000 1.000 1.000 1.000 0.878 0.904
V
d
r
1.741 1.684 1.763 1.717 1.705 1.792
V
d
i
1.678 1.622 1.701 1.655 1.650 1.735
r
36.65
0
38.23
0
28.5
0
28.5
0
30.9
0
27.4
0
i
18.0
0
18.0
0
18.0
0
18.0
0
18.0
0
18.0
0
Table 3.3: Points of collapse for V
g
1
= V
g
2
= V
g
3
= 1. Voltages and currents are
in p.u. for base quantities V
n
= 550 kV, I
n
= 2.5 kA, and S
n
= 100 MVA.
63
60
40
20
0
20
40
60
2 1.5 1 0.5 0 0.5 1
o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o
o
o
o
o
o
o
o
o
o
o o o
o
o
o
o
o
o
o
o
o
o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o
o
o
o
o
o
o
o
o
o
o o o
o
o
o
o
o
o
o
o
o
oooooooooooooooooooooooooooooooooooooooo o o o
o o o o o o o
o
o
o
oooooooooooooooooooooooooooooooooooooooo o o o
o o o o o o o
o
o
o
x
x
x
x
x x x x x x x x x x x
x
x x
x
x
x
x
x
x x x x x x x x x x x
x x x x x
x
x
x
x
x
x
x
x x
x
x
x
x
x
x x
x x x x x
x x x
x
xx x x x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x x x x
x
x
x
x
x
x
x
x
x
x
x
x
x
x x
x x x x x
x
x
x
xx x x x
x x
x
x
x
x
x
x
x
x x x x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
Real
I
m
a
g
i
n
a
r
y
Figure 3.12: Eigenvalues for the stable (o) and unstable (x) equilibria for active
power changes at bus 2. ESCR=6.2, VDCOL included.
lossless system models [54]. However, it has been observed in more detailed ac sys
tem models that consider transfer conductances and exciter dynamics [27]. This
is also signicant because it indicates that a global Lyapunov function does not
exist for the ac/dc dynamic model used in this thesis [55]. Reference [28] presents
some computational techniques for directly nding these kind of local bifurcations
in nonlinear power system models.
Chapter 4
Transient Energy Function
4.1 Introduction
Transient energy function methods for stability analysis in power systems are based
on the Lyapunov theorems for establishing asymptotic stability and regions of attrac
tion for equilibria [20]. Lyapunovs direct method considers nonlinear autonomous
systems of the form
z = s(z) z R
n
(4.1)
such that s(z
0
) = 0, and establishes stability properties through a scalar function
of the state V (z), V : R
n
R, that is locally positive denite (l.p.d.f.) at the
asymptotically stable equilibrium z
0
, with time derivative
c
= z R
n
: V (z) V
cr
can be used to decide whether the trajectory starting at this point would
approach z
0
when compared to V
cr
, i.e., if
V (z
) V
cr
(4.2)
the trajectory converges to the s.e.p. When the test fails, it does not necessarily
imply that the system state is unstable.
Hence, the idea is to nd the scalar function V (z) (Lyapunov function) and its
upper bound, so that one can plug in the system state at any point in time in this
function and determine whether it is stable. Although this technique cannot be
used to conclude instability of the system state, the energy functions V (z) used
in ac power systems, and the methods developed for determining the upper bound
66
V
cr
have proven in practice very accurate in predicting the behavior of the system
during rst swing transient stability analyses, so much so that it is currently being
used to estimate the critical clearing times in fault studies of ac networks [56, 57].
This chapter rst presents an overview of energy functions for structure preserv
ing ac system models, and briey describes the techniques used for approximating
the boundary of the stability region of a s.e.p. in an ac power network, namely,
the controlling u.e.p. and the Potential Energy Boundary Surface (PEBS) methods
[17, 18, 19]. Secondly, a new energy function for the proposed ac/dc system models
is studied and applied to rst swing transient stability and voltage collapse studies
of the sample system in chapter 3.
4.2 AC Energy Function
During the last decade, many researchers have thoroughly analyzed the use of energy
functions for the direct stability assessment of ac networks. Particularly, the TEF for
structure preserving ac system models has been widely studied [33, 34, 35, 36, 37].
These models are based on the representation of the ac system suggested in chapter
2, with the only dierence that the active part of the load has no dependence on
voltage, i.e., equations (2.3) can be rewritten as:
P
l
(
l
) = P
0
l
D
l
(
l
+
n
G
)
Q
l
(V
l
) = Q
l
1
_
V
l
V
0
l
_
2
Q
l
2
_
V
l
V
0
l
_
Q
l
3
where P
l
(
l
) and Q
l
(V
l
) are power injections at the load bus l. The parameter in
(2.3) is assumed constant for the time scale of interest in rst swing stability studies;
this denes the loading level of the power network.
67
As a rst approximation of the eect of HVDC links in transient stability stud
ies, we will consider a dc system model whose terminal behavior is represented by
constant active and reactive load injections. Hence, vector equations (2.6) can be
replaced by (4.3) below [33, 34], assuming
n
G
to be the phase angle reference
1
G
= T
G
G
(4.3)
M
G
G
= T
T
G
f
G
(, V) +T
T
L
f
L
(, V) D
G
G
D
x
V
G
= g
G
(, V) = U
1
G
g
G
(, V)
D
L
L
= f
L
(, V) +D
L
T
L
G
0 = g
L
(, V) = U
1
L
g
L
(, V)
Here matrix T
G
= [I
n
G
1
[1
n
G
1
]
(n
G
1)n
G
(1
m
= [1 1 1]
T
1m
), and matrix T
L
= [0[
1
n
L
]
n
L
n
G
. U
G
and U
L
are positive denite diagonal matrices with diagonal entries
equal to the respective voltage magnitudes. Vectors f
G
, f
L
, g
G
, and g
L
are the active
and reactive power bus mismatch equations dened in (2.7). Note that the HVDC
converter buses are treated as load buses with D
l
small, so that D
L
is a diagonal
positive denite matrix.
In [37] it is proposed to singularly perturb the algebraic constraints in (4.3) to
have a well posed nonlinear system, i.e., system trajectories for all feasible initial
conditions are well dened at all times, even when the linearization of the algebraic
constraints at some time t, D
V
L
g
L
[
t
, is singular. Therefore, equations (4.3) take the
form
G
= M
1
G
[T
T
G
f
G
(, V) +T
T
L
f
L
(, V) D
G
G
] (4.4)
G
= T
G
G
1
The assumption of generator n
G
being an innite bus is not needed in this case, as
long as the transfer conductances are neglected. If transfer conductances are considered,
the system equilibria do not meet the condition
G
= 0, unless either an innite bus or
generator governor dynamics are dened.
68
L
= D
1
L
f
L
(, V) +T
L
V
G
= D
1
x
g
G
(, V)
V
L
= 1/ I
n
L
g
L
(, V)
where is a small positive parameter.
If the transfer conductances in (4.4) are neglected, the following candidate Lya
punov function can be dened for the stable equilibrium (0,
0
, V
0
):
V
ac
(
G
, , V) =
_
(
G
,,V )
(0,
0
,V
0
)
_
(M
G
)
T
f
T
(, u) g
T
(, u)
_
_
_
d
d
du
_
_
where f()
= [f
T
G
() f
T
L
()]
T
and g()
= [ g
T
G
() g
T
L
()]
T
. The integral dened
above is path independent if
2
V
ac
(
G
, , V) is symmetric, which is the case when
neglecting transfer conductances. Therefore, the candidate Lyapunov function can
be expressed in closed form as shown in (4.5).
V
ac
(
G
, , V) =
1
2
T
G
M
G
G
+
1
2
n
k=1
n
j=1
B
kj
V
0
k
V
0
j
cos(
0
k
0
j
)
1
2
n
k=1
n
j=1
B
kj
V
k
V
j
cos(
k
j
)
k=1
P
0
k
(
k
0
k
)
n
k=1
_
V
k
V
0
k
Q
k
()
d (4.5)
Recall that G
kj
+jB
kj
is the kj term of the bus admittance matrix, and P
0
k
+jQ
k
(V
k
)
is the complex power injected at bus k with phasor voltage V
k
k
. Notice that V
ac
()
is independent of the values of D
G
, D
L
, D
x
, and .
It can be readily shown that equations (4.4) are represented by
_
_
G
V
_
_
= A
_
_
M
G
G
f(, V)
g(, V)
_
_
= A
T
V
ac
(
G
, , V)
69
where A is the nonsingular matrix
A =
_
_
M
1
G
D
G
M
1
G
M
1
G
T
T
G
M
1
G
T
T
L
0 0
T
G
M
1
G
0 0 0 0
T
L
M
1
G
0 D
1
L
0 0
0 0 0 D
1
x
0
0 0 0 0 1/ I
n
L
_
_
with (A+A
T
) positive semidenite. Hence, theorem 4.2 applies in this case.
Theorem 4.2 [58] Let z = [
T
G
T
V
T
]
T
, and
s() = [(M
G
G
)
T
f
T
() g
T
()]
T
Then, if the linearization D
z
s[
0
is stable at the equilibrium z
0
= (0,
0
, V
0
), and
(i) z = s(z) = A
T
V
ac
(z), where A R
mm
(m = 3n
G
+ 2n
L
+ 2n
dc
1).
(ii) det A ,= 0.
(iii) (A+A
T
) is positive semidenite.
(iv) V
ac
(z)(A+A
T
)
T
V
ac
(z) = 0 only for trivial trajectories z(t) z
e.p.
, where
s(z
e.p.
) = 0.
it follows that V
ac
(z) is a global Lyapunov function of (4.4), i.e., V
ac
(z) is l.p.d.f. at
z
0
,
V
ac
(z) 0 z R
m
, and
V
ac
(z) = 0 only for trivial trajectories z(t) z
e.p.
.
Proof: From (i) and (iii),
V
ac
(z) = V
ac
(z) s(z)
=
1
2
V
ac
(z)(A+A
T
)
T
V
ac
(z) 0
70
Moreover, from the structure of (A+A
T
) it follows that
0 = V
ac
(z)(A+A
T
)
T
V
ac
(z)
=
T
G
D
G
G
+f
T
L
()D
1
L
f
L
() + g
T
G
()D
1
x
g
G
() + 1/ g
T
L
() g
L
()
Hence,
G
= 0, f
L
() = 0, g
G
() = 0, and g
L
() = 0. Furthermore,
G
= 0 f
G
() = M
G
G
If f
G
(z(t)) 0, then the trajectory z(t) z
0
. However, f
G
(z(t
n
)) ,= 0 (
G
(t
n
) ,= 0)
and
V (z(t
n
)) = 0 (
G
(t
n
) = 0) at some time t
n
, imply
V (z(t
n
+ t)) ,= 0, since
G
(t
n
+ t) ,= 0. This leads to the following conclusion:
V
ac
(z) = 0 only for trivial trajectories z(t) z
e.p.
On the other hand, from (i) it follows that
2
V
ac
(z
0
) = A
1
D
z
s[
0
Since the linearization of the nonlinear system is stable at the equilibrium z
0
, and
det A ,= 0, then the Hessian
2
V
ac
(z
0
) is nonsingular. Furthermore, suppose that
the Hessian has a negative eigenvalue, then there exists points z
R
m
arbitrarily
close to the equilibrium such that V
ac
(z
) < 0. However,
V
ac
(z) 0 z R
m
,
then a system trajectory starting at z
k=1
n
j=1
B
kj
V
0
k
V
0
j
cos(
0
k
0
j
)
1
2
n
k=1
n
j=1
B
kj
V
k
V
j
cos(
k
j
)
k=1
P
0
k
(
k
0
k
)
n
k=1
_
V
k
V
0
k
Q
k
()
d
+
n
k=1
n
j=1
G
kj
V
0
k
V
0
j
cos(
0
k
0
j
)(
k
0
k
)
+
n
k=1
n
j=1
G
kj
V
0
j
sin(
0
k
0
j
)(V
k
V
0
k
) (4.6)
Including these correction terms into the energy function has given good results
in many system simulations, especially for trajectories close to the stable equilibrium
point. In this case, one can formally prove that the energy function is decreasing
along trajectories close to the s.e.p. [59]. Figure 4.2 depicts both TEFs, with and
without the correction terms, along a system trajectory for the ac power system
shown in gure 4.1. Notice the monotonically decreasing energy when the transfer
conductance terms are included in the ac TEF.
Singularly perturbed models of voltage dependent reactive loads are not practi
cal for the kind of numerical methods typically used in time simulations of power
networks; therefore, one is forced to deal with the nonlinear algebraic constraints
72
G
1
G
2
2
Figure 4.1: Example form [60].
73
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05
0 0.2 0.4 0.6 0.8 1 1.2
T
E
F
(
p
.
u
.
)
time (sec.)
TEF no G
TEF G
Figure 4.2: AC TEF during recovery period from a fault at bus 2 in system 4.1.
74
produced by the reactive power mismatches at the load buses. Furthermore, treat
ing active loads as constant powers or having connecting buses (i.e., buses with
no load) in the system structure, introduces additional implicit variables into the
analysis. This poses a problem since one can have instantaneous changes on bus
voltage phasors due to system modications, like applying and clearing faults, which
in turn yields instantaneous changes in the system energy. This can be avoided by
projecting the fault trajectories back to the postfault system structure. The
process simply takes the system state at each point along the faulton trajectory
and computes the corresponding values for implicit variables that would result if
the fault were cleared at that instant; the energy value is then calculated based
on this choice of implicit variable values. In essence, this ensures that the system
energy is evaluated only for values of state and implicitly dened variables that
are consistent with the postfault system structure for which the energy function
is derived. Failure to take this projection into account signicantly degrades the
quality of critical clearing time predictions of TEF methods. For a threephase fault
at bus 2 in the ac system of gure 4.1, the calculated critical clearing time from
simulations is t
sim.
cc
= 0.0495 sec., whereas the controlling u.e.p. and PEBS energy
function methods, together with the projection process, yield t
u.e.p.
cc
= 0.002 sec.
and t
PEBS
cc
= 0.0468 sec., respectively. When this technique is not used, these TEF
methods would erroneously predict a completely unstable system, i.e., t
cc
= 0.
The controlling u.e.p. and the PEBS energy function methods are techniques
developed for determining the approximate size of the stability region in ac systems,
for a postfault asymptotic stable equilibrium. In other words, both techniques
yield a value for V
cr
in (4.2). The former consists on nding the u.e.p. closest to
the point where the faulton system trajectory leaves the stability region, so that
V
cr
= V
ac
(z
u.e.p.
). (Notice that for heavily loaded systems where there are only
75
one or two u.e.p.s, the controlling u.e.p. method often produces an underestimate
of the critical clearing time, as shown in the example above.) The latter tries to
determine the boundary of the stability region by looking for the local maximum of
the potential energy, V
P
ac
= V
ac
(z) 1/2
T
G
M
G
G
, in the direction of the faulton
trajectory. Of this two methods, the controlling u.e.p. is the more cumbersome
and inaccurate; however, the PEBS sometimes overestimates the clearing time, as
demonstrated in [61] and by the results depicted in section 4.5 and chapter 5.
4.3 DC Energy Function
For xed ac voltages V
r
and V
i
at the ac/dc converter buses, the dynamics of the
current controllers at both ends the dc link become piecewiselinear. Hence, for each
control mode we may construct a state equation and a Lyapunov equation [20] of
the form:
x
c
dc
= A
c
x
c
dc
+b
c
V
c
dc
(x
c
dc
) =
1
2
(x
c
dc
x
c
dc
0
)
T
P
c
dc
(x
c
dc
x
c
dc
0
)
P
c
dc
A
c
+A
cT
P
c
dc
= I
2
(
V
c
dc
(x
c
) < 0) (4.7)
where x
c
dc
= [x
r
I
d
]
T
= x
r
dc
for rectier current control, or x
c
dc
= [I
d
x
i
]
T
= x
i
dc
for
inverter current control.
The complete behavior of the two controllers is determined by the limit functions,
which cause a switch between the control modes. To dene a candidate quadratic
Lyapunov function with P
dc
R
33
, we combine components of matrices P
r
dc
and
76
P
i
dc
as shown in (4.8), where = T
r
22
/T
i
11
and x
dc
= [x
r
I
d
x
i
]
T
.
V
dc
(x
dc
) =
1
2
x
T
dc
_
_
T
r
11
T
r
12
0
T
r
12
T
r
22
T
i
12
0 T
i
12
T
i
22
_
_
. .
P
dc
x (4.8)
In order to prove that V
dc
(x
dc
) is a Lyapunov function for the HVDC link, the
various terms in matrix P
dc
must be calculated by solving equation (4.7). Thus, we
have that
T
r
11
=
1
2K
I
_
K
I
+ K
1
a
r
V
r
K
1
K
P
a
r
V
r
+ K
2
+
K
1
K
P
a
r
V
r
+ K
2
K
1
a
r
V
r
_
(4.9)
T
r
12
=
1
2K
1
a
r
V
r
T
r
22
=
K
I
+ K
1
a
r
V
r
2K
1
a
r
V
r
(K
1
K
P
a
r
V
r
+ K
2
)
T
i
11
=
K
I
+ K
1
a
i
V
i
2K
1
a
i
V
i
(K
1
K
P
a
i
V
i
+ K
2
)
T
i
12
=
1
2K
1
a
i
V
i
T
i
22
=
1
2K
I
_
K
I
+ K
1
a
i
V
i
K
1
K
P
a
i
V
i
+ K
2
+
K
1
K
P
a
i
V
i
+ K
2
K
1
a
i
V
i
_
where
K
1
=
3
2
L
d
K
2
=
1
L
d
_
3
(X
cr
X
c
i
) + R
d
_
Theorem 4.3 The function V
dc
: R
3
R in (4.8) is a Lyapunov function for the
system of dierential equations (2.4) with constant ac converter voltages V
r
and V
i
,
assuming that either the rectier or inverter is controlling the dc current, and that
77
these two controllers do not operate at the same time, i.e., one is saturated while
the other is controlling current and vice versa.
Proof: Since V
dc
(x
dc
) is positive denite by construction, only the time deriva
tive along trajectories of (2.4) must be examined. For rectier control the inverter
is saturated at its minimum extinction angle
i
min
, which is its normal operating
condition, then x
i
= 0 and
V
dc
(x
dc
) =
1
2
[x
r
I
d
0] (P
dc
A+A
T
P
dc
)
_
_
x
r
I
d
0
_
_
=
1
2
[x
r
I
d
] (P
r
dc
A
r
+A
rT
P
r
dc
)
_
_
x
r
I
d
_
_
=
1
2
[x
r
I
d
]
_
_
x
r
I
d
_
_ 0
For inverter control, the rectier controller is saturated at its minimum ring angle
r
min
. The function V
dc
(x
dc
) can be rewritten in this case as
V
dc
(x
dc
) =
1
2
T
r
11
x
2
r
+T
r
12
x
r
I
d
+
2
[I
d
x
i
] P
i
dc
_
_
I
d
x
i
_
_
where x
r
is a positive constant due to the saturation of the rectier current con
troller. Then it follows that
V
dc
(x
dc
) = T
r
12
x
r
I
d
2
[I
d
x
i
]
_
_
I
d
x
i
_
_
From (4.9) it follows that > 0, and T
r
12
< 0. On the other hand, since the rectier
controller is saturated and the inverter is controlling the current, for any trajectory in
78
this mode of operation cos
r
= cos
r
min
, cos
i
< cos
i
min
, and I
d
< I
d
0
(assuming
that the rectier is controlling the current at the s.e.p. x
dc
0
). Hence,
3
(a
r
V
r
cos
r
min
a
i
V
i
cos
i
) >
3
(a
r
V
r
cos
r
0
a
i
V
i
cos
i
min
)
On the other hand, at x
dc
0
it follows from (2.4) that
3
(a
r
V
r
cos
r
0
a
i
V
i
cos
i
min
) =
_
3
(X
cr
X
c
i
) + R
d
_
I
d
0
= K
2
I
d
0
Thus, assuming K
2
> 0 (usually X
cr
> X
c
i
by design),
3
(a
r
V
r
cos
r
min
a
i
V
i
cos
i
) > K
2
I
d
L
d
I
d
=
3
(a
r
V
r
cos
r
min
a
i
V
i
cos
i
) K
2
I
d
> 0
Therefore,
I
d
> 0,
V
dc
(x
dc
) 0 x
dc
R
3
, and
V
dc
(x
dc
) = 0 if and only if
x
dc
(t) x
dc
0
.
To test how this function behaves when the ac voltages are allowed to change, the
ac systems at both rectier and inverter sides were modelled as xed voltage sources
behind constant reactances. The capacitive voltage support at the ac converter buses
was also taken into account. In this case the system equations are nonlinear, and
function (4.8) serves as a candidate local Lyapunov function for the ac/dc system
modelled. The positive deniteness of V
dc
(x
dc
) is obvious, but its behavior along
system trajectories must be examined. Several ac balanced faults were simulated on
this system by instantaneously changing the Short Circuit Ratio (SCR) at either side
of the dc link, and tracking the dc energy function for dierent ac system strengths
with SCR ranging form 3 to 8.
Figures 4.3 and 4.4 depict the HVDC state variables and the corresponding
dc energy function for an ac fault simulated by reducing the rectier short SCR
79
0.86
0.88
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
xr
xi
Id
time (sec.)
p
.
u
.
Figure 4.3: HVDC state variables for a fault at the ac rectier side and clearing
time of 0.02 sec.
80
0
0.5
1
1.5
2
2.5
x10
4
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
T
E
F
(
p
.
u
.
)
time (sec.)
tc
Figure 4.4: DC energy function for state variables in gure 4.3.
81
from 8 to 3 while the inverter remains at a SCR of 8. Here one can observe that
the energy, V
dc
(x
dc
), smoothly increases as the system state moves away from
the stable equilibrium during the fault conditions, and after clearing the fault, when
the system recovers its original state, V
dc
(x
dc
) monotonically decreases towards zero.
This is the kind of behavior expected from the system energy. The same phenomenon
was observed for a large number of simulated ac system strengths and faults. In
all these cases the overlap angles
r
and
i
where traced to detect any undesirable
converter mode changes, which are considered to be unstable states since they
render the dc model unrealistic
2
. Moreover, this is the only mechanism for losing
stability in this case, due to the constant voltage source models used to simulate
the ac systems at both converter ends.
4.4 Coupling AC and DC Dynamics in the TEF
The next step is to analyze the independent behavior of the dc and ac energy
functions dened above, in an ac/dc system including generator dynamics. To that
purpose, both energy functions were traced for the ac/dc test system of gure 3.5,
during the recovery period from several ac balanced faults and arbitrary initial
conditions.
The dc active and reactive powers are treated now as voltage dependent loads
of the ac system, similar to the approach used in [31, 32] where the energy function
accounts for the power demands of the dc line. Hence, the terms in the ac TEF (4.6)
corresponding to these dc powers are changed to account for this fact, as shown in
2
Changing the dc system equations when the converter abandons the rst mode of
commutation operation, may render the system stable. However, these new modes of
operation are not considered in this thesis.
82
(4.10).
V
dc
ac
= P
r
(V
r
, x
r
, I
d
)(
r
0
r
) + P
i
(V
i
, x
i
, I
d
)(
i
0
i
) +
_
Vr
V
0
r
Q
r
(, x
r
, I
d
)
d +
_
V
i
V
0
i
Q
i
(, x
i
, I
d
)
d (4.10)
The integral expressions in (4.10) can be found from equations (2.4), where we have
that
Q
r
(V
r
) =
3
V
n
I
n
S
n
I
d
_
a + bV
r
+ cV
2
r
a = X
2
cr
I
2
d
b = 2
2a
r
X
cr
I
d
cos
r
c = 2a
2
r
(1 cos
2
r
)
Thus, the integral term for the rectier reactive power can be shown to be
_
Vr
V
0
r
Q
r
()
d = q
r
(V
r
) q
r
(V
0
r
)
q
r
() =
3V
n
I
n
I
d
S
n
_
U +
b
2
c
ln
_
2
cU + 2c + b
_
a sin
1
_
b + 2a
d
__
where d = 8a
2
r
X
2
cr
I
2
d
, and U = a + b + c
2
. The corresponding integral for the
inverter side can be represented by a similar expression.
The total TEF for an ac/dc system with n
dc
HVDC links is dened then as
V (z)
= V
ac
ac
(z) +
n
dc
i=1
_
V
dc
ac
i
(z) +
i
V
dc
i
(x
dc
i
)
_
where z = [
T
G
T
V
T
x
T
dc
1
x
T
dcn
dc
]
T
, V
ac
ac
() is the ac TEF shown in (4.6), and V
dc
ac
i
()
corresponds to the correction terms (4.10) for the i
th
dc link. The weighting coe
cients
i
are restricted to an interval such that the ac/dc energy function is locally
83
positive denite at the s.e.p. Hence, assume for n
dc
= 1 that the ac only system
is stable when the HVDC link is replaced by the equivalent active and reactive ac
converter power injections at the s.e.p.. Suppose also that the inverter current con
troller remains saturated, and the VDCOL is not operational for trajectories close
to the this equilibrium. Then, one can dene a decoupled ac energy function
V
ac
that is locally positive denite at the corresponding s.e.p. [58], where its Hessian
has the following structure:
V
ac
[
0
=
_
_
H
1
H
2
H
T
2
H
_
_ (4.11)
Furthermore, since
2
V [
0
=
_
_
H
1 H
2
0
H
T
2
0
H
_
_
a necessary condition for V to be locally positive denite is the positive deniteness
of the six by six block shown in (4.12).
H =
_
_
A 0 C
0 B D
C
T
D
T
P
r
dc
_
_
(4.12)
By neglecting transfer conductances, matrices A, B, C, and D can be shown to
be equal to
A =
_
_
Q
0
r
V
0
r
2
B
rr
P
0
r
V
0
r
+
P
r
V
r
P
0
r
V
0
r
+
P
r
V
r
0
1
V
0
r
Q
r
V
r
Q
0
r
V
0
r
2
B
rr
_
_
84
B =
_
_
Q
0
i
V
0
i
2
B
ii
P
0
i
V
0
i
+
P
i
V
i
P
0
i
V
0
i
+
P
i
V
i
0
1
V
0
i
Q
i
V
i
Q
0
i
V
0
i
2
B
ii
_
_
C =
_
_
P
r
x
r
0
P
r
I
d
0
1
V
0
r
Q
r
x
r
0
1
V
0
r
Q
r
I
d
0
_
_
D =
_
_
0
P
i
I
d
0
0
1
V
0
i
Q
i
I
d
0
_
_
and P
r
dc
is the positive denite matrix that comes from the solution of matrix
equation (4.7) for rectier current control.
Using basic properties of the Euclidean norm [49], one can readily prove that if
A and B are positive denite, then
y
T
Hy = [
r
V
r
[
i
V
i
[ x
r
I
d
]
. .
y
T
H
_
_
y
1
y
2
y
3
_
_
= y
T
1
Ay
1
+ 2y
T
1
Cy
3
+y
T
2
By
2
+ 2y
T
2
Dy
3
+
y
T
3
P
r
dc
y
3
m
(A)  y
1

2
2
M
(C)  y
1
  y
3
 +
m
(B)  y
2

2
2
M
(D)  y
2
  y
3
 +
m
(P
r
)  y
3

2
85
_
_
_
m
(A)  y
1

m
(P
r
)
2
 y
3

_
_
2
+
_
_
_
m
(B)  y
2

m
(P
r
)
2
 y
3

_
_
2
+
_
_
2
m
(A)
m
(P
r
) 2
M
(C)
_
 y
1
  y
3
 +
_
_
2
m
(B)
m
(P
r
) 2
M
(D)
_
 y
2
  y
3

Closed form equations for the minimum and maximum singular values (
m
and
M
, respectively) of matrices A, B, C, and D can be found by assuming I
0
d
= 1pu
and V
0
r
1pu. Thus,
m
(A) B
rr
R
cr
P
0
r
Q
0
r
> 0
m
(B) B
ii
+ R
c
i
P
0
i
Q
0
i
> 0
M
(C) <
_
_
1 +
P
0
r
2
Q
0
r
2
_
_
n
2
r
+ Q
0
r
2
+ (R
cr
+ K
P
n
r
)
2
_
M
(D)
_
_
1 +
P
0
i
2
Q
0
i
2
_
_
Q
0
r
2
+ R
2
cr
_
where
R
cr
=
3
V
n
I
n
S
n
X
cr
R
c
i
=
3
V
n
I
n
S
n
X
c
i
n
r
=
3
V
n
I
n
S
n
a
r
Since matrices A and B are positive denite, the ac/dc TEF can be locally
86
positive denite if inequalities (4.13) and (4.14) hold.
>
2
_
n
2
r
+ Q
0
r
2
+ (R
cr
+ K
P
n
r
)
2
_
_
1 +
P
0
r
2
Q
0
r
2
_
m
(P
r
)
_
B
rr
R
cr
P
0
r
Q
0
r
_
(4.13)
>
2 (Q
0
i
2
+ R
2
cr
)
_
1 +
P
0
i
2
Q
0
i
2
_
m
(P
r
)
_
B
ii
+ R
c
i
P
0
i
Q
0
i
_
(4.14)
These inequalities represent a necessary but not sucient condition for local positive
deniteness of V . However, due to the highly diagonal dominant structure of matrix
H
1
in (4.11), one can reasonably expect the minimum eigenvalues of H
1
and H to
dominate the maximum singular value of matrix H
2
. For the sake of completeness,
an additional numerical test of positive deniteness of
2
V [
0
is performed after the
candidate value of has been chosen.
Note that (4.13) and (4.14) primarily depend on the dc power levels at the s.e.p.
and the design parameters of the dc line. This proves helpful when determining
energy proles for voltage instability studies in several ac/dc networks, since the
value of needs not to change with system loading.
The inequalities above yield a minimum value for the coecient , so that the
positive deniteness requirement for the ac/dc TEF can be met. However, a sec
ond property desired of the system energy function is that it be decreasing along
trajectories. Classical vector energy function theory uses comparison functions to
guarantee this property, and to calculate the value of the coupling constant [18, 19].
However, the mathematical complexity of the proposed ac/dc TEF does not allow
for this type of analysis; a more practical approach is used instead. Figure 4.5 clearly
shows for the 6 bus test system, that the ac and dc TEFs have opposite behavior,
87
suggesting that an energy interchange takes place between the ac and the dc sys
tems. The same behavior was observed for a variety of system trajectories in all the
ac/dc systems tested. Therefore, the choice of can be further constrained by im
posing the restriction of nonincreasing along trajectories at a sequence of sampling
times, i.e.,
V (t
j
) V (t
j+1
)
j
V
ac
(t
j+1
) V
ac
(t
j
)
V
dc
(t
j
) V
dc
(t
j+1
)
(4.15)
An average value produced by (4.15), subject to (4.13) and (4.14), is selected
for the constant . In constructing the set of values to be averaged, a maximum
threshold is used so that all values greater than this limit are discarded. This is
necessary because the dc energy function can pass through or near zero at some
sample points. The resulting ac/dc energy function V is shown in gure 4.5 for the
u.e.p.s.e.p. trajectory in the 6 bus system. Notice that it decreases towards zero
as the system trajectory approaches the s.e.p.; however, there are places where the
energy function actually experiences a relatively small increase. Similar behavior was
observed for several ac balanced faults and rectier system strengths in the sample
system, especially when the system trajectory is far from the stable equilibrium.
Hence, the proposed ac/dc TEF could be considered to be a local Lyapunov function
of the ac/dc system. For networks where the ac elements primarily govern the overall
ac/dc system dynamics, a better behavior of the energy function was observed.
In systems with multiple HVDC lines, the same criteria can be applied to de
termine the value of the
i
corresponding to each dc link. However, as it is shown
in chapter 5, using a single coupling constant between the ac and dc dynamics in
the TEF yielded good results in all systems tested.
88
4
2
0
2
4
6
8
10
12
14
16
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
T
E
F
(
p
.
u
.
)
time (sec.)
TEF
TEFac
1.3e4*TEFdc
Figure 4.5: AC and dc energy functions along the trajectory between the closest,
and unique, u.e.p. and the s.e.p. for the 6 bus ac/dc sample system.
89
4.5 Application to Sample System
Two types of studies were done on the sample network of gure 3.5 to assess the
usefulness of the proposed ac/dc TEF. First, three dierent faults were applied
to this system to compare the critical clearing times obtained by time simulation
to those obtained using the ac/dc energy function. AC faults at the rectier and
inverter were represented by reducing the ESCR of the ac converter side, whereas at
bus 2 a threephase solid fault was applied. Second, the energy function was used to
determine system proximity to voltage collapse for either active power or reactive
power increments at the load bus. VDCOL was not considered in these studies.
For a rectier ESCR of 21.4, inequalities (4.13) and (4.14), and condition (4.15)
applied to the closest u.e.p.s.e.p. system trajectory, yielded a weighting coecient
of = 1.310
4
, for both power load levels shown in table 4.1. For the ESCR of 6.2,
the coecient value produced by the proposed method was = 3.15 10
4
. Notice
that the dierent load levels did not alter the value of , since the HVDC variables
are not signicantly aected by changes at the load bus.
Table 4.1 depicts the clearing times for dierent balanced ac faults simulated in
the test system. The controlling u.e.p. and the PEBS techniques, together with the
faulton trajectory projections, were used to determine the critical clearing times by
means of the ac/dc TEF. Controlling u.e.p. methods underestimate the values of
the clearing times, which is also a problem in ac only system models with just one
u.e.p. neighboring the s.e.p. On the other hand, the PEBS technique yields better
estimates for ac faults away from the dc converters, but it fails to produce accurate
results in some study cases for the 6 bus test system. However, this is not the case
in larger and more realistic systems, where the PEBS and the proposed ac/dc TEF
consistently produced good estimates of clearing times.
90
ESCR P
L
Fault Time Controll. PEBS
rectif. MW BUS Simul. u.e.p.
1
0.028
0.030
0.0092
475 2
0.028
0.030
0.0067 0.029
3
0.022
0.023
0.0181
21.4 1
0.035
0.038
0.0262 0.05
2975 2
0.0040
0.0045
0.0024 0.0043
3 > 0.1
0.0127
Stable
Unstable
Converter mode change
Table 4.1: Critical clearing times (in seconds) from TEF and time simulations for
the test system.
91
0
5
10
15
20
25
30
35
40
45
50
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
T
E
F
(
p
.
u
.
)
TEF
10*TEF
Loading Factor (MW)
Figure 4.6: TEF at various unstable equilibria for active power changes at the load
bus. The rectier ESCR is 21.4.
92
0
2
4
6
8
10
12
0 500 1000 1500 2000 2500 3000
T
E
F
(
p
.
u
.
)
Loading Factor (MVAR)
Figure 4.7: TEF at various unstable equilibria for reactive power changes at the
load bus. The rectier ESCR is 21.4.
93
0
10
20
30
40
50
60
70
0 500 1000 1500 2000 2500 3000 3500 4000 4500
T
E
F
(
p
.
u
.
)
TEF
10*TEF
Loading Factor (MW)
Figure 4.8: TEF at various unstable equilibria for active power changes at the load
bus. The rectier ESCR is 6.2.
94
0
2
4
6
8
10
12
0 500 1000 1500 2000 2500
T
E
F
(
p
.
u
.
)
Loading Factor (MVAR)
Figure 4.9: TEF at various unstable equilibria for reactive power changes at the
load bus. The rectier ESCR is 6.2.
95
In general, the results above lead to conclude that the value of the proposed
TEF at the closest u.e.p. gives a relatively good notion of the dynamic stability of
the ac/dc system. This idea can be used to evaluate the vulnerability of the system
to voltage collapse, as proposed in [15] and [16] for ac only systems. Figures 4.6
through 4.9 show the value of the ac/dc energy function V for all possible system
u.e.p.s; the arrows in the graphs indicate how the equilibria bifurcate as the load
power increases. Figures 4.7 and 4.9 present a linear behavior somewhat similar
to the one presented in [15, 16], which apparently is related to the quadratic
shape of the voltage proles shown in chapter 3. However, gures 4.6 and 4.8
exhibit a much more complicated structure. Here the energy function for a specic
unstable equilibrium has been scaled; this u.e.p. is associated with a second stable
equilibrium corresponding to inverter current control that bifurcates twice with the
u.e.p. Eventually the system model loses all power ow solutions when the TEF
becomes zero (point of voltage collapse).
The appearance of a second stable equilibrium where the inverter takes over the
current control suggests the possibility of intentionally forcing the ac/dc system to
have a second s.e.p. by changing the settings and control modes of the HVDC links.
This could prove useful as a mechanism of maintaining the voltage stability of the
network, especially for systems close to a saddlenode bifurcation.
Chapter 5 shows the same applications of the proposed energy function to assess
stability on several ac/dc power networks with up to 121 buses and multiple dc
links.
Chapter 5
Results
5.1 Introduction
This chapter shows applications and comparisons of the PoC, continuation, and
TEF methods for voltage and transient stability analyses of ac and ac/dc systems
ranging from 14 buses up to 2158 buses. A brief description of the characteristics of
the dierent systems utilized throughout this chapter, precedes the discussion and
presentation of the various results obtained with the help of the PoC power ow and
the TEF transient stability programs. The various features of these two programs
are presented and explained in appendices B and C, respectively.
5.2 Test Systems
The 14 bus system depicted in gure 5.1 was designed based on [63, 64], to replicate
some of the characteristics of power systems in the western part of the United
States. The system consists of two asynchronous ac systems and two bipolar dc
links. In both HVDC lines, the rectier side is relatively strong as compared to the
corresponding inverter end. The complete set of this systems data is presented in
96
97
appendix A.
Figure 5.2 shows the oneline diagram corresponding to the 173 bus ac/dc system
used in this thesis. This system is a simplied version of a real 2158 bus network,
with ac transmission voltage levels of 500 kV, 345 kV and 230 kV. It was developed
mostly in CEPEL, Brazil, as part of EPRIs project RP 26754, and it has similar
characteristics regarding the relative ac system strength of HVDC rectier and in
verter sides as the 14 bus system described above. Both rectier stations are weakly
coupled through a long 500 kV transmission system, whereas the inverter ends are
tightly coupled through a 500 kV345 kV230 kV ac network. The dc link on the
left of gure 5.2 is a 12 pulse bipolar HVDC system with a 500 kV dc voltage
level and 3000 MW capacity; the second HVDC system is also a 12 pulse bipolar
dc link designed for 500 kV and 2000 MW nominal levels. The ac system has 29
generators, 203 transmission lines and transformers, and 52 series capacitors. If the
intermediate buses due to series compensation are eliminated, the system can be
reduced to 121 buses. Also, gure 5.2 shows the 4 buses chosen to simulate solid
3phase faults for rst swing transient stability studies. The data utilized in this
chapter for the 173 bus system can be found in appendix A.
The 133 bus ac/dc system is actually an expanded version of the 173 (121) bus
system, where the series capacitors have been added to the respective transmission
line impedances. Overall, it has similar qualitative characteristics as the 173 system
depicted above. The dc systems are identical in both cases, but the 133 bus network
has more generators (41) and ac transmission lines and transformers (213).
Finally, the 2158 bus ac/dc system is a realistic large system. It was used as a
base for developing the 173 and 133 bus systems, using the same two HVDC links. It
has 384 generators and 3402 ac elements (including ac transmission lines, transform
ers, and series capacitors). The system is divided in 23 areas with controlled power
98
13.8 KV
200mi
550KV
200mi
550KV
2000MW
+/ 600KV
600 mi
+/ 600KV
3000MW
400 mi
3000 MVA (0.9 pf)
550 KV 550 KV
800MVA
BUS_3
BUS_5
BUS_2
4500 MVA (0.9 pf)
BUS_1G
5000MVA
BUS_1
BUS _2G
700MVA (0.9 pf)
2200 MVA (0.9 pf)
BUS_5G
2500MVA
BUS_4
500 KV  30 mi
550 KV  80 mi
550 KV  20 mi 500 KV 20 mi
5000 MVA (0.95 pf) 6000 MVA (0.95 pf)
BUS_9G
BUS_9
BUS_6 BUS_8
BUS_7G
BUS_7
50 mi 50 mi
3000 MVA (0.9 pf)
500 MVA (0.95 pf)
13.8 KV 13.8 KV
13.8 KV
13.8 KV
Figure 5.1: 14 bus ac/dc system.
99
98
51
159
17
Figure 5.2: 173 bus ac/dc system. Series capacitances are eliminated for TEF
studies, reducing the number of ac buses to 121.
100
interchanged among them. It also has 62 voltage regulator transformers (Load Tap
Changers or LTCs), and 8 P (phase shifters) and Q (LTCs) regulator transform
ers. Typical V and Q limits are included, as well as tap limits for the regulator
transformers.
The 133 and 2158 systems were used only for voltage stability studies due to time
limitations in transient stability simulations. In all cases, HVDC control switching
and limits were considered; however, VDCOL was not included in the analyses.
In order to carry out some additional testing, in some cases the HVDC links were
eliminated by replacing them with the corresponding constant active and reactive
powers at the normal operating conditions. For all bifurcation analyses, constant
load power at all buses was assumed to increase keeping the initial power factor
unchanged. Also, all generators were given a participation factor according to their
power capacity so that generation was evenly distributed across the networks (dis
tributed slack bus).
5.3 Voltage Proles and PoC points
Figures 5.3 through 5.6 show the voltage proles (bifurcation diagrams, PV curves,
or nose curves) obtained using the parameterized continuation method described
in chapter 3, for all the systems depicted above. In all of these gures, the sharp
turning points represent load levels at which dierent reactive generation limits (Q
limits) are reached. These Qlimits signicantly reduce the loadability margin of
the system, as demonstrated in [11] for a purely reactive network; for example, the
maximum loading factor
0
for the 173 bus system without Qlimits is 6.0725%
above the initial system load, whereas when these limits are included
0
= 2.0275%.
PV curves 5.4 and 5.6 for the 133 and 2158 bus systems, respectively, depict
101
another interesting phenomenon characterized by immediate loss of stability when a
Qlimit is attained. It has been demonstrated by Dobson et al [29], using transcritical
bifurcation analysis, that at suciently high loading levels when a generator exciters
current reaches a limit, which is represented in steady state as a limit in the reactive
power at the generators terminals, the system loses stability due to loss of voltage
control. The PoC method cannot detect this type of instabilities. Nevertheless, from
the results shown in table 5.1 one can observe that for all cases tested this problem
develops rather close to the bifurcation point, so that for all practical purposes both
points are at the same loading level.
The HVDC commutation transformer tap limits and the converter control mode
also have an eect on the collapse point. To illustrate the eect of tap limits on
saddlenode bifurcations, the converter transformer taps in the 173 bus network
were xed at their base case value for both dc links; this increased
0
from the
original 2.0275% to 2.1618%. Observe that, contrary to what one would expect,
xing the HVDC tap limits actually increases the maximum loading margin of the
system in this case. On the other hand, to demonstrate the dierences between
constant current control and constant power control in the dc links, both HVDC
rectier ends of the 14 bus system were set in constant power control mode. This
decreased
0
, as expected [4, 26], to 11.2692%, from 11.5683% for constant current
control. The same exercise for the 173 bus system did not produce any signicant
changes in
0
, since in this case the converter voltage remains almost unchanged as
the system load increases.
Table 5.1 shows maximum loading factors and time performances for dierent
ac/dc and ac systems obtained in a SUNSPARC IPC Unix workstation. The times
shown include all program operations, including input and output, which in a ded
icated Unix workstation correspond to real times. The PoC method consistently
102
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
1.05
0 2 4 6 8 10 12
V
o
l
t
a
g
e
(
p
.
u
.
)
Loading Factor (%)
V12
V8
V7
V11
Figure 5.3: Voltage proles for 14 bus ac/dc system. Sharp corners correspond to
generator Qlimits.
103
0.7
0.75
0.8
0.85
0.9
0.95
1
1.05
1.1
0 1 2 3 4 5
V
o
l
t
a
g
e
(
p
.
u
.
)
Loading Factor (%)
V51
V120
V99
Figure 5.4: Voltage proles for 133 bus ac/dc system. Immediate Qlimit instabil
ities are shown.
104
0.75
0.8
0.85
0.9
0.95
1
1.05
0 0.5 1 1.5 2 2.5
V
o
l
t
a
g
e
(
p
.
u
.
)
Loading Factor (%)
V127
V110
V135
V140
Figure 5.5: Voltage proles for 173 bus ac/dc system. Sharp corners correspond
to generator Qlimits.
105
0.96
0.98
1
1.02
1.04
1.06
1.08
0 2 4 6 8 10 12 14 16 18
V
o
l
t
a
g
e
(
p
.
u
.
)
Loading Factor (%)
V20
V200
V400
V600
Figure 5.6: Voltage proles for 2158 bus ac/dc system. Immediate Qlimit insta
bilities are shown.
106
performed faster, by about a factor of 2, than the continuation methods. However,
the continuation method has the advantage of producing voltage proles, and as a
byproduct yields unstable equilibrium points that can be useful for transient sta
bility analyses, as illustrated in section 5.4. Another advantage of the continuation
method is that it is able to detect immediate Qlimit instabilities, although in prac
tice these points and the bifurcations are rather close. Finally, the automatic step
cutting technique used in the continuation methods to improve convergence has the
disadvantage in some cases of yielding slightly inaccurate values for the loading mar
gin; this is shown in table 5.1 by . This problem consistently occurred in systems
with sharp turning points.
Another important feature of the PoC method is that it yields the right or left
eigenvectors corresponding to the zero eigenvalue at the bifurcation point. This
information can be used to either detect the area most sensible to the voltage col
lapse problems (right eigenvector), or to calculate with an iterative method the load
direction that takes the system to the closest saddlenode bifurcation in parameter
space (left eigenvector) [65, 66], i.e., nd for all loads the values of P
l
and Q
l
in
equation (2.3) that yield the minimum loading margin
0
min
. In the 173 bus system,
the maximum element values of the saddlenode right eigenvector point to a lack of
reactive power support in the area conformed by buses 74, 75 and 76 as the source
of the voltage stability problem. Hence, increasing the shunt reactive support from
the original 550 MVAR to 850 MVAR in bus 75, moves the collapse point from
0
= 2.0275% to
0
= 2.3117%. (For examples on possible applications of the left
eigenvector see [65, 66].)
107
System P.F. Continuation PoC
time
0
[%] time
0
[%] time
14 AC buses 0.4s 11.567 7.9s 11.5683 3.6s
2 DC bipoles
21 AC elem.
5 Gen.
2 Areas
173 AC buses 6.5s 2.7591 43.3s 2.76086 25.4s
255 AC elem.
29 Gen.
173 AC buses 5.7s 2.0254 44.1s 2.0275 23.3s
2 DC bipoles
255 AC elem.
29 Gen.
133 AC buses 4.3s 4.9983
5h 15.2986 1h
2 DC bipoles 11.0m 31.2m
3402 AC elem.
384 Gen.
62 V reg. trf.
8 PQ reg. trf.
23 Areas
2158 AC buses 2.8m 32.688
1h 32.6786 29.5m
3402 AC elem. 5.5m
384 Gen.
Qlimit instability Convergence problems
Table 5.1: continuation and PoC method results for SUN SPARCIPC Unix work
station.
108
5.4 Clearing Times and TEF Proles
In this section, the only HVDC limits considered were the ring angle and extinc
tion angle limits. Furthermore, generators were modelled using a constant terminal
voltage and swing dynamics, since this is a better way of representing generator
dynamics for faults relatively far from the terminal buses. The loads were simulated
as frequency and voltage dependent.
To demonstrate one of the usual applications of the TEF in stability analysis,
several critical clearing times (t
cc
) were calculated for 4 dierent 3phase solid faults
in the 121 (173) bus system. The PEBS technique was used together with the
projection of the potential energy faulton trajectory on the constraint manifold.
These times were compared against actual time domain simulations, as shown in
table 5.2. Notice the good estimates generated by the TEF, although in some cases
this technique yields wrong values for t
cc
. This overestimation can be attributed to
the transfer conductance correction terms in the energy function, since they produce
a slight shift in the local maxima of the TEF at the unstable equilibrium points.
The same problem can be observed in aconly systems (e.g., for the simple 5 bus ac
system depicted in gure 5.7 and table 5.3, t
sim.
cc
= 0.1075 sec. and t
PEBS
cc
= 0.1150
sec.).
The TEF can be demonstrated to be a global Lyapunov function for a particular
ac system model, as shown in chapter 4, and, hence, it can be used in this specic
case to predict system instabilities. However, one has to be careful when using this
technique for predicting critical clearing times in a more complete network model,
since the energy function becomes only a local Lyapunov function. However, the
TEF presents a good way of measuring relative stability of the system, especially
when used as a measure of proximity to voltage collapse. In this case the value
109
Fault t
cc
t
cc
BUS Simul. PEBS
17
0.020
0.025
0.025
51
0.060
0.065
0.069
98
0.060
0.065
0.067
159
0.035
0.040
0.039
Stable Unstable
Within TEF G terms error margin
Table 5.2: Critical clearing times t
cc
for 121 (173) bus ac/dc system.
Bus H
g
(sec.) D
g
(sec.) T
/
d
0
(sec.) X
d
(p.u.) X
/
d
(p.u.)
BUS 1 1 10
5
1 10
5
1 10
5
1 10
5
1 10
5
BUS 3 18.0 0.1 1.00 0.15 0.05
Table 5.3: Generator dynamic data for 5 bus test system (M
g
=
Hg
f
). Bus 1 is
modelled as an innite bus.
110
HDG
AC 5 bus test system.
WSCC power flow data format.
August 1991.
BAS
C Bus Data:
C Name><kV <Pl<Ql<Psh<Qsh<Pmx<Pg<Qmax<Qmin<V
BQ BUS 1 138 65 30 0 0 0 0 300 3001.10
B BUS 2 138 115 60 0 0 0 0 0 01.00
BQ BUS 3 138 70 40 0 0 226225.5 300 3001.10
B BUS 4 138 70 30 0 0 0 0 0 01.00
B BUS 5 138 85 40 0 0 0 0 0 01.00
C
C Line Data:
C Name1><kV1 Name2><kV2 <R<X<G1<B1
L BUS 1 138 BUS 2 138 0.042 0.168 0.01025
L BUS 1 138 BUS 5 138 0.031 0.126 0.00775
L BUS 2 138 BUS 3 138 0.031 0.126 0.00775
L BUS 3 138 BUS 4 138 0.084 0.336 0.02055
L BUS 3 138 BUS 5 138 0.053 0.21 0.01275
L BUS 4 138 BUS 5 138 0.063 0.252 0.01525
C
C Terminate all data with a ZZ card
ZZ
C
C Slack bus
SOL 15 BUS 1 138
END
Figure 5.7: WSCC power ow data format [67] for 5 bus test system.
111
of the energy function at the closest u.e.p. gives the user a good idea of the size
of the stability region, particularly for heavily loaded systems, since the region of
attraction, or the system well, shrinks at the same time as unstable equilibria
around the s.e.p. disappear (bifurcate) when the system is loaded [17, 52].
Figures 5.10 and 5.11 show the TEF proles for the 14 bus and 121 (173) bus
systems, respectively. In these two cases, a single weighting coecient was used
for both HVDC links. For the 14 bus network, was initially determined to be
100, by using the u.e.p.s.e.p. trajectory and then testing for positive deniteness
of the energy functions Hessian at the s.e.p. In the case of the 121 bus system,
the positive denite condition prevailed over the u.e.p.s.e.p. trajectory, obtaining
= 1.110
4
. Figures 5.8 and 5.9 show the ac, dc, and ac/dc energy functions along
the closest u.e.p.s.e.p. trajectory for the 14 bus and 121 (173) bus system. Notice
once more the energy interchange between the ac TEF and the dc TEF.
An interesting observation for the 173 bus system was that the series capacitance
produced negative eigenvalues in the TEF Hessian
2
V [
0
, for both the decoupled
ac energy function (HVDC links are replaced by constant power injections) and the
ac/dc TEF. Hence, eliminating intermediate buses resolved this particular problem
and also reduced the 173 bus system to 121 buses.
112
0
5
10
15
20
25
30
35
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
T
E
F
(
p
.
u
.
)
time (sec.)
V
Vac
100*Vdc
Figure 5.8: AC/DC energy functions along closest u.e.p.s.e.p. trajectory for the
14 bus ac/dc system.
113
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
T
E
F
(
p
.
u
.
)
time (sec.)
V
Vac
1.1e4*Vdc
Figure 5.9: AC/DC energy functions along closest u.e.p.s.e.p. trajectory for the
121 (173) bus ac/dc system.
114
0
5
10
15
20
25
0 2 4 6 8 10 12 14 16 18 20
Loading Factor (%)
T
E
F
(
p
.
u
.
)
Figure 5.10: AC/DC TEF proles for 14 bus ac/dc system.
115
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0 1 2 3 4 5 6 7 8 9
Loading Factor (%)
T
E
F
(
p
.
u
.
)
Figure 5.11: AC/DC TEF proles for 121 (173) bus ac/dc system.
Chapter 6
Conclusions
A thorough analysis of bifurcation phenomena for a particular model of ac/dc sys
tems is presented. Point of Collapse and Continuation methods are shown to be
computational feasible as a means to determine acceptable load increase before
encountering voltage collapse (saddlenode bifurcations) in power networks. An im
plementation on C code of these two methods is presented and applied to bifurcation
studies of ac/dc systems of up to 2158 buses, including HVDC converter tap limits,
dc control mode switching, regulating transformers, active power control between
areas, and V and Q limits at generator buses. The results obtained with this PoC
power ow program depict the way these two techniques complement each other.
This thesis also introduces a new method of determining energy functions for
power system models, particularly when dc links are included in the simulations.
The proposed technique for choosing the weighting factor that couples the ac and dc
energy functions is thoroughly studied, to the extent of the limitations imposed by
the complexity of the mathematical models used to represent the HVDC lines. The
resulting ac/dc TEF is then utilized in several sample systems for voltage collapse
and transient stability assessments with promising results.
116
117
A possible computational shortcoming of the TEF methods is the need for run
ning a complete time simulation of the ac/dc system in order to determine the
weighting coecient . However, the results presented in this thesis show that un
less the stable operating conditions of the HVDC system are signicantly changed,
will not change for dierent load levels of the ac network, hence, greatly reducing the
need for carrying out full simulations to determine the closest u.e.p.s.e.p. system
trajectory. On the other hand, projecting the fault trajectories into the manifold
dened by the postfault algebraic constraints, is necessary in both ac/dc and ac
only systems, so that instantaneous changes in the energy function are avoided and
better critical clearing time estimates can be obtained.
Considering VDCOL for transient stability studies does not seem to present
major diculties. This control mode could be introduced in the dynamic model
without changing the proposed ac/dc energy function. However, this idea has to
be tested and validated with time domain simulations. Finally, the possibility of
controlling voltage collapse problems by steering the system with changes in the
ac/dc limits and control modes is worth further study. In this area, the potential
use of right and left eigenvector information for detecting areas needing reactive
support and for rescheduling generation needs to be explored in more detail.
Most of the material presented in this thesis has been summarized in the following
recent publications [68, 69, 70, 71].
Appendix A
System Data
The generator dynamic data for the 14 bus ac/dc network of gure 5.1 are depicted
in table A.1. For transient stability studies, the dc current controller gains for all
HVDC converters in the two systems tested are K
P
= 0.4KA
1
and K
I
= 30KA
1
.
The power ow data in WSCC/ETMSP format [22, 67] for the 14 bus system is as
follows:
HDG
14 BUS AC/DC SYSTEM
WSCC/ETMSP DATA FORMAT
AUGUST 1991
BAS
C
C First AC System
C
BQ 1 BUS_1G 13.8 1 46004000.2000.20001.00
B 1 BUS_1 550. 1 1540
T 1 BUS_1G 13.8 BUS_1 550. 5000 0.0 .0016 13.8 550.
L 1 BUS_1 550. BUS_2 550.1 3030 .0025 .03678 0.0 2.3408200.
L 1 BUS_1 550. BUS_2 550.2 3030 .0025 .03678 0.0 2.3408200.
BQ 1 BUS_2G 13.8 1 700 475. 500. 5001.00
B 1 BUS_2 550. 1 475.156.1
T 1 BUS_2G 13.8 BUS_2 550. 800 0.0 .01 13.8 550.
L 1 BUS_2 550. BUS_3 550.1 3030 .0025 .03678 0.0 2.3408200.
L 1 BUS_2 550. BUS_3 550.2 3030 .0025 .03678 0.0 2.3408200.
B 1 BUS_3 550. 1 1368
B 1 BUS_4 550. 1 912
L 1 BUS_3 550. BUS_4 550.1 5520 .00020.00497 0.0 .39033 30.
L 1 BUS_3 550. BUS_4 550.2 5520 .00020.00497 0.0 .39033 30.
L 1 BUS_3 550. BUS_6 550.1 5520 .00033.00829 0.0 .65055 50.
L 1 BUS_3 550. BUS_6 550.2 5520 .00033.00829 0.0 .65055 50.
118
119
L 1 BUS_4 550. BUS_8 550.1 5520 .00033.00829 0.0 .65055 50.
L 1 BUS_4 550. BUS_8 550.2 5520 .00033.00829 0.0 .65055 50.
B 1 BUS_6 550. 14750.1561.
BQ 1 BUS_7G 13.8 1 30502400.1800.18001.00
B 1 BUS_7 550. 1
T 1 BUS_7G 13.8 BUS_7 550. 3000 0.0 .00267 13.8 580.
L 1 BUS_7 550. BUS_6 550.1 3030 .00038.00552 0.0 .35112 30.
L 1 BUS_7 550. BUS_6 550.2 3030 .00038.00552 0.0 .35112 30.
L 1 BUS_6 550. BUS_8 550.1 5520 .00052.01326 0.0 1.0409 80.
L 1 BUS_6 550. BUS_8 550.2 5520 .00052.01326 0.0 1.0409 80.
B 1 BUS_8 550. 15700.1874.
BS 1 BUS_9G 13.8 1 30502400.1800.18001.00
B 1 BUS_9 550. 1
T 1 BUS_9G 13.8 BUS_9 550. 3000 0.0 .00267 13.8 580.
L 1 BUS_9 550. BUS_8 550.1 3030 .00025.00368 0.0 .23408 20.
L 1 BUS_9 550. BUS_8 550.2 3030 .00025.00368 0.0 .23408 20.
C
C Second AC System
C
T 1 BUS_5G 13.8 BUS_5 550. 2500 0.0 .0032 13.8 550.
BQ 1 BUS_5G 13.8 2 22202000.1000.10001.00
B 1 BUS_5 550. 2 1026
C
C Slack and Swing Buses
C
A REST_SYST BUS_9G 13.8 1974.1 1
A BUS_5_GEN BUS_5G 13.8 1974.1 2
C
C DC Link Bus_1Bus_3
C
BD POS_R1 GROUND BUS_1 550.1 1 1 47.40 .4542 1.25 0.9 1.1 5. 90. 18.
BZ POS_R1 GROUND BUS_1 550.1IDAL 2500. 20.R
BD POS_I1 GROUND BUS_3 550.1 1 1 46.914.4243 1.25 0.9 1.1 90.180. 18.
BZ POS_I1 GROUND BUS_3 550.1VDGA 1134. 18.I
LD POS_R1 POS_I1 1 1 1 1 13.72 4359.7 8.06 11040.
C
C DC Link Bus_5Bus_4
C
BD POS_R2 GROUND BUS_5 550.1 1 3 411.1 .4541 1.25 0.9 1.1 5. 90. 18.
BZ POS_R2 GROUND BUS_5 550.1IDAL 1690. 20.R
BD POS_I2 GROUND BUS_4 550.1 1 1 410.35.4242 1.25 0.9 1.1 90.180. 18.
BZ POS_I2 GROUND BUS_4 550.1VDGA 1134. 18.I
LD POS_R2 POS_I2 1 1 1 1 20.18 493956 12.1 11040.
ZZ
END
For the 173 bus ac/dc system depicted in gure 5.2, the ac generator dynamic
data are depicted in table A.2. The power ow data in WSCC/ETMSP format is:
HDG
173 BUS AC/DC SYSTEM
120
Bus H
g
(sec.) D
g
(sec.) T
/
d
0
(sec.) X
d
(p.u.) X
/
d
(p.u.)
BUS 1G 315.0 2.39 10.0 0.0244 0.0089
BUS 2G 88.0 1.17 8.43 0.0627 0.0118
BUS 5G 21.0 0.37 5.40 0.1971 0.0371
BUS 7G 132.0 1.59 6.00 0.0367 0.0133
BUS 9G 132.0 1.59 6.00 0.0367 0.0133
Table A.1: Generator dynamic data for 14 bus system (M
g
=
Hg
f
).
WSCC/ETMSP DATA FORMAT
AUGUST 1991
BAS
B BUS_1 345 1750.56.0
BQ BUS_2 20.0 1000700.0300.0300.1040
B BUS_3 345 2350.127.
BQ BUS_4 22.0 1200998.0400.0400..950
B BUS_5 500 113
B BUS_6 345 239. 56.0 155
BQ BUS_7 22.0 2060.700.0500.1.00
B BUS_8 230 139.723.8
BQ BUS_9 20.0 30002000.900.0900..980
B BUS_10 500 90.0070.00 190
BQ BUS_11 26.0 16001590.700.0280.1000
B BUS_12 500 391
BQ BUS_13 24.0 25402540.1300.900..960
B BUS_14 500 793.4 207. 146
B BUS_15 345 840. 5.0
BQ BUS_16 22.0 1000862.0300.0300.1050
B BUS_17 500 617. 69.0 427
T BUS_3 345 BUS_9 20.01 .01500 3450020.00
L BUS_3 345 BUS_18 3451 .00811.13690 1.2174
L BUS_6 345 BUS_1 3451 .00179.01988 1.2880
L BUS_6 345 BUS_15 3451 .0005 .00530 .04410
L BUS_15 345 BUS_3 3451 .00977.11000 1.0000
B BUS_19 500
B BUS_20 500
B BUS_21 500
B BUS_22 500
L BUS_10 500 BUS_19 5001 00634
L BUS_19 500 BUS_20 5001 .00077.01804 .69921
L BUS_20 500 BUS_12 5001 00634
L BUS_10 500 BUS_21 5001 01188
L BUS_21 500 BUS_22 5001 .00241.05865 243280
L BUS_22 500 BUS_17 5001 01188
B BUS_23 500
B BUS_24 500
B BUS_25 500
121
B BUS_26 500
L BUS_12 500 BUS_23 5001 00826
L BUS_23 500 BUS_24 5001 .00179.04244 169610
L BUS_24 500 BUS_17 5001 00826
L BUS_12 500 BUS_25 5001 01795
L BUS_25 500 BUS_26 5001 .00207.04959 197580
L BUS_26 500 BUS_27 5001 01795
L BUS_14 500 BUS_17 5001 .00040.00960 .45190
L BUS_14 500 BUS_17 5002 .00040.00960 .45190
B BUS_28 500
B BUS_29 500
L BUS_5 500 BUS_28 5001 00408
L BUS_28 500 BUS_29 5001 .00177.04189 167230
L BUS_29 500 BUS_12 5001 00612
T BUS_14 500 BUS_13 24.01 3066 .00006.00539 5300023.00
T BUS_1 345 BUS_2 20.01 .01200 3550020.00
T BUS_15 345 BUS_16 22.01 .00600 3600022.00
T BUS_5 500 BUS_6 3451 .01100 5315034500
T BUS_5 500 BUS_6 3452 .01100 5315034500
T BUS_6 345 BUS_7 22.01 3000 .00000.00590 3450022.00
T BUS_6 345 BUS_8 2301 430. .00028.01380 3450023000
T BUS_6 345 BUS_8 2302 430. .00029.01390 3450023000
T BUS_10 500 BUS_11 26.01 2000 .00666 5400026.00
BQ BUS_30 20.0 79004400.4000.40001000
B BUS_31 500 44001000. 0.
B BUS_32 500
B BUS_33 230
B BUS_34 230 3600 700. 0.
BQ BUS_35 20.0 50004600.5320.35001020
L BUS_31 500 BUS_32 5001 .0035 .07000 2.303
L BUS_33 230 BUS_34 2301 .0020 .02000 .4000
T BUS_32 500 BUS_33 2301 .01000 5500023000
T BUS_31 500 BUS_30 20.01 .00150 52500 2000
T BUS_34 230 BUS_35 20.01 .00200 23000 2000
BQ BUS_36 22 20001590.600.0525.1009
B BUS_37 500 912.
B BUS_38 500
B BUS_39 230
BQ BUS_40 18.0 424.400.268.0134.1020
B BUS_41 230 135.027.00
B BUS_42 230
BQ BUS_43 18.0 460.160.0300.0220.1000
B BUS_44 345 52.532.9 430.
BQ BUS_45 26.0 20001680.850.0440.1050
B BUS_46 230 72.817.
BQ BUS_47 11.5 113.110.0100.0100.1020
B BUS_48 230 121. 25.00
B BUS_49 500 80.
B BUS_50 230 320.065.00
B BUS_51 138 237.263.2
B BUS_52 287
B BUS_53 287
B BUS_54 230 138.028.00
122
B BUS_55 230 907.8132.1
B BUS_56 500
B BUS_57 230 117.024.00
B BUS_58 230 121.025.00
B BUS_59 230 987.7 6.2
B BUS_60 230 2146
B BUS_61 230 401. 80.6
B BUS_62 230 205.217.6
B BUS_63 287 129. 32.2 108
B BUS_64 500
L BUS_52 287 BUS_63 2871 .01070.07905 .18335
L BUS_53 287 BUS_63 2871 .01070.07905 .18335
L BUS_55 230 BUS_41 2301 .00047.00723 .00812
L BUS_55 230 BUS_58 2301 .00119.01244 .01399
L BUS_55 230 BUS_58 2302 .00119.01244 .01399
L BUS_57 230 BUS_42 2301 .00201.03074 .03443
L BUS_57 230 BUS_54 2301 .00073.01025 .01279
L BUS_57 230 BUS_54 2302 .00073.01025 .01279
L BUS_57 230 BUS_58 2301 .00110.01189 .01257
L BUS_62 230 BUS_55 2301 .00128.00979 .01060
L BUS_64 500 BUS_49 5001 .00083.01884 .83334
L BUS_41 230 BUS_58 2301 .00035.00536 .00602
L BUS_37 500 BUS_38 5001 .00074.01861 .70132
L BUS_37 500 BUS_56 5001 .00082.01668 .59401
L BUS_37 500 BUS_64 5001 .00159 .06001
L BUS_37 500 BUS_64 5002 .00159 .06001
L BUS_42 230 BUS_58 2301 .00281.04296 .04824
L BUS_48 230 BUS_62 2301 .00138.01116 .01235
L BUS_48 230 BUS_62 2302 .00138.01116 .01235
L BUS_50 230 BUS_42 2301 .00220.03422 .03858
L BUS_50 230 BUS_42 2302 .00238.03669 .04142
L BUS_50 230 BUS_57 2301 .00037.00366 .00415
L BUS_50 230 BUS_58 2301 .00055.00586 .00623
L BUS_48 230 BUS_55 2301 .00229.01583 .01530
L BUS_48 230 BUS_55 2302 .00229.01583 .01530
L BUS_39 230 BUS_46 2301 .00221.03346 .03669
L BUS_39 230 BUS_48 2301 .00290.03800 .04120
L BUS_39 230 BUS_59 2301 .00309.04677 .05040
L BUS_39 230 BUS_60 2301 .00226.03422 .03753
L BUS_48 230 BUS_46 2301 .00029.00434 .00475
L BUS_48 230 BUS_59 2301 .00141.00967 .00970
L BUS_48 230 BUS_59 2302 .00141.00967 .00970
L BUS_48 230 BUS_59 2303 .00161.00971 .00964
L BUS_48 230 BUS_59 2304 .00161.00971 .00964
L BUS_48 230 BUS_60 2301 .00027.00393 .00459
L BUS_48 230 BUS_60 2302 .00027.00393 .00459
L BUS_48 230 BUS_60 2303 .00027.00393 .00459
T BUS_64 500 BUS_63 2871 .00020.02330 4902828750
T BUS_44 345 BUS_45 26.01 .00000.00520 3536226.00
BQ BUS_65 20.0 3100 28601500.10001000
B BUS_66 500 1700. 300. 0. 1000
T BUS_66 500 BUS_65 20.01 500 54500 2000
B BUS_67 115 160.031.25 1035
123
B BUS_68 230 67.5160.0 83.6 1043
B BUS_69 500 44.222.00
BQ BUS_70 13.8 15001201.692.0711.1055
B BUS_71 230 792.
B BUS_72 500 462.
B BUS_73 500 3200.1100. 1200 1085
BS BUS_74 13.8 60004045.2649.18501000
B BUS_75 500 3500.500.0 550.
BQ BUS_76 20.0 99999900.5780.20001000
B BUS_77 500 5000.400.0 1200
B BUS_78 500 220
B BUS_79 500 66.6 97. 674
B BUS_80 500 339.119. 110
B BUS_81 500 220
B BUS_82 345 610. 414. 870
B BUS_83 500
L BUS_75 500 BUS_73 5001 .00120.02316 .8576
L BUS_75 500 BUS_73 5002 .00030.02000 1.800
L BUS_77 500 BUS_75 5001 .00020.00820 .6500
L BUS_77 500 BUS_75 5002 .00020.00820 .6500
L BUS_68 230 BUS_71 2301 3020 .00006.00131 .00189
L BUS_68 230 BUS_71 2302 3020 .00006.00116 .00166
L BUS_69 500 BUS_72 5001 3450 .00001.00030 .00717
L BUS_69 500 BUS_72 5002 3450 .00001.00030 .00922
L BUS_69 500 BUS_73 5001 2175 .00023.00451 .16660
L BUS_69 500 BUS_73 5002 2175 .00020.00446 .15250
L BUS_79 500 BUS_73 5001 3450 .00063.01412 .54878
L BUS_79 500 BUS_73 5002 3020 .00109.02408 .77771
L BUS_79 500 BUS_73 5003 3020 .00108.02409 .77674
B BUS_84 500
B BUS_85 500
L BUS_79 500 BUS_84 5001 3450 .00041.00737 .36347
L BUS_84 500 BUS_85 5001 2000 01263
L BUS_85 500 BUS_83 5001 3450 .00060.01036 .50728
B BUS_86 500
B BUS_87 500
L BUS_80 500 BUS_86 5001 3600 .00072.01382 .63786
L BUS_86 500 BUS_87 5001 2000 00858
L BUS_87 500 BUS_83 5001 3600 .00012.00238 .10963
B BUS_88 500
B BUS_89 500
B BUS_90 500
B BUS_91 500
B BUS_92 500
B BUS_93 500
B BUS_94 500
B BUS_95 500
L BUS_79 500 BUS_88 5001 3020 .00066.01266 .47988
L BUS_88 500 BUS_89 5001 2400 01263
L BUS_89 500 BUS_90 5001 3020 .00074.01428 .5411
L BUS_90 500 BUS_91 5001 2400 01263
L BUS_91 500 BUS_80 5001 3020 .00078.01502 .56905
L BUS_79 500 BUS_92 5001 3020 .00066.01266 .47988
124
L BUS_92 500 BUS_93 5001 2000 01263
L BUS_93 500 BUS_94 5001 3020 .00074.01428 .5411
L BUS_94 500 BUS_95 5001 2000 01263
L BUS_95 500 BUS_80 5001 3020 .00074.01413 .53317
B BUS_96 500
L BUS_78 500 BUS_96 5001 3600 .00264.05356 264533
L BUS_96 500 BUS_81 5001 1732 02667
T BUS_81 500 BUS_82 3451 1500 .00720 5250034500
T BUS_77 500 BUS_76 20.01 .00250 53300 2000
T BUS_72 500 BUS_71 2301 2500 .00200 5373024150
T BUS_68 230 BUS_67 1151 250. .00085.02845 2328011790
T BUS_69 500 BUS_68 2301 1008 .00018.01071 5375024150
T BUS_69 500 BUS_68 2302 1300 .00008.00667 5375024150
T BUS_73 500 BUS_74 13.81 5000 .00410 5250013.20
T BUS_68 230 BUS_70 13.81 2000 .01130 2300013.20
B BUS_97 200 43.3 20.0
B BUS_98 200 210.4 77.
B BUS_99 500 50.0025.00
BQ BUS_100 25.0 1180600.0330.0310.0.98
B BUS_101 500 305. 7.6 91.
B BUS_102 200 27.50 0.1
B BUS_103 200 8.010
B BUS_104 500 55.6 329. 327
B BUS_105 200 777.6 32.6 130
B BUS_106 500 189.61.5
BQ BUS_107 20.0 1500957. 400. 400.1.02
B BUS_108 200 148. 128
B BUS_109 500 91.
B BUS_110 500 0.7 118.5 91.
BQ BUS_111 20.0 454. 300. 300.1.05
B BUS_112 200 884. 54.8 32.
BQ BUS_113 20.0 3367.2500.10001000
B BUS_114 500 5661.3491. 1500
B BUS_115 500
B BUS_116 500
B BUS_117 500
B BUS_118 500
L BUS_106 500 BUS_115 5001 2667 01000
L BUS_115 500 BUS_116 5001 2894 .00076.01952 .91225
L BUS_116 500 BUS_117 5001 2667 01000
L BUS_117 500 BUS_118 5001 2894 .00082.02119 .99210
L BUS_118 500 BUS_114 5001 2667 01000
B BUS_119 500
B BUS_120 500
B BUS_121 500
B BUS_122 500
L BUS_109 500 BUS_119 5001 1800 .0000200998
L BUS_119 500 BUS_120 5001 2396 .00140.02338 .73750
L BUS_120 500 BUS_110 5001 1800 .0000100666
L BUS_109 500 BUS_121 5001 1800 .0000200998
L BUS_121 500 BUS_122 5001 2396 .00140.02338 .73750
L BUS_122 500 BUS_110 5001 1800 .0000100666
B BUS_123 500
125
B BUS_124 500
B BUS_125 500
B BUS_126 500
L BUS_110 500 BUS_123 5001 2667 .0000101120
L BUS_123 500 BUS_124 5001 2450 .00154.03409 115570
L BUS_124 500 BUS_114 5001 2667 .0000101120
L BUS_110 500 BUS_125 5001 2667 .0000100720
L BUS_125 500 BUS_126 5001 2450 .00095.02102 .71260
L BUS_126 500 BUS_114 5001 2667 .0000100360
L BUS_98 200 BUS_108 2001 752. .01113.06678 .03643
L BUS_98 200 BUS_108 2002 602. .01050.06540 .03430
L BUS_98 200 BUS_108 2003 752. .01105.06642 .03580
L BUS_98 200 BUS_112 2001 747. .03903.27403 .15536
L BUS_98 200 BUS_97 2001 838. .02482.16938 .10116
L BUS_97 200 BUS_112 2001 838. .0148 .10101 .06033
L BUS_98 200 BUS_102 2001 747. .01382.09268 .05530
L BUS_102 200 BUS_112 2001 747. .03058.20460 .12236
L BUS_98 200 BUS_103 2001 838. .01668.11381 .06804
L BUS_103 200 BUS_112 2001 838. .02235.16106 .09171
B BUS_127 500
B BUS_128 500
B BUS_129 500
L BUS_114 500 BUS_127 5001 .0000100755
L BUS_127 500 BUS_128 5001 2450 .00165.05719 123870
L BUS_128 500 BUS_104 5001 1800 .0000201331
L BUS_114 500 BUS_129 5001 2450 .0000101098
L BUS_129 500 BUS_101 5001 2450 .00093.03644 .69475
B BUS_130 500
L BUS_101 500 BUS_130 5001 2450 .00072.01600 .54395
L BUS_130 500 BUS_104 5001 2450 .0000200998
L BUS_101 500 BUS_99 5001 .00079.01937 .66425
L BUS_99 500 BUS_104 5001 .00087.02087 .72855
L BUS_99 500 BUS_104 5002 .00087.02087 .72855
T BUS_107 20.0 BUS_108 2001 .0192 20.0 218.
T BUS_108 200 BUS_109 5001 840. .00009.01578 2350052500
T BUS_113 20.0 BUS_114 5001 .004 20.0 529.
T BUS_111 20.0 BUS_112 2001 .015 20.0 220.
T BUS_112 200 BUS_114 5001 1120 .00018.01134 2350052500
T BUS_105 200 BUS_104 5001 840. .00027.01578 2350052500
T BUS_105 200 BUS_104 5002 1120 .00018.01079 2350052500
T BUS_99 500 BUS_100 25.01 .00000.00980 5250025.00
B BUS_131 500 856. 19.6
B BUS_132 230 175. 18.0
BQ BUS_133 20.0 882.7300. 300.1.02
B BUS_27 500 902.311.4 319
BQ BUS_134 20.0 3055.2000.900.1.02
B BUS_135 230 3191. 630.
B BUS_136 500 204.228.2
B BUS_137 230 377.464.5
BQ BUS_138 20.0 1570.900.0400.1.05
B BUS_139 500 3098.1189. 400.
B BUS_140 230
B BUS_141 500 196
126
BQ BUS_142 22.0 15801580.700.0300.1050
BQ BUS_143 20.0 2100.600. 600.1010
B BUS_144 230 3118.78.0
B BUS_145 500 1230. 72.8
B BUS_146 500 406. 41.0
B BUS_147 500
B BUS_148 230 1066.10.8 190
L BUS_136 500 BUS_147 5001 3600 .00044.01125 .41460
L BUS_136 500 BUS_147 5002 3600 .00044.01125 .41460
L BUS_136 500 BUS_141 5001 3600 .00190.03100 207010
L BUS_27 500 BUS_136 5001 3600 .00193.02779 233560
L BUS_141 500 BUS_27 5001 3600 .00056.01415 .52145
L BUS_131 500 BUS_146 5001 3600 .00042.00905 .33397
L BUS_136 500 BUS_145 5001 3600 .00060.01280 .47310
L BUS_139 500 BUS_145 5001 3600 .00021.00457 .16168
L BUS_145 500 BUS_146 5001 3600 .00040.00930 .34280
L BUS_136 500 BUS_139 5001 3600 .00028.00753 .25868
L BUS_136 500 BUS_139 5002 3600 .00035.00750 .27680
L BUS_144 230 BUS_148 2301 2320 .00285.03649 .06328
L BUS_144 230 BUS_148 2302 2320 .00138.03399 .05626
L BUS_132 230 BUS_137 2301 2320 .00190.02580 .04920
L BUS_132 230 BUS_144 2301 1160 .00845.07034 .07977
L BUS_135 230 BUS_137 2301 2320 .00110.01270 .02400
L BUS_148 230 BUS_137 2301 2320 .00320.03950 .07200
L BUS_140 230 BUS_137 2301 2320 .00138.05399 .07626
T BUS_147 500 BUS_148 2301 1120 .01149 5315623000
T BUS_147 500 BUS_148 2302 1120 .01149 5315623000
T BUS_147 500 BUS_148 2303 1120 .01149 5315623000
T BUS_133 20.0 BUS_27 5001 .015 20.0 502.
T BUS_141 500 BUS_142 22.01 .00980 5250022.00
T BUS_134 20.0 BUS_135 2301 .0035 20.0 235.
T BUS_138 20.0 BUS_139 5001 .005 20.0 508.
T BUS_139 500 BUS_140 2301 .005 5000023000
T BUS_52 287 BUS_51 1381 .00059.01491 2875013800
T BUS_53 287 BUS_51 1381 .00059.01491 2875013800
T BUS_54 230 BUS_51 1381 .00030.01330 2300013800
T BUS_54 230 BUS_51 1382 .00030.01340 2300013800
T BUS_56 500 BUS_55 2301 .00013.01386 5053023000
T BUS_56 500 BUS_55 2302 .00013.01386 5053023000
T BUS_38 500 BUS_48 2301 .00013.00693 5250023000
T BUS_42 230 BUS_43 18.01 .00058.02535 2413018.00
T BUS_49 500 BUS_48 2301 .00026.01386 5250023000
T BUS_48 230 BUS_47 11.51 .00499.11473 2410011.50
T BUS_39 230 BUS_40 18.01 .00050.02380 2300018.00
T BUS_60 230 BUS_61 2301 .00120 2280022500
T BUS_143 20.0 BUS_144 2301 .010 20.0 233.
B BUS_149 345 457.7 81.7 60.
B BUS_150 345 33.9 11.9
B BUS_151 230 148. 7.9
B BUS_152 345 116.1 38.4 220
BQ BUS_153 20.0 1590.9999.99991.05
B BUS_18 345 62. 12.8
B BUS_154 230 255. 100.
127
BQ BUS_155 20.0 370.09999.99991.00
B BUS_156 345
B BUS_157 345 31.6 11.5 18.
B BUS_158 345 141.271.4
B BUS_159 345 379. 43. 50.
B BUS_160 345 185. 78.5
L BUS_150 345 BUS_160 3451 .00160.02260 .19050
L BUS_149 345 BUS_160 3451 .00080.01060 .10195
L BUS_150 345 BUS_149 3451 .00240.03320 .29245
L BUS_149 345 BUS_18 3451 .00170.02250 .19960
L BUS_149 345 BUS_18 3452 .00210.02380 .19225
L BUS_152 345 BUS_157 3451 .00960.08780 .71325
L BUS_149 345 BUS_152 3451 .00520.06020 .50500
L BUS_149 345 BUS_152 3452 .00490.05370 .44215
L BUS_149 345 BUS_158 3451 .00120.01720 .14935
L BUS_152 345 BUS_159 3451 .00340.03740 .31040
L BUS_152 345 BUS_159 3452 .00340.03720 .30910
L BUS_18 345 BUS_159 3451 .00380.03400 .29120
L BUS_18 345 BUS_159 3452 .00320.03490 .28610
L BUS_151 230 BUS_154 2301 .01080.09650 .16480
L BUS_152 345 BUS_158 3451 .00340.03920 .32620
T BUS_3 345 BUS_4 22.01 .01238 3450022.00
T BUS_152 345 BUS_153 20.01 .00020.00580 3400020.00
T BUS_157 345 BUS_156 3451 .01950 3450034500
T BUS_150 345 BUS_151 2301 .00030.01810 3450023000
T BUS_154 230 BUS_155 20.01 .00050.01410 2435220.00
L BUS_31 500 BUS_77 5001 83.02390 1.650
L BUS_75 500 BUS_66 5001 70.07400 2.435
B BUS_161 500
B BUS_162 500
B BUS_163 500
B BUS_164 500
B BUS_165 500
B BUS_166 500
L BUS_80 500 BUS_161 5001 1800 00720
L BUS_161 500 BUS_162 5001 3020 .00103.02338 .79020
L BUS_162 500 BUS_109 5001 1800 00720
L BUS_80 500 BUS_163 5001 1800 00864
L BUS_163 500 BUS_164 5001 3020 .00107.02470 .76350
L BUS_164 500 BUS_109 5001 1800 00720
L BUS_80 500 BUS_165 5001 01000
L BUS_165 500 BUS_166 5001 .00103.03230 139800
L BUS_166 500 BUS_106 5001 01000
B BUS_167 500
B BUS_168 500
B BUS_169 500
B BUS_170 500
B BUS_171 500
B BUS_172 500
L BUS_104 500 BUS_167 5001 2134 00935
L BUS_167 500 BUS_168 5001 3600 .00123.02659 .99351
L BUS_168 500 BUS_147 5001 2134 00935
L BUS_104 500 BUS_169 5001 2134 00944
128
L BUS_169 500 BUS_170 5001 3600 .00123.02662 .99440
L BUS_170 500 BUS_147 5001 2134 00935
L BUS_104 500 BUS_171 5001 2134 00935
L BUS_171 500 BUS_172 5001 3600 .00112.02517 .91793
L BUS_172 500 BUS_147 5001 2100 00840
L BUS_136 500 BUS_64 5001 .00020.00410 .14810
L BUS_27 500 BUS_64 5001 .00179.02524 .26773
L BUS_27 500 BUS_64 5002 .00179.02524 .26773
L BUS_144 230 BUS_61 2301 3070 .00065.01187 .02336
L BUS_144 230 BUS_61 2302 3070 .00065.01187 .02336
L BUS_132 230 BUS_61 2301 3070 .00140.02640 .05100
L BUS_10 500 BUS_27 5001 1630 .00280.02110 .50970
L BUS_14 500 BUS_131 5001 1800 .00259.02967 107650
L BUS_14 500 BUS_131 5002 1800 .00259.02967 107650
B BUS_173 500 220
L BUS_78 500 BUS_173 5001 1732 02667
L BUS_173 500 BUS_83 5001 3600 .00122.02373 110355
L BUS_150 345 BUS_82 3451 .00620.06730 .55780
L BUS_44 345 BUS_18 3451 .00180.02450 .21960
L BUS_44 345 BUS_18 3452 .00180.02450 .21960
L BUS_156 345 BUS_6 3451 .00480.04360 .35390
T BUS_82 345 BUS_36 221 2000 .00000.00460 3450022.00
BD DC1RP+GROUND BUS_71 2301 117.6 1.843 .001 .80 1.40 5. 100. 0. 5.E3
BD GROUND DC1RPBUS_71 2301 117.6 1.843 .001 .80 1.40 5. 100. 0. 5.E3
BD DC1IP+GROUND BUS_60 2301 117.3 1.843 .001 .80 1.40 90. 170.18. 5.E3
BD GROUND DC1IPBUS_60 2301 117.3 1.843 .001 .80 1.40 90. 170.18. 5.E3
BZ DC1RP+GROUND BUS_71 2301IDAL 3100. 15. R
BZ GROUND DC1RPBUS_71 2301IDAL 3100. 15. R
BZ DC1IP+GROUND BUS_60 2301VDGA 447. 18. I
BZ GROUND DC1IPBUS_60 2301VDGA 447. 18. I
LD DC1RP+DC1IP+1 19. 1300.
LD DC1RPDC1IP1 19. 1300.
BD DC2RP+GROUND BUS_44 3451 124.7 1.194 .001 .92 1.15 5. 100. 0. 4.E3
BD GROUND DC2RPBUS_44 3451 124.7 1.194 .001 .92 1.15 5. 100. 0. 4.E3
BD DC2IP+GROUND BUS_37 5001 123.7 0.808 .001 .92 1.20 90. 170.16. 4.E3
BD GROUND DC2IPBUS_37 5001 123.7 0.808 .001 .92 1.20 90. 170.16. 4.E3
BZ DC2RP+GROUND BUS_44 3451IDAL 1920. 15. R
BZ GROUND DC2RPBUS_44 3451IDAL 1920. 15. R
BZ DC2IP+GROUND BUS_37 5001VDGA 485. 16. I
BZ GROUND DC2IPBUS_37 5001VDGA 485. 16. I
LD DC2RP+DC2IP+1 7.82 693.
LD DC2RPDC2IP1 7.82 693.
ZZ
END
129
Bus H
g
(sec.) D
g
(sec.) T
d
0
(sec.) X
d
(p.u.) X
d
(p.u.)
BUS 2 24.129600 0.484891 3.90 0.203501 0.027352
BUS 4 38.836800 0.789407 4.00 0.121640 0.016465
BUS 7 84.063600 1.304006 8.40 0.071603 0.011595
BUS 9 73.500000 1.591546 4.50 0.056667 0.008333
BUS 11 127.08600 1.878024 4.20 0.049718 0.006215
BUS 13 119.41940 1.654147 8.20 0.063502 0.012027
BUS 16 54.020200 1.089678 3.80 0.081305 0.012658
BUS 30 390.77360 4.776759 8.00 0.009996 0.002777
BUS 35 179.95000 3.236143 4.60 0.029918 0.004770
BUS 36 41.064000 0.939012 4.10 0.116949 0.015819
BUS 40 32.050000 0.265258 10.8 0.199000 0.039000
BUS 43 22.302000 0.286478 4.80 0.314815 0.047407
BUS 45 57.081600 1.051481 4.70 0.100404 0.012361
BUS 47 4.2940000 0.059948 7.40 0.547788 0.210619
BUS 65 104.27300 1.636640 4.78 0.040065 0.007131
BUS 70 35.970000 0.636088 4.00 0.074229 0.030859
BUS 74 198.95070 2.875923 8.00 0.017156 0.006586
BUS 76 449.80000 6.896698 7.20 0.004908 0.001754
BUS 100 58.301000 0.893918 6.12 0.126350 0.027715
BUS 107 44.865800 0.542187 4.00 0.102740 0.031311
BUS 111 37.232000 0.474811 7.80 0.083799 0.035754
BUS 113 261.28800 3.628724 4.50 0.024854 0.003655
BUS 133 127.71280 1.116204 7.80 0.035646 0.015209
BUS 134 253.80000 4.774637 5.30 0.018000 0.002722
BUS 138 82.500000 1.591546 8.00 0.065000 0.011333
BUS 142 63.448200 0.964477 8.40 0.096810 0.015677
BUS 143 75.750000 1.326288 4.30 0.071600 0.008800
BUS 153 62.857800 1.182518 3.90 0.073576 0.010767
BUS 155 25.043200 0.441389 5.21 0.216106 0.038942
Table A.2: Generator dynamic data for 173 bus system (M
g
=
Hg
f
).
Appendix B
The PoC Power Flow
This program is a portable C code implementation of the PoC and Continuation
methods described in chapter 3. It has been successfully tested in several Unix
workstations, namely, SUN, HP and DELL, and is also running in PC machines un
der DOS 3.3 (a Windows 3.0 version is currently under development). The program
is capable of handling any size ac/dc or ac systems, limited only by the available
memory and swap space. It reads WSCC [67], EPRI [22], and Common Format
[72] input les, and produces a variety of ASCII depending on the user dened op
tions shown below. It was designed to be a production type software for research
and commercial applications; hence, speed and versatility were the main concerns
during the development of this program.
The program solves the ac/dc power ow using a NewtonRaphson iterative
process with automatic stepsize adjustment, similar to the idea proposed in [73];
this makes the solution process numerically robust. The factorization routines where
directly taken from the Sparse Matrix Manipulation System (SMMS) [74]. The PoC
and Continuation methods were implemented with all the practical adjustments
explained above, so that a solution to the voltage collapse problem can be obtained
for systems with a diversity of operational limits.
130
131
To test the validity and performance of the program, results were compared
against three commercial software packages: EPRIs power ow version 3 [67] for
SUN and version 5 [22] for DELL, and Electrocons power ow version 3.10 for PC
DOS machines [75]. The following are a brief description and a complete list of the
options of the PoC program :
AC/DC Power Flow (c)1991 C. Canizares and F. Alvarado
Usage: pflow [aAbBcCFGhHiIjJkKlLMnNpPqQrRstTvVwW] [<]input_file [>]output_file
The input_file could be in WSCC format (or Anaredes variation) or IEEE
common format (or Electrocons variation) for the AC system. The DC
data must be in multiterminal ETMSP format, although it just deals with
the standard twoterminal HVDC problem.
The program writes the solution into the output_file in ASCII. It
can also write (W option) the solved case in a file in IEEE common
format (the HVDC links are written in ETMSP format).
The AC/DC power flow is solved with simultaneous NR, allowing for asyn
chronous systems, area interchange, remote voltage control, and regula
ting transformers (LTCs and phase shifters).
Options: a Turns off tap and angle limits in regulating transformers.
A Turns off interchange area control.
b Solve base case before changing the loading factor lambda.
Bnum PQ bus number num where the voltage is fixed in order to
find the loading factor (lambda) for voltage collapse studies.
Must be used with K and v options.
cfile Increases the loading factor lambda using a modification of
Ibas homotopy continuation method for finding voltage profiles.
The output (optional file) is a list of 8 random AC voltages
plus 3 additional variables for each DC bus. Must be used with
K option.
Cfile Point of Collapse studies, i.e., find the maximum loading factor
lambda for a given generation and load direction. The base case
loading can be initialized using the L option; nevertheless, the
program calculates an initial loading of the system before the
PoC method is applied. The left eigenvector is written in file
(optional). Must be used with K option.
Fval Stability/sparsity value val for factorization (def. 0.01).
A value of 0 means choose a pivot based on sparsity only;
a value of 1 means choose a pivot based on stability only.
G Do not enforce Q limits during the iteration process. The
limits will be applied after a base solution has been found.
h Prints this message in standard output.
Hfile Increases the loading factor lambda using a parameterized
homotopy continuation method for finding voltage profiles.
The output (optional file) is a list of 8 random AC voltages
plus 3 additional variables for each DC bus. Must be used with
K option.
ifile List of bus numbers and names in file for printing voltage
profiles with the H option. The input format is:
BusNumber BusName; use zero when either the number or the name
132
are unknown. If BusName has spaces, wrap it in double or single
quotes.
I Input data in IEEE common format.
kval Factor val used in the homotopy continuation method for
finding the increments in the loading factor lambda (def. 1).
Must be used with the H option.
Kfile Read generation and load distribution factors from file.
The data is all p.u. and must be separated by spaces, i.e.,
BusNumber BusName DPg DPl DQl. If the input variables are
unknown give them a value of zero. The generation factors
are normalized for each area, i.e., Pg=1 in each area.
Buses not in the list are assumed to have zero distribution
factors. If BusName has spaces, wrap it in double or single
quotes.
jfile Write the Jacobian of the solved case in I J VALUE format in
file.jac. The equation mismatches and the system variables
are also written in file.mis and file.var, respectively.
If no file is given the program writes to standard output.
Jfile Similar to j option, but in this case the Jacobian corresponds
to the system without the loading factor as a variable.
lfile Write standard error output to file (log file).
Lval Loading factor val (def. 0). Simulates load changes in
conjunction with the load distribution factors (K option).
Mnum Number num of max. NR iterations, overriding input data
(default 50).
n Turns off all AC system limits.
N Turns off all AC system controls.
p Turns off P and Q limits in regulating transformers.
P Turns off P and Q control by regulating transformers.
q Turns off Q limits in PV buses.
Q Turns off remote voltage generator control. The generators
will just control their terminal voltage to its initial value.
r Turns off V limits in regulating transformers and PV buses.
R Turns off V control by regulating transformers.
s Suppress ASCII output_file.
ttol If the relative error of two consecutive iteration mismatches
is larger than tol, voltage limits and regulating transformer
limits are applied (default 0.1).
Ttol P.U. tolerance tol for NR method (default 1e4).
vmag Voltage magnitude mag at the first PQ bus (unless otherwise
specified by B option) to find the corresponding loading factor
for voltage collapse studies. Must be used with K option.
Vfile Read initial guesses for AC bus voltages from file.
The data must be separated by spaces, i.e.,
BusNumber BusName V_mag V_ang(deg). If the input variables
are unknown give them a value of zero. Buses not in the list
are given a flat start. If BusName has spaces, wrap it on
double or single quotes.
wfile Write solved case into file in IEEE CARD common format.
If no file is given the program writes to standard output.
Wfile Similar to w option, but the solved case is written in IEEE
TAPE common format.
133
Notice that the load distribution factors (i.e., P
l
and Q
l
) and the generation
factors (participation factors for distributed slack bus) dened using the K option,
are necessary in order to dene a direction in demand space and generation space to
solve the voltage collapse problem as dened in this thesis. However, these directions
can also be used to solve the base case power ow problem with a distributed slack
bus and dierent loading levels.
Appendix C
TEF Transient Stability Program
This program is a simple portable C code implementation of the dynamic power
system model described in chapter 2, and has been successfully tested in several
Unix workstations (SUN, HP, DELL) and PCDOS machines. The generators can
be modelled as round rotor synchronous machines with constant eld voltage, as
constant voltage sources behind transient reactance, or as constant terminal volt
ages; in all these cases the swing dynamics are simulated using classical second order
dierential equations. The user has the option of modelling the loads as constant
impedance or constant power, or as a mixture of both. Also, the active power at
the load can be represented using a dynamic frequency dependent model, whereas
the reactive power can be simulated using a dynamic voltage dependent model (sin
gular perturbation techniques). Finally, the HVDC dynamics are modelled using
saturable PI current controllers and RL circuits for the dc transmission line. The
line options that allow the user to chose between these dierent dynamic models are
shown below.
The s.e.p. is initially obtained running a complete power ow, including regulat
ing transformers, area interchange control, and generator V and Q limits. However,
for the dynamic simulation only HVDC converter ring angle and extinction angle
134
135
limits are considered. The network can be perturbed from the initial s.e.p. by ap
plying a solid 3phase fault at any system bus, by reading the initial conditions from
an ASCII le, or by nding the corresponding u.e.p. using the Continuation method
described in chapter 3. This last option allows the user to do voltage collapse studies
in the dened dynamic model. The output of the program is generated either as
a plain ASCII le or as an ASCIIMatlab le [76], depending on the user dened
options. A brief description of the input data and options of the TEF stability
program follows:
AC/DC Transient Stability and TEF (c)1991 C. Canizares and F. Alvarado
Usage: tef [aABcCDeEfFgGhHiIkKlLmMnNOpPqQrRstTvVxX] pf_data ts_data [>]output
Basic ac/dc transient stability and TEF analysis package.
The program executes a complete power flow to find the s.e.p., and then
either applies a fault (c and f options), or calculates the initial
conditions (u.e.p.).
The AC power flow data (pf_data) could be in WSCC format (or Anaredes
variation) or IEEE common format (or Electrocons variation). The DC
data must be in multiterminal ETMSP format, although it just deals with
the standard twoterminal HVDC problem.
The dynamic data (ts_data) must have the following format:
 The first line defines the time interval (dt), the total simulation
time (tf), and the base frequency (f). (Default: dt=0.001 sec.,
tf=1 sec., f=60 Hz.)
 For each generator bus define: BusNumber BusName H D Tdo Xd Xd,
where: H > Inertia in sec. H=MW*sec/MVA_system_base
D > Damping in sec. D=D_pu/2*PI*f
Tdo > daxis opencircuit time constant in sec.
Xd > daxis synchronous reactance in p.u.
Xd > daxis transient reactance in p.u.
If either the BusNumber or BusName are unknown give them a value
of zero (0). If BusName has spaces, wrap it in double or single
quotes. To simulate an infinite bus make H=D>=1e5, Xd=Xd<=1e5.
To eliminate generator field dynamics make Xd=Xd, or use x option.
 For each dc link define: 0 BusName KPr KIr KPi KIi Ld, where KP and
KI are the proportional and integral gains for the rectifier and
inverter PI controllers in 1/KAmp, respectively, and Ld is the dc
line inductance in mH. If Ld is already defined in the power flow
data, give it a value of zero. BusName could be either the recti
fier or inverter bus name. If BusName has spaces, wrap it in double
or single quotes.
The program writes the ac and dc TEFs in the file output, in ASCII.
In addition to this it outputs the frequency, phase angle and internal
voltage for several generators, voltages and angles for several load
buses, and rectifier and inverter angles and currents for each dc link.
The AC/DC power flow is solved with simultaneous NR, allowing for asyn
chronous systems, area interchange, remote voltage control, and regula
136
ting transformers (LTCs and phase shifters). The dynamic simulation is
done using NR trapezoidal integration, assuming a third order model
for generators and a linear third order dynamic model with firing angle
limits for each dc link.
Options: a Turns off tap and angle limits in regulating transformers.
A Turns off interchange area control.
Bval Percentage of load to be simulated as constant reactance
(def. 0, i.e., the reactive load is assumed to be constant Q).
cval Clearing time for 3phase solid fault (default 10*dt).
Cval Apply homotopy continuation method to find u.e.p. using
val as the step (def. 4). The program outputs the voltage
profile for 8 random buses in the file output.
eval Time interval dt. Supersedes value in ts_dat.
Eval Total simulation time tf. Supersedes value in ts_dat.
Dval Define the time constant for frequency dependent simulation
of active power at load buses (default 0.1). To turn off
this type of load model make val=0.
fnum Bus number num at which a 3phase solid fault is applied.
Must be used with the c option.
Fval Stability/sparsity value val for factorization (def. 0.01).
A value of 0 means choose a pivot based on sparsity only;
a value of 1 means choose a pivot based on stability only.
g Do not enforce Q limits during the iteration process. The
limits will be applied after a base solution has been found.
Gval Percentage of load to be simulated as constant resistance
(def. 0, i.e., the active load is assumed to be constant P).
h Prints this message in standard output.
Hfile Write in file the AC/DC TEF Hessian at the s.e.p. in Matlab
format for singular value calculations. It also writes a list
of the state variables and their corresponding s.e.p. values.
ifile Read initial guesses for AC bus voltages from file, to
calculate u.e.p. The data must be separated by spaces, i.e.,
BusNumber BusName V_mag V_ang(deg). If the input variables
are unknown give them a value of zero. Buses not in the list
are given a s.e.p. start. If BusName has spaces, wrap it on
double or single quotes.
I Input data in IEEE common format.
kval Start simulation at a percentage val off the s.e.p. or
u.e.p. (default 4%).
Kfile Read generation and load distribution factors from file.
The data is all p.u. and must be separated by spaces, i.e.,
BusNumber BusName DPg DPl DQl. If the input variables are
unknown give them a value of zero. The generation factors
are normalized for each area, i.e., Pg=1 in each area.
Buses not in the list are assumed to have zero distribution
factors. If BusName has spaces, wrap it in double or single
quotes.
lfile Write standard error output to file (log file).
Lval Loading factor val (def. 0). Simulates load changes in
conjunction with the load distribution factors (K option).
m Write file otput in Matlab format.
Mnum Number num of max. NR iterations, overriding input data
(default 50).
137
n Turns off all AC system limits.
N Turns off all AC system controls.
Ofile List of ac buses in file for printing frequencies and angles
(for generator buses), or voltages and angles (for load buses).
The input format is: BusNumber BusName; use zero when either
the number or the name are unknown. If BusName has spaces,
wrap it in double or single quotes. When this option is not
defined, the program randomly chooses 8 gen. and load buses.
p Turns off P and Q limits in regulating transformers.
P Turns off P and Q control by regulating transformers.
q Turns off Q limits in PV buses.
Q Turns off remote voltage generator control. The generators
will control their terminal voltage to its initial value.
r Turns off V limits in regulating transformers and PV buses.
R Turns off V control by regulating transformers.
s Suppress ASCII output_file.
ttol If the relative error of two consecutive iteration mismatches
is larger than tol, voltage limits and regulating transformer
limits are applied (default 0.1).
Ttol P.U. tolerance tol for NR method (default 1e4).
vval Define the time constant for voltage dependent simulation
of reactive power at load buses (default 0.01). To turn off
this type of load model make val=0.
Vfile Read P.F. initial guesses for AC bus voltages from file.
The data must be separated by spaces, i.e.,
BusNumber BusName V_mag V_ang(deg). If the input variables
are unknown give them a value of zero. Buses not in the list
are given a flat start. If BusName has spaces, wrap it on
double or single quotes.
x Turns off generator field dynamics, i.e., the generator is
modelled as a constant voltage source behind transient
reactance.
X The terminal voltage of the generator is assumed constant
and the field dynamics are neglected (i.e., Xd=Xd=1e7).
Bibliography
[1] A. Kurita and T. Sakurai, The Power System Failure on July 23, 1987 in
Tokyo, Proceedings of 27
th
IEEE Conference on Decision and Control, Austin,
Texas, December 1988.
[2] C. Barbier and JP Barret, Analysis of Phenomena of Voltage Collapse on a
Transmission System, Revue Generale de lelectricite, Vol. 89, October 1980,
pp. 672680.
[3] Proceedings: Bulk Power System Voltage PhenomenaVoltage Stability and
Security, EPRI EL6183, January 1989.
[4] B. Franken and G. Anderson, Analysis of HVDC converters connected to Weak
AC Systems, IEEE Transactions on Power Systems, Vol. 5, No. 1, February
1990, pp. 235242.
[5] J. Guckenheimer and P. Holmes, Nonlinear Oscillations, Dynamical Systems,
and Bifurcations of Vector Fields, SpringerVerlag, New York, USA, 1986.
[6] R. Seydel, From Equilibrium to ChaosPractical Bifurcation and Stability
Analysis, Elsevier Science Publishers, NorthHolland, 1988.
[7] H. G. Kwatny, Steady State Analysis of Voltage Instability Phenomena, pp.
5.15.22 in [3].
138
139
[8] I. Dobson and H. D. Chiang, Towards a Theory of Voltage Collapse in Electric
Power Systems, Systems & Control Letters 13, 1989, pp. 253262.
[9] F. L. Alvarado and T. H. Jung, Direct Detection of Voltage Collapse Condi
tions, pp. 5.235.38 in [3].
[10] V. Ajjarapu, Identication of SteadyState Voltage Stability in Power Sys
tems, IASTED Proceedings of International Conference on High Technology
in the Power Industry, Phoenix, Arizona, March 1988, pp. 244247.
[11] T. Van Cutsem, A Method to Compute Reactive Power Margins with respect
to Voltage Collapse, IEEE Transactions on Power Systems, Vol. 6, No. 2,
February 1991, pp. 145156.
[12] K. Iba, H. Suzuki, M. Egawa, T. Watanabe, Calculation of Critical Loading
Condition with Nose Curve Using Homotopy Continuation Method, IEEE
Transactions on Power Systems, Vol. 6, No. 2, May 1991, pp. 584593.
[13] H. D. Chiang, W. Ma, R. J. Thomas, J. S. Thorp, A Tools for Analyzing
Voltage Collapse in Electric Power Systems, Proceedings of the Tenth Power
System Computation Conference, Graz, Austria, August 1990, pp. 12101217.
[14] V. Ajjarapu and C. Christy, The Continuation Power Flow: A Tool for Steady
State Voltage Stability analysis, IEEE PICA Conference Proceedings, May
1991, pp. 304311.
[15] C. L. DeMarco and T. J. Overbye, An Energy Based Security Measure for
Assessing Vulnerability to Voltage Collapse, IEEE Transactions on Power
Systems, Vol. 5, No. 2, May 1990, pp. 419427.
140
[16] T. J. Overbye and C. L. DeMarco, Voltage Security Enhancement Using En
ergy Based Sensitivities, IEEE Transactions on Power Systems, Vol. 6, No. 3,
August 1991, pp. 11961202.
[17] P. Varaiya, F. F. Wu, R. L. Chen, Direct Methods for Transient Stability
Analysis of Power Systems: Recent Results, Proceedings of the IEEE, Vol. 73,
No. 12, December 1985, pp. 17031715.
[18] M. A. Pai, Power System Stability: Analysis by the Direct Method of Lya
punov, NorthHolland Publishing Company, Systems and Control Series, Vol.
3, Netherlands, 1981.
[19] M. A. Pai, Energy Function Analysis for Power System Stability, Kluwer Aca
demic Publishers, Boston, 1989.
[20] M. Vidyasagar, Nonlinear Systems Analysis, PrenticeHall, 1978.
[21] Methodology of Integration of HVDC Links in Large AC SystemsPhase 1:
Reference Manual, EPRI EL3004, March 1983.
[22] Methodology of Integration of HVDC Links in Large AC SystemsPhase 2:
Advanced Concepts, EPRI EL4365, Vol. 1, April 1987.
[23] J. Arrillaga, High Voltage Direct Current Transmission, Peter Peregrinus, Lon
don, UK, 1983.
[24] J. Arrillaga, C. P. Arnold, B. J. Harker, Computer Modelling of Electrical Power
Systems, John Wiley & Sons, UK, 1983.
[25] A. E. Hammad and W. Kuhn, A Computation Algorithm for Assessing Voltage
Stability at AC/DC Interconnections, IEEE Transactions on Power Systems,
Vol. 1, No. 1, February 1986, pp. 209216.
141
[26] L. A. S. Pilotto, M. Szechtman, A.E. Hammad, Transient AC Voltage Related
Phenomena for HVDC Schemes Connected to Weak AC Systems, IEEE/PES
91 SM 3053PWRD, July 1991.
[27] E. H. Abed and P. P. Varaiya, Nonlinear Oscillations in Power Systems,
International Journal of Electric Power & Energy Systems, Vol. 6, 1984, pp.
3743.
[28] F. L. Alvarado, Bifurcations in Nonlinear Systems: Computational Issues,
Proceedings of ISCAS, May 1990, pp. 922925.
[29] I. Dobson and L. Lu, Immediate Change in Stability and Voltage Collapse
when Generator Reactive Power Limits are Encountered, Bulk Power System
Voltage Phenomena, Voltage Stability and Security NSF Workshop, Deep Creek
Lake, Maryland, August 47, 1991.
[30] M. A. Pai, K. R. Padiyar, C. Radhakrishna, Transient Stability of Multi
Machine AC/DC Power Systems Via EnergyFunction Method, IEEE Trans
actions on Power Apparatus and Systems, Vol. PAS100, No. 12, December
1981, pp. 50275035.
[31] YiXin Ni, A. A. Fouad, A Simplied TwoTerminal HVDC Model and its
use in Direct Transient Stability Assessment, IEEE Transactions on Power
Systems, Vol. 2, No.4, November 1987, pp. 10061012.
[32] K. R. Padiyar and H. S. Y. Sastry, Direct Stability Analysis of AC/DC Power
Systems Using a Structure Preserving Energy Function, Proceedings of Inter
national Conference on Computers, Systems & Signal Processing, Bangalore,
India, December 1984, pp. 16201624.
142
[33] A. R. Bergen and D. J. Hill, A Structure Preserving Model for Power System
Stability Analysis, IEEE Transactions on Power Apparatus and Systems, Vol.
PAS100, No. 1, January 1981, pp. 2535.
[34] D. J. Hill and A. R. Bergen, Stability Analysis of Multimachine Power Net
works with Linear Frequency Dependent Loads, IEEE Transactions on Cir
cuits and Systems, Vol. 29, No. 12, December 1982, pp. 840848.
[35] N. Narasimhamurthi and M. R. Musavi, A Generalized Energy Function for
Transient Stability Analysis of Power Systems, IEEE Transactions on Circuits
and Systems, Vol. 31, No. 7, July 1984, pp. 637645.
[36] A. Tsolas, A. Aropastathis, P. Varaiya, A Structure Preserving Energy Func
tion for Power System Transient Stability Analysis, IEEE Transactions on
Circuits and Systems, Vol. 32, No. 10, October 1985, pp. 10411049.
[37] C. L. DeMarco and A. R. Bergen, A Security Measure for Random Load Dis
turbances in Nonlinear Power System Models, IEEE Transactions on Circuits
and Systems, Vol. 34, No. 12, December 1987, pp 15461557.
[38] D. B. Giesner and J. Arrillaga, Operating Modes of the 3Phase Bridge Con
vertor, International Journal of Electrical Engineering Education, Vol. 8, 1970,
pp. 373388.
[39] A. R. Bergen, Power Systems Analysis, PrenticeHall, New Jersey, 1986.
[40] S. N. Chow and J. Hale, Methods of Bifurcation Theory, SpringerVerlag, New
York, 1982.
[41] W. Rudin, Principles of Mathematical Analysis, Third Edition, McGrawHill,
USA, 1976.
143
[42] C. L. DeMarco and A. R. Bergen, Application of Singular Perturbation Tech
niques to Power System Transient Stability Analysis, Proceedings of ISCAS,
May 1984, pp. 597601.
[43] K. Jimma, A. Tomac, K. Vu, C. C. Liu, A Study of Dynamic Load Models for
Voltage Collapse Analysis, Bulk Power System Voltage Phenomena, Voltage
Stability and Security NFS Workshop, Deep Creek Lake, Maryland, August
47, 1991.
[44] M. Huneault, A. FahmidehVojdani, M. Juman, R. Calderon, F. D. Galiana,
The Continuation Method in Power System Optimization: Applications to
EconomicSecurity Functions, IEEE Transactions on Power Apparatus and
Systems, Vol. PAS104, No. 1, January 1985, pp. 114124.
[45] S. Sastry and P. P. Variaya, Hierarchical Stability and Alert State Steering
Control of Interconnected Power Systems, IEEE Transactions on Circuits and
Systems, Vol. 27, No. 11, November 1980, pp. 11021112.
[46] P. W. Sauer and M. A. Pai, Power System SteadyState Stability and the
Load Flow Jacobian, IEEE Transactions on Power Systems, Vol. 5, No. 4,
November 1990, pp. 13741383.
[47] A. Spence and B. Werner, Nonsimple Turning Points and Cusps, IMA Jour
nal of Numerical Analysis 2, 1982, pp. 413427.
[48] S. D. Conte and Carl de Boor, Elementary Numerical AnalysisAn Algorithmic
Approach, Third Edition, McGrawHill, New York, USA, 1980.
[49] D. F. Delchamps, State Space and InputOutput Linear Systems, Spinger
Verlag, New York, 1988.
144
[50] F. L. Alvarado, SolverQ Instruction Manual, Software Development Distribu
tion Center, University of WisconsinMadison, 1987.
[51] F. L. Alvarado and D. J. Ray, Symbolicallyassisted Numeric Computation
in Education, International Journal of Applied Engineering Education, Vol. 4,
No. 6, 1988, pp. 519536.
[52] Y. Tamura, K. Sakamoto, Y. Tayama, Current Issues in the Analysis of Voltage
Instability Phenomena, pp. 5.395.53 in [3].
[53] I. Dobson, Observations on the Geometry of Saddle Node Bifurcations and
Voltage Collapse in Electric Power Systems, Report ECE905, University of
WisconsinMadison, August 1990, to appear in IEEE Transactions of Circuits
and Systems.
[54] H. D. Chiang and F. F. Wu, Stability of Nonlinear Systems Described by a
SecondOrder Vector Dierential Equation, IEEE Transactions on Circuits
and Systems, Vol. 35, No. 6, June 1988, pp 703711.
[55] H. D. Chiang, Study of the Existence of Energy Functions for Power Systems
with Losses, IEEE Transactions on Circuits and Systems, Vol. 36, No. 11,
November 1989, pp. 14231429.
[56] A. A. Fouad, V. Vittal, S. Rajagopal, V. F. Carvalho, M. A. ElKady, C. K.
Tang, J. V. Mitshe, M. V. Pereira, Direct Transient Stability Analysis Using
Energy FunctionsApplication to Large Power Networks, IEEE Transactions
on Power Systems, Vol. 2, No. 1, February 1987, pp. 3744.
[57] Evaluation of Transient Energy Function Method Software for Dynamic Secu
rity Analysis, EPRI EL7357, July 1991.
145
[58] C. L. DeMarco, A New Method of Constructing Lyapunov Functions for Power
Systems, Proceedings of ISCAS, 1988, pp. 905908.
[59] R. Qian, Stability Analysis and Security Assessment in Uncertain Power System
Models, Ph. D. Thesis, University of WisconsinMadison, 1990.
[60] M. A. Pai, Computer Techniques in Power System Analysis, Tata McGrawHill,
New Delhi, 1979.
[61] H. D. Chiang, F. F. Wu, P. P. Varaiya, Foundations of the Potential En
ergy Boundary Surface Method for Power System Transient Stability Analy
sis, IEEE Transactions on Circuits and Systems, Vol. 35, No. 6, June 1988,
pp. 712728.
[62] I. Dobson and L. Lu, Immediate Change in Stability and Voltage Collapse
when Generator Reactive Power Limits are Encountered, Bulk Power System
Voltage Phenomena, Voltage Stability and Security NFS Workshop, Deep Creek
Lake, Maryland, August 47, 1991.
[63] Transmission Line Reference Book/345KV and above, EPRI EL2500R, Second
Edition, 1987.
[64] Transmission Line Reference Book/HVDC to 600KV, EPRI EL0100.2.
[65] I. Dobson and L. Lu, Using an Iterative Method to Compute a Closest Saddle
Node Bifurcation in the Load Power Parameter Space of an Electric Power Sys
tem, Bulk Power System Voltage Phenomena, Voltage Stability and Security
NFS Workshop, Deep Creek Lake, Maryland, August 47, 1991.
[66] I. Dobson and L. Lu, Computing and Optimal Direction in Control Space to
146
Avoid Saddle Node Bifurcation and Voltage Collapse in Electric Power Sys
tems, Bulk Power System Voltage Phenomena, Voltage Stability and Security
NSF Workshop, Deep Creek Lake, Maryland, August 47, 1991.
[67] Extended TransientMidterm Stability Package: Users Manual for the Power
Flow Program, EPRI EL2002, January 1987.
[68] C. A. Ca nizares, F. L. Alvarado, C. L. DeMarco, I. Dobson, W. F. Long, Volt
age Collapse and Transient Energy Function Analyses of AC/DC Systems,
Report ECE907, Department of Electrical and Computer Engineering, Uni
versity of WisconsinMadison, November 1990.
[69] C. A. Ca nizares, F. L. Alvarado, C. L. DeMarco, I. Dobson, W. F. Long, Point
of Collapse Methods Applied to AC/DC Power Systems, IEEE/PES 91 SM
4911PWRS, July 1991.
[70] C. L. DeMarco and C. A. Ca nizares, A Vector Energy Function Approach for
Security Analysis of AC/DC Systems, IEEE/PES 91 SM 4218PWRS, July
1991.
[71] C. A. Ca nizares and F. L. Alvarado, Computational Experience with the Point
of Collapse Method on Very Large AC/DC Systems, Bulk Power System Volt
age Phenomena, Voltage Stability and Security NFS Workshop, Deep Creek
Lake, Maryland, August 47, 1991.
[72] IEEE Committee Report, Common Format for Exchange of Solved Load Flow
Data, IEEE Transactions on Power Apparatus and Systems, Vol. PAS92, No.
6, November/December 1973, pp. 19161925.
[73] S. Iwamoto and Y. Tamura, A Load Flow for IllConditioned Power Systems,
147
IEEE Transactions on Power Apparatus and Systems, Vol. PAS100, No. 4,
April 1981, pp. 17361743.
[74] F. L. Alvarado, Manipulation and Visualization of Sparse Matrices, ORSA
Journal of Computing 2, 1990, pp. 180207.
[75] TRANSMISSION 2000 Power Flow Program Users Reference Manual, Elec
trocon International, Ann Arbor, Michigan, July 1989.
[76] PROMATLAB Users Guide, The MathWorks, Natick, Maryland, USA, Jan
uary 1990.