512
February 1995
ON BIFURCATIONS, VOLTAGE COLLAPSE AND LOAD MODELING
Claudio A. Caiiizares, Member
University of Waterloo
Elec. & Comp. Eng. Dep.
Waterloo, ON, Canada N2L 3G1
Abstract: This paper discusses the relation between bifurcations
and power systems stability through a thorough analysis of sev-
eral examples, to clarify some ideas regarding the usefulness and
limitations of bifurcation theory in network studies and operation,
particularly in voltage stability related issues. Different types of
load models are used in a sample system to analyze their effect on
system stability and bifurcation. Finally, the Ecuadorian National
Interconnected System (SNI in Spanish) is used to depict and dis-
cuss the effect of load modeling in saddle-node bifurcation analysis
of real power systems.
Keywords: bifurcations, voltage collapse, load modeling, stability,
eigenvalue analysis.
1 INTRODUCTION
During the past decade utilities have reported serious compli-
cations in maintaining network stability in their power systems,
particularly voltage stability, as some events occur and parameters
change in the system [1]-[3]. The study of several cases has led
the power systems community to identify different causes for this
problem.
Hopf and saddle-nodebifurcations have been recognized as some
of the reasons, albeit not the only ones, for voltage stability prob-
lems in a variety of power system models [4]-[9]. Local bifurcations
are detected by monitoring the eigenvaluesof the current operating
point. As certain parameters in the system change slowly, allow-
ing the system to quickly recover and maintain a stable operating
point, the system eventually turns unstable, either due to one of
the eigenvaluesbecoming zero (saddle-node, transcritical, pitchfork
bifurcations), or due to a pair of complex conjugate eigenvalues
crossing the imaginary axes of the complex plane (Hopf hifurca-
tion). The instability of the system is reflected on the state vari-
ables, usually represented by frequency, angles and voltages, by an
oscillatory behavior or a continuos change (voltage decrease, i.e.,
collapse, and frequency and angle increase, i.e., loss of synchrw
nism). In some cases these bifurcations can be associated to the
power transfer limit of the transmission system; in other instances
the bifurcations appear due to voltage control problems, like fast
acting automatic voltage regulators (AVR) in the generator [IO], or
voltage dependent current order limiters (VDCOL) in HVDC links
[7].In all cases these bifurcations occur on very stressed systems,
i.e., the region of stability for the current operating point (stable
equilibrium point or s.e.p.) is small, hence, the system is not able
to withstand small perturbations and becomes unstable. Although
there are reports of these bifurcations occurring in unstressed sys-
tems [ll], this camiot be considered typical, since power system
controls are designed so that eigenvaluesof several operating points
are well into the left half complex plane.
Some voltage collapse problems can also be associated to voltage
control devices like under-load tap changer (ULTC) transformers or
94 SM 512-4 PWRS A paper recommended and approved
by t h e IEEE Power System Fngineering Committee of t h e
IEEE Power Engineering S o c i e t y for p r e s e n t a t i o n a t
t h e IEEEIPES 1994 Summer Meeting, San Francisco, CA,
-
July 24 28, 1994. Manuscript submitted December 23,
1993; made a v a i l a b l e f o r p r i n t i n g May 3, 1994.
0885-8950/95/$04.00 0 1994 IEEE
~ ~~
IEEE Transactions on Power Systems, Vol. 10, No. 1.
1L6 VLOO n
U
Load 1
Fig. 1: Sample s y s f e m f o r bifurcation analysis.
AVRS [12]-[15]. In some of these cases, but not all, the voltage con-
trols force the eigenvaluesto instantaneonslyjump into the unstable
region, making the system immediately unstable. This phenomena
is not directly associated to a bifurcation, since the eigenvaluesdo
not go through zero or the imaginary axis. Nevertheless, transcriti-
cal bifurcation theory can be used to explain the phenomena when
AVR limits are assumed to apply gradually [13].
Recent literatureshows that there is not a consensus in the power
system community regarding the importance and usefulness of bi-
furcation analysis of power systems (e.g., [16]). Therefore, this pa-
per concentrates mainly on clarifying some issues regarding local
saddle-node bifurcations, by clearly depicting their effect and rela-
tionship to the voltage collapse phenomena for different load mod-
els in several sample systems. This does not rule out the fact that
other typeof bifurcations, i.e., transcriticalor Hopf, do also occur in
real systems, with similar catastrophic results for system (voltage)
stability. Although bifurcations cannot be considered as the only
cause for voltage stability problems, as discussed above, the effect
of saddle-nodebifurcations in system stability cannot be overlooked
based on this sole argument.
The final section of this paper concentrates on applying the re-
sults obtainedfor the samplesystems to the 115 bus SNI Ecuadorian
system [lq, to discuss some of the applications and shortcomings
of bifurcation theory in a more realistic environment.
2 EXAMPLES
In this section the stability of four sample systems is thoroughly
analyzed. Local saddle-node bifurcations are depicted and discussed
to highlight some of their main characteristics. All four cases dis-
cussed here are based on the system depicted in Fig. 1.
The generator is modeled with the classical second order me-
chanical differential equations. The damping torques are assumed
to have a relatively large value to indirectly simulate the effect of
damper windings in the generator. The AVR is modeled by keep
ing the generator terminal voltage constant at 1 pu, but no limits
are included in this simple model. Although this is uot a detailed
model of the generatorby any means, it gives a good starting point
for analyzing the significanceof saddle-node bifurcations in system
stability, without adding unnecessary complexities that ouly ob-
scure some of the questions that are being addressed in tlus paper.
For an analysis of bifurcations in detailed generator models review
reference [I]].
The transinissionsystem is simply modeled as a constant lumped
reactance, which is a typical representation of these elements in
transient stability studies, where the system frequency oscillates
slowly around the nominal value (e.g., 60Hz), allowing for a quasi-
steady state or phasor analysis of the system. Consequently, this
type of methodology cannot be used to simulated high frequency
voltage problems in the system.

Fig. 2: ( a ) M i n i m u m magnitude of the s . e . p . (u.e.p.) eigenvalues,
and ( b ) m i n i m u m singular value of the s . e . p . as a function of P .
Notice the sharp change in the eigenvalues magnitude and singular
values when the system approaches bifurcation.
The four study cases are obtained by changing the load model at
the receiving end of the transmission line, as shown below. Some of
these examples are discussed from a different point of view in [18].
2.1 Generator-Infinite Bus
The first load model is an infinite bus, which is drawing active
power P > 0 from the generator at a constant voltage V > 0. This
is the classical generator-infinite bus model utilized to introduce
basic ideas of angle and frequency stability in power systems, hence,
the same example is used here to introduce some basic concepts of
bifurcation analysis. For this simple case, differential equations ( 1 )
represent the system.
s = w (1)
X
AU equilibrium points can be easily found by solving equations
(1) with [6 &IT = 0 , yielding equilibria (So,O), and ( f r r - S0,O)
with 60 = sin-'(PX/V) (other equilibrium points are not of inter-
est). Hence, all the eigenvaluesand eigenvectorsassociated to these
equilibria can be calculated from the Jacobian matrix (2).
From the eigenvaluesof J it can be readily shown that (So, 0) is a
s.e.p., i.e., all eigenvaluesare in the left halfconiplex plane, whereas
( i r r - So,O) are unstable equilibrium points (u.e.p.) of type one,
i.e., only one eigenvalueis on the right half plane. If P changes, i.e.,
the system load changes, then for any value of P > Pmal= V / X ,
one eigenvalue of J becomes zero ( J is singular) and no equilibria
exist. When P = P m a Z , the maximum power trausfer point for the
transmission system, J is also singular and two equilibrium points
( r / 2 , 0 ) and (-3rr/2,0) exist, i.e., the s.e.p. and one u.e.p. have
merged. The point (rr/2,0,Pm,,) corresponds to a saddle-node
bifurcation, siuce it meets all the transversality conditions required
for this type of bifurcation in state-parameter space (7, 19, 201.
Figures 2, 3, and 4 were obtained from equations (1) by setting
M = D = O.lpu, X = k = O.Spu, and V = lpu. In Fig. 2 the
minimum magnitude of the complex eigenvaluesand the minimum
singular value of the s.e.p. (S0,O) are plotted for a continuously
increasing P load; the u.e.p. (rr - So,O) presents the same profiles.
At the value Pmal= 2pu one eigenvalue (singular value) of the
s.e.p. and the u.e.p. becomes zero. Observe that the profile of these
two measures is highly nonlinear aud varies only slightly up to the
load level of P x 1.8, decreasing rapidly after that. Similar profiles
can be observed in larger and more detailed modeled systems [4, 9,
211, making these measures inadequate to monitor system proximity
to bifurcation.
Other tednuques have been proposed to avoid using eigenvalues
or singular values for measuring proximity to bifurcation (see refer-
ence [22] for a summary). In particular, Transient Energy Function
(TEF) techniques have been shown to have a "linear" profile and
yield additional information regarding the stability of the system
[23], making these methods more suited for measuring proximity
to bifurcation but at a higher computational cost. For the sample
513
Fig. 3: Bifurcation diagrams: (a) in ( S , w , P ) azes; ( b ) P - 6 node
curve; ( c ) potential energy W p between s . e . p . and closest U . C . P .
Right v and left w eigenvectors at the bifurcation point are also
depicted; note that v i s tangent to the bifurcation diagram.
system under discussion, the euergy can be defmed as a summa-
tion of the kinetic and the potential energy as show11 in (3). The
potential energy can be approximated in this case by the integral
of the power-angle product, when assmniiig only small variations
of the system frequency (a typical conjecture in transient stability
studies).
V(63~9P) = W ~ ( WW +
) p(6,P) (3)
= iMwZ+ l60
hT(zL,P)du
+ lo[
6 V
x !Mwz - sin(u) - P ] du
X
= -1M w Z
2
+ P(60 - 6 ) + -
V [cos(So) - cos(S)]
X
By using the potential energy distance V($,,O,P) between the
s.e.p. and the "closest" u.e.p., as depicted in Fig. 3(c), one can
get a good idea of the distance to bifurcation, due to the quasi-
linear profile of this measure, and also visualize the relative size of
the stability region, including all system nonlinearities. The latter
property is rather important, since no technique based on eigeii-
values or linearizations of the system around equilibrium poiiits
can give the user this information, due to the elimination of the
nonlinearities that characterize the stability of the system for large
perturbations like line openings at fixed load levels. It is interesting
to observe that this particular energy function V ( . ) is the integral
of the area of the bifurcation diagrams depicted in Figs. 3(a) and
(b), for a chosen load level P (i.e., equal area criterion), being tlus
the reason for the "linear" profile of the TEF; this property has
been demonstrated for other system models and energy functions
in reference [24]. Similar TEF profiles have also been observed for
highly nonlinear AC/DC system models when close to bifurcation
[25]. Hence, the energy measurement clearly depict the reduction
of the stability region as the system approaches bifurcation, making
it more sensitive to small perturbations that can easily drive it to
instability.
The continuos lines in Figs, 3(a) y d (b) represent the s.e.p.s,
and the dashed lines the correspondmg "closest" u.e.p.s for each
value of the parameter P. Figure 3(b) is the projection of Fig.
3(a) in the ( 6 , P ) plane, yielding a typical "nose" curve for angle
variables and a different view of the classical P-6 stability curve.
Some interesting observations can be made in these figures:
e The uormalized left eigenvector v at the bifurcation poilit is
tangent to the bifurcation diagram [26]; hence, by looking
514

Fig. 4: T i m e domain simulation for a change in P f r o m 1.99

t o 2.01 at t = 2s. Observe the slow initial change in the state
variables (6,w).

at the maximum entries in tlus vector, one can pinpoint the

areas (state variables) that are dianging the most and driving
the system to bifurcation, so that corrective measures can be
taken to avoid stability problems 19, 271. This eigenvector
also yields information regarding the initial dynamics of the
system at the bifurcation point [SI.

b The normalized right eigenvector w at the bifurcation does
not have a direct geometric interpretation in bifurcation dia-
grams, but it has been shown to yield information regarding
the equations that are most sensitive to changes at the bifur-
cation point [9, 271, and can be used to devise a technique
to move away from tlus instability [28]. Tlus eigenvector de-
pends on the time constants ( M in the sample system) of the
differential equations [26], whereas v is independent of them.
Fig. 5: Bijurcation diagrams: ( a ) i n ( 6 , V , P ) azes; ( 6 ) minimum
magnitude of s.e.p. eigenvalues as a function of load level P ; ( 6 )
P - V node curve; ( c ) potential energy W , between s.e.p. and closest
u.e.p. Right v and leji w eigenvectors at the 6ifurcation point are
depicted. T h e vector w' is the projection of w onto the (6,V) plane.

The bifurcation diagram can be used to determine the dis-

tance to the bifurcation point, so that an operator knows in
state and parameter space how close the system is to the in-
stability point. Nevertheless, tlus diagram by itself does not

;.I rr //
yield the size of the stability region associated to the different
s.e.p.s depicted in it, and cannot give the operator a sense
of the type of perturbations that the system c a n withstand
at some specific load level, unless it is used in conjunction
with and "energy" measure that takes into consideration the
nonlinearities in the system.
Finally, Fig. 4 depicts the dynamics of the system as the load
is increased slightly beyond the bifurcation value; a rather small
system perturbation. Notice the separation or lost of synchronism
of the generator with respect to the infinite bus, with initial slow
ji
IO
,
, ~ ,

dynamics of the state variables (6, w ) [a]. This particular example

does not present any stability problems that could be characterized
Fig. 6: Time domain sirnolation f o r a change i n P from 0.61 to
as voltage collapse, however, it clearly depicts an instability due to
0.62 at t = 28. Observe the fast voltage collapse and the mechanical
a bifurcation problem, and if voltages were allowed to change the
instalility reflected in 6 and w.
angular-frequency instability would necessarily force the voltages to
collapse. This is the reason why bifurcation studies should be done

considering all system dynamics and not only voltage controls, to
correctly capture all the bifurcation phenomena.
2.2 PQ Dynamic Load
The infinite bus load model is now replaced by a more "realistic"
constant active and reactive power model. This model is meant to
simulate the behavior of typical static loads when ULTCs are taken
into consideration 1121. Furthennore, introducinga dynamic voltage
term in the reactive power load roughly simulates the response of
induction motors to sudden voltage changes under normal voltage
operatingconditions [29,30]. This cannot be considered an accurate
model of the actual system load, but allows to introduce and depict
some of the stability problems, particularly voltage collapse, that
occur when a saddle-node bifurcation is encountered by the system.
The differential equations used to model this sample system are
equations (1) plus equation (4) below, and the Jacobian correspond-
ing to an equilibrium point (60,0,Vo, P) of tlus set of equations is
represented by (5).
[ ]
1 0
J = -V/M;cos(60) -DIM -1/MXsin(60) (5)
- V / T X sin(&) 0 - 1 / r X [ 2 V - sin(S,)]
Here k is a scalar representing constant power factor, and r is the
time constant of the dynamic voltage term. When r -t 0, equation
(4) becomes an algebraic constraint, however, the dynamics of the
system lose physical meaning since this set of differential equations
and algebraic constraints cannot be qualified as a "causal" system
according Kwatny's et al definition [5]. Furthennore, this system
presents a degenerate bifurcation point in state-paranleter space
where two stable equilihrinnipointscoalesce. For a detailed analysis
of these types of system models review references [31]-[33].
Figures 5 and 6 were obtainedusing the same constant values as
in the previous example, and setting T = lins to simulate fast load
dynamicsin the context of a typical transient stability study. Figure
5(a) depicts a saddle-node bifurcation diagram for 6 and V (w = 0
for all equilibria), i.e., a s.e.p. and a u.e.p. coalesce at a singular
point (Fig. 5(b)) for the parameter value P,,, = 0.618034, with
equilibria disappearing for values of P > P,. P,, corresponds

1 (4)
again to the point of maximum power transfer, as it is shown in
the next section with an R-L load, Figure 5(c) depicts the classical
 515

P - V nose curve, wludi is a projection of the bifurcation diagram

in the ( P , V ) plane. A similar result as the one depicted in Fig.
3(b), i.e., a P-6 nose curve, can also be obtainedhere by projecting
the bifurcation diagrani in the ( P ,6) plane; however, in tlus case
the bifurcation occurs at a value of 6 ES 30' << 90'. Observe that
the right eigenvector v is also tangent to the bifurcation diagram
at the saddle-node bifurcation point, with the entries in this vector
pointing to a voltage problem regardless of the value of r > 0.
The energy function for this particular system is represented by
equation (6) [34]. The results of evaluating the energy difference
between the s.e.p.s and corresponding u.e.p.s in the bifurcation di-
agram of Fig. 5(a) are depicted in Fig. 5(d). Notice once more the
quasi-linear profile of the TEF, representing the reduction of the
stability region as the system approaches bifurcation.

V(S,w,V,P) =

+&(Vz
X
V
l M w z + P ( 6 0 - 6 ) + --[c0s(60)-cos(6)]

- V z ) - kP[ln(Vo) - W')]

Figures 6 depict the time domain simulation of the differential

(6) :::v
;L \
0.1 o.* 0.1 0.4 0.5 0 . s Q 0.1 0.2 0.3 0.4 0.5 0.6

equations (1) and (4) for a slight increase in load power P . In

this plots the same patterns of the previous example appear, i.e., Fig. 7 : Bifurcation diagrams: ( a ) in (6,V,G) azes; ( b ) minimum
the state variables start cliaiging slowly to then rapidly render the magnitude of s.e.p. eigenvalues as a function of the parameter C:;
system nnstable. In this case, due to the small voltage time constant ( c ) P - V nose curve; ( d ) P - G curve. Notice fhat all equilibrium
as compared to the machine inertia, the voltage collapses at t x 5s points are stable f o r any value of the parameter G, even though the
rather quickly when compared to the angle and frequency. Tlus P - V nose curve i s ezacily the same as in the case with the P Q load
particular voltage behavior due to the differences in time constants, model.
have led many researchers to only consider voltage dynanucs for
the analysis of bifurcations problems, ignoring frequency dynamics.
However, the previous example clearly shows that this assumption
is not completely justifiable.
This sample system is a classical bendunark for voltage stability
studies, due to the characteristic voltage behavior discussed in this
section.

2.3 RL Loads
Now the PQ dynamic load is replaced by a11 RL dynamic load,
i.e., the load is represented by a constant power factor impedance,
with the same dynamic voltage term as in the previous case. Hence,
differential equations (7) are used to model this particular system,
with G representing the load conductance and the slow varying
parameter in tlus example.

s = w (7)

& = 1
M
[V'G- xsin(6)
X
- Dw
v cos(6)
I Fig. 8 : Time domain simulation f o r a change in G f r o m 1.5 to 2.5
at t = 25, faking the systems around the maximum power transfer
V = 1 [-kVzG - -+ -
v2
x x 1 point P,,, = 0.618034pu. Observe that all state variables ( S , w , V )
remain sfable.

Using the same constant values as in the previous examples, one

obtains Figs. 7 and 8. Figure 7(a) depicts the bifurcation diagram
for changes in the parameter G; observe that no bifurcations occur
in this case, as shown by the eigenvaluesplot of the system equilibria
in Fig. 7(b). Furthermore, only one s.e.p. can be obtained for
each value of G 2 0. Nevertheless, due to the relation between
conductance G and the active power P depicted in Fig. 7(d), this
system presents exactly the same P-V nose curve as in the case
of the PQ load model (Fig. 7(c)), with all equilibria in tlus curve
being stable. The tip of this nose curve is not a bifurcation point,
although it corresponds to the system's maximum power transfer
point Pmas, i.e., IG(1 +jk)( = 1/X (G = 4/& = 1.78885, and
Pmaz= 0.618034). The nose curve in this case cannot be considered
a bifurcation diagram, since G is the parameter that changes in the
system and not P.
Figures 8 corroborate the previous discussion. A relatively large
change in G, wlucli takes the system around the maximum power
transfer point P m a z , does not render the system unstable, taking
it to a new s.e.p. in spite of the large perturbation.
A n exact energy function cannot be found for tlus sample system,
due to the presence of a resistive term in the model [35]. Although
an approximate TEF could be constructed in tlus case, it cannot be
readily used as in previous examples to obtain a relative measure
of the stability region as the conductance G changes.
Finally, the system load is changed to a static RL load model,
elinunating the dynamic voltage dependence in equations (7) by
setting T = 0, i.e., the tlurd differential equation becomes an alge-
braic constraint. Despite this change in load model, a i d contrary
to what happens in the case of the PQ load model when r -+ 0, the
results remain exactly the same with the only difference that V is
allowed to change instantaneously.
3 APPLICATIONS
The foregoing discussion can be directly extended to larger and
more complex power system models. Transient stability analysis is
t y p i d y done on a balanced phasor model of the network, wlud~
can be represented by the set of nonlinear differential and alge-
braic constraints (S), where x E 8" stands for the state variables,
y E 8"' portrays a set of iniplicitly defiled variables, and X E W
is a parameter used to simulate slow changing conditions (small
516
perturbations) in the system.
This model has the difficulty of not fully representingreal physical
quantities, which is reflected in the presence of the algebraic con-
straints g(.) and the use of reactive powers throughout the equa-
tions. Nevertheless, this set of equations can be used to represent
any degree of complexity in modeling and system size, within the
bounds of the assumptions used to build the model.
For equations ( 8 ) , the algebraic constraints must meet certain
conditions in order for the model to approximate physical reality
[5, 7, 26, 31, 321. Particularly, at equilibrium points (xo,yo,X),
such that f(xo,yo,X) = 0 and g(Xo,yo,X) = 0 , the Jacobian
of these constraints D y g ~ ~ l o , yshould
o , ~ ) be invertible, rendering
the system causal [5], otherwise singular perturbations or noise
techniques must be used to analyze the dynamics of the system
[31? 321. Furthermore, if D,g(.) is invertible (nonsingular) along
system trajectories, the system can be represented by the set of
differential equations
X = f(x,y-(x, X),X) = F(x,X)
This reduction is actually unnecessary, since one only needs to check
for and remove singularities in the algebraic equations Jacobian.
For a detailed discussion on this issues and their relation to bifur-
cation analysis in power systems review [7,261.
For bifurcation analysis of this system model, the eigenvalues
and eigenvectors of the system Jacobian (IO) must be monitored.
For saddle-node bifurcations in particular, one needs only to look
for a singularity of this Jacobian and test for the corresponding
transversality conditions [7,261. The latter is usually unnecessary
since saddle-node bifurcations occur generically in differential equa-
tions (9) (191; nevertheless, under certain operating conditions like
remote voltage control, which give the system equations a special
symmetry, transcritical bifurcations can also be encountered.
Embedded in this set of equations are the algebraic and/or dif-
ferential equations representing the load. For this paper a mixed
load model represented by equations (11) is used, since it has been
shown to adequately represent a variety of system loads in typical
transient stability studies [29, 30, 371.
9 = -(Pi,,+ Api,X) - ( 4 ,+ A f i l x ) ( K / V ? ) 2
-(p12+A~,,x)(KIII;O)-T~,,(~~- ~ n G ) - ~ l ~ p ~
QI = -(Qio + AQi,X) - (QI] + AQi,X) (Vtlxo)2
+
- ( Q I ~ AQi,A) (Vt/xo)- (61 - - 7ifQ WnG) 7tVQ Vi
Here Pi and Qi are the powers injected by the load, and hL6t
is the load phasor voltage at bus 1. P i , , Pi1,Pi,, Qi,,, Qil, and
Qi, are constant weighting factors that define the steady state base
load. q f P ,q V p 71fQ,
, and riVQ represent the time constants of
the frequency and voltage dependent dynamic terms, and is
the system reference frequency; note that these time constants can
be set to zero to represent static load models. APl,, Apt,, APi, ,
AQi,, AQI,, andAQ1, areusedtomodeladirectionofloadchange
as the parameter X slowly varies. For most bifurcation studies it is
assumed that the pattern of load change can be represented with
one degree of freedom (A), and that this evolution of load drives the
system to bifurcation. Notice that differential equations (11) can
be rewritten as
where 6Pi(6,V,X) and SQi(6,V,X) are the active and reactive
power mismatches at the load buses, respectively.
Although the issue of whether this load model is a realistic
representation of the actual system load does not pertain to this
particular paper, it is a very valid question, since as demonstrated
in this paper and others before (e.g., [36]), the results of stabil-
ity studies are closely related to system modeling, particularly load
(8) modeling [371. This is an old discussion that has not been resolved
to the total satisfaction of all the Dower system commuuity. Nev-
ertheless, the load model must be & i d in-the context of tce tools
that are used for stability studies; for instance, very fast dynamics
like thyristor switching of motor drives cannot be fully represented
in typical transient stability program, since these tools have been
designed to deal with quasi-steady state models, i.e., balanced pha-
sor models that treat transformersand transmission lines as lumped
one-phase impedances without a n y frequency dependence. For fast
load dynamics a very detailed model of the whole system is needed,
being toolslike the EMTP more appropriate enviromnents for ana-
lyzing stability of these models; however, these types of studies are
very expensive, with costs growing exponentially with system size.
Hence, a compromise and a critical view of the results yielded by
some of these programs is needed, and bifurcation analysis is not
an exception, i.e., the output and cost of t h e e studies are closely
related to the models used.
The SNI Ecuadorian system is used in this paper due to its par-
ticular voltage stability problems, and also to depict the effect of
(9) different load models in saddle-node bifurcation analysis of real
systems. The results shown here have been used to c o i f i n sus-
picions regarding the possible causes of the voltage problems in
the system, and to study some possible solutions. The generators
in this case are simply model using the second order differential
equations with constant terminal voltage used in the previous ex-
amples, which is an acceptable model to address the main issues in
the present discussion. A continuation method was used to obtain
the bifurcation diagrams in Figs. 9, 10 and 11 [15,20, 221. This
method allows to trace bifurcation manifolds (diagrams) for any
type of system model; however, the technique is particularly suited
to detect saddle-node bifurcations, yielding a close approximation
of the bifurcation point (singularity point) and the corresponding
right eigenvector v [%I, without having to actually calculate and
trace the system eigenvalues and eigenvectors as the parameter X
changes. The problem with this approach is that it cannot detect
Hopf bifurcations, nor differentiate between transcritical and pitch-
fork bifurcations. For the latter, one needs to check for the transver-
sality conditions of the system at the bifurcation point, which is a
costly task. The fonner needs of the computation of the system
equilibria eigenvalues to detect the crossing of the imaginary axis,
since the Jacobian J in (10) does not become singular in this case;
a very expensive procedure when dealing with relatively large sys-
tems. Furthermore, to discern wluch type of Hopf has taken place
(11) (subcritical or supercritical), one needs to run a costly transient
simulation or use special techniques that take into considerations
additional nonlinear terms in the system [8, 19, 201. Nevertheless,
for PQ load models and typically small resistive losses in the trans-
mission system, Hopf bifurcations are unlikely to occur [lo].
Figures 9 and 10 depict the projections of the bifurcation di-
agram onto different ( P , V ) planes ( P - V nose curves) for various
system models. Due to system size, detecting possible occurrences
of Hopf bifurcations before a saddle-node was not pursued, since
Hopf bifurcations are beyond the scope of this paper. For this rea-
son the author has chosen not to differentiate between s.e.p.s and
u.e.p.s in these hal figures; nevertheless, one typically expects to
have stable equilibria above the saddle-node bifurcation for these
system models. In all these figures, the voltage profiles in buses
15 (a 138KV bus located in Machala, a city in the south-west of
Ecuador) and 34 (a 138KV bus located in Manabi, a province in the
central west coast of Ecuador), were plotted to depict two distinct
system nose curves. Bus 34 was particularly chosen since the right
and left eigenvectors at the bifurcation point, pinpoint that partic-
ular area as the one with the largest reactive power deficits, fully
matching the operationalknowledgeof the actualsystem. Although
the bifurcation diagram projections in Fig. 9(a) do not yield mudi
information regarding the size of the stability region by themselves,
the area on the diagrams can be associated to an energy measure
due to the P Q load model used in this case, giving an idea of how
stable the system is at a particular load level. Furthennore, the
parameter X in this case can be associated to a total MVA sys-
tem load (1292MVA), yielding a maximum loadability margin of

0,40 o.l 0.2 0.3 0 0.2 0.4 0.6
lama lambda
Fig. 9: Bifurcation diagvams of the SNZ Ecuadorian system f o r
( a ) PQ and (b) mixed load models.
la) (b)
m
0.9
lama lambda
Fig. 10: k?ijurcation diagrams of the SNZ Ecuadorian system f o r
( a ) P Q and (b) mized load models when Q limits are considered in
the generators.
296MVA (A, = 0.229133).
In Fig. 9(b), the mixed load model (11) was used to obtain
new bifurcations diagrams and a different saddle-node bifurcation.
Notice again the wide changes in voltage magnitude at bus 34 for
chauging loading conditions, indicating once more the lack of reac-
tive power support that is the characteristic of this system. The
value of ,A,, = 0.417712 is practically twice as mush as the one in
the previous case, since now the loads reduced their power demand
as the voltages decrease, allowing the system to handle heavier
loading conditions. The latter comment should be regarded with
caution, since the loadiiig level indicated by X does iiot have a one-
to-one correspondence to MVA loading levels as in the previous
figure. Due to the relatively large resistive terms in this system
model, one cannot use the simple analogy of the bifurcation dia-
gram area to convey an idea of the size of the stability region for
each value of A; for this particular load model only time simulations
can yield the robustness of the system to large perturbations at a
given load level. Nevertheless, close to the saddle-node bifurcation
oue expects the system to have a small stability region due to the
merging of the s.e.p and a u.e.p.
Figures. 10 depict the effect of generator Q limits in saddle-
node bifurcations. These Q limits were roughly modeled by fix-
ing the reactive power output of the generator at its limits [12],
which is not an accurate model, but the approach allows for a first
uuderstanding of the general effects of AVR limits in saddle-node
bifurcations for different load models. In Figs. 10 one can clearly
observe that the Q limits significantlyreduce the loadabilitymargin
in both cases, since major generating facilities reach maximum re-
active power limits, i.e., vapor turbines in El Salitral (located near
the largest load center in Ecuador, the coastal city of Guayaquil)
at X = 0.05109 = 66MVA in Fig. lO(a), and the Agoyan hydro
station (a relatively large generation station in the central Andean
region) at X = 0.1747 in Fig. 10(b). Furthermore, Fig. lO(a) de-
picts an innnediate system instability at A = 0.0615 = 79.5MVA,
due to some generators readcng maximum reactive limits at the
largest Ecuadorian generating station (Paute, a 1000 MVA hydro
complex located in the south central Ecuadorian Andes) and com-
pletely losing voltage control of the system. Notice the rather small
loadability margin of the system in Fig. lO(a). These particular
problems have been detected during the actual system operation,
517
I
0.60 0.05 0.1 0.15 0.J
0 0.1 0.2 0.3
lama lambda
Fig. 11; Bifurcation diagrams of the SNZ Ecuadorian system when
a 70 MVAR synchronous compensator is placed on bus 54. Figure
( a ) corresponds to PQ load models, and figure (b) t o mixed load
models. Q limits are enforced in the generators.
and they are the reason for the repetitive voltage problems through-
out the network.
Based on the results of the bifurcation analyses of the previous
cases, a 70MVAR synchronous compeusator is placed 011 bus 34.
which is one of the possible solutions under consideration at the
Ecuadorian utility (INECEL in Spanish) to improve the stability
of the SNI system. The new system yields Figs. 11; observe the
significant improvement in system loadability. However, Fig. 11(a)
still presents a Q limit instability at X = 0.132 = 170.5MVA, due
to a maximum reactive limit in the hydro generating statio11 of
Agoyan.
4 CONCLUSIONS
The paper has thoroughly discussed the limitations and typi-
cal uses of saddle-node bifurcations. The relation between these
particular bifurcations and voltage collapse bas also been analyzed
for several system models. Particularly, the effect of differeut load
models in saddle-node bifurcations has been addressed in a variety
of test systems, including a real 115 bus system.
The results from bifurcation studies presented in the paper cam
not be dismissed based solely on the idea that the load and system
model are not realistic. Observe that similar arguments can be
used to ignore the output of Power Flow studies; research and ex-
perience, however, have demonstrated the validity and limitatious
of these types of analyses. This paper answers the latter two ques-
tions pertaining saddle-node bifurcation studies, and deinoustrates
that these analyses and the related methods are thoroughly valid,
as the results for a real system show when compared to the ac-
tual operational knowledge of it. Nevertheless, as with any other
system analysis technique, the user must discern the usefulness of
these results based on the methods and models used.
REFERENCES
[l] L. H. Fink, ed., Proceedings: Bulk PoWeT System voltage
Phenomena-Voltage Stability and Security, EPRI EL-6183,
January 1989.
[2] L. H. Fink, ed., Proceedings: Bulk Power System Voltage
Phenomena-Voltage Stability and Security, ECC/NSF Work-
shop, Fairfax, VA, Ecc. Inc., August 1991.
[3] Modelling of Voltage Collapse Including Dynamic Phenom-
ena, CIGRE Report, Task Force 38-02-10, Draft 3, June 1992.
[4] M. M. Begovic and A. G. Phadke, Dynamic Simulation of
Voltage Collapse, ZEEE Trans. Power Systems, Vol. 5, No.
1, February 1990, pp. 198-203.
[5] H. G. Kwatny, A. K. Pasrija, L. Y. Baliar, Static Bifurcations
in Electric Power Networks: Loss of Steady-State Stability and
Voltage Collapse, ZEEE Trans. Circuits and Syst., Vol. 33,
No. 10, October 1986, pp. 981-991.
[GI I. Dobson and H. D. Chiang, Towards a Theory of Voltage
Collapse in Electric Power Systems, Sysiems U Conirol Let-
ters, Vol. 13, 1989, pp. 253-262.
518
[7] C. A. C d e a r e s , F. L. Alvarado, C. L. DeMarco, 1. Dobson, I. Dobson, Computing and Optimum Direction in Control
W. F. Long, Point of Collapse Methods Applied to AC/DC Space to Avoid Saddle Node Bifurcation and Voltage Collpase
Power Systems, IEEE Trans. Power Systems, Vol. 7, No. 2, in Electric Power Systems, IEEE Trans. Automatic Control,
May 1992, pp. 673-683. Vol. 37, No. 10, October 1992, pp. 1616-1620.
[SI V. Ajjarapu and B. Lee, Bifurcation Theory and its Appli- K. Walve, Modelling of Power System Components at Severe
cation to Noillinear Dynanical Phenomena in an Electrical Disturbances, Proc. Int. Conf. Large High Voltage Efechic
Power System, IEEE Trnas. Power Systems, Vol. 7, Febru- Syst.-CIGRE, August 1986, paper 38-18.
ary 1992, pp, 424-431. K. Jimma, A. Tomac, K. Vu, C.-C. Liu, A Study of Dynamic
B. Gao, G. K. Morrison, P. Kundur, Voltage Stability Eval- Load Models for Voltage Collapse Analysis, in [2].
uation Using Modal Analysis, IEEE Trans. Power Systems, S . S. Sastry, The Effects of Sinall Noise on Implicitly De-
Vol. 7, No. 4, November 1992, pp. 1529-1542. lined Nonlinear Dynanucal Systems, IEEE TTans. Circuits
E. H. Abed and P. P. Varaiya, Nonlinear Oscillations in Power and Syst., Vol. 30, No. 9, September 1983, pp. 651-663.
Systems, International J o u r n a l of Electric Power U Energy C. L. DeMarco and A. R. Bergen, Application of Singular
Systems, Vol. 6, 1984, pp. 37-43. Perturbation Techniques to Power System Transient Stability
P. W. Sauer, B. C. Lesieutre, M. A. Pai, Dynamic vs. Static Analysis, Proc. ISCAS, May 1984, pp. 597-601.
Aspects of Voltage Problems, in [2], pp. 207-216. V. Venkatasubramanian, H. Schattler, J. Zaborszky, A Tax-
C. C. Liu and K. T. Vu, Types of Voltage Collapse, in [2], onomy of the Dynamics of the Large Power Systems with Em-
pp. 133-141. phasis on Its Voltage Stability, in [2], pp. 9-52.
C. L. DeMarco, A New Method of Constructing Lyapnnov
I. Dobson and L. Lu,Voltage CollapsePrecipitatedby thehn-
Functions for Power Systems, Pror. ISCAS, 1988, pp. 905-
mediate Change in Stability When Generator Reactive Power
908.
Limits are Encountered, IEEE Trans. Circuits and Syst.-l,
Vol. 39, No. 9, September 1992, pp. 762-766. M. A. Pai, Energy Function Analysis f o r Power System Sta-
bility, Kluwer Academic Publishers, 1989.
G. K. Morrison, B. Gao, P. Kundur, Voltage Stability Anal-
ysis Using Static and Dynamic Approadies, IEEE Trans. M. K. Pal, Voltage Stability Coiiditiois Considering Load
Power Systems, Vol. 8,No. 3, August 1993, pp. 1159-1171. Characteristics, IEEE Transactions Power Systems, Vol. 7,
No. 1 , February 1992, pp. 243-249.
C. A. Caiiizares and F. L. Alvarado, Point of Collapse and
Continuation Methods for Large AC/DC Systems, IEEE Load Modeling f o r Power Flow and Transient Stability Com-
Trans. Power Systems, Vol. 8, No. 1 , February 1993, pp. 1-8. puter Studies, Vol. 2 , EPRI Report EL-5003,January 1987.
M. K. Pal, Discussions to [7, 9, 14, 15, 181.
Claudio A. Caaizares (5387, M91) was bornin Mexico, D.F. in
National Interconnected System Load Flow: Peak Demand, 1960. In April 1984, he received the Electrical Engineer diploma
INECEL, December 1991, in Spanish. froin the Escuela PolitCcnica Nacional (EPN), Quito-Ecuador,
J. Deuse and M. Stubbe, Dynamic Siinnlationof Voltage Col- where he was a Professor for 9 years, and the MS (1988) and
lapses, IEEE Trans. Power Systems, Vol. 8, No. 3, August PhD (1991) degrees in Electrical Engineeringfrom the University of
1993, pp. 894-904. Wisconsin-Madison. Dr. CGzares is currently an Assistant Pro-
fessor at the University of Waterloo, a t the Electrical and Computer
J. Guckenheinier and P. Holmes, Nonlinear Oscillations, Dy- Engineering Department, and Ius research activities are mostly con-
namical Systems, and Bifurcations of Vector Fields, Springer- centrated in the analysis of stability issues in ac/dc systems.
Verlag, New York, 1986.
[20] R. Seydel, From Equilibrium to Chaos-Practical Bifurcation
and Stability Analysis, Elsevier Science Publishers, Nortli-
Holland, 1988.
[21] P. A. Laf, T. Smed, G . Anderson, D.J. Hill, Fast Calculation
of a Voltage Stability Index, IEEE Trans. Power Systems,
Vol. 7, No. 1, February 1992, pp. 54-64.
C. A. Caiizares, W. F. Long, F. L. Alvarado, C. L. DeMarco,
Techniquesfor Detecting Proximity to Voltage Collapse in
AClDC Systems, Proc. 111 SEPOPE, Belo Horizonte, Brazil,
May 1992, paper IP-18.
T. J. Overbye anf C. L. DeMarco, Voltage Security Enhance-
ment Using Energy Based Sensitivities, IEEE Trans. Power
Systems, Vol. 6, No. 3, August 1991, pp. 1196-1202.
T. J. Overbye, 1. Dobson, C. L. DeMarco, Q-V Curve
Interpretations of Energy Measures for Voltage Security,
IEEE/PES Winter Meeting, Columbus, OH, January 1993,
paper WM 184-2-PWRS.
C. L. DeMarco and C. A. CaFFzares, A Vector Energy Func-
tion Approach for Security Analysis of AC/DC Systems,
IEEE Trans. Power Systems, Vol. 7, No. 3, August 1992, pp.
1001-1011.
C. A. CGzares, Saddle-Node Bifurcations in Power Sys-
tems, Proc. JIEE, Quito, Ecuador, July 1993, pp. 222-229.
I. Dobson, Observations on the Geometry of Saddle Node Bi-
furcations and Voltage Collapse in Electrical Power Systems,
IEEE Trans. Circuits and Syst.-I, Vol. 39, No. 3, March 1992,
pp. 240-243.

Discussion
Peter W. Sauer (University of Illinois at Urbana-
Champaign, Urbana, IL): This paper presents
results for a system which is shown as a single
machine, single line and single load. In order to
understand and appreciate the results, it is
necessary to clearly understand the model. Figure
1 of the paper shows the generator terminal voltage
to have magnitude 1.0 with angle delta. The
magnitude of 1.0 is consistent with the system
dynamic model description of section 2 which
states that the AVR maintains this constant at 1.0.
In order to achieve this, the generator "internal"
voltage magnitude must vary instantaneously to
maintain constant terminal voltage. The angle
associated with the internal voltage is normally the
dynamic state which is related to the integral of the
generator shaft speed. The section 2 description
and figure 1 indicate that the transmission system
is modeled by a reactance X which is used in the
dynamic model. The paper does not mention the
internal reactance of the machine. Since the load
voltage angle is shown to be zero, the dynamic
angle delta must be some type of difference angle
which makes the load voltage angle zero at all
times. Would the author please provide additional
information on the relationship between figure 1
and the dynamic model of equation (I)?
Specifically, where is the internal generator
reactance, and how is the generator terminal
voltage angle related to the generator internal angle
which is normally related to the speed integral?
That is, how did the generator terminal voltage
angle become a dynamic state, and how is the load
voltage angle maintained at zero?
Section 2.1 describes a load model as an infinite
bus which normally means constant voltage
magnitude and angle (this would be a valid way to
maintain the load voltage angle at zero). With this
constraint, the real power which is drawn by this
infinite bus normally cannot be constant during
dynamic analysis. The shaft torque can be made
constant by proper turbinelgovernor assumptions,
but the infinite bus power will change with time
during any transient. Thus it does not seem
appropriate to say that "the first load model is an
infinite bus, which is drawing active power P". It
will do so at an initial equilibrium point, but not
519
after any disturbance. Would the author please
clarify this power P?
Section 2.2 introduces a PQ dynamic load model
stated to be "constant active and reactive power".
If the load is constant real power, then the output
of the generator must be. constant real power since
the line is lossless. If the classical dynamic model
is used with constant input power then there will
be no shaft dynamics. Could the author please
explain this load model further by giving a
complete mathematical description of the load in
terms of reallreactive power and voltage? From
equation (4), there appears to be an error in the 3,3
entry of equation (5) (i.e. cos rather than sin).
Was this error present in the results shown?
Section 2.3 is titled RL loads. The dynamic model
of equation (7) shows a voltage dependent shaft
input torque. What assumption has been made
about the turbine/governor that has led to this
model? If this voltage dependent power is meant
to be the load, then I do not see how it enters the
generator dynamic model in this way.
A clarification of these modeling issues would
greatly enhance the papers contribution.
Manuscript received August 4, 1994.
S.C. SRIVASTAVA, K.N. SRIVASTAVA, S.N. SINGH
and S.K. JOSH1 (Department of Electrical
Engineering, Indian Institute of Technology,
Kanpur, INDIA): The discussers would like to
commend the author for his insightful
presentation on bifurcation, voltage
stability and effect of load modeling in
power system networks. Application of
bifurcation theory in power system has added
a new dimension to the study of power system
stability issues. The author's paper assumes
greater significance in view of the fact
that many a utility is facing the possible
threat of voltage collapse due to lack of
generation and transmission facility coupled
with environmental and economic pressure.
However, a few clarifications/queries as
follows are in order and the discussers
would appreciate author's clarifications on
them.
1. Referring to eq (1) of the paper, it is
found that the active power load 'P' being
drawn by the infinite bus has been taken
same as the mechanical power input to the
synchronous generator. Under what
assumptions this will be valid?
520
2. In section 3, the author has made a
remark that under certain operating
conditions like remote voltage control,
which give the system equations a special
symmetry, transcritical bifurcation can also
be encountered. While a transcritical
bifurcation is not generic, it will be of
interest if author could elucidate it with
the help of an example.
3. How will the load model of eq (11)
affect the study if all the time constants
pertaining to a few load buses are equal or
SlfP/~lf=
P T i p /TIq ? generator terminal voltage is admissible only when the load
4. In section 3, the author further made a
statement that for PQ load models and
typically small resistive losses in the
transmission system, Hopf bifurcations are
unlikely to occur. However, in ref [ A ] , it
has been shown that even with generators'
being modeled by their swing equations and
transmission system having small resistance
the jacobian matrix becomes highly
asymmetric. This leads to an increased
probability of large number of complex
eigenvlues and, thus, Hopf bifurcation.
[AI C Rajagopalan, P.W. Sauer and M.A. Pai,
'Analysis of voltage control system
exhibiting Hopf bifurcation', Proc. of
28th IEEE Conf. on Decision and Control,
Tampa, FL, USA, Dec 19989, pp. 130-136.
Manuscript received August 22. 1994.
M. K. Pal (Consultant, Edison, NJ): In the introduction it is
stated that the main purpose of the paper is to clarify the
relationship between bifurcations and power system stability,
particularly voltage stability. A careful reading of the paper
suggests just the opposite. We support this by reviewing some of
the analyses presented in the paper.
In the system chosen for bifurcation analysis (Fig. l), it is
stated that the AVR is modelled by keeping the generator terminal
voltage constant. We note that this is not an admissible model for
studying the generator-infinite bus system - the first example of the
paper. (It is not clear why the infinite bus has been classified as
load; no known load behaves like an infinite bus.) This particular
example illustrates a purely angle stability problem. The
bifurcation (stability limit) occurs when the angle between the
generator terminal bus and the infinite bus is 90".This is well
established. However, due to the invalid system model (constant
generator terminal voltage), the stability limit thus obtained is well
beyond the stability limit in a real power system.
In the second example using the P,Q dynamic load model, we
comment that the load model given in equation (4) is realistic in
the sense that any first order dynamic load model derived from
physical laws can be reduced to this form. A load model in this
form may also be derived from field measurements. The model
uses time derivative of a variable which is not a true state variable,
and therefore can have discontinuities. Computational difficulties
may be encountered in general simulation studies unless some kind
of transformation is applied. For small signal analyses, as in the
present example, this model would be as good as any other first-
order load model.
Our concern here is, however, the author's interpretation of the
effect of the magnitude of the time constant. First, we note that
since there is only one generator and no reference bus (e.g. an
infinite bus), the question of angle instability does not arise, and
therefore the swing equation is redundant. However, with a load
time constant as small as lms (used in the author's example), it is
imperative that generator flux dynamics along with AVR
representation be included in order for the results to be valid. The
authors may wish to verify the truth of the above statement by
repeating the analysis using a detailed generator model. Constant
dynamics is slow compared with AVR response. The main
justification for not considering angle and other generator internal
dynamics in voltage stability studies lies in the overall slow
dynamics of the composite load, i.e., just the opposite of what the
authors have stated.
For the system and load models used in the second example,
stability limit occurs at the point of maximum power transfer. The
bifurcation point corresponds to the singular point of the power
flow Jacobian. As noted above, the system model is not valid for
loads with small time constants. Also, the stability conditions
would be completely different if the load had a different
characteristic. These have been shown in a rigorous way, for the
first time, in Reference 36 of the paper, and again, in somewhat
more detail, in [A]. In most of the literature on voltage stability,
voltage stability limit has been (arbitrarily) associated with the
singularity of the power flow Jacobian. The singular point of the
power flow Jacobian was identified as the bifurcation point, even
when no dynamics of the system components responsible for
voltage instability (i.e., the loads) were considered. The myth was
clarified in [B]. Although a simple system was used for the
purpose of illustration, the analysis readily extends to larger
systems.
As regards the two examples using RL loads, we should point
out that static loads, unless rendered dynamic by some means, e.g.
LTCs, cannot cause voltage instability. The third equation in (7)
therefore has no significance, and its inclusion only helps confuse
the issues.
In the section on applications to the SNI Ecuadorian system,
since the generators were modelled by constant terminal voltage,
the mathematical steps shown in equations @)-(lo)are irrelevant,
as pointed out earlier. The author would have obtained identical
results from the power flow Jacobian after modifying it to account
for the voltage dependent parts of the load.
The author mentions Hopf bifurcation in several places in the
paper. We would like to point out that, in the context of voltage
stability, Hopf bifurcation will only be observed when an
unrealistic load model, e.g., a static constant power load model is
used in a dynamic system formulation. All instances of Hopf
bifurcation reported in the literature on voltage stability can be
traced to unrealistic load modelling. We hope this should save the
author future work in the pursuit of something nonexistent.
In conclusion, the points the author set out to make in this
paper were already made in a more rigorous and straightforward
manner. This paper only adds to the confusion that exists in the
literature.
M.K. Pal, "Voltage Stability: Analysis Needs, Modelling
Requirement and Modelling Adequacy," IEE Proc. C,
Vo1.140, No.4, pp.279-286, July 1993.
M.K. Pal, Discussion of "An Investigation of Voltage
Instability Problem," by N, Yorino, et al. IEEE Trans. on
Power Systems, Vo1.7, No.2, pp.600-611, May 1992.
Manuscript received September 2, 1994.
~~

Claudio A. Cafiizares (University of Waterloo): The author
would like to thank the discussers, particularly Profs. Srivastava
et al and Prof. Sauer for their interest and constructive discus-
sion which should certainly help enhance the paper. Comments
to each one of the discussers remarks and questions follow:
Prof. S. C. Srivastava et al:
The equations used in the paper to crudely model the gener-
ator is a slight modification of the typical second order or swing
equation representation of the mechanical dynamics of the syn-
chronous machine, as described in standard power systems text-
books, e.g., [A]. This model assumes that the mechanical torque
remains constant, where the generated power is the same as the
network demand in steady-state. Local bifurcation analyses as-
sume that a system parameter changes very slowly, and that
the system responds to these changes by eventually reaching a
steady state, until the system equilibria become non-hyperbolic
at the bifurcation point, i.e., the eigenvalues reach the imagi-
nary axis of the complex plane. In the textbook examples shown
in the paper, this parameter is aswmed to be the power system
demand P, i.e., the generator outbut in this lossless system. In
other words, the generator shaft torque is the changing param-
eter at synchronous speeds in this case; changes to this torque
produce small oscillations of the generator angle 6 according
to equations (1)until and equilibrium point 6, is reached, assum-
ing that P is changing gradually. Since the analyses presented
in the paper concentrate only in the system equilibria, which
is the standard way of locating and studying local bifurcation
problems, one needs not to worry about the oscillations of 6 and
study only the characteristics of the equilibrium points.
With regard to transcritical bifurcations in power systems,
the author has published a paper illustrating this problem in
an ac/dc system where remote voltage control is shown to be
the cause of this particular bifurcation. Since this topic is be-
yond the scope of the present paper, and due to limitations on
the length of this closure, the author would kindly refer the dis-
cussers to reference [B] for a detail analysis of this problem.
The load model represented by equation (11) does not present
any difficulties during simulation for any values of the time con-
stants TI,?, TI., , TI,^, and TI, . The discussers concern seems to
be associated to the following matrix defined on the paper:
[ :; ::; ]
which would be singular for the given condition T I ~ ~ = / T
T ~ . ~ / T ! , ~Since
. this is used only in a mathematical represen-
tation of the load model that is useful when studding transver-
sality conditions at bifurcations of power system models, as
demonstrated in references [26] and [C], one should not be con-
cerned during simulations with particular values of these con-
stants when using robust numerical integration techniques.
Finally, one has to be careful when studying Hopf bifurca-
tions and avoid associating these to the degree of asymmetry of
the Jacobian matrix only. Notice that although the Jacobian for
power systems modeled using swing equations is highly asym-
metric, even for zero transfer resistance as shown in Jacobian
equations (2) and (5) in the paper, it has been formally demon-
strated in [lo], [D], and other articles that such lossless systems
cannot have Hopf bifurcations, regardless of the structure of the
Jacobian. In [lo], Abed and Varaiyashow that Hopf bifurcations
for these system models occur only when transfer resistances are
521
included. However, since these resistances are typically small,
one generically expects to encounter saddle-node bifurcations
rather than Hopfs, although technically one cannot rule out the
latter.
Prof. P. W.Sauer:
It is important to start this discussion by repeating some of
the disclaimers made throughout the paper regarding the mod-
els used. The intention here was not to duel on modeling issues,
since that has been discussed at length in many other publi-
cations before. Furthermore, the three rather simple systems
illustrated in section 2 of the paper are meant to be taken only
as teztbook examples of bifurcation analysis, since very crude
system models are used. The author only wanted to demon-
strate the effect of different load types in the same simple gener-
ator model, so that the points brought up in the paper are not
obscured by model complexities. Nevertheless, Prof. Sauers
comments on basic modeling issues are very valid, and a clarifi-
cation of certain assumptions made by the author ought to be
addressed in order to make more sense of the examples shown.
On the issue of modeling the AVR and the internal genera-
tor impedances in the second order generator model chosen, the
implicit assumption in the paper is obviously that of ignoring
the internal generator impedance, not a realistic one unless one
makes some additional assumptions regarding generator loading
and rating, but in such a coarse system model this certainly does
not represent a significant issue. For the reader uncomfortable
with this assumption, one can go back to the idea of lumping
the generator reactance with the transmission line impedance,
and talk about a constant internal generator voltage, which is
certainly the classical swing equation model, and view the paper
examples and results in that context. Notice that for all three
system models represented by Fig. 1, 6 corresponds to the angle
difference between the transmission line terminal voltages, which
is the angle that defines the power flowing through the line, and
since these models correspond to quasi-steady-state phasor rep-
resentations of the original equations, one can assume, for the
load models used in the examples and without loss of general-
ity, that the load voltage is the system reference, i.e., the load
voltage angle is 0.
Regarding the meaning of P in the generator-infinite bus
model of section 2.1, it represents, as Prof. Sauer mentions in
his discussion, the steady state power demand of the infinite bus.
Hence, P corresponds to the generator mechanical power in the
lossless system under study. This power is the system parameter
~ ~ ~
changed for the bifurcation studies illustrated in the paper. The
infinite bus demand V / X sin(6) would certainly change accord-
ing to the dynamic changes in 6. Notice that for one-parameter
local bifurcation analysis, one is mostly worried about tracing
the system equilibria, since gradual changes of P shift system
equilibria until the bifurcation point is encountered.
For the PQ dynamic model, the author must recognize that
this name is misleading and it is not descriptive of the load
model used in the example of section 2.2, and Prof. Sauers
comments offer an opportunity to clarify this point. The idea in
this case is to add one degree of complexity to the load model
used in the previous section, i.e., allow the voltage V in equation
(1) to dynamically change. Hence, the load model in this case
is assumed to be drawing active and reactive powers
Pr. = P + D L ~ L
QL = k P + T V
522

with large damping DL,i.e., the load voltage angle 6~ is as-
sumed to remain constant at 0. Evidently, the active and re-
active powers are only constant in steady state. Although this
load can be hardly justified and labeled in practical terms, the
resulting simple system presents certain futures that make it
very appealing to demonstrate certain issues in the studies of
dynamical systems, such as Lyapunov functions, bifurcations,
and differential-algebraic systems.
Before commenting on the RL load example in section 2.3,
the author would like to correct some typographical errors in
equations (5), (6), and (7) of the paper. These corrections will
help explain the voltage dependency of the shaft torque in (7).
Thus, Jacobian equation (5) should read:
in this paper will not occur in real systems, and therefore these
should be of no concern to the power systems community. The
discusser writes several paragraphs criticizing the models used in
the paper, missing the point that these are very simple systems
meant to be used in a teaching environment, and duplicated
with little effort by a y o n e interested in the subject, with any
values for the models time constants. He suggests that bifurca-
tion analysis is a waste of time, since the particular modeling
assumptions used in this paper and others before seem to be the
only reason for the appearance of the phenomena. Obviously
the discusser is not familiar with the vast literature in bifur-
cation theory of nonlinear systems, not only in power systems,
but many other engineering and scientific areas. He would be
surprise on how generalized this phenomenon is, and there is

1
0 1 0

[
J = -(Vo/MX)cos(60) -DIM
-(VO/rX)sin(60) 0
-(I/MX)sin(60)
-(1/7X)[ZVo - cos(ao)]
(5) no reason for the power system to be the exception. Although
this paper was not written to demonstrate the existence of bifur-
cations in real power systems (there are many other articles
the Lyapunov function defined in equation (6) should be: which concentrate on this particular issue), the results shown in
1 1 the paper for the very real SNI Ecuadorian system have been
V(6,w,V,P) = ZMwZ + P(& - 6) + -[VOCOS(&)- V C O S ( ~ ) ] certainly corroborated in practice, where there is no need for
X
1 any modeling assumptions.
-,,(Vi - Vz) - kP[ln(Vo)- h ( V ) ] (6) The discusser mentions that Hopf bifurcations will only be
observed when an unrealistic load model is used, as he sates on
and the second equation in (7) should read:
his discussion. Well, if the discusser would read some of the ref-
.
w = IV~~G
1
- -sln(6)
v .
X
- I
DW
erences cited in this paper, he would realize that his statement is
incorrect. The author would like to invite this discusser to read
papers [lo], [B], and [D], where it is shown that the phenomenon
These typos are not part of the simulations and results shown has very little to do with load modeling.
ir the paper. Finally, this author would like to encourage the discusser to
Now, coming back to the assumptions in the case of the RL submit his results for publication, so that the rest of the research
load, once more the idea is to make an additional change to the community can be enlighten on the unique views that he has in
system equations, without significantly affecting the final results this particular area of system stability, because it seems that the
to allow for direct comparisons between the different sample myth which the discusser says to have clarified in one of his
systems. In equation (7b) above, the shaft torque is a quadratic many discussions is still in several researchers minds.
function of the steady state voltage, so that the torque remains
constant for system variations other than those associated to the [A] A. R. Bergen, Power Systems Analysis. Prentice-Hall, New
slow varying parameter G; this way the generator steady state Jersey, 1986.
output is adjusted to the system demand changes, as it was
done in all previous cases, since one is mostly interested in the [B] C. A. Cafiizares and S. Hranilovic, Transcritical and Hopf
characteristics of the changing system equilibria for bifurcation bifurcations in ac/dc systems, in [E].
studies.
[C] C. A. Cafizares, Conditions for saddle-node bifurcations in
ac/dc power systems, accepted for publication in the Int.
e Mr. M. K. Pal:
J. of Electric Power d Energy Systems, July 1994.
This discusser comments are surprisingly blunt, which seems
to be his style as one can see from previous discussions to many [D] H. D. Chiang and F. F. Wu, Stability of nonlinear sys-
papers in bifurcation theory applied to power systems, e.g., [16]. tems described by a second-order vector differential equa-
From his remarks one gathers that either he knows something tion, IEEE Trans. Circuits and Systems, vol. 35, no. 6, June
many researchers, including this author, do not know about bi- 1988, pp. 703-711.
furcation related problems in nonlinear dynamical systems, as [E] L. H. Fink, editor, Proc. Bulk Power System Voltage Phe-
one gathers from his final comment about the confusion that nomena Ill-Voltage Stability and Security, ECC Inc., Fair-
exists in the literature, or he is just talking about different fax, VA, August 1994.
phenomena.
Reading his comments, one concludes that the discusser is Manuscript received October 24, 1994
certainly convinced that the system stability problems described