Itage

CHIN-WOO TAN, MATTHEW VARGHESE, PRAVIN VARAIYA, FELLOW, IEEE, AND

FELIX F. WU, FELLOW, IEEE

Invited Paper

A model of a power system with load dynamics is studied by implies a change in the number of equilibria. A dynamic
investigating qualitative changes in its behavior as the reactive bifurcation can be a Hopf, cyclic fold, or period doubling
power demand at a load bus is increased. In addition to the
bifurcation. A Hopf bifurcation causes the emergence or
saddle node bifurcation often associated with voltage collapse, the
system exhibits sub- and supercritical Hopf bifurcations, cyclic disappearance of a periodic solution from an equilibrium
fold bifurcation, and period doubling bifurcation. Cascades of point [25].A cyclic fold bifurcation occurs when a stable
period doubling bifurcations terminate in chaotic invariant sets. limit cycle and an unstable limit cycle collide and disappear.
The presence of these new bifurcations motivates a reexamination A period doubling bifurcation refers to the emergence of a
of the saddle-node bifurcation as the boundaly of the feasible set
of power injections. periodic orbit near another periodic orbit. Static and Hopf
bifurcations are local phenomena. Chaos emerges from a
global bifurcation. This means there is a qualitative change
I. INTRODUCTION
in system trajectories not restricted to a small neighborhood
Our objective is to study bifurcations in the behavior
of an equilibrium point.
of a power system as a system parameter (e.g., load) is
The classical (swing equation) model of a power system
quasistatically varied. By definition, bifurcations occur at
exhibits both local and global bifurcations. Static bifurca-
those parameter values where there is a change in the
tions are always present because the power flow equations
system's qualitative properties. Power systems are operated
at a stable operating point and, under normal conditions, the have multiple solutions. In power systems where voltages
operating point varies smoothly with changes in the system are determined by reactive power flows, these bifurcations
i
parameter, so the variation can be tracked by local, linear can shrink the region of stability. An electrical subsystem,
analysis. However, the linear analysis breaks down when such as the exciter, can interact with machine dynamics
bifurcations occur, and one must resort to more global, to produce oscillations via a subcritical Hopf bifurcation.
nonlinear analysis. Bifurcations also profoundly affect sys- This, too, shrinks the region of stability. As will be seen,
tem behavior even under normal conditions, because they the occurrence of chaos can similarly affect the behavior
constrain the region of attraction of a stable operating around an operating point.
point. Thus bifurcations influence both the range of stable The paper is organized as follows. In Section 11, we
operating points as well as the transient stability properties review bifurcations that are relevant to our study. In Section
of those operating points. This is what makes the study of HI, we discuss chaos and related complicated dynamic be-
bifurcations useful from an engineering viewpoint. havior. Where appropriate, we draw examples from power
We study static bifurcations, dynamic bifurcations and, systems. In Section IV, simulation results of a classical
especially, chaos' in power systems. A static bifurcation power system model augmented by a dynamic load are
Manuscript received June 18, 1994; revised August 9, 1995. This work
presented. Others have studied the same model [3], [ 5 ] ,[6].
was supported by the Electric Power Research Institute. The global asymptotic picture of even this simple system is
C.-W. Tan is with GeoStar Technology Inc., Santa Clara, CA 95051 incomplete, although local bifurcations near an equilibrium
USA.
M. Varghese, P. Varaiya, and F. F. Wu are with the Electrical En- or a periodic orbits are well understood. Some concluding
gineering and Computer Sciences Department, University of California, remarks are collected in Section V.
Berkeley, CA 94720 USA. The classical models with positive damping and loss-
IEEE Log Number 9414947.
Chaos is a concept describing bounded, unstable behavior. More less transmission lines have relatively simple behavior.
detailed definitions follow in Section 111. Because they possess energy functions that can also serve as

0018-9219/95$04,00 0 1995 IEEE

1484 PROCEEDINGS OF THE IEEE, VOL 83, NO. 11, NOVEMBER 1995

Authorized licensed use limited to: PONDICHERRY ENGG COLLEGE. Downloaded on August 9, 2009 at 01:17 from IEEE Xplore. Restrictions apply.
 Lyapunov functions, they have no limit cycles2Trajectories arbitrarily long time intervals. Thus the occurrence of
either converge to equilibrium points or are unbounded. The these behaviors must be inferred indirectly. Often they are
existence of unbounded trajectories depends on whether manifested as experimental or numerical anomalies. For
machine angles are measured [0,27r) or (-00,$00). The example, some bifurcations can produce numerically and
same conclusion may be drawn for more general struc- experimentally unobservable quantities such as unstable
ture preserving systems (with damping), because they too limit cycles and fractals like the Cantor set.6 Even if one
possess energy functions. General algorithms for the con- were to accept these models as valid, it is usually not
struction of these functions are available in [19], [21], and possible to prove the occurrence of chaos, and one has
[31]. For the classical and structure preserving models, to resort as we do to interpretations of simulation results
saddle-node bifurcation (SNB)3 is the only possible local that lend support to the claim of occurrence of chaos. Thus
bifurcation. Analytical characterizations of the exact bound- much theoretical and experimental work remains to be done
ary of the stability region of a stable equilibrium point are before one can accept that certain complicated behaviors are
available in [12], [35]. For some systems, this boundary characteristic of power systems.
has a "tangled" shape [15]. Details on how this affects the
construction of Lyapunov functions are given in [33]. 11. REVIEWOF BIFURCATION
TERMINOLOGY
Global bifurcations can be direct consequences of local
The mathematical literature on bifurcation theory is im-
bifurcations. Unstable limit cycles contribute to this tran-
mensely rich and diverse. A large number of definitions and
sition by forming homoclinic orbits with other limit sets.
concepts have been developed to deal with the qualitative
A list of such transitions is given below. More detailed
behavior of the orbit structure of parametrized families of
definitions can be found in Sections I1 and 111.
autonomous dynamical systems. Static bifurcations were
1) Cyclic Fold Bifurcation: This is an important bifur- first related to voltage collapse in [23] and nonlinear
cation for the power system example considered in oscillations in power systems were identified with Hopf
Section IV. An unstable limit cycle and a stable bifurcations in [ 11, [4], and [27]. A companion paper in this
limit cycle are created. The shrinkage of the stability volume also analyzes local bifurcations. An EPRI report
domain of the stable operating point commences provides a review of the literature [32]. In this section, we
with the appearance of a pair of unstable-stable limit present the basic definitions of cyclic fold, period doubling
cycles. and the formation of invariant tori. These phenomena can be
2 ) Period Doubling: The study in [4] for a pair of viewed as local bifurcations on a suitably chosen Poincar6
generators connected to a lossy line shows that in the section.
case of a supercritical Hopf bifurcation, the stability
of the periodic orbit is short-lived. When the stability
of the periodic orbit is lost, another periodic orbit A. Cyclic Fold, Period Doubling, and Invariant Tori
of twice its period becomes stable. If this pattern A cyclic fold bifurcation occurs when a stable limit
continues, it is known as period doubling. cycle collides with an unstable one, and both disappear as
3) Homoclinic Chaos: In the classical power system the system parameter is further varied. A period doubling
model with no damping, it is possible to obtain bifurcation refers to the emergence of a stable periodic
complicated trajectories provided that the machine orbit near another periodic orbit rl,where the period
angles are measured in [0,27r). Suppose a saddle of r2 is approximately twice of that of rl. The orbit rl
connection4 in the machine dynamics of a single is stable before the bifurcation, but loses its stability after
swing equation in [0,27r) x 8 is perturbed by a the bifurcation. The bifurcation of a periodic orbit for a
periodic solution from another part of the power differential equation can be studied as a local bifurcations
system. Suppose the periodic forcing perturbation of a fixed point7 of the Poincar6 map P. We first review
term breaks up the saddle connection so as to produce the idea of a Poincar6 map for a periodic orbit.
a homoclinic tangle.5 This global bifurcation results Consider a differential equation k = f ( x ) , IC E 92".
in chaos. Denote its solution starting at II: by 4t(x)= q5(t,x). The
As mentioned, certain types of bifurcations cannot occur map 4 : ( t , x ) -+ $ ( t , x ) is called the flow of the vector
in the classical models. There is a question about validation field f ( z ) . Let be a periodic orbit of period T of the
of models which do exhibit those behaviors. In the last flow 4. We first take a local cross section S c X",of
instance, validation should be based on observed data. How- dimensional n - 1, such that 1) the flow 4 is everywhere
ever, observed data are collected over a finite time interval, transverse' to S , and 2) S intersects at a unique point
whereas phenomena like chaos characterize behaviors over 6A fractal is a set with a fractional dimension. The dimension of a set
is defined to be D if the following covering of the set is implemented.
'If the machine angles are measured in [0,2a),there may be limit Suppose one were to cover the set with hypercubes of side t. Then,
cycles.
SNB occurs when a stable-unstable pax of equilibrium points collide
and disappear.
N(E) -
N ( E ) the
, minimum number of such hypercubes needed, would scale as
when t + 0. For a line segment D is 1, for a plane D is
2. For a Cantor set D is an irrational number between 1 and 2.
4 A saddle connection occurs when trajectories connect pairs of saddle 7 A point zo is a fixed point of a map z H f(z)if f(zo)= zo.
equilibrium points. 8This means the inner product < f ( z ) , n ( z ) ># 0 for all z E S,
A homoclinic tangle is defined in Section 111. where n ( z ) is the normal to S at z.

TAN et al.: BIFURCATION, CHAOS, AND VOLTAGE COLLAPSE IN POWER SYSTEMS 1485

Authorized licensed use limited to: PONDICHERRY ENGG COLLEGE. Downloaded on August 9, 2009 at 01:17 from IEEE Xplore. Restrictions apply.
 p . Let U c S be a neighborhood of p . For q E U t h e j h t of two hyperbolic fixed points, one stable and the other
return or Poincare' map P : U 4 S is defined by unstable, as p approaches po. The stable (respectively,
unstable) fixed point lies on a stable (respectively, unstable)
P ( 4 ) := 4 ' ( 4 ) ( d = 4 ( T ( 4 ) , 4 ) periodic orbit. Hence a cyclic fold bifurcation of periodic
where ~ ( qis) the time it takes for the orbit based at q to first orbits corresponds to a (local) saddle-node bifurcation of
return to S.Thus ~ ( p=) T and ~ ( q4 ) T as 4 -+ p . Also
the fixed point solution for the PoincarC map.
p is a fixed point of P.In local coordinates, P is a map Cyclic fold bifurcation occurs for the power system
from U c RnP1to PP1, so D P ( p ) has n - 1 eigenvalues, example in Section IV.
called the Floquet multipliers (or characteristic multipliers) Case 2: [Period doubling] X ~ ( p 0 = ) -1 Suppose po is
associated with the periodic orbit I?. a stable fixed point for Pp0.The implicit function theorem
Theorem 2.1: The periodic orbit I? is orbitally asymptot- gives a smooth fixed point solution p i p ( p ) through
ically stable if all the Floquet multipliers have magnitude (PO,po). However, the solution branch loses its stability
less than one. Moreover, p is a stable fixed point for the and a stable periodic orbit consisting of two periodic points9
PoincarC map if r is stable. 0 of period 2 is present for either p > po or p < po
depending on whether $X1(po)I > 0 or &lXl(po)I <
Consider a one-parameter family of differential equa-
tions li. = f ( z , p ) . Suppose it has a periodic orbit for 0. We say that the fixed point solution has a pitchfork
a parameter value p. Denote the Floquet multipliers by bifurcation at po. Suppose p near po is a parameter after
, XnL-l(p).If the periodic solution is stable, then the bifurcation has occurred. Let p 1 ( p ) , p z ( p ) be the
m 2.1 A 1 ( p ) , . . . , X n P 1 ( p )are all inside the unit two period 2 periodic points for the PoincarC map P,.
circle. The multipliers are functions of the parameter p. Since PP(Pl(PC1)) = P 2 ( P ) and q L ( P z ( P ) ) = P1(PI9 a
As p is varied, some of them may cross the unit circle at new periodic orbit of approximately twice the period of
a critical value po. A multiplier crossing the unit circle is the periodic orbit which contains p ( p ) is created, and it
called a critical multiplier. Several types of bifurcation may intersects the PoincarC section at p 1 ( p ) and p 2 ( p ) . Thus a
occur depending on how a multiplier or a pair of complex period doubling bifurcation of a periodic orbit corresponds
conjugate multipliers crosses the unit circle. There are three to a (local) pitchfork bifurcation of the fixed point solution
typically possible cases: for the Poincark map.
The power system model in Section IV as well as an
1) Only one multiplier crosses the unit circle along the
example involving two generators in [4] both undergo
positive real axis with &(PO) = 1 for some i.
period doubling bifurcation.
2) Only one multiplier crosses the unit circle along the
As an example, consider the logistic map [26] defined by
negative real axis with X,(po) = -1 for some i.
3) Only one pair of complex conjugate multipli- F,(z) := pz(1-2) , 0 5 12: 51 , 15 p 5 4. (1)
ers crosses the unit circle with IX,(po)l = 1,
For 1 5 p 5 4, F, maps [0,1] to [0,1]. Solving the
Im(X,(po)) > 0, and &lX,(po)l # 0 for some i.
equation F,(z) = z gives the fixed points 0 and z * ( p ) =
If p E r,r a periodic orbit for a parameter p, we say that 5. For z * ( p ) E [0,1], we need p 2 1. We find also
p is a hyperbolic fixed point for the (appropriate) PoincarC
that &EF,(O) = p and &F,(z*(p)) = 2 - p. So 0 is
map P, if none of the multipliers XI ( p ) ,
always unstable (since p 2 1) and 12:*(p)is stable only for
on the unit circle. All the above three cases refer to a
1 < p < 3. There is a saddle-node bifurcation at p = 1
loss of orbital stability when p approaches po from the
when z = 0 and 12:*(p)annihilate each other. At po = 3,
appropriate side (Conversely, the system gains stability if
critical multipliers enter the unit circle). Thus there is a
&F,(z*(po)) = -1, so z * ( p ) undergoes a pitchfork
(period doubling) bifurcation and becomes unstable for
bifurcation at po. Exactly like in the case for differential
p 2 3. Moreover, a stable 2-cycle" is born. The two period
equations, the lack of hyperbolicity causes bifurcations for
2 periodic points 21, 5 2 satisfy the equations
a mapping. Without loss of generality, we assume X l ( p 0 )
is the critical multiplier in both case 1 and case 2, and IC2 = F,(IC1) = p21(1- 21)
X ~ ( p o )= %(PO) is the conjugate pair of multipliers in z1 = F,(12:2) = pz2(1- .2)
case 3. Proofs of the following results may be found in [8]
and [34]. resulting in
Case I : [Cyclic fold] Xl(p0) = 1 A fixed point for P,
is a point on a periodic orbit that solves the equation

P,(q)-4=0 > ClE S, Since x1,xz form a 2-cycle for the map F,, each is a fixed
point for the iterated map F i . Also the term pz - 2p - 3
Suppose p o is a fixed point for Pp0.Since X l ( p 0 ) = 1,
9 A penodic point of penod n for a map f is a point zo which satisfies
DqPp0( p o ) - I is not invertible, where I is the (n-1)x (n- f" ( Z O ) = ZO, where f" = f o o f (n times) If n is a penod for z g ,
1) identity matrix. So we cannot apply the implicit function then k n is also a penod for 2 0 , k = 1,2, The smallest of all periods
theorem to get a smooth fixed point solution p ( p ) with po = for zo is called the fundamental or prime period
"A m-cycle for a map f is a set of m distinct points {ZI, , z m }
p ( p 0 ) . Indeed the fixed point solution undergoes a saddle- such that f ( z Z=) z,+1, 1 5 z 5 m - 1, and f ( z m ) = z1 Hence each
node bifurcation at po. This corresponds to a coalescence I, is a penodic point of period m for f

1486 PROCEEDINGS OF THE IEEE, VOL. 83, NO 11, NOVEMBER 1995

Authorized licensed use limited to: PONDICHERRY ENGG COLLEGE. Downloaded on August 9, 2009 at 01:17 from IEEE Xplore. Restrictions apply.
 in (2) is negative for p < 3, which explains why period 2
fixed points do not exist for p < 3. p i = 3.0 p5 = 3.568759. . .
Case3: [Invariant tori] lA1(po)l = IAz(p0)l = 1, pz = 3.449490. . . p6 = 3.569692. . .
Im(AI(p0)) = -Im(Az(po)) > 0 The implicit function p3 = 3.544090. . . p7 = 3.569891.. .
p4 = 3.564407. . . p7 = 3.569934. . .
theorem gives a smooth curve p + p ( p ) through ( P O ,PO).
However, the fixed point solution p ( p ) undergoes a
bifurcation at po with the emergence of a continuous
one-parameter family of circles C, around p ( p ) , for will discuss how chaos disappears in Section 111. Table 1
p E (PO,po +
E ) and small E > 0. The circle C, is lists the first eight period doubling bifurcation values for
invariant" under the Poincark map P,, hence the forward F,. The sequence of bifurcation values {pn} converges
trajectories Uo
, 4: (C,) form an invariant 2-dimensional geometrically to po3 = 3.569945. . ..
torus for the flow of x = f,(z). So if I? is the periodic
orbit being considered before the bifurcation, then after the 111. CHAOS, CRISES, AND HOMOCLINIC TANGLES
bifurcation there are orbits emerging around I? which form
In this section we define chaos and discuss aspects of
a thin torus. We call this a secondary Hopf bijiurcationI2
sudden changes in chaotic attractors which occur as a
for the periodic orbit I?. Similar to the Hopf bifurcation for
system parameter is varied. These changes are classified as
an equilibrium solution in the continuous case, a secondary
a boundary crisis or an interior crisis. In the power system
Hopf bifurcation is supercritical (subcritical) if the invariant
example of Section IV, we only find boundary crises. We
circle is attracting (repelling) under the flow $,: and I?
also discuss homoclinic tangles proposed as models for
becomes unstable (stable).
chaos in power systems in [22] and [30]. Denote a chaotic
Invariant tori have recently been detected for a 39-bus
invariant set by A. The term chaos is associated with the
power system in [l 11.
following properties.
1) Sensitivity to initial conditions: Trajectories starting at
B. Period Doubling as a Route to Chaos
arbitrarily close initial points will eventually diverge,
The classic example of a cascade of period doubling making prediction impossible.
bifurcations leading to chaos is the logistic map F, defined 2) Periodic orbits: A contains periodic orbits of all
in (1). Recall that F, has a fixed point given by x*(p) = periods. It also contains infinitely many nonperiodic
5. Initially the attracting set consists of the single point orbits.
x * ( p ) that bifurcates into two points z1,2(p) at p1 = 3.0, 3) Dense orbit: There is an orbit which is dense in A.
where 4) Cantor set: A is a Cantor set.
The last property also implies that A is a fractal, i.e.,
a subset of the state space of fractional dimen~ion.'~ To
better understand the dynamics after the onset of chaos,
we appeal to a result due to the Russian mathematician
A simple calculation shows that $F;(x~(p)) = Sarkovskii. The result begins with an ordering of the natural
&Fz(z2(p))= -1 whenp = p2 = l+a = 3.44949... . numbers. The Sarkovskii ordering of the natural numbers
So the map F : undergoes a period doubling bifurcation is the following:
at pz. Since both z1(p), zz(p) are fixed points for F:,
they lose their stability and bifurcate into four points
at p2. These four points are stable fixed points for the
iterated map F,: and thus are stable periodic points of
period 4 for F,. For example, at p = 3.5, the steady
state solution (or attracting set) alternates between the four ...........................
values of 0.82694, 0.50088, 0.87500, and 0.38282. As we . . . D 25 D 24 D 23 D 2' D 2~ 1.
increase p further, the number of alternating steady state
values increases with 2%, the interval between successive That is, first list all the odd numbers except 1, followed by
bifurcation values decreases, and the distance between 2 times the odds, 22 times the odds, 23 times the odds, etc.
neighboring periodic points decreases until eventually This exhausts all the natural numbers with the exception of
what looks like a chaotic attracting set appears. This is the powers of 2 which we list last, in decreasing order.
called a period doubling cascade to chaos. The chaotic Theorem 3.1: (Sarkovskii Theorem) Suppose f : !I? + 8
region is interspersed with bands. In the bands only a is continuous and has a periodic point of prime period14 k .
small number of periodic points form the attracting set. We If k D 1 in the Sarkovskii ordering, then f has a periodic
" A set Y C X is invariant under a map f : X + X if f ( y ) E Y for point of prime period 1. 0
all y E Y .
I2In proving the existence of invariant circles C,, we need to assume l 3 See footnote one for the definition of fractional dimension.
the eigenvalue A 1 (PO)is not a k t h root of unity, IC = 1 , 2 , . . . , 5 . That I4A point 20 is a periodic point of prime period IC for a map f if it is
is, eJkX1(fiO) # 1, IC = 1 , 2 , . . . , 5 . A proof of the result can be found a periodic point of period IC for f and IC is the smallest positive integer
in [25]. for which f'((a0) = 20 holds.

TAN et al.: BIFURCATION, CHAOS, AND VOLTAGE COLLAPSE IN POWER SYSTEMS 1487

Authorized licensed use limited to: PONDICHERRY ENGG COLLEGE. Downloaded on August 9, 2009 at 01:17 from IEEE Xplore. Restrictions apply.
 Returning to our logistic map example, we see that for -$ + 42 + c which is ~tab1e.l~ As c is increased to
p l < p < p ~there
, are two period 2 points x~(p),zz(p) 0.75 (corresponding to p = 3.0), ys(c) undergoes a period
given by (3). So by the Sarkovskii theorem there must doubling bifurcation, followed by a cascade of period
be a period 1 point (or a fixed point), which we know doubling leading to chaos.
is x*(p) = 9. ’ ~ p 2 p1.
The point .*(,U) is ~ n s t a b l e for In Section I1 we reviewed bifurcations of equilibrium
Sarkovskii’s theorem implies that as p is increased, the points and periodic orbits. Here we draw comparisons
logistic map Fp must have periodic points of all periods between bifurcation of a fixed point and that of a chaotic
of the form 2” before it can have infinitely many periodic attractor. A simple definition of a crisis [17] is as follows:
points with distinct periods. It also hints at the possibility A crisis is a collision between a chaotic attractor and a
of small windows of stable periodic behavior after the coexisting unstable Jixed point or periodic orbit. Much of
onset of chaos. For ,LL rs 3.83, there is a stable 3-cycle. the classification of a crisis given below is the result
So Sarkovskii’ s theorem implies there are periodic points of extensive numerical experiments on the quadratic map
of all periods. This suggests that after the onset of chaos
for the power system example in Section IV there exist
Boundary crisis: The sudden destruction of the chaotic
small windows of stable periodic orbits that are not easily
attractor and its basin of attraction is symptomatic of
detected by numerical experiments.
a boundary crisis. For parameter values just past the
The bifurcation diagram for the logistic map is inter-
crisis point, the attractor no longer exists. Trajecto-
spersed with bands, and in the bands only a small number
ries starting in the region that was occupied by the
of periodic points form the attracting set. So there is a
destroyed attractor would appear to be chaotic, in a
transition from a chaotic attracting set containing infinitely
fashion similar to what it was before the crisis, but
many periodic points to an attracting set with only a small
only for a finite time after which the trajectory would
number of periodic points. The simulations in Section
leave the region. One is led to speculate what happens
IV reveal the annihilation of the chaotic attractor as the
after the destruction of the chaotic attractor in the
reactive power at the load bus is increased. Next, we discuss
power system model to be considered in Section IV.
“crises” involving chaotic attractors. These phenomena
One possible explanation is that an unstable voltage
characterize when and how a chaotic attractor is destroyed.
trajectory may “wander” in the vacancy created by the
To understand how the chaotic attractor vanishes, we first
destroyed attractor for a long time before escaping to
consider an equivalent form of the logistic map. The logistic
f o o . For the quadratic map G,(y), the unstable fixed
map I?@ (x)= px( 1- x) can be transformed to a quadratic
point yU is on the boundary of the basin of attraction
map G,(y) := c- y2 via the coordinate change y = ,UZ- $ when c = 2. As c is increased beyond 2, the chaotic
and the parameter change c = 5
- $ (or equivalently,
attractor is destroyed and all trajectories diverge to
p = 1+2&). There are good reasons to call G,(y) the -W. A boundary crisis occurs when the unstable limit

logistic map. For example, the Henon map H ( z ,y) defined set, such as a saddle periodic orbit, that collides with
by the chaotic attractor is on the boundary of the basin of
attraction of the attractor.
Interior crisis: This is different from the boundary
crisis because collision occurs within the basin of
attraction. Consider the map G,(y). A bifurcation at
reduces to G,(y) when b 0. Also, for small values of b, c = 1.755 leads to the birth of a stable 3-cycle and an
the dynamics of H ( x ,y) becomes invariant on the manifold unstable 3-cycle. The stable 3-cycle becomes chaotic
{ (x,y) : y = c - x2}, which may explain why the shape via period doubling. Since each individual “branch”
of the chaotic attractor of the Henon map resembles a of the 3-cycle undergoes period doubling, the chaotic
quadratic curve.16 More importantly, the simple form of attractor lies in three pieces. At c M 1.79 the unstable
G,(y) makes the computation of bifurcations and basins of 3-cycle collides with the three-piece attractor. The
attraction of fixed points much easier. three chaotic bands widen to form a single band. An
Since 1 5 p 5 4 for Pp, the parameter change implies interior crisis is signified by a discontinuous widening
_ _ 5 c 5 2. For c < -2.
no fixed point exists
and all trajectories go to -CO. At c = -:
a saddle-
of the attractor.
Fix b at 0.3 and vary c. As c is increased, period doubling
node bifurcation occurs. The bifurcation creates a pair of
of the stable 6-cycle leads to chaos. At c z 1.0807 a bound-
stable and unstable fixed points. These fixed points are
yu(c) = -&j- 6 which is unstable, and ys(c) =
ary crisis and therefore a homoclinic tangency18 occurs

17They correspond to the unstable fixed point z = 0 and stable fixed

”The Sarkovskii theorem does not predict which periodic points are point z*(,u) = 5 for the map F p
stable. I8Let p be a saddle equilibrium point (fixed point) for a differential
l6We will discuss in Section IV-D that the chaotic attractor of the power equation z = f(x) (a map z H f(z)) Suppose its stable manifold
system model is “close” to the constraint manifold obtained by treating W s ( p ) intersects transversely with its unstable manifold W u ( p ) at a
the load variables as very fast system variables. An explanation for that point q, i e , the tangents to the two manifolds at q span the state space
phenomenon may help us to understand why the chaotic attractor of the The point q is a homoclinic point, or q is said to be homoclinic to p The
Henon map is “close” to and resembles the quadratic curve. coincidence of W”( p ) and W”( p ) is known as a homoclinic tangency

1488 PROCEEDINGS OF THE IEEE, VOL. 83, NO 11, NOVEMBER 1995

Authorized licensed use limited to: PONDICHERRY ENGG COLLEGE. Downloaded on August 9, 2009 at 01:17 from IEEE Xplore. Restrictions apply.
 when the unstable 6-cycle on the boundary collides with the
chaotic attractor. One can verify that the unstable 6-cycle ...................................
! E = generator angle :
is on the boundary by observing whether some component
of its unstable manifold converges to the attractor. Unlike
G,(y), H ( z ,y) is a diffeomorphism.”
Consider a chaotic attractor which emerges as a result of (slack bus)

a period doubling cascade. Are period doubling cascades

prerequisites for homoclinic tangles?20From the definition
of a boundary crisis, it is clear that a homoclinic tangency is Fig. 1. A three-node power system model
a precursor to a boundary crisis. It seems plausible to assign
the following sequence to these global bifurcations:21
We use the parameter values in [3]. This gives rise to a
system of four differential equations of the form:
Period doubling + Chaos + Homoclinic tangles
+ Homoclinic tangency --+ Boundary crisis.

Suppose one has an explicit expression for a homoclinic

orbit. Then the method of Melnikov, which uses pertur-
bation analysis about the homoclinic orbit, can be used
to prove the existence of homoclinic tangles. A classical
three node power system where the mass of one generator
is considerably larger than the other generator, thereby
reducing the system to Sitnikov’s three body problem,
is analyzed in [22]. A classical power system, without
generator damping, has a Hamiltonian structure. For such The variables 0 and w are the machine angle (in radians)
a system, it is shown in [30] that a form of chaos called and angular velocity (in radians per second). V and 6 are
Arnold diffusion can occur. Both studies use Melnikov’s the voltage magnitude (in pu) and angle of the load bus. The
method. It is not possible to use Melnikov’s method for the normalized nonslack generator damping coefficient is d,.
model power system in our work, because for the example The terms in square brackets are the load flow equations
in Section IV we do not have an explicit expression for the for a constant PQ load, where P and Q are the real and
homoclinic orbit. reactive power demand at the load bus. The electrical real
power P, and reactive power Q , delivered to the load are
Iv. SIMULATION RESULTSFOR A POWER SYSTEM MODEL
P, = -ylVsin(6) - yZVsin(6 - 8) (9)
A bifurcation analysis of a three-bus power system
model, shown in Fig. 1 and previously proposed in [14] Q , = fyiVcos(6) + ~ ~ V C O- S0) (-~y3V2 (10)
and [ 3 ] , is presented. The bifurcation diagrams show that
the system exhibits both static and dynamic bifurcations, where yi are products of conductances and constant volt-
and cascades of period doubling bifurcations leading to ages behind reactance. The coefficients Kig are derived
chaos. We also study the effects of these bifurcations on from the induction motor model proposed in [14] and [3].
system trajectories by investigating the behavior of the The constant Kqwis small so that the (6, V) dynamics are
trajectories near the stable operating point for parameter faster than that of (0, w ) . The real and reactive powers P,
values corresponding to the various types of bifurcation. and Qe demanded by the induction motor are modeled as

A. System Model
P, = Kpw&+ KpvV + KpvTV+ P (11)

The system model consists of a load bus and two gen-

+
Q e = Kqw& KqvV Ku,V2 Q. + + (12)
erator buses, one of which is a slack bus (infinite bus).
The other generator bus has a fixed voltage magnitude and Note that (12) is the same as (7). The variable & can be
swing equation dynamics as given below in (5) and (6). eliminated by substituting (12) into (11) to obtain (8).
The load is modeled as an induction motor.
B. Bifurcation Analysis
Three sets of bifurcation diagrams are drawn for different
19A diffeomorphism is a differentiable map whose inverse exists and
is also differentiable. The Henon map H ( x , y) is an area contracting values of machine damping coefficient d,. The parameter
diffeomorphism if I b I < 1. varied in each diagram is Q , the reactive power demand.
*OA homoclinic tangle is a set of homoclinic points. The bifurcation diagram that exhibits the richest qualitative
*]The order of occurrence depends on the direction in which the behavior, including two period-doubling routes to chaos, is
parameter is varied. for the value d, = 0.05. When d, is increased to 0.1,

TAN et al.: BIFURCATION, CHAOS, AND VOLTAGE COLLAPSE IN POWER SYSTEMS 1489

Authorized licensed use limited to: PONDICHERRY ENGG COLLEGE. Downloaded on August 9, 2009 at 01:17 from IEEE Xplore. Restrictions apply.
 Equilibrium number 1 (xJ:
4- , I
Y
a \
t
g 35-
U
lu

(two complex eigen dues in O W ) i 3-

(two real and negahve eigenvalues) Q = 11 41136
P
BP4
ORHP open right half complex plane h
251
a z
one eigenvalue is zero

Equilibrium number 2

SNB

I
unstable (type-1 equilibnum point)
(one real and positive eigenvalue)
II
Q = 11.41136
Q = 10.5

UHB: Q = 10.94681
SHB:Q = 11.40665
SNB: Q = 11.41136

Fig. 2. Stability of equilibrium points showing local bifurcation

values. real part of eigenvalue #2
Fig. 3. Location of the complex eigenvalues as Q changes.
all period-doubling bifurcations are suppressed. At d, =
0.114 the only bifurcation is the saddle-node bifurcation The Jacobian at 2, has a pair of complex eigenvalues
(SNB), defined in Section I. For d, = 0.05 we observe six and two real and negative eigenvalues. Fig. 3 shows the
types of bifurcations: movement of the pair of complex eigenvalues, as Q is
1) C F B , Cyclic Fold Bifurcation: Q = 10.81813, increased from 10.5 to Q S N B = 11.41136. The eigenvalues
2 ) P D B 1 , Period Doubling Bifurcation: Q = 10.87327, cross the jw axis transversely at Q = Q U H B so that
3) UHB, Unstable or Subcritical Hopf Bifurcation: Q = IC, becomes unstable for Q > Q U H B , and recross it at
10.94681, Q = Q S H B so that x, regains stability for Q > Q S H B .
4) P D B 2 , PeriodDoubling Bifurcation: Q = 11.38779, At Q = Q U H B , the complex eigenvalues are f j w o =
5 ) S H B , Stable or Supercritical Hopf Bifurcation: Q = fj3.746. By the Hopf bifurcation theorem, the unstable
11.40665,
6 ) S N B , Saddle-Node Bifurcation: Q = 11.41136.
oscillations corresponding to Q values slightly less than
Q U H Bhave periods approximately equal to We check
that for Q = 10.9459, a parameter value slightly less than
z.
All the simulation results are obtained using the bifurca-
tion analysis software package AUTO [9] and the nonlinear Q U H B = 10.94681, there is an unstable periodic solution

system simulation toolkit DSTOOL [7].22 oscillating at a period of T = 1.677 M seconds. For Q
Fig. 2 shows the location of the Hopf bifurcations with near Qs~-,TB, the real part of the eigenvalue is very sensitive
respect to the position of the eigenvalues for d, = 0.05. to changes in Q. At Q = Q S N B , one of the two real
For the range of Q values considered, the system has and negative eigenvalues crosses the J W axis and becomes
two equilibrium points. One of them is always unstable. positive. This reactive power demand corresponds to the
The Jacobian at this equilibrium has one real and positive system steady-state operating limit, and the system has no
eigenvalue and three open left half plane eigenvalues, so operating solution for Q > Q S N B .
it is a type-1 unstable equilibrium. The other equilibrium
is stable and denoted z,. At Q = 10.94681, there is a C. The Esfeects of Damping on the
U H B with the emergence of an unstable limit cycle around Occurrence of Bifurcations
x, for Q < Q U H B , hence by the exchange of stability Figs. 4-6 show bifurcation diagrams for three values of
property given in [25], x, loses stability for Q > Q U H B . machine damping: d, = 1) 0.050, 2) 0.100, or 3) 0.114.
At Q = 11.40665, there is a S H B , so x, regains stability From (5)-(8) we know that different values of d, have
for Q > Q S H B . At Q = 11.41136, a S N B occurs and no effects on the locations of the two equilibrium points.
the equilibria coalesce and become a single equilibrium However, the Jacobian evaluated at the equilibrium points
at which the Jacobian is singular and has a simple zero will be different. In Fig. 5 there are no period doubling
eigenvalue. bifurcations. Also implied is the absence of chaos, since
the route to chaos (for the system being considered here) is
"For details, email the authors at [email protected],comell.edu. followed via a sequence of period doubling bifurcations as

1490 PROCEEDINGS OF THE IEEE, VOL 83, NO. 11, NOVEMBER 1995

Authorized licensed use limited to: PONDICHERRY ENGG COLLEGE. Downloaded on August 9, 2009 at 01:17 from IEEE Xplore. Restrictions apply.
 U n S t a M stationary branch

O.*
0.d
t
10.6
'
10.7
'
10.8
'

10.9
/

II
I

11.1
'
11.2
'
11.3
'
11.4
'
11.5
0.85
11.25 11.3 11.35
UNtable stationary kanch

11.4 11.45

reactive power demand Q reactivepowerdemand Q

S: stable periodic branch
S: stable periodic branch
U: unstable periodic branch
U: unstable periodic branch
Fig. 5. Bifurcation values for V, damping = 0.1.
Fig. 4. Bifurcation values for V, damping = 0.05.

1.2, I I , I , I

will be seen later. It may be possible as in [4] to calculate

the value of d, at which no period doubling bifurcation can
1.1 -
occur. Fig. 5 also shows that a UHB occurs at Q = 11.3365
h
and a SHB occurs at Q = 11.3975. The unstable and a
i
stable limit cycles emerging from these two bifurcations 9 I- stable stationary branch
collide at a CFB at Q = 11.3295. Also, the distance
between the two values Q U H B and Q S H B for d, = 0.1
is smaller than that for d, = 0.05. In Fig. 6, all the
dynamic bifurcations C F B , U H B and S H B disappear.
There is only a static bifurcation SNB at Q = 11.41136.
In fact, as d, is increased, U H B and S H B will come
closer until they coincide with each other and disappear at
some value Q = Q H B . This corresponds to a degenerate
Hopf bifurcation [lo] and an eigenvalue plot similar to
that in Fig. 3 will show a nontransversality in the pair of
eigenvalues crossing the j w axis.
061
10.2
"

104 10.6
"
10.8 11
"
112 11.4
'
116

naclive pwa demandQ

D. Period Doubling Routes to Chaos
Fig. 6. Bifurcation values for V, damping = 0.114.
For d, = 0.05, numerical results show that there is cas-
cade of period doubling bifurcations starting at Q P D B= ~
10.87327. Similar to that for the logistic map discussed twice the "fundamental period" where W O = 3.78 E,
in Section 11, this sequence of bifurcation values con- is the imaginary part of the pair of complex eigenvalues
verges quickly to a value around 10.89. Fig. 7 shows at Q = 10.87857. Further period doubling bifurcations
the bifurcation into doubly periodic oscillations in state occur as Q is increased. These period doublings accumulate
space and Fig. 8 shows voltage magnitude versus time of in a dense fashion on the attractor shown in Fig. 9 for
the period doubled bifurcation. A periodic double-loop is Q = 10.894. The chaotic motion has a crude orbit structure,
clearly observed in Fig. 7. Fig. 8 is the corresponding plot and its "fundamental period" is approximately 1.67 s. As
of V versus time and shows that the "lower envelope" implied by the Silnikov theorem [18], this period is roughly
exhibits modulation between the two values 0.83 and 0.855. a
U1 ' where w1 = 3.771 is the imaginary part of the complex

The period of this periodic double-loop is 3.31 s, roughly eigenvalues at Q = 10.894. Therefore the approximate

TAN et al.: BIFURCATION, CHAOS, AND VOLTAGE COLLAPSE IN POWER SYSTEMS 1491

Authorized licensed use limited to: PONDICHERRY ENGG COLLEGE. Downloaded on August 9, 2009 at 01:17 from IEEE Xplore. Restrictions apply.
 1.15
1

11

21 05
.=.
C

8 1
c
c
995
m
OI

42 0.9
>
-0
3085

08

Generator angle (in rad.) 07%1 0 0'1 0'2 0'3 0'4

Generator angle (in rad )
0'5 0'6 0'7 '
Fig. 7. Bifurcation of a periodic solution (dashed) into a doubly
periodic oscillation (solid), Q = 10.87857. Fig. 10. The two chaotic attractors: (1) Q = 10 894, and (2)
Q = 11.377

I " " " " ' I

other PDBs
other PDBs I

+
Q = 11.411$6

4 0 32 34 38 40 42
time (in seconds)
&r 46 & a! CFB Q = 1081813, PDB, Q = 10 57327 , UHB Q = 1094681
chaos IO 89 5 Q 5 10 594
Fig. 8. Voltage versus time of a periodic solution (solid) and its
doubly periodic oscillation after a PDB (dashed), Q = 10.87857. SNB Q=1141136. SHB Q=1140665, PBD2 Q=1138779
chaos 11 377 5 Q 5 11 38

Fig. 11. Partition of the parameter space with respect to bifurca-

Qon values.

in size than the one obtained for Q = 11.377. Fig. IO

shows a comparison of the two attractors in the ( 0 , V )
space. Finally, Fig. 11 shows the partition of the parameter
space, 10.5 5 Q 5 11.41136 = Q S N B , with respect to the
various bifurcation values considered in this section.

E. Chaotic Dynamics and the Constraint Manifold

Suppose we set Kpw= 0 in (7) and (8) so that the load
Fig. 9. Chaotic dynamics due to successive PDB's, Q = 10.894. variables (6,V ) are infinitely fast. Then the system equa-
tions will consist of two differential equations constrained
by two algebraic equations:
frequency of the chaotic oscillation is between 1 and 2 Hz.
The chaotic structure is first observe at about Q = 10.89,
but disappears in a boundary crisis when the reactive power Q = W (13)
demand is increased to Q = 10.894. = -d,w - Vsin(0 - 6) + P, (14)
When Q is decreased beyond PDB2, there is a similar 0 = gi(Q, 6,V ;Q) (15)
cascade of period doubling bifurcations. Similar to the other 0 = 92(Q, 6,V ;Q ) (16)
chaotic attractor that exists for 10.89 5 Q 5 10.894, this
attractor exists for Q in the range [11.377, 11.381. We where gI(0, 6, V; Q) and g2(0,6, V; Q) are the right-hand
observed that the strange attractor at Q = 10.894 is larger sides of (7) and (S), respectively. The dynamics of (13)-(16)

1492 PROCEEDINGS OF THE IEEE, VOL. 83, NO. 11, NOVEMBER 1995

Authorized licensed use limited to: PONDICHERRY ENGG COLLEGE. Downloaded on August 9, 2009 at 01:17 from IEEE Xplore. Restrictions apply.
 1.151

1.1- 1.1 -

-1.05- -1.05-
a n ,
E. 1 - 6 1
U m
2 U
‘E 0.95- g0.95-
0 0

s
2
0.9- 5 0.9- ;
- -P I
9 0.85- g0.85-
U
B
3 0.8- 3 0.8-
0.75- 0.75 -

Olz.6 -0:4 -0:2 0 0:2 0:4 0:6 ( ‘$6 -0:4 -0:2 0 0:2
Generator angle (in rad.)
014 0:6

Fig. 14. Chaotic trajectory and constraint manifold, Q = 10.894.

voilage
magnitude
(In P.u.)
+
1.15’

-1.05-
1-11 , i v componed
of stable
equilibrium
(1.0997 P.U.)
a stable
6 1- double-loop
U
U , limit cycle
0.90-

.g0.95-
I
m
5
0
0.9- ; direction of
voltage collapse
L : V component of
3 0.85- ‘i , type-1 equilibrium
P 0.70- ’ (07812P.u.)
3 0.8-
0
b iou dw dw & dm
0.75t
0.
-$!E -0:4
U1
-012 0 0:2 014 016 0!6 Fig. 15. Voltage collapse: divergence of the unstable manifold of
time
(in seconds)

Generator angle (in rad.) the type-1 equilibrium, Q = 10.945.

Fig. 13. Stable double-loop limit cycle and constraint manifold,

Q = 10.87858. is bom. The two loops “collapse” into a single loop at
points near Ms. The same is true as Q is further increased.
are confined to lie on the two-dimensional constraint man- Fig. 14 shows that period doubling bifurcations result in a
ifold “thickening” of the trajectory at points close to Mu. The
orbits appear to collapse into one loop near Ms. These
M := {(e,W , 6,V ) : gI(O,S, V ;Q ) = g2(8,S, Vi Q ) 1 0 ) . observations are yet to be mathematicized.
(17)
A companion paper in this session discusses systems with F. Voltage Collapse
constraints. The model given by (5)-(8), however, admits a
realistic, small positive parameter Kqw.For nonzero K,,, For some values of Q smaller than Q U H B= 10.94681
it is plausible that singular perturbation results in [13] and but larger than Q = 10.894 at which chaos gets annihilated,
[29] can be modified to show that trajectories of (5)-(8) the unstable manifold of the type-1 equilibrium does not
are close to the stable componentz3 (denoted M s ) of M . converge to zs. Fig. 15 shows that for Q = 10.945,
We postulate that trajectories going through period doubling the voltage component of the unstable manifold oscillates
bifurcations will lie close to the stable component of M , around and away from the stable equilibrium zs,and
and will be repelled from the unstable component (denoted later experiences a sudden sharp drop-a voltage collapse.
Fig. 16, for Q = 11.378, shows that the voltage compo-
MIA).
When Q = 10.85582, a value smaller than Q P D B ~ , nent of the unstable manifold of the type-1 equilibrium
Fig. 12 shows M and the stable limit cycle rS and it can converges to the chaotic attractor. The presence of this
be seen that rs lies close to Ms. As Q is increased beyond stable structure (the chaotic attractor) appears to “stabilize”
QPDB~ Fig.
, 13 shows that a stable double-loop limit cycle the system, preventing voltage collapse. Increasing Q after
encountering the second chaotic attractor, we observe in-
23The stable (unstable) component of M is the set of verse period doubling at Q = 11.38779. This culminates
points ( @ , U 6,
, V) E M where all (some) eigenvalues of
[D(s,v)gi,(@, v;
6 &) I D ( s , v ) g 2 ( @ , 6V, ;Q)] are in the open in a supercritical Hopf bifurcation at Q = 11.40665. The
left-half (nght-half) complex plane. saddle-node bifurcation occurs at Q = 11.41136.

TAN et al.: BIFURCATION, CHAOS, AND VOLTAGE COLLAPSE IN POWER SYSTEMS 1493

Authorized licensed use limited to: PONDICHERRY ENGG COLLEGE. Downloaded on August 9, 2009 at 01:17 from IEEE Xplore. Restrictions apply.
 Yoitage until Q S N B . Between Q C F B and Q U H B ,the unsta-
ble limit cycle emerging from the subcritical Hopf
bifurcation at QUHBis labeled as U,. Since the
two chaotic attractors are formed as a result of two
cascades of periodic doubling bifurcations, there is
c 1736 F .,
probably a countable but infinite number of unstable
limit cycles for Q values belonging to two narrow
nonintersecting intervals which lie in between the
\
chaotic attractor values Q P D B ~ and Q p D B 2 . Between Q U H B and
’ V component of
Q S H B ,E , turns into a type-2 unstable equilibrium
type-1 equiiibrium
(0.7905 p U-)
point, which we label as U,. After Q S H B only U ,
remains.
lime
(in seconds) 7) A component of the unstable manifold of U, con-
b too 400 $00 400 do0 dm 650 verges to Rz for Q 5 10.894. The load demand
Fig. 16. Reappearance of chaotic attractor, Q = 11.378 Q = 10.894 is where the chaotic attractor that occurs
after PDBl vanishes. At Q = 11.37, another chaotic
attractor makes its appearance. For 10.894 5 Q 5
V. CONCLUDING REMARKS 11.37, the same component of the unstable manifold
Bifurcations occur at parameter values where there is a of Ul converges to RI (infinity). For 11.37 5 Q 5
qualitative change in system behavior. For the three-bus 11.38, this component converges to R3,which is
model studied in Section IV, this change can characterized now a chaotic attractor. It is clear that for parameter
in terms of attractors. intervals between these “switchings” of convergence
There are at most three attractors for any value of to various R,, there must exist scenarios that violate
Q E [10.5,Qs~ =~11.411361, denoted RI, Ra and transversality . Examples of nontransversal scenarios
R3. We consider infinity ( f m ) to be an attractor. are trajectories connecting pairs of unstable limit sets.
Label f c c as RI. In seeking explanations other than the saddle-node bifur-
The attractor labeled Rz is the one of most engi- cation to voltage collapse, we have presented a numerical
neering interest. For Q < Q U H B and Q > Q S H B , example in Section IV where voltage collapse occurred
Rz is xs,the stable operating point. For values of because the chaotic attractor is destroyed in a boundary
Q between Q U H Band Q S H B , Rz disappears since crisis. Prior to voltage collapse, we have identified, using
2 , becomes unstable and the Jacobian has a pair of the results of Silnikov in [lx], that the approximate fre-
complex open right half plane eigenvalues. quency of chaos is related to the imaginary part of the
R3 makes its appearance at Q = QCFB as a stable complex eigenvalues of the Jacobian linearized about the
limit cycle. R3 undergoes several dynamic bifur- operating point. This approximate frequency is between
cations after Q P D B ~The . period doubling cascade 1 and 2 Hz. Whether or not such an oscillation is ac-
transforms RS into a chaotic attractor; a possible ceptable to the power system depends on the mechanical
global bifurcation eliminates R3 at a Q value near frequency of generators connected to the network. Electrical
10.894. R3 reappears just before Q P D B ~as another subsynchronous res~nances~~have been known to damage
chaotic attractor. Between Q P D B ~and Q S H B , R3 is the mechanical shaft of generators. Although the example
a stable limit cycle. presented in Section IV is small, the same conclusions seem
For values of Q between 10.894 (which corresponds to hold for a 39-bus power system [ll]. This confirms
to the disappearance of the first chaotic attractor) and that Hopf bifurcations and chaos do occur in large scale
11.377 (which corresponds to the appearance of the electric networks operating under stressed conditions. Our
second chaotic attractor), numerical results show that simulation results show that the second chaotic attractor
RIis the only attractor. These parameter values imply serves to “stabilize” the system after voltage collapse. This
transient instability. Also voltage collapse can take can be viewed as an advantage of chaos. The deleterious
place. effects of chaos are that the magnitude of the chaotic oscil-
There must exist at least one unstable limit set on lation may be unacceptable for thermal overload reasons or
the boundary of each of the attractors R,,1 5 a 5 3. that its harmonic content could damage generators or other
Denote the boundary of R,by d ( R , ) . We can test this components connected to the network.
by checking whether some component of the unstable Since the period-doubling route to chaos is the outcome
manifold of the unstable limit set converges to R,. of a cascade of bifurcations commencing with the Hopf
We restrict ourselves to “some component” because bifurcation, the prevention of chaos is directly connected to
an unstable limit set could be on a boundary d(R,) the prevention of the Hopf bifurcation. The Hopf bifurca-
which is “shared” by more than one attractor. tion was shown to exist because of voltage exciter systems
For the example considered in Section IV, the follow-
ing unstable limits sets, denoted U,, can be identified. 24 A subsynchronous resonance signifies an electrical oscillation below
U , is a type-1 unstable equilibrium point that persists the line frequency of 60 Hz.

PROCEEDINGS OF THE IEEE, VOL 83, NO. 11, NOVEMBER 1995

Authorized licensed use limited to: PONDICHERRY ENGG COLLEGE. Downloaded on August 9, 2009 at 01:17 from IEEE Xplore. Restrictions apply.
 in [27]. As shown in Figs. 4-6, the Hopf bifurcation can N. Kopell and R. B. Washburn, “Chaotic motions in the two-
be prevented by an increase in machine damping. This may degree-of-freedom swing equations,” IEEE Trans. Circ. and
Syst., vol. CAS-29, pp. 738-746, 1982.
also be achieved either by power system stabilizers or by H. G. Kwatny, A. K. Pasrija, and L. Y. Bahar, “Static bifur-
the use of high power solid state switches as described in cations in electric power systems: Loss of steady state stability
and voltage collapse,” IEEE Trans. Circ. and Syst., vol. CAS-33,
[201. pp. 981-991, 1986.
E. N. Lorenz, “Deterministic nonperiodic flow,” J. Atmospheric
Sci., vol. 20, pp. 130-141, 1 9 6 f
ACKNOWLEDGMENT J. E. Marsden and M. McCracken, The HoDf Bifurcation and Its
The authors are grateful for comments on earlier drafts Applications. New York Springer-VerlG, 1976.
R. M. May, “Simple mathematical models with very compli-
by Profs. Eyad Abed and Hsiao-Dong Chiang, and Dr. Tom cated dynamics,” Nature, vol. 261, pp. 459-467, 1976.
Schneider of EPRI. C. Rajagopalan, P. W. Sauer, and M. A. Pai, “Analysis of
voltage control systems exhibiting voltage collapse,” in Proc.
28th Con$ on Decision and Contr., Tampa, FL, 1989, pp.
REFERENCES 332-335.
D. Ruelle and F. Takens, “On the nature of turbulence,”
E. H. Abed and P. P. Variaya, “Nonlinear oscillations in power Mathematical Phys., vol. 20, pp. 167-192, 1971.
systems,” Int. J. Elecrric Power and Energy Syst., vol. 6, pp. S. S. Sastry and C. A. Desoer, “Jump behavior of circuits
3 7 4 3 , 1984. and svstems.” IEEE Trans. Circ. and Svst.. vol. CAS-28. vv.
~I ,

H. 0. Wang, E. H. Abed, and A. M. A. Hamdan, “Bifurcations, 1109-1124, Dec. 1981.

chaos and crises in voltage collapse of a model power system,” F. M. A. Salam. J. Marsden. and P. P. Varaiva. “Amold
IEEE Trans. Circ. and Syst.-I: Fund. Theory and Applic., vol. diffusion in swing equations of a power system,” iEEE Trans.
41, pp. 294-302, Apr. 1994. Circ. and Syst., vol. 30, pp. 673-688, Aug. 1984.
E. H. Abed et al., “Dynamic bifurcations in a power system N. A. Tsolas, A. Araposthatis, and P. P. Varaiya, “A structure
model exhibiting voltage collapse,” Int. J. Bifurcation and preserving energy function for power system transient stabil-
Chaos, vol. 3, no. 5, pp. 1169-1176, Oct. 1993. ity analysis,” IEEE Trans. Circ. and Syst., vol. CAS-32, pp.
J. C. Alexander, “Oscillatory solutions of a model system of 1041-1049, Oct. 1985.
nonlinear swing equations,” Int. J. Electric Power and Energy P. P. Varaiya, F. F. Wu, and H. D. Chiang, Bifurcation and
Syst., vol. 8, no. 3, pp. 130-136, 1986. Chaos in Power Systems: A Survey, EPRI Rep. TR-100834,
V. Ajjarapu and B. Lee, “The application of bifurcation the- Aug. 1992.
ory to study the nonlinear Phenomena in an electrical power M. Varghese, H. D. Chiang, and J. S. Thorp, “Computer
system,”lEEE Trans. Power Syst., vol. 7, pp. 424432, Feb. algorithms in power systems: from constructive methods to
1992. mncated fractals,” IEEE Trans. Educ., vol. 36, pp. 36-41, Feb.
-, “Period doubling route to chaos in an electrical power 1993.
system,”lEE Proc. Part C, vol. 104, pp. 490-496, Nov. 1993. S. Wiggins, Introduction to Applied Nonlinear Dynamical Sys-
A. Back et al., “DSTOOL: Computer assisted exploration of
dynamical systems,” Computers and Mathematics, vol. 39, no. Springer-Verlag, 1990.
__
tems and Chaos, Texts in Avvl. Mathematics, vol. 2. Berlin:

4; pp. 303-309, Apr. 1992. J. Zaborskv. G. Huang. B. Zheng, and T. C. Leuna, “On the
Y I I

R. L. Devaney, An Introduction to Chaotic Dynamical Systems. phase portraits of a class of large nonlinear dynamical systems
New York: Cummings, 1986. such as the power system,” IEEE Trans. Autom. Contr., vol.
E. J. Doedel and J. P. KemCvez, “AUTO: Software for contin- 33, pp. 4-15, Jan. 1988.
uation and bifurcation problems in ordinary differential equa-
tions,’’ Applied Mathematics Rep., Calif. Inst. Tech., 1986.
R. L. Chen and P. P. Varaiya, “Degenerate Hopf bifurcation
in power systems,” IEEE Trans. Circ. and Syst., vol. 35, pp.
818-824, July 1988.
H. D. Chiang, private communication.
H. D. Chiang, M. Hirsch, and F. F. Wu, “Stability regions of
nonlinear autonomous dynamical systems,” IEEE Trans. Autom.
Contr., vol. 33, pp. 16-27, Jan. 1988.
C. De Marco and A. Bergen, “A security measure for random Chin-Woo Tan received the B.Sc. and Ph.D. degrees in electrical engi-
load disturbances in nonlinear power system models,” IEEE neering, and the M.A. degree in mathematics, all from the University of
Trans. Circ. and Syst., vol. CAS-34, pp. 1546-1557, Dec. 1987. California at Berkeley.
I. Dobson and H. D. Chiang, “Toward a theory of voltage From 1991 to 1994 he was a Researcher and Lecturer at the University
collapse in electric power systems,” Syst. Contr. Letts., vol. 13, of California at Berkeley. Since 1994 he has been working at GeoStar
pp. 253-262, 1989. Technology in Santa Clara, CA. His research interests are nonlinear
I. Dobson and D. F. Delchamps, “Basin boundaries in the pen- dynamics and signal processing.
dulum with nonperiodic forcing,” in Proc. 23rd Con$ Inform.
Sci. and Syst., Baltimore, MD, 1989.
M. J. Feigenbaum, “Universal behavior in nonlinear systems,”
Nonlinear Dynamics and Turbulence, G. I. Barenblatt, G. Iooss,
and D. D. Joseph, Eds., pp. 101-138, 1983.
C. Grebogi, E. Ott, and J. A. Yorke, “Crises, sudden changes in
attractors, and transient chaos,” Physica, vol. 7D, pp. 181-200,
1983.
J. Guckenheimer and P. J. Holmes, Nonlinear Oscillations,
Dynamical Systems, and Bifurcations of Vector Fields, Ap- Matthew Varghese received the degrees of
plied Mathematical Sciences, vol. 42. Berlin: Springer-Verlag, B.Tech. (honors), M.S., and Ph.D. degrees in
1986. electrical engineering from IIT Kharagpur,
D. Hill and A. Bergen, “Stability analysis of multimachine Drexel University, and Cornel1 University,
power networks with linear frequency dependent loads,” IEEE respectively. At Cornell, he was a graduate
Trans. Circ. and Syst., vol. CAS-29, pp. 804-848, Dec. 1982. fellow of the Mathematical Sciences Institute.
N. G. Hingorani, “Flexible AC transmission systems,” IEEE He is currently an Assistant Research Engi-
Spectrum, vol. 30, pp. 4 0 4 5 , Apr. 1993. neer with the Electronics Research Laboratory
I. Hiskens and D. Hill, “Energy functions, transient stability and at the University of California at Berkeley. Prior
voltage behavior in power systems with nonlinear loads,” IEEE to that, he was a faculty member at Clarkson
Trans. Power Syst., vol. 4, pp. 1525-1533, Nov. 1989. University and Harvey Mudd College.

TAN ef al.: BIFURCATION, CHAOS, AND VOLTAGE COLLAPSE IN POWER SYSTEMS 1495

Authorized licensed use limited to: PONDICHERRY ENGG COLLEGE. Downloaded on August 9, 2009 at 01:17 from IEEE Xplore. Restrictions apply.
 Pravin Varaiya (Fellow, IEEE) is James Fife Felix F. Wu (Fellow, IEEE) is currently Professor and Vice Chairman
Professor of Electrical Engineering and Com- of the Electncal Engineenng and Computer Sciences Department at the
puter Sciences at the University of California University of California at Berkeley. He has held visiting appointments
at Berkeley, and Director of Califonia PATH, at Eidgenossische Technische Hochschule-Zurich, Shangha Jiao Tong
a multiuniversity program of research in Intel- Umversity, Beijing Tsinghua University, University of Tokyo, and Uni-
ligent Transportation Systems. From 1975 to versita Degli Studi de Cassino. He was a consultant to Pacific Gas and
1992 he also was Professor of Ekonomcs at Electnc Company, Kansas Gas and Electric, Norwegian State Power
Berkeley. He has taught at MIT and the Federal Board, Taiwan Power Company, Electric Power Research Institute, and
University of B o de Janelro. He was a technical the Chinese Electnc Power Research Institute
staff member at AT&T Bell Laboratones dunng Dr. Wu was C h m of Tokyo Electric Power Company’s “Frontier
1962-1963. His research interests are c o m m u r ~ Technology for the Future Electric Energy Systems,” in 1991 He is now
cation networks, transportatlou systems, and electric power systems. serving as Chairman of the Fellows Working Group of the Power System
Dr Varaiya has held a Guggenheim Fellowship and a Miller Research Englneenng Commttee He has served on the Editorial Board of the IEEE
Professorship Proceedings IEEE Award Board-Prize Paper Commttee, and other IEEE,
IFAC, and CIGRE commttees

1496 PROCEEDINGS OF THE IEEE, VOL 83, NO 11, NOVEMBER 1995

Authorized licensed use limited to: PONDICHERRY ENGG COLLEGE. Downloaded on August 9, 2009 at 01:17 from IEEE Xplore. Restrictions apply.