Professional Documents
Culture Documents
expressly forbidden. Terms & Conditions of access and use can be found at
http://www.tandfonline.com/page/terms-and-conditions
WILSUN XU
Department of Electrical and Computer Engineering
University of A lberta
Edmonton, Alberta
Canada, T 6G 2G7
T his pape r pre sents an enhanced method fo r po we r syste m vo ltage stability
asse ssm ent. K e y feature s o f this method are the inc lusio n of the charac teristic s o f key dynamic devices and an e c ie nt method to so lve the assoc iated
alge braic and die rential equation se ts in steady- state co nditio ns. It oers an
acc urate re prese ntatio n o f equipment o perating c haracte ristics fo r vo ltage stability margin calc ulatio n. T he importance o f the propo sed modeling approac h
is demonstrated with simulation results.
Introduction
349
350
da Silva et al.
solve the enhanced power ow model that includes algebraic and dierential equations. T he proposed approach, in eect, simulates a systems steady-state responses
when it has been stressed by load increases or contingencies. As a result, problems
such as how to scale up generator real power output to meet load increase are
automatically solved with the inclusion of governor characteristics. T he unrealistic slack and PV buses assumptions are eliminated. Simulation results are used to
demonstrate the importance of the proposed modeling approach.
T he electric power system with m generators and n buses can be fully modeled
with an algebraic-dierential equation set as follows:
X = F ( X , Y , , U ) ,
( 1)
0 = G ( X , Y , , U ) ,
where X represents the state variable vector, Y is the algebraic variable vector,
is the system loading factor, and U is the control system setpoint vector. In
system ( 1) , the dierential equation set ( F ) describes the dynamic behavior of
synchronous generators, its voltage and load-frequency control systems, and the
load dynamic behavior [8].
T he two-axis synchronous machine model with the direct and quadrature axis
currents eliminated [5] is represented by
d@i =dt = (!i
d!i =dt = ( P m i
x di
dE q i =dt = E f d i
Pgi =
Q gi =
x qi
x q i
Vi
x di
Vi
x di
x di
sin 2( @i
E q i cos( @i
V i2
2
{1
x q i
x qi
E q i sin( @i
V i2
x di
E q i +
x q i
V i sin( @i
i )
x q i
i )
i ) +
cos 2( @i
( 2)
x q i
x di
i ) }
x q i
i = 1, . . . , m ,
( 5)
i )
( 6)
sin( @i
E di
( 4)
i ) =T q oi ,
cos( @i
E di
Vi
i ) =T d oi ,
Vi
x q i
( 3)
V i cos( @i
x di
i = 1, . . . , m,
"
x di
E q i +
1, R + 1, . . . , m,
i = 1, . . . , m ,
D i!i ) =M i ,
Pgi
"
=dt =
dE di
i = 1, . . . , R
!R )!o ,
i )
x di
V i2
x di
( 7)
where the machine rotor angle at bus R is chosen as angle reference. Consequently,
equation ( 2) is not applied to this machine. In this work, the generator with largest
inertia constant is chosen as angle reference. T he generator terminal voltages are
351
controlled by exciters, which are modeled using the following IEEE type 1 excitation
system [9]:
dE f di =dt = ( K E i E f d i + V r i ) =T E i ,
dV r i
dt
dR f
dt
"
K A iK f i
Vr i + K A i R f i
Tf
i = 1, . . . , m ,
Rf i +
Kf
Tf i
Ef
i = 1, . . . , m ,
di
i = 1, . . . , m ,
E f di + K A ( V r e f
( 8)
V i ) =T A i ,
( 9)
=T F i
with V r i, min
Vr i
V r i, max ,
( 10)
where V r i, min and V r i, max represent the excitation system limits. T he prime mover
and speed governor are modeled as [9]
dP m i =dt = ( P m i + P sv i ) =T c h i ,
dP sv i
dt
"
P sv i + P g i
with 0
1
r ef
P sv i
Ri
i = 1, . . . , m ,
!i
!sy n
P sv i, max ,
( 11)
i = 1, . . . , m ,
=T sv i ,
( 12)
where P sv , max represents turbine mechanical torque limit. T he dynamic load model
as presented in [8] is given by
T pi
Tq i
dx pi
dt
dx q i
dt
= P si
x pPti,
( 13)
= Q si
x q Q ti,
( 14)
2
P si = P o ( ap + bpV + cpV
P t i = ( ep + f pV + gpV
+ hpV
Q si = Q o ( aq + bqV + cqV
Q t i = ( eq + f qV + gqV
+ dpV n p ) ,
mp
+ dqV
+ hqV
mq
( 16)
),
nq
( 15)
),
),
( 17)
( 18)
where P S and P t represent the steady-state and transient load characteristics, respectively, both voltage dependent.
Equations ( 2) to ( 18) describe the dynamic behavior of key devices in details.
Dynamics of OLT Cs, SVCs, AGC, etc, could also be included in the same way.
In system ( 1) , the algebraic equation set ( G ) represents the transmission network.
T hese equations are obtained through the active and reactive power balance in each
system bus, which is necessary to assure the equilibrium between the amount of
generated and demanded powers. T he nodal active and reactive power balances for
a system with n buses are given by
Pgi
Q gi
P S i
P i ( , V ) = 0,
i = 1, . . . , n,
( 19)
Q S i
Q i ( , V ) = 0,
i = 1, . . . , n,
( 20)
Pi =
V i V m ( G im cos im + B im sin im ) ,
( 21)
V i V m ( G im sin im
( 22)
m2 K
Qi =
m2 K
B im cos im ) ,
352
da Silva et al.
where P g i and Q g i represent the generated active and reactive powers at bus i,
P S i and Q S i are the steady-state demanded active and reactive powers at bus i,
which are parameterized by a loading factor , P i and Q i represent the injected
active and reactive powers owing into the transmission lines, and K represents
the set of buses directly connected to bus i, including bus i. Each loading increase
is automatically shared by all generators according to their load-frequency control
system characteristics.
0 = F ( X , Y , , U ) ,
0 = G ( X , Y , , U ) .
( 23)
T his equation is then solved iteratively for a given loading factor ( ) with an
expanded Newton method:
" # "
# "
#"
#
X
X0
J1 J2
F ( X 0 , Y 0 , , U )
,
=
( 24)
Y
Y0
J3 J4
G ( X 0 , Y 0 , , U )
where J1 , J2 , J3 , and J4 are Jacobian submatrices of the dynamic system ( 1) .
Equation ( 24) makes it possible to compute, in steady-state conditions, all system variables ( algebraic and dynamic ) in a more realistic framework. T he load
increments are automatically distributed among the generators according to their
load-frequency control system characteristics. So, it is not necessary to adjust the
specied active power generation in each PV bus in order to match the demand
increase.
T he control systems setpoint ( denition of vector U ) is performed by using a
single run of conventional load ow method at base case ( = 1) . It means that only
at base case, the terminal voltages of generators will be exactly equal to its specied
values. In cases where =
/ 1, the terminal voltages will show smaller values than
the one specied when the load increases, or bigger values when the load decreases
from the base case. T his is due to steady-state error introduced by the automatic
voltage regulator.
Direct solution of equation ( 24) can be quite time-consuming due to large system sizes. Fortunately, a major portion of the equations can be analytically solved
and substituted into the more complex equations. For equations ( 8) and ( 10) , a
steady-state condition yields
Vr = K E E f d ,
( 25)
R f = K f E f d =T f .
( 26)
V i ) =K E ,
( 27)
353
1
ref
Ri
!i
!sy n
( 28)
( 29)
x qi
x q i
x qi
V i sin( @i
( 30)
i ) ,
KA
KE
1
r ef
(Vr e f
Ri
i
!R
!sy n
Vi )
1
x di
x di
E q i +
Pgi
x di
x di
x di
i = 1, . . . , m ,
Pg i =
Qgi =
Vi
x d i
Vi
x d i
E q i sin( @i
i ) +
E q i cos( @i
i )
V i2
2
V i2
2
sin 2( @i
{1
i = 1, . . . , m,
D i!R ,
i )
cos 2( @i
V i cos( @i
i ) ,
x qi
i ) }
x d i
x qi
( 31)
( 32)
( 33)
x di
V i2
x di
( 34)
3.1
Generator Limits
T he generator excitation system has the function of keeping the terminal generator voltage around a specied value and the function of protecting the generator
eld and armature windings against current overloading [10]. W hen the saturation
of the excitation system is not considered, the eld current limit can be directly
controlled through the Automatic Voltage Regulator ( AVR ) , since, under this conditions, the output AVR variable ( V r ) is proportional to the eld voltage ( E f d ) ,
which is, in steady-state conditions and in p.u., equal to the eld current. T herefore, V r , max can be used to model the eld current limit. Once this maximum value
is reached, the variable E f d becomes a xed parameter ( E f d, max = V r, max =K E ) ,
354
da Silva et al.
and consequently, equation ( 32) for this generator must be modied to reect its
new operation condition:
0 = Ef d
max
x di E q i =x di + (x di
x d i ) V i cos( @i
i ) =x di .
( 35)
3. 2
max
Pgi
D i!R .
( 36)
Continuation M ethod
T he conventional Newton-based load ow methods encounter convergence di culties due to Jacobian matrix singularity at the nose point. T his paper makes use of
a continuation method with predictor step, corrector step, parameterization, and
step-size control [11,12,13,14] to avoid the singularity. As a result, accurate computation of maximum loading condition ( nose point ) and points on lower part of the
PV curve can be achieved. In this work we chose the voltage magnitude or angle
of the weakest system bus as the continuation parameter [11]. T he weakest system
bus is the one with the biggest change in voltage magnitude or angle for a given
loading increase [12].
4
4.1
T his system contains a generator of 247.5 MVA feeding a base case load of 190 +
j 90 MVA, through a lossless transmission line with reactance of 0.1 p.u. Figure 1
shows the dependence of equilibrium points with the AVR static gain K A . W hen the
static gain is increased, the steady-state error on the controlled voltage decreases,
and consequently, the maximum system loadability increases.
T he curve corresponding to innite static gain ( zero steady-state error) is equivalent to the conventional load ow response ( the PV bus representation) . T his dependence of the AVR steady-state error with static gain can be analytically shown
by using the equivalent AVR equation. Rewriting equation ( 27) to have the generator terminal voltage at the left side results as follows:
Vi = V r e f
KEi
KAi
Ef
di ,
( 37)
355
= Vi
KEi
BC
Ef
di B C
BC
Ef
KAi
( 38)
BC
KE
KA
( E f di
di ) ,
( 39)
which is the steady-state value of the controlled voltage in a general loading condition related with the base case. From equation ( 39) the AVR steady-state error can
be assessed analytically. A load increase from the base case would demand a bigger
eld voltage, and the term E f di B C E f di would be negative, forcing a lower value
to the controlled voltage than the one specied. Similarly, a load decrease from the
base case would force a bigger value to the controlled voltage than the specied
value.
Figure 2( a) shows the eect of the generator eld current limit on the system
maximum loadability. T he eld current limit is reached at point A . Beyond this
point the terminal generator voltage becomes an uncontrollable variable, which
starts to decline as loading increases. T his characteristic causes an equally strong
decline in the load bus voltage. Consequently, the maximum system loadability is
drastically decreased. T he results clearly show the importance of eld current limit
representation. Figure 2( b ) shows the eld current limiter action, which freezes the
AVR output when the eld current limit is reached at point A .
Figure 3( a) shows the armature current limit eect on the system steady-state
behavior; this limit is reached at point B . As it can be noted, the system maximum
loadability is considerably decreased with the armature current limit representation.
Beyond the point B , the generator works as a constant current source, as shown in
Figure 3( b ) . T he eld current must be adequately reduced in each curve point by
the armature current limiter, as it is shown in Figure 3( c ) .
For this simple system, the armature current limit have a more restrictive eect
on the generator operating conditions. Once the limit is reached, the generator does
not support load increases since the armature current must be kept constant.
356
da Silva et al.
4. 2
T his commonly used test system consists of three generators and three load areas
[15]. It is used here to show the joint action of armature and eld current limiters, as
well as the discrepancies between the proposed methodology and the conventional
load ow approach.
Figure 4 shows the PV curves of buses 1 ( slack bus) , 2, and 3 ( PV buses) .
As it can be observed, these voltage magnitudes show dierent values than the
specied ones when the system loading is changed, due to the steady-state error of
automatic voltage regulators. T his is the rst discrepancy detected in relation to
the conventional load ow, which considers the controlled voltages constant for all
loading conditions.
Figure 5 shows a single run of the proposed methodology by considering the
armature and eld current limiters action, as well as the generator s active power
limits. Figure 5( a) shows the PV curve of bus 2 ( PV type) and the points where
several limits are reached. Figures 5( b ) and ( c ) show the AVR s outputs ( excitation control) and speed governors outputs ( load-frequency control) for the three
generators. It can be noted from these gures that, at the points A 1 and A 2 , the
generators G 3 and G 2 reach their respective active power limits. T he speed governors outputs are kept on their maximum values, and the generator G 1 assumes
the entire load increase. Even on the active power limit, the generator G 2 continues
357
supplying reactive power until the point B , where its eld current limit is reached.
Beyond this point, the excitation signal of G 2 is kept constant. Consequently, its
terminal voltage becomes a noncontrolled variable and starts to decrease rapidly,
and the armature current starts to increase to its limit at the point C . Beyond
this point, it is necessary to reduce the eld current in order to keep the armature
current xed on its maximum value.
358
da Silva et al.
It is important to note that the active power limit of generator G 1 has been not
reached, and therefore this generator could supply additional loading requirements;
however, the conditions of G 2 and G 3 drastically reduce the system reactive power
load-ability, characterizing a typical voltage stability scenery.
Figure 6 pictures a comparison of the proposed methodology with the conventional load ow formulation results. As it can be seen, the static voltage stability
limit, corresponding to the system maximum loadability point, is optimistically
computed by the conventional load ow method, even with transformer tap limits and the generator reactive power limits included. In addition to the AVR s
steady-state error, the following items also cause discrepancies between the proposed methodology and the conventional load ow formulations:
a) T he generators reactive power limits usually are kept constant on conventional approaches, but in reality, they are dependent on the generator
terminal voltage and active power supplying conditions [1]. It is therefore
almost impossible for the conventional load ow formulation to respect the
generators capability limits accurately.
359
Conclusions
T his paper presented a systemic approach to model power system component dynamics for the steady-state assessment of voltage stability. T he models have eliminated arbitrariness existing in conventional power owbased PV curve calculations. Examples are the scaling of generator real power output and the accurate
representation of generator real and reactive power limits. Furthermore, an e cient
continuation-based method is developed to compute the system PV curve trajectory
with various device models. T he simulation results show that the conventional load
ow method leads to optimistic voltage stability limit. Main sources of discrepancies
have been discussed and demonstrated analytically.
A cknowledgments
360
da Silva et al.
References
[1] P. A. Lof, G. Andersson, and D. J. Hill, 1995, Voltage dependent reactive power
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
limits for voltage stability studies, IEEE Trans. on P o we r Syste ms, Vol. 10, pp. 220
228.
Z. Feng, V. Ajjarapu, and B. Long, 2000, Identication of voltage collapse through
direct equilibrium tracing, IEEE Trans. on P ower Syste ms, vol. 15, pp. 342349.
V. Ajjarapu and Z. Feng, 1998, A novel approach for voltage collapse analysis and
control, PowerCon98, Beijing, China, pp. 14991503.
E. Bompard, E. Carpaneto, G. Chicco, and R. Napoli, 1996, A dynamic interpretation of the load ow Jacobian singularity for voltage stability analysis, E lec trical
P ower & Energy Syste ms , Vol. 18, No. 6, pp. 385395.
L. C. P. da Silva, C. M. Moreira, and V. F. da Costa, 1998, Inuence of Generators and Loads Characteristics on Power System Voltage Stability, VI SEPOPE,
Salvador-Brazil.
P. W. Sauer and M. A. Pai, 1990, Power system steady-state stability and the load
ow Jacobian, IEEE Trans. o n P o we r Systems, Vol. 5, pp. 13741383.
C. A. Canizares, 1995, Conditions for saddle node bifurcations in AC/ DC power
systems, Elec trical P ower & Energy Syste ms, Vol. 17, No. 1, pp. 6168.
W . Xu and Y. Mansour, 1993, Voltage stability analysis using generic dynamic load
models, IEEE Transac tio ns o n P o we r Syste ms, Vol. 9, pp. 18.
P. W. Sauer and M. A. Pai, 1998, P o we r Syste m D ynamics and Stability , 1st edn.
Prentice Hall: New Jersey.
P. Kundur, 1994, P owe r Syste m Stability and C ontro l, McGraw-Hill, New York.
D. A. Alves, L. C. P. da Silva, C. A. Castro, and V. F. da Costa, 1999, Parameterized
Fast Decoupled Load Flow for Tracing Power Systems Bifurcation Diagrams, IEEE
PES Summer Meeting, Edmonton, Canada, pp. 708713.
Y. Mansour, 1993, Suggested techniques for voltage stability analysis, IEEE Working Group on Voltage Stability, Publication 93T H0620-5-PW R.
V. Ajjarapu and C. Christy, 199 2, The Continuation power ow: A tool for steady
state voltage stability analysis, IEEE Trans. on P ower Systems, Vol. 7, pp. 416423.
H. D. Chiang, A. J. Flueck, K. S. Shah, and N. Balu, 1995, CPFLOW : A practical tool for tracing power system steady-state stationary behavior due to load and
generation variations, IEEE Trans. o n P o we r Syste ms, Vol. 10, pp. 623634.
P. M. Anderson and A. A. Fouad, 1977, P ower Syste m Co ntro l and Stability , Iowa
State University Press.