Sie sind auf Seite 1von 7

1

Critical Eigenvalue Trajectory Tracing for


Power System Oscillatory Stability
Assessment
Xiaoyu Wen, Student Member,IEEE, and Venkataraman Ajjarapu, Senior Member,IEEE

one can calculate all the eigenvalues of any dynamical system.


Abstract This paper presents an integration based approach However it is very computationally expensive to calculate all
to trace a critical eigenvalue to identify oscillatory instability. these values for practical power systems. We may need only
Indices are developed to identify critical eigenvalue to be traced one eigenvalue which crosses the imaginary axis first.
for further analysis. This index takes into account the rate of
change as well as direction of the movement. The rate of change
References [13] and [14] discuss the robustness and
is with respect to any parameter variation of interest. This efficiency of existing dominant eigenvalue-computing
method in combination with maximum real part calculation methods and provide new alternatives. Since only one
provides reliable information related to margins with respect to eigenvalue is calculated each time, the algorithm is very fast.
oscillatory stability and minimum damping requirements. Reference [15] applies the power method with bilinear
Eigenvalue and eigenvector sensitivities (with respect to any transformation to calculate the dominant eigenvalue. In
explicit /implicit parameters) are by products of this approach.
reference [13] the power method is implemented and
Index Terms Damping margin, Hopf bifurcation, Eigenvalue
compared with other dominant eigenvalue-computing
trajectory, Oscillatory stability algorithms. The author finds that this method belongs to the
linear convergence algorithm. Reference [13] also
implemented Newtons method, inverse power, Rayleigh
I. INTRODUCTION quotient iteration, and other methods. The various algorithms
are compared and evaluated with regards to convergence,
O SCILLATORY instability is an inherently nonlinear
phenomenon that is related to bifurcation from the
viewpoint of nonlinear dynamic systems. Substantial research
performance and applicability. The conclusion of reference
[13] points out that linear convergence algorithms, like the
has been conducted to help understand and analyze the power method with bilinear transformation, are more robust
mechanism of this type of instability based on Hopf than the higher-order methods. However, higher-order
bifurcation theory [1-7]. The essential information of a Hopf methods, such as the quadratic and cubic convergence
bifurcation can be obtained in terms of eigenproperties of the methods, are much faster. A better result is achieved by
reduced power system matrix of a structure-preserving power combining these two types together.
system model described by differential algrbraic equations Reference [16] describes new matrix transformations suited
[11]. This matrix is called the dynamic system state matrix. to the efficient calculation of the dominant eigenvalue of
Damping also plays an important role in power system large-scale power system dynamic models. Since only the
oscillations. Margin related to damping can be defined as the most critical eigenvalue is calculated, all the other eigenvalue
amount of additional load on a specified pattern of load information is not provided.
increase that would cause the damping ratio to reach its References [17, 18] describe the algorithm which
minimum limit. In the oscillatory stability assessment, in efficiently computes the dominant poles of any specified high-
order to keep the system far away from the minimum damping order transfer function. It has the numerical properties of
limit, the damping ratio margin needs to be checked for each global and ultimately cubic convergence. A numerical
contingency and scenario. WSCC [12] recommend the various example is provided to study low-frequency oscillations in
criteria in determining the safe operating limits with respect to electrical power systems.
damping: In the large-scale power system simulation, using more than
Oscillatory stability assessment with damping information one processor will obviously speed up the calculation.
needs eigenvalue estimation. In the literature very extensive References [19, 20, 21] present the application of parallel
work is available for estimating the eigenvalues. In principle computing in eigenvalue calculation. Parallel processing
introduces increased complexity in software and algorithm
Xiaoyu Wen is with the Iowa State University, Ames, IA 50011 USA. (E- strategy. Therefore, the task of converting a sequential
mail: xwen@iastate.edu). algorithm into an efficient parallel procedure is always
Venkataraman Ajjarapu is with Iowa State University, Ames, IA 50011 USA challenging.
(E-mail: vajjarap@iastate.edu).
2

Reference [22] presents two sparsity-based eigenvalue eigenvalue and eigenvector vs. the parameter value can be
techniques for oscillatory stability analysis of large-scale obtained. Thus, the complete information about how the
power systems. dominant eigenvalue approaches and crosses the imaginary
Above critical eigenvalue calculation algorithms can axis can be obtained.
provide snap shot at a particular operating point. To find the References [29, 30] provide an approach for developing the
stability margin, we have to repeat this process at a series of eigenvalue and eigenvector differential equation for the
operating points. parameterized matrix J(). For any eigenvalue of J(), and
Reference [15] combines dominant eigenvalue calculation corresponding eigenvector , the following relation is well
with an eigenvalue-based iterative algorithm that calculates known.
the Hopf bifurcation-related segment of the feasibility J ( ) v = v (2.1)
boundary for a realistically large power system model. Each T
v v = 1
iteration consists of the parameter estimator and the dominant
If you differentiate (2.1) with respect to , then
eigenvalue calculation, also called the eigenvalue corrector. In
dJ dv d dv
d v + J d = d v + d
reference [15], the secant method is used for the parameter
estimator. dv T
Another way to identify Hopf bifurcation is through the dv
v + vT =0
direct method [23]. The direct method can identify the d d
oscillatory stability margin directly without computing any d dv dv dJ
intermediate operating points. The Hopf bifurcation point is d v + d J d = d v (2.2)
the solution of a set of algebraic nonlinear equations. T
dv dv
References [24] and [25] first applied the direct method in v + vT =0
d d
the identification of the power system oscillatory stability
(2.2) can be simplified into:
margin. Solving of these nonlinear algebraic equations is also
d dv dv dJ
complex and sometimes the traditional Newton-based
v+ J = v
techniques can lead to difficulties or failure. This approach is d d d d (2.3)
dvT
very fast for a close initial guess. Reference [26] extended the v=0
direct method to include damping margin as a constraint. This
d
reference utilized dominant eigenvalue/eigenvector [18] as an
Q dvd v = dv
T n
dv )
initial guess. Reference [26] rightly pointed out that this area
( v =v
i T

d d
i
i =1
needs further work. For example there can be a case where the
By rearranging the terms in (2.3) we get the derivatives of
dominant eigenvalue at base case is moving away from the
and with respect to as shown in equation (2.4)
imaginary axis and may lead to non convergence.
Reference [27] proposed a manifold based indirect method ( I J ) v v& J&v
= (2.4)
that does not need eigenvalue estimation to identify Hopf vT 0 & 0
bifurcation. This method looks for singularity of (Atotal+AtotalT). where a dot denotes differentiation with respect to .
Reference [28] presents an approach for tracking all the Here J depends explicitly on . In power systems, J
eigenvalues of a dynamic system Jacobian matrix. It employs corresponds to the Jacobian matrix. However the Jacobian
an iterative method to update the eigenvalue. As the authors elements in power systems are not explicitly expressed in
mentioned it may diverge for repeated eigenvalues. terms of .
In this paper we are proposing a new approach that is based For this case, the differential equation, similar to (2.4), is
on integration to track one or several eigenvalues/eigenvectors derived to solve the implicitly expressed parameterized
of interest. This approach is combined with dominant Jacobian matrix.
eigenvalue calculation to identify Hopf bifurcation. Next For power system differential and algebraic equations
section provides the details about this approach. (DAE) model:
.

II. NEW APPROACH FOR OSCILLATORY STABILITY MARGIN AND X = F ( X , Y , )


DAMPING MARGIN IDENTIFICATION 0 = G( X , Y , )
X and Y are state and algebraic variable vectors
A. Oscillatory stability margin identification
respectively. is the parameter to represent load level of the
A new eigenvalue-tracking approach [29, 30] for power entire system. The Jacobian Matrix of power system DAE
system applications is proposed. This method involves a set of model becomes
differential equations. The derivative in the differential F ( ) FY ( )
Atotal = X
equation denotes the differentiation of the eigenvalue and G X ( ) GY ( )
eigenvector with respect to the system parameter. By Thus Asys = FX ( ) FY ( )GY1 ( )G X ( )
integrating in the parameter domain, the curve of the
3

Then (2.1) becomes is 2(n+m+1).


Asys ( ) v = v
The step length selection for the integration can be obtained
T

v v = 1 by using the derivative d R . In the Newton method, the
d
and v are the eigenvalue and the right eigenvector that we
( n +1) is predicted by ( n +1) = ( n ) R
(n)
want to trace. To preserve the sparsity, an extended . (See Fig. 2.1)
eigenvector is defined as u = GY 1 ( )G X ( )v . Then, d R / d
(n)

The integration step length can be obtained


FX v + FY u = v
with ( n+1) ( n) . By the eigenvalue tracing method, the
G X v + GY u = 0 dominant eigenvalue at ( n+1) can be obtained by integration.
In this case, we derived the following differential equation, For a case shown in Fig.2.1, with the same initial operating
similar to (2.4) point (n ) and ( n +1) , the secant method will predict the
I FX FY v v& F& X v + F&Y u parameter value ' (in Fig. 2.1) as well as the next parameter
(2.5)

GX GY 0 u& = G& X v + G& Y u value. However the Newton method estimated parameter
v

T
0 0 & 0

value ( n+ 2) is closer to * than ' . Thus, the Newton
Since ( ( ), v ( ), u ( )) T is in C C n C m , method can reduce the iteration number for the most of the
cases.
Here the dimension of state variable vector is m, and the
From the above formulation (2.6), one can trace all the
algebraic vector is n.
eigenvalues or any specified subset or a single eigenvalue of
Define
interest. To detect the Hopf bifurcation, we are interested in
v = vR + jvI ; vR , vI R n the complex eigenvalue which crosses the imaginary axis first.

u = uR + ju I ; Where u R , u I R m To identify this eigenvalue an index is derived.
= + j ; , R Eigenvalue Real Part
R I
R I 0 * 1 2
With the above notation, equation (2.5) can be further 0
extended to the following form.
R I FX I I FY 0 vR vI v&R F& v
X R + F&Y uR
&

I I R I FX 0 FY vI vR v&I
X I
F v + F&Y uI (2.6)
GX 0 GY 0 0 0 u&R G & v + G&Y uR
=
X R

0 GX 0 GY 0 0 u& &
I GX vI + G&Y uI
v T vI
T
0 0 0 0 & 0
R R

&I
T T
vI vR 0 0 0 0

0 Fig. 2.2 Eigenvalue real part vs. parameter
th th
FX (ij ) is the element at i row, j column in matrix FX . Fig 2.2 shows a conceptual variation of real parts of three
FX ( ij ) X FX (ij ) Y FX ( ij ) ; eigenvalues with respect to parameter . With the equation
F& X ( ij ) = + +
X Y (2.6), any eigenvalue real part derivative can be obtained. The
The above formulation leads to an algorithm for tracing any dashed lines in Fig.2.2 indicate the slopes at base case
eigenvalue and eigenvector of a parameterized matrix, which parameter value of 0. This information is used to define an
retains sparsity. index to estimate which eigenvalue will cross imaginary axis
first. This index is given by equation (2.7)
Real part of dominant eigenvalue
Re Re (2.7)
IndexEigen = =
(n)
(n+1)
(n+2) *
dRe &Re
0 d

If we assume all the eigenvalues at the base case are on the
left half of the complex plane, then Index positive means the
Hopf corresponding eigenvalue moves towards the imaginary axis
and negative means it moves away from the imaginary axis.
Low positive index value relative to others indicates the
corresponding eigenvalue is critical and may be the first one
Slope is d R = &R
d to cross the imaginary axis.
In Fig.2.2, the eigenvalues corresponding to points A and C
have positive index value, where as point B has negative
Fig 2.1 Tangent information is used for predicting parameter value
index value. Out of A and C, C has less positive index value.
In the eigenvalue-tracing method, at each operating point In this particular scenario, Hopf bifurcation can be identified
the eigenvalue information is obtained through numerical by tracing the real part of this eigenvalue that corresponds to
integration. The dimension of the linear equation to be solved point C.
4

B. Damping margin identification 12


Damping ratio is defined as
I 10 C
D = cos(arctan( )) (2.8)
B
R
8
Since &R and &I are available, damping ratios derivative to

Imaginary Part
A
6
parameter can be derived and is given by (2.9)
d d
( R I I R )
4
dD I d d
= sin(arctan( )) (2.9)
d R (2I + 2R ) 2
D E FG H
The damping index can be obtained by 0
-2.5 -2 -1.5 -1 -0.5 0
D D0 Real Part
Index damping =
dD Figure 3.1 All complex eigenvalues in complex plane at base case
d
Following similar approach to Hopf identification, damping
ratio of any eigenvalue can be traced with a given damping Table 3.1 Index of all complex eigenvalues in Fig 3.1
threshold D0. Fig 2.3 shows the process of searching damping Index Symbol Eigenvalue Index Value
margin. Ranking
1 A -0.14272394j6.2611139 0.22210698
Damping Ratio 2 B -0.66343979j10.138474 0.91222316
3 C -0.66402544j9.3350234 0.93246577
D -2.1290647j1.3082278 -1.3789475
Slope is dD
d E -0.92005709j0.91899444 -2.6854991
F -0.44958111j0.65411231 -9.4661737
G -0.42909641j0.63828035 -5.8927809
Damping Margin H -0.050857671j0.18008881 -10.775733

Table 3.1 shows index value for all the complex


eigenvalues in Fig 3.1. The complex eigenvalues D,E,F,G, and
D0
(n) (n+1) (n+2) * H have negative index value. We included these in the table to
0 show the importance of the index. Eigenvalue H is very close
Fig. 2.3 Tangent information is used for searching damping margin to the imaginary axis compared to any other ranked
Slope of damping ratio curve in Fig 2.3 can be used to eigenvalues. However it is moving away from the imaginary
predict parameter value in damping margin identification. axis for increasing load. By using the index we eliminated this
eigenvalue for further tracing.
III. SIMULATION RESULTS We traced the ranked number one eigenvalue through
equation (2.6) by fourth order Runge-Kutta method. To
The algorithm is tested on a 10-bus, 4-generator system
prevent error caused by too large an integration step length,
with two-axis generator model, IEEE DC-I excitation system
0.08 is used as step length maximum limit for parameter . It
and governor model, which can be found in reference [11].
represents entire system load increase of 8%. Table 3.2 shows
There are 9 state variables for each generator. The total
each iteration result during the integration. At each
system Jacobian dimension is 5555 (35 state variables and
intermediate operating point, MATLAB provided function is
20 algebraic variables). Load consists of 50% constant power,
used to obtain the actual eigenvalue for comparison.
30% constant current, and 20% constant impedance. Load on
all buses will increase with same percentage. Sum of initial Table 3.2: Hopf bifurcation search for each iteration
Iter. value R R R
load on all the buses is 1973.727MW. No Integrated Actual Error for
The total number of eigenvalues at any given operating 0 0 -0.14272394 -0.14272394 0
equilibrium is 35. Out of which 19 are real and 8 are complex 1 0.08 -0.10026188 -0.10019053 7.13419E-05
pairs. 2 0.16 -0.059541278 -0.059424926 1.16E-04
Fig3.1 shows all complex eigenvalues in upper-half of the 3 0.24 -0.019699047 -0.019541067 1.58E-04
complex plane for base case (total load: 1973.727MW). With 4 0.27967539 0.000247638 0.000208616 3.90226E-05
(2.6), each eigenvalues &R and &I can be calculated. The 5 0.27917908 -2.72461E-06 -3.92981E-05 3.65735E-05
arrows show the moving direction of each eigenvalue when
parameter increases. The length of arrow shows the speed In Table3.2, integrated eigenvalue real part and its actual
of movement with respect to . value are shown. The corresponding error is listed. In the first
three steps, the parameter is increased by 0.08, which
5

corresponds to 1973.727MW 8% = 157.898MW. From Table 3.3: Damping ratio search for each iteration
Table 3.2, we can see that the accumulated error is under value Integrated Actual Error for
control very well. Fig 3.2 shows the eigenvalues position for Iter Damping Damping (%) Damping
No. (%)
every step in the complex plane. 0 0 2.2789375 2.2789375 9.76996E-15
Fig3.3 shows the eigenvalues real part with the parameter
1 0.08 1.611725 1.6105401 1.18484E-03
. Eigenvalue position in Steps 4 and 5 are very close.
2 0.1548193 1.0061742 1.0053922 7.81999E-04
6.3 3 0.15558904 0.99987697 0.99921253 6.64443E-04

6.25
2.6
Imaginary Part

2.4
6.2
2.2

6.15

Damping Ratio (%)


2

1.8
6.1
1.6

1.4
6.05
-0.16 -0.14 -0.12 -0.1 -0.08 -0.06 -0.04 -0.02 0 0.02
1.2
Real Part
Fig. 3.2 Integrated eigenvalue in complex plane 1

0.8
0 50 100 150 200 250 300 350
Entire System Load Increase (MW)
0.02
Fig. 3.4 Damping ratio vs. entire system load increase
0

-0.02
Figure 3.4 shows the each iterations load increase and
Eigenvalue Real Part

-0.04
damping ratio. The actual damping margin is computed by
-0.06 calculating the actual eigenvalue as 15.54718% load increase.
-0.08 There is an error of 1.172E-4 in value, which equal to
-0.1 0.231MW (=1.172E-41973.727MW).
-0.12 If we compare oscillatory stability margin with damping
-0.14
margin 306.859MW (=1973.727MW15.54718%), there is
-0.16
244.185 MW (551.044MW -306.859MW) difference. When
0 100 200 300 400 500 600 we use D0=0%, the corresponding damping margin becomes
Entire System Load Increase (MW)
oscillatory stability margin.
Fig. 3.3 Integrated eigenvalue with total load increase

IV. DISCUSSION AND CONCLUSIONS


With the MATLAB provided eigenvalue calculation In this paper a method based on eigenvalue tracing is
method, the exact Hopf bifurcation appears at a load increase proposed to obtain both oscillatory stability and damping
of 27.91894%. This is equal to 1973.727MW * 27.91894%. = information. An eigenvalue index is defined to rank the
551.044MW load increase. This compares well with our value eigenvalues. This index is helpful for identifying the rate of
of 1973.727MW * 27.917908%=551.023MW. The error is change of movement and the direction of movement of these
0.021MW. eigenvalues with respect to change in any parameter of
For each load increase step, we have to solve 2(m+n)+2 interest.
dimension linear equation four times. This approach is used to identify Hopf bifurcation. It is also
Similar to oscillatory stability margin identification, the extended to satisfy minimum damping margin constraint.
damping margin identification also is obtained. Table 3.3 This method is very fast for local analysis. In the previous
shows the simulation results for a damping ratio limit (D0) of section we calculated all the eigenvalues and their indices to
1%. Here the same ranked number one eigenvalue damping trace any specific eigenvalue of interest. In general we do not
ratio is traced. need to calculate all the eigenvalues. In the previous example
the ranked number 1 eigenvalue at the base case is finally
crossing the imaginary axis first. However this may not
always be true for all the operating conditions. Flow chart in
Fig. 4.1 describes the steps involved for any general starting
point. First we calculate the eigenvalue with maximum real
part for any given equilibrium point. As discussed in the
introduction, powerful methods are proposed in the
6

literature[13, 14, 16, 17, 22, 18] to calculate these dominant REFERENCES
eigenvalues. Then the index for this eigenvalue is calculated. [1] P. Kundur, Power system stability and control, McGraw-Hill, NewYork,
If this index is negative, this eigenvalue should not be traced 1994
(since it moves away form the imaginary axis). We trace only [2] M. Ilic, J. Zaborszky, Dynamics and contrl of large electric power
systems, John Wiley & Sons, New York, 2000
the eigenvalue with positive index until it crosses imaginary
axis ( we also trace the index. If the index becomes negative [3] L. H. Fink, ed. Proceeding: Bulk power system dynamics and control IV
restructuring, Santorini, Greece, August 23-28, 1998
before it crosses the imaginary axis , again another dominant
[4] B. D. Hassard, N. D. Kazarinoff and Y. H. Wan, Theory and
eigenvalue with positive real part is calculated and traced). applications of Hopf bifurcation, Cambridge University Press,
Start Cambridge, 1981
[5] I. Dobson, F. Alvarado and C.L DeMarco, Sensitivity of Hopf
bifurcations to power system parameters, Proceedings of the 31st IEEE
Conference on Decision and Control, 1992 vol. 3, pp. 2928 - 2933
Compute Eigenvalue with
Maximum Real Part [6] C.D. Vournas, M.A. Pai, P.W. Sauer, The effect of automatic voltage
regulation on the bifurcation evolution in power systems, IEEE
Transactions on Power Systems, vol. 11, Nov. 1996, pp.1683-1688
[7] W. Zhu, R.R. Mohler, R. Spee, W.A. Mittelstadt, D. Maratukulam, Hopf
Index (0, index -limit]? No Compute Next
Eigenvalue with
bifurcations in a SMIB power system with SSR, IEEE Transactions on
Power Systems, vol. 11, Aug. 1996, pp. 1579- 1584
Maximum Real [8] V. Ajjarapu, B. Lee. Bifurcation theory and its application to nonlinear
Part dynamical phenomena in an electrical power system,IEEE Transactions
Yes
on Power Systems, vol. 7, pp. 424-431 Feb. 1992
Trace this eigenvalue with [9] K. Kim, H. Schattler, V. Venkatasubramanian, J. Zaborszky, P. Hirsch,
integration method Methods for calculating oscillations in large power systems, IEEE
Transactions on Power Systems, vol. 12, Nov. 1997, pp.1639-1648,
[10] F. Howell, V. Venkatasubramanian, Transient stability assessment with
unstable limit cycle approximation, IEEE Transactions on Power
Yes Verification & Systems, vol. 14, May 1999, pp. 667 -677
Re < Tolerance ?
Output [11] Zhihong Feng, Venkataramana Ajjarapu, and Bo Long, Indentification of
Voltage Collapse Through Direct Equilibrium Tracing, IEEE
Transactions on Power Systems, vol. 15, no. 1, Feb2000. pp.342-349.
No
[12] Dmitry Kosterev, Bill Mittelstadt, Mike Viles, Brian Tuck, Jim Burns,
Mike Kwok, Jorge Jardim, and Gerry Garnett, Model Validation and
Yes
Index (0, index -limit]? Analysis of WSCC System Oscillations following Alberta Separation on
August 4, 2000, Final Report, January 25, 2000,
www.transmission.bpa.gov/orgs/opi/Wide_Area/August4Report.doc.
No [13] George Angelidis and Adam Semlyen, Improved methodologies for the
calculation of critical eigenvalues in small signal stability analysis IEEE
Fig 4.1 Flow chart for oscillatory stability margin estimation
Transaction on Power System, vol. 11, no. 3, August 1996, pp.1209-1217.
To make sure a particular eigenvalue A finally crosses [14] George Angelidis and Adam Semlyen, Efficient calculation of critical
imaginary axis first, maximum real part test at this crossing eigenvalue clusters in the small signal stability analysis of large power
systems, IEEE Transaction on Power System, vol. 10, no. 1, Feb 1995,
point is conducted. We have solution if there is no other pp.427-432.
eigenvalue with real part greater than A. If there is another [15] K. Kim, H. Schattler, V. Venkatasubramanian, J. Zaborszky, P. Hirsch,
eigenvalue which is already crossed the imaginary axis before Methods for calculating oscillations in large power systems, IEEE
A, then we have to apply backward tracing on this eigenvalue. Transactions on Power Systems, vol. 12, Nov. 1997, pp.1639-1648,
As a by-product, eigenvalue index (see (2.7)) and [16] Leonardo T.G. Lima, Licio H. Bezerra, Carlos Tomei,and Nelson Martins,
New method for fast small-signal stability assessment of large scale
eigenvalue derivative &Re can be also used in eigenvalue power systems, IEEE Transaction on Power System, vol. 10, no. 4, Nov
1995, pp.1979-1985.
analysis at any given operating point. Combined with
[17] Nelson Martins, Leonardo T.G. Lima, and Jerminio J.C.P. Pinto,
participation factor and eigenvalue sensitivity, we can get not
Computing dominant poles of power system transfer function, IEEE
only mode related to the least damping eigenvalue, but also Transaction on Power System, vol. 11, no. 1, Feb 1996, pp.162-170.
the mode related to the eigenvalue that moves fastest or most [18] N. Martins, The dominant pole spectrum eigensolver, IEEE Transactions
likely to cross the imaginary axis. on Power Systems, vol. 12, no. 1, Feb. 1997, pp. 245 -254
The eigenvector derivatives can be used to obtain [19] Jorge M. Campagnolo, Nelson Martins, Jose L.R. Pereira, Leonardo T.G.
participation factor derivatives. The participation factors are Lima, Herminio J.C.P. Pinto and Djalma M. Falcao, Fast small-signal
stability assessment using parallel processing, IEEE Transaction on
generally indicative of the relative participations of the Power System, vol. 9, no. 2, May 1994, pp.949-956.
respective states in the corresponding modes (or eigenvalues) [20] Jorge M. Campagnolo, Nelson Martins and Djalma M. Falcao,
[1]. So the derivatives of these factors can further show how Refactored Bi-Iteration: A high performance eigensolution method for
this association will change for any parameter variation of large power system matrices IEEE Transaction on Power System, vol. 11,
no. 3, August 1996, pp.1228-1235.
interest.
[21] Jorge M. Campagnolo, Nelson Martins and Djalma M. Falcao, An
efficient and robust eigenvalue method for small-signal stability
7

assessment in parallel computers IEEE Transaction on Power System,


vol. 10, no. 1, Feb 1995, pp.506-511.
[22] L. Wang, and A. Semlyen, Application of sparse eigenvalue techniques to
the small signal stability analysis of large power systems, IEEE
Transactions on Power Systems, vol. 5 no. 2, May 1990, pp. 635 642.
[23] M. Kubicek, M. Marek, Computational Methods in Bifurcation Theory
and Dissipative Structures, Springer Series in Computational Physics,
1983 by Springer-Verlag New York Inc.
[24] Yuri V. Makarov, Zhao Yang Dong, and David J. Hill, A general method
for small signal stability analysis IEEE Transaction on Power System,
vol. 13, no. 3, August 1998, pp.979-985.
[25] David J. Hill, and Zhao-Yang Dong, Nonlinear Computation and Control
for Small Disturbance Stability, Panel Session on Recent Applications of
Small-Signal Stability Techniques, Proc. of IEEE Summer Meeting,
Seattle, July 2000.
[26] S. Gomes Jr., N. Martins, and C. Portela, Computing small-signal
stability boundaries for large-scale power systems, IEEE Transactions on
Power Systems, , vol. 18, no. 2 , May 2003, pp. 747 -752
[27] Yuan Zhou; and V. Ajjarapu, Identification and tracing of voltage and
oscillatory stability margins, Power Engineering Society Winter Meeting,
2002. IEEE , vol. 1 , 27-31 Jan. 2002 pp. 250 -255
[28] R. T. H. Alden, and F. A. Qureshy, Eigenvalue tracking due to parameter
variation, IEEE Transaction on automatic control, vol. AC-30, no. 9,
September 1985, pp.923-925.
[29] Robert Kalaba, Karl Spingarn, and Leigh Tesfatsion, Variational
Equations for the eigenvalues and eigenvectors of nonsymmetric matrices,
Journal of Optimaization Theory and Applications: vol. 33, no. 1, Jan
1981, pp.1-8.
[30] Robert Kalaba, Karl Spingarn, and Leigh Tesfatsion, Individual tracking
of an eigenvalue and eigenvector of parameterized matrix, Nonlinear
Analysis, Theory, Methods & Application, vol. 5, no. 4, pp.337-340.

Das könnte Ihnen auch gefallen