Professional Documents
Culture Documents
AbstractHarmonic resonance is closely related to the singularity of a network admittance matrix. The smallest eigenvalue
of the matrix defines the mode of harmonic resonance. This
paper applies this eigenvalue theory and proposes a method to
determine which network components have significant contributions to a harmonic resonance phenomenon. The basic idea is to
calculate the sensitivities of a resonance mode to the parameters
of network components. The sensitivity results are then ranked
to quantify the impact of each component. In this paper, the
eigen-sensitivity theory as applied to harmonic resonance mode
analysis is presented. Case studies are used to verify the theory. A
practical example is given to illustrate the application of the proposed method. In addition, this paper further conducts extensive
comparative analysis on three types of network-oriented modal
analysis techniques. The results have clarified the similarities and
differences among the techniques.
Index TermsEigenvalue sensitivity, harmonic resonance, harmonics, modal analysis, power quality.
I. INTRODUCTION
Manuscript received October 25, 2005; revised July 4, 2006. This work was
supported by the Natural Science and Engineering Research Council of Canada
and was conducted at the University of Alberta. Paper no. TPWRS-00646-2005.
Z. Huang is with the Energy Science and Technology Directorate, Pacific
Northwest National Laboratory, Richland, WA 99354 USA (e-mail: zhenyu.
huang@pnl.gov).
Y. Cui and W. Xu are with the Department of Electrical and Computer Engineering, University of Alberta, Edmonton, AB T6G 2V4, Canada (e-mail:
wxu@ ece.ualberta.ca; yucui@ece.ualberta.ca).
Digital Object Identifier 10.1109/TPWRS.2006.883678
(2)
HUANG et al.: APPLICATION OF MODAL SENSITIVITY FOR POWER SYSTEM HARMONIC RESONANCE ANALYSIS
where
is the diagonal eigenvalue
and
matrix, and
are the left and right eigenvector matrices, respectively. Superscript indicates matrix transpose. The diagonal entries
of have the unit of admittance. Its inverse is called modal
. The th modal impedance is defined as follows:
impedance
223
(3)
A singular
matrix means that one of the eigenvalues of
the is close to zero, i.e., the corresponding modal impedance
approaches infinity. This eigenvalue is called the critical resonance mode at this particular frequency. It is this mode that
causes the observed resonance phenomenon. The above theory
can be further extended to determine the impact of various network components on the resonance mode. This can be achieved
using eigenvalue sensitivity (modal sensitivity) analysis as explained below.
For the th eigenvalue in (2) , its right and left eigenvectors
satisfy the following equations:
and
(5)
and applying the first two equations in
(6)
Equation (6) implies that the eigenvalue sensitivity to a parameter can be obtained by the matrix sensitivity to the same
parameter.
A. Eigenvalue Sensitivities With Respect to
is the
(10)
(4)
Suppose is a network component parameter, e.g., capacitance of a shunt. Differentiating the third equation in (4) with
respect to yields
Post-multiplying by
(4) , one can obtain
(9)
Matrix Entries
One of the immediate applications of (6) is to find the sensitivity of the critical mode with respect to a particular entry of
. Obviously, the derivative of matrix with
the matrix
has a value of 1 at the position of the entry and
respect to
zeros at all other positions
(7)
where
represents other shunt components at bus . Thus,
is
from (6), the eigenvalue sensitivity to
(11)
One can see that eigenvalue sensitivities with respect to shunt
network components are actually same as bus participation factors [1].
, and
are complex values. This is not
In (11), ,
desirable in resonance studies as the eigenvalue magnitudes are
of much greater concern. Moreover, the derivative to a complex
is also confusing. To address this problem, an altervalue
native sensitivity form is derived as follows. Assume that
224
Consider there is a
change
Let
Similarly
Convert
back to
(12)
(13)
The above sensitivity indexes reflect the relationship of critical eigenvalue magnitudes with respect to actual network components. Both of them are with actual units. Under certain situations, it could be more desirable to have normalized sensitivities that are sensitivity indexes with relative units instead
of actual ones, that is, 10% of modal impedance magnitude
increase with 1% of capacitor capacitance increase compares
capacitance
with 0.1 increase of modal impedance with 1
increase. Another advantage of using normalized sensitivities
is to make indexes with different components comparable. For
example, it may not make sense if one says the impact of 0.1
H inductance change is greater than that of 0.001 F capacitance
change, but it can reveal the relative impact by comparing 1%
inductor change with 1% capacitor change. Normalized sensitivities can be expressed as follows:
(17)
(14)
It can be seen that eigenvalue sensitivities with respect to series network components are certain combinations of the entries
, eigenvalue magniof the sensitivity matrix . Similar to
tude sensitivity with resistive and reactive components can also
be obtained. For a series network component, it is usually repinstead
resented by an equivalent impedance
of an admittance
. Thus, it will be more convalues. The
venient to obtain the sensitivity with the and
new form of sensitivity is derived as follows.
Assume that
where
and
represent the relative changes of the
eigenvalue magnitude and a network component parameter, respectively. Equation (17) shows that normalization can be done
by simply multiplying the sensitivity with the original value of
the network component parameter divided by the eigenvalue
magnitude.
D. Summary
Eigenvalue sensitivity concepts are introduced in this section,
including sensitivity matrix (sensitivity to matrix entries) and
sensitivity to actual network components. Normalized eigenvalue magnitude sensitivities with respect to shunt components
and series components are of the most interest for harmonic resonance studies and mitigation. The procedure to calculate these
sensitivity indexes is summarized as follows.
1) Find the sensitivity matrix using (9).
2) Obtain eigenvalue magnitude sensitivities by (12), (13),
(15), and (16).
3) Calculate normalized sensitivities with (17).
Normalized sensitivities are used throughout the remaining
parts of this paper.
III. CASE STUDY RESULTS
The IEEE 14-bus test system [5] is used to illustrate the application of the eigen-sensitivity theory to harmonic resonance
analysis. Fig. 1 and Table I show the resonance mode information of the test system [1]. In Fig. 1, each peak on the curves
indicated one resonance mode. Table I also shows the most participating bus for each resonance frequency, which were identified by the largest participation factors.
HUANG et al.: APPLICATION OF MODAL SENSITIVITY FOR POWER SYSTEM HARMONIC RESONANCE ANALYSIS
225
TABLE I
RESONANCE MODE INFORMATION OF THE TEST SYSTEM
Entries
Eigen-sensitivity studies have been carried out for all the resonance modes shown in Fig. 1. Due to limited space, only re, 10.7, and 25.3 p.u., which represent different
sults for
frequency ranges, are presented in Figs. 24. The figures show
the normalized sensitivities of critical eigenvalue magnitudes
with respect to series reactances or shunt capacitance. The bar
charts show the sensitivity indexes to shunt or series components in a descending order. Five components with the largest
indexes are labeled. The bubble charts provide the same information in a more visualized way with the topology of the IEEE
14-bus system as background. Bubbles on a bus or a line indicate whether the sensitivity index is with respect to a shunt or
a series component. The size of a bubble is in proportion to the
magnitude of the sensitivity index.
From these figures, the following observations can be
obtained.
Harmonic resonance modes are mainly affected by only
a few components, while most components have little or
insignificant impact on the resonance modes.
The sensitivities of critical eigenvalue magnitudes with respect to network components may not always be consistent
with bus participation factors. It is true that buses with large
participation factors are generally heavily involved in a resonance condition. However, this does not mean that the
resonance can be effectively mitigated by adjusting components at those buses.
226
HUANG et al.: APPLICATION OF MODAL SENSITIVITY FOR POWER SYSTEM HARMONIC RESONANCE ANALYSIS
227
TABLE IV
MODAL IMPEDANCE CHANGE WITH RESPECT TO 5% COMPONENT
PARAMETER INCREASE AT h = 5:1 p:u:
Fig. 10. Participation factors and normalized sensitivities for the resonance
mode h = 5:65 p:u.
Fig. 9. Resonance path for the resonance mode h = 5:10 p; u.
Fig. 8 shows participation factors and normalized eigen-senresonance mode. The participasitivities for the
tion factor chart reveals that Bus 3 also has a noticeable participation in this mode, which implies that the resonance can be
excited and observed at bus 3 to some extent. The modal sensitivity chart provides more insights for this observation. The
chart suggests that the path consisting of transformers 4, 3, and
5 resonates with capacitors 3 and 4, with transformer 4 and capacitor 4 as the main resonating pair. This situation is depicted
in Fig. 9.
Capacitors 3 and 4 have the largest sensitivities so they are
resthe best candidates to adjust to mitigate the
onance. Changing the parameters of transformer 4 (such as its
size and short-circuit impedance) will also work, but it is not a
common practice. Table IV shows the impact of 5% component
.
parameter increase on the modal impedance at
One can see that the modal impedance change is consistent
with the sensitivity results shown in Fig. 8. As a further validation, the combined change of driving point impedances at
buses 3 and 4, both of which are heavily involved with the
resonance, is also given in Table IV. Driving
point impedance is a direct indicator of harmonic resonance
severity. The consistency between the modal impedance change
and the driving point impedance change further confirms the
usefulness of the eigen-sensitivity information for harmonic resonance mitigation.
, the most participating bus is bus 3. DeFor
tailed participation factor and sensitivity results show that this
mode involves both buses 3 and 4 almost equally (see Fig. 10).
The components contributing most to the resonance are capacitors 3 and 4 as well as transformers 3 and 4. The resonance path
is shown in Fig. 11. This mode has a frequency away from har-
Fig. 11. Resonance path for the resonance mode h = 5:65 p:u.
228
Fig. 12. Participation factors and normalized sensitivities for the resonance
mode h = 6:20 p:u.
Fig. 13. Resonance path for the resonance mode h = 6:20 p:u.
(20)
determination of network eigenvalues have been published. The
first type uses the state-space model of an RLC network and
analyzes system resonance characteristics based on eigenvalues
of the state matrix. This is a well-known technique originally
developed to study the transient performance of a network [2],
[4]. Reference [3] adopted it for harmonic analysis, and [7]
attempted to improve its applicability to power systems. The
second type conducts eigen-analysis on the
matrix at harmonic frequencies only [10], [11]. With the development of the
resonance mode analysis method, it has become necessary to
determine the similarities and differences among the methods.
The objective of this section is to answer such questions.
A. State-Space Model-Based Technique
The state-space analysis of a linear network has been well
documented in various textbooks. It is the foundation for network transient analysis. According to the theory, the state variables of a linear network are the currents of inductors and voltages of capacitors. A power system can thus be modeled as
(18)
where is the state vector, the state matrix, the excitation
vector, and is a matrix representing contribution of to state
variable derivatives. This equation can be solved to determine
the transient responses of a network. The solution has the following form:
can only be determined using the eigen-decomposition technique. The results, eigenvalues of , are called eigen-modes of
(21)
It can be seen that
actually determines the transient
, the frequency of transient reresponse of the system.
sponse, represents the system natural frequency.
reflects
the decaying speed of the transient component and is called the
damping coefficient. Inverse of
is the time-constant of the
network. As a result, one can conclude that
is a parameter
associated with the transient responses of a network. Accordingly, a more suitable topic to apply the matrix concept is
the transient disturbances rather than harmonics, which is a
steady-state phenomenon.
It is important to note if a sinusoidal excitation has a frequency close to
, the system tends to exhibit a stronger
response. This can be understood by analyzing parameter
,
which can be rearranged as
(22)
Equation (22) reveals that
can become quite large if approaches
. If
,
can approach infinity. This is another way of explaining resonancea resonance occurs when
the frequency of excitation approaches the natural frequency of
a system. From this perspective,
can also be applied to investigate resonance phenomenon.
The modes defined by the proposed resonance mode analysis
technique are characterized by two parameters: the frequency
HUANG et al.: APPLICATION OF MODAL SENSITIVITY FOR POWER SYSTEM HARMONIC RESONANCE ANALYSIS
229
Fig. 16. Modal scan result for the five-bus test system shown in Fig. 6.
where the network driving point impedance approaches maximum and the impedance at that frequency
The results are in (23) and (24), shown at the bottom of the page,
where subscript indicates that the modes are determined
matrix.
is the frequency of the resofrom the network
nance mode, and
is the magnitude of the modal impedance.
represents the degree of amplification of a current excitation.
.
This pair of parameters represents resonance mode
It can be seen from (23) that
. The two frequen. Equation (24)
cies become equal only when damping
indicates that
has no simple relationship with the damping
coefficient . This analysis suggests that there is no one-to-one
correspondence between
and . Knowledge with one mode
does not necessary lead to understanding the other. This is especially true for the damping parameter
and . The difference
between the two modes has also been compared numerically.
Fig. 15 shows the ratio
as a function of
. It can be
seen that the difference increases noticeably when
is large.
The same figure also plots a scalar product
. Note that a
larger
represents a smaller damping in the harmonic domain,
and a smaller
represents a smaller damping in the transient
domain. So
is a measurement of the difference between
the two damping parameters. The result shows that
is a
constant only for very small damping values.
Although the above analysis is illustrated by a simple sample
system, the conclusions are general. This is because the sample
system represents one pair of modes for multistate or multinode
systems. From this comparative analysis, one can see that
and
are different concepts, although they are related to each
other. Each concept has its own physical meaning and explanation. When system damping is small, one can be used to approximate the other. It is worthwhile to note that the differences
between these two concepts have been recognized in modern
circuit analysis and are a subject of active research [12], [13].
(23)
(24)
230
where
means the adjunct matrix of
. With this
transformation, the modes can also be understood as the poles
of a system. The transfer function between an input such as a
nodal current and an output such as component voltage can be
expressed as
where
denotes zeroes of the transfer function. Unlike the
poles, zeros are input and output dependent. They represent the
modes of series resonance. It can be seen that besides the zeroes,
the -domain and transfer function formulation does not offer
any new information. They just represent an alternative view on
the transient modes of a network. In fact, eigen-decomposition
of the matrix is still the most effective way to find poles and
zeros. An example is the small-signal stability analysis program
SSAT [14].
The resonance mode analysis technique can also reveal the
modes of series resonance. For example, if one wants to check
if a series resonance can be excited from node 1, a voltage source
with a magnitude of 1.0 per-unit can be applied to this node (see
Fig. 17).
The nodal matrix of Fig. 17 can be rearranged as
The modes of
To directly solve the above highly nonlinear equation is an extremely complex task. Therefore, [8] formulates the problem in
the form of a transfer function between a current injection and
a bus voltage and then applies the NewtonRaphson method to
analysis provides same eigenvalue incalculate poles. This
formation as the eigen-analysis on the matrix. It is also important to point out that the eigenvectors associated with the
have different meaning from those of the
matrix. In the
case, the
and
vectors do not represent sinusoidal responses since the real part of is not zero.
3) Modeling Limitations: The state-space-based formulation has at least two modeling limitations. One is the modeling
of distributed parameter lines. To our knowledge, there is no
method available to include such a model in the state-space
formulation was developed to address this
matrix. The
limitation. Unfortunately, the approach makes the numerical
solution of modes even more complicated.
Another limitation is to model frequency-dependent imped. Such impedances are common in harmonic analances
ysis. For example, a supply system can be represented by such
an impedance. It is not possible to directly model such frematrix or
maquency-dependent components in the
trix formation. One has to create equivalent RLC networks first
using network synthesis techniques.
C.
HUANG et al.: APPLICATION OF MODAL SENSITIVITY FOR POWER SYSTEM HARMONIC RESONANCE ANALYSIS
231
that the
technique cannot perform harmonic resonance
analysis. Notwithstanding the above problems, the technique
might support harmonic power flow analysis by identifying the
impact of harmonic load currents on bus voltage distortions.
D. Summary
Based on the extensive analysis and comparison presented in
this section, the following main conclusions are obtained.
The matrix-based modal analysis is more suitable to investigate the propagation of transients in a network.
matrix-based modal analysis is the best choice for
The
harmonic resonance analysis.
matrix-based modal analysis is not suitable for
The
harmonic resonance analysis. It might be useful for harmonic power flow analysis.
matrix-based modal analysis does not offer more
The
information than already available from the matrix or
matrix-based analysis. Its computational burden is actually
more than other methods.
VI. CONCLUSIONS
This paper proposes and develops a modal sensitivity technique to assist the analysis of power system harmonic resonance. Modal sensitivity concepts are introduced, including sensitivity matrix and sensitivity to actual network components,
and their normalized forms. Sensitivity indexes can help identify critical network components involved in each resonance
mode. Such information is very useful for designing strategies to
mitigate harmonic resonance problems. This paper further conducted comparative analysis on the three types of network-oriented modal analysis techniques. The main contributions and
conclusions of this paper are summarized as follows.
The most useful sensitivity information is the normalized
sensitivity of critical modal impedance (magnitude) with
respect to the component parameters. This information can
help identify the components most involved in a particular
resonance mode.
Sensitivities of the critical eigenvalues to the diagonal elements of the admittance matrix are shown to be modal
participation factors. One can therefore conclude that participation factors are a type of modal sensitivity.
Comparative analysis of three modal analysis techniques
shows that the state-space-based modal analysis is more
suitable to analyze the propagation of transients in a
Yu Cui received the B.Eng. degree from Tsinghua University, Beijing, China,
in 1995, the M.Sc. degree from the Institute of Electrical Engineering, Chinese
Academy of Sciences, Beijing, in 2000, and the M.Sc. degree from University
of Saskatchewan, Saskatoon, SK, Canada, in 2003. He is currently pursuing the
Ph.D. degree at the University of Alberta, Edmonton, AB, Canada.
His research areas include power quality and distributed generation.