This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE SYSTEMS JOURNAL 1
A Coherency-Based Approach for Signal Selection
for Wide Area Stabilizing Control in Power Systems
Bibhu Prasad Padhy, S. C. Srivastava, Senior Member, IEEE, and Nishchal K. Verma, Member, IEEE
AbstractIn this paper, a systematic procedure to select wide-
area input/output control signals based on a signal coherency
approach to damp out inter-area mode of oscillations is proposed.
The coherent signal groups are chosen based on a data clustering
approach. The input/output signal selection is carried out in two
steps. First, the data is transformed into orthogonal space to
make correlated variables uncorrelated, by applying principal
component analysis. Then, the principal component vectors are
given as input to a self-organizing map for nal data clustering.
From each clustered signal group, a signal out of those with
common features is selected. For clustering, the data are collected
from the system dynamic simulation considering few critical line
contingency. The efcacy of the proposed feature selection based
method is veried by comparing the results with a geometric-
based approach and a pole vector direction based signal selection
approach on 39-bus New England and 68-bus New England New
York power systems.
Index TermsInputoutput selection, inter-area oscillations,
principal component analysis, robust controller, wide area
controller.
I. Introduction
W
ITH THE GROWING electricity demand and the aging
utility infrastructure, the present-day power systems
are operating close to their maximum transmission capacity
and stability limit. One of the main steps is to employ wide
area coordinated controllers to minimize blackouts and, hence,
to enhance the system security. In the past few decades, the
angular instability, caused by small signal oscillations, has
been observed in the power systems under certain system con-
ditions, such as during the transmission of a large amount of
power over long distance through relatively weak tie lines and
under use of high gain exciters. These instability conditions
make the system vulnerable to the blackouts. The traditional
approach to damp out these oscillations is through conven-
tional power system stabilizer (CPSS) [1], forming part of the
generator excitation system or exible ac transmission system
controllers. These controllers usually employ local signals as
inputs and may not always be effective to damp out the inter
area modes of oscillations. Furthermore, the local controllers
lack the global observation. The synchrophasor technology
Manuscript received March 27, 2012; revised August 30, 2012; accepted
February 20, 2013. This work was supported in part by the Department of
Science and Technology, New Delhi, under Project DST/EE/20100258.
The authors are with the Department of Electrical Engineering, Indian
Institute of Technology Kanpur, Kanpur 208016, India (e-mail: bibhupee@
iitk.ac.in; [email protected]; [email protected]).
Color versions of one or more of the gures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/JSYST.2013.2249241
based wide area monitoring and control system forms an
important part of the smart grid to enhance the grid security
[2]. This employs phasor measurement units, which provide
synchronized and time stamped real-time measurement data
from remote locations to a phasor data concentrator at the
control center. These measurements can be effectively utilized
to design wide area damping controllers [3][8]. Selection
of input/output control signals is an important issue, while
designing an effective wide area controller. Wide area signals
to be selected should have a better damping effect on the
critical inter-area modes of oscillations. In recent years, few
methods have been proposed for this propose. A comparative
study of geometric and residue approach for selection of the
wide area signals has been provided in [9], where input/output
signals have been selected based on joint controllability and
observability index. This paper also concludes that the geo-
metric approach performs better than the residue approach. A
relative gain array based method has been proposed in [4],
and [10][12]. In all these methods, the signal is selected for
the intact system and at an operating point, which may be
quite different during large change in the system operating
conditions or under large disturbance in the system.
In this paper, coherent generator groups have been identied
by using a cluster algorithm, and only one generator in each
group has been selected to receive output signal from the con-
troller. The generator has been selected in each group based on
largest variation in the speed deviation subject to a disturbance,
which can be considered the weakest generator in the group.
The electric power ow in the tie lines has been considered
input signal to the controller. In this paper, from a large set
of input signals, few signals are selected based on a feature
selection approach using cluster algorithm. The signals with a
similar control property can be clustered together. One signal
from each cluster has been selected and the others provide
redundant information. A combination of principal component
analysis (PCA) [13] and self-organizing map (SOM) [14] has
been used in a sequence to identify these cluster groups. First,
the nonlinear data vectors of speed deviation of generators and
power line deviations are converted into linear orthonormal
vectors by applying the PCA. The SOM has been used to
make the nal cluster group, which has the advantage of
being mathematically simple and computationally efcient to
implement.
This paper is organized as follows. Sections II and
III describe the mathematical theory for the pole vector
and geometric approaches. Section IV provides the detailed
1932-8184/$31.00 c 2013 IEEE
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
2 IEEE SYSTEMS JOURNAL
description of the proposed method for signal selection. In
Section V, implementation results of the proposed method for
signal selection on 39-bus and 68-bus power systems have
been presented. Section VI describes the results for the pole
vector based signal selection approach. Section VIII describes
the results for the validation of the proposed signal selection
approach.
II. Selection of Wide-Area Control Signal by
Using Pole Vectors
In this approach [15], the classical concepts of state con-
trollability and observability are used to select inputs such
that the unstable or critical states can be easily controlled,
and also to select the outputs such that the unstable or critical
states are easily observed. This motivates one to consider the
input pole vectors and the output pole vectors for selecting
the input and the output signals to the stabilizing controller.
The approach also makes it possible to consider the inputs
based on state controllability and the outputs based on state
observability, separately. The state space model of a power
system, in general, can be written as
x = Ax+Bu
y = Cx
(1)
where x R
n
is the state vector, u R
M
is the input vector,
y R
P
is the output vector, A R
nn
is the state matrix,
B R
nM
is the input matrix, and C R
Pn
is the output
matrix of the system. Assume that the A matrix has
i
and
i
as the left and right eigenvectors, respectively, associated with
an ith mode of oscillation. The input and output normalized
pole vector P
i
, associated with the ith mode of oscillation,
can be dened as
u
P
i
=

B
T
j

i

_
_
B
T
j

i
_
_
(2a)
and, y
P
i
=

C
j

_
_
C
j

i
_
_
(2b)
where |Z| and Z are the modulus and Euclidean norm of Z.
If B
T
= 0, the corresponding pole is not state controllable,
and if C = 0, the corresponding pole is not state observable.
The magnitudes of the elements in the input pole vector u
P
i
give information about the input through which the ith mode
is the most controllable, and the magnitude of the elements in
the output pole vector y
P
i
give information about the output
through which the ith mode is the most observable. Hence,
in a multi-input, multi-output plant, the pole vectors provide
a very useful tool for selecting the inputs and the outputs for
the wide-area controller design. For selection of the inputs and
the outputs, the following procedure has been applied.
1) Scale the inputs and the outputs such that a unit change
in each input and output are of equal importance in the
objective.
2) Calculate the input pole vector u
P
i
and the output pole
vector y
P
i
for the ith mode of oscillation.
3) Select inputs corresponding to the largest value of the
input pole vector u
P
i
.
4) Select outputs corresponding to the largest value of the
output pole vector y
P
i
.
III. Selection of Wide-Area Control Signal by
Using a Geometric Approach
A joint controllability/observability index has been used for
selection of the stabilizing signals. The controllability index
(COI
j
) and observability index (OBI
k
) for the ith mode can
be dened [16] as
COI
j
= cos((
i,
B
j
)) =

B
T
j

i

_
_
B
j
_
_

(3a)
and, OBI
k
= cos((
i,
C
k
)) =

C
j

_
_
C
j
_
_

(3b)
where B
j
is the jth column of matrix B and C
k
is the kth row
of the matrix C. The joint controllability/observability index
can be written as
Joint

index
jk
= COI
j
OBI
k
. (4)
A high value of the index means better suitability of the
signal to be selected. A detailed discussion of this method has
been presented in [6], [9], and [16].
IV. Proposed Method for Wide-Area Signal
Selection
A. Dynamic Contingency Ranking Index
The contingency analysis is required to be performed to
assess the power system security. The security analyses mostly
consider the steady-state operating limits following a contin-
gency and do not, generally, consider the dynamic stability of
the power system. In this section, a dynamic contingency index
has been proposed to select the critical line contingencies
in the system. The proposed index considers the stability
condition of the power system based on relative change in
the damping ratio of the system following a contingency. The
proposed dynamic contingency index is dened as percentage
decrease in critical damping ratio from the intact system condi-
tion to when it is subjected to a contingency j. Mathematically,
this index is dened as
DCI
J
=

I
sys

J
sys

I
sys
(5)
where
I
sys
is the critical damping ratio for the pre-fault
operating condition with mode of oscillation having frequency
less then 2 Hz and
J
sys
is the critical damping ratio for
the post-fault operating condition with modes of oscillation
having frequency less then 2 Hz following a contingency J.
A owchart for selection of the critical line contingencies is
given in Fig. 1.
B. Principal Component Analysis (PCA)
PCA [13], [17] is a mathematical tool that applies an orthog-
onal transformation to convert a set of correlated variables into
a set of uncorrelated variables called principal components.
The number of principal components is less than or equal
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
PADHY et al.: COHERENCY-BASED APPROACH FOR SIGNAL SELECTION 3
Fig. 1. Flowchart for selection of critical line contingencies.
to the number of original variables. This transformation is
dened in such a way that the rst principal component has as
high a variance as possible, and each succeeding component,
in turn, has a lower variance as compared to the preceding
components and it is orthogonal to the preceding components.
The principal components can be derived from the singular
value decomposition (SVD) as follows.
Consider a matrix X, of dimension (mn), where m is the
number of samples and n the number of variables. The PCA
transforms the data matrix X, as a linear weighted sum, given
by
X = SP
T
(6)
where S is dened as principal component scores hav-
ing dimension of (m m). The ith column of S can be
written as ( s
1,i
s
2,i
..... s
m,i
)
T
and P is dened as
the principal component loading, which are the normalized
eigenvectors of the matrix X
T
X. The ith row of P
T
is
( p
1,i
p
2,i
..... p
m,i
)
T
. The data matrix, X, can be de-
composed as follows by using SVD:
X = UWV
T
(7)
where the matrix W is diagonal and its elements are the
positive square roots of the covariance matrix, X
T
X, from
(6) and (7), the principal component scores, S = UW and
P
T
= V
T
. The scores are the data formed by transforming the
original data into the space of the principal components. The
detailed description of the PCA can be found in [13].
C. Data Clustering Using SOM
Kohonens SOM [14] is the most widely used unsupervised
learning neural network approach to data clustering. The basic
structure of SOM is that it consists of a layer of input units to
receive input patterns and propagates to the output units. The
algorithm for the Kohonens SOM is as follows.
1) Set the iteration number K = 0. Initialize the weight
of all the neurons, W, with small random value. Also,
initialize the learning rate and set the total number of
iterations equal to say T.
2) Repeat the steps from 3 to 6 if K is less than T.
3) Select an input vector X
i
from the input training data
set.
4) Compute the Euclidian distance between X
i
and the
entire weight vector. The winning neuron unit N
j
will be
the closest to the input vector X
i
. The closest distance
D
min
(K) can be calculated as
D
min
(K) = min
i
(D
min
(K)) = min
i
_
_
X
i
(K) W
N
j
(K)
_
_
.
5) Update the weight for both the winning neuron unit N
j
and its neighbor neuron units as
W
N
j
(K+1) = W
N
j
(K)+(K)(K){X
i
(K)W
N
j
(K)}
where (K) is the neighborhood kernel function and
(K) = exp
_

D
2(K)
2
_
. D is the Euclidian distance be-
tween the neighborhood and the winning node. The (K)
and (K) are the learning rate and neighborhood radius
function, which exponentially decreases with time. The
mathematical expressions used for (K) and (K) are
given by
(K) =
0
exp
_

K
T
_
and (K) =
0
exp
_

K
T
_
.
6) Set K = K + 1 and if K < T, go to step 3, else STOP.
The network of the SOM, after adequate self-training, tends
to be spatially organized and structured according to the input
pattern of the data set applied. The location of the neuron cells
will be ordered in accordance with the topological relations
among all the patterns in the input set.
V. Implementation of the Proposed Method on
39-Bus and 68-Bus Power System
A. Calculation of Contingency Ranking Index
For the dynamic contingency ranking, only the single-
line outage contingencies are considered in this paper. The
values of the DCI
J
index and relative ranking of the line
contingencies for the two systems are given in Table I. Table
I shows that in the 39-bus New England system, outage of
line connecting buses 28 and 29 is the most severe and in the
68-bus New England New York power system, the outage of
line connecting buses 1 and 2 is the most severe. The results
also show that, for the 39-bus New England system, the index
value for line outages L
12
, L
139
, L
2829
are quite close to
each other. Hence, these contingencies are also considered for
the data generation in the 39-bus system for further analysis.
Similarly, in the 68-bus New England New York power system,
index values for line contingencies L
89
and L
225
are also
high. Hence, these contingencies are also considered for the
data generation in this system.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
4 IEEE SYSTEMS JOURNAL
TABLE I
Dynamic Contingency Index Values
Line Outage Index Value
L
2829
1.2463
39-bus New L
139
1.2073
England System L
12
1.1953
L
225
0.3666
L
1621
0.1280
68-bus New L
12
0.096
EnglandNew L
89
0.064
York System L
225
0.058
L
2122
0.039
L
1621
0.0374
Fig. 2. Gen rotor speed in a 39-bus system for line L
12
outage.
Fig. 3. Gen rotor speed in a 68-bus system for line L
12
outage.
B. Calculation of the PCA Vectors
The required data set has been generated for the two sys-
tems through dynamic simulation of the critical contingencies
found, as discussed in an earlier section. These data have been
acquired at a sampling rate of 0.01 s, and about 3000 samples
have been collected for speed variation of each generator and
power ow in all the interconnected tie-line lines. The plots of
all the generators rotor speed in the 39-bus and the 68-bus sys-
tems, after the application of the critical contingency, are given
in Figs. 2 and 3, respectively. The generator speed data matrix
X, with the application of each critical contingency has been
constructed with 3000 rows and ten columns, i.e., having size
300010 and the line power ow matrix having size 300033
for the 39-bus system. Similarly, for the 68-bus New Engalnd
and New York interconnected power system, the generator
speed and the power ow data matrix have the size 300016
and 300064, respectively. The PCA has been applied to
Fig. 4. Speed deviation PCA score vectors in a 39-bus system for line
contingency L
12
.
Fig. 5. Speed deviation PCA score vectors in a 68-bus system for line
contingency L
12
.
calculate the principal component scores S and the matrix W.
The PCA score vectors have been truncated to three dimen-
sions, and the score plot for the generator speed data and the
power ow data considering line contingency L
12
, in both the
39-bus and the 68-bus systems, are shown in Figs. 47, respec-
tively. These truncated PCA score vectors S have been given
as input to the SOM network to select the nal control signals.
C. Application of SOM and Control Signal Selection in
39-Bus and 68-Bus Systems
For detection of coherent groups, the generator PCA score
vectors matrix of dimension 3 10 and line power PCA
score vectors matrix of dimension 333 have been applied as
input to the SOM network in the case of the 39-bus system.
Similarly, in the 68-bus New England and New York intercon-
nected power system, the generator PCA score vectors matrix
of dimension 316 and line power ow matrix of dimension
of 364 have been applied to the SOM network. The neuron
dimension of the SOM network for generator and line power
clustering has been taken as 3-D lactic space, and the number
of neurons has been decided by the number of cluster groups
to be formed. The results of the generator cluster groups in the
case of the 39-bus power system for the three critical contin-
gencies applied are shown in Table II. This table shows that,
for all the three cases, the number of cluster groups has been
found to be ve and the cluster groups are almost the same
for all three contingency cases. Only one of the generators
has been selected from the cluster of the generator groups
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
PADHY et al.: COHERENCY-BASED APPROACH FOR SIGNAL SELECTION 5
Fig. 6. Power line PCA score vectors in a 39-bus system for line contingency
L
12
.
Fig. 7. Power line PCA score vector in a 68-bus system for line contingency
L
12
.
based on the truncated PCA score vector length. The generator
with the highest PCA score vector length has been selected
from the group. These generators are highlighted in bold italic
underlined letters in Table II. Thus, the generators selected are
G3, G5, G9, and G10. The generator G1 has not been selected
due to its size and large inertia constant of 50 at 1000-MVA
base, which is approximately ten times larger as compared to
the other generators present in the system. It has been veried
from Fig. 2 that due to the line contingency L
12
, the generator
G1 oscillates against the rest of the generators in the system.
The line power cluster groups for the 39-bus system are shown
in Table IV. The lines with the highest PCA score vector length
are highlighted in bold underlined italic letters in this table. It
has been found that ve cluster groups have been formed with
the application of the contingencies. The signals, which are
selected, have the highest truncated PCA score vector length
within the cluster group. It has been found that the number
of feature signals becomes large even by selecting one signal
from each cluster group, which may increase the complexity
of the controller implementation. Hence, to reduce the number
of feature signals, some of the most common signals with high
frequency of occurrence within the cluster groups have been
selected. By applying this methodology, the signals selected
are L
1727
, L
1619
, L
939
, L
225
, and L
89
. In the 68-bus
system, the best feature group has been selected by picking
up a generator from the clusters based on the length of the
truncated PCA score vectors. Thus, the generators selected are
G1, G3, G5, G9, G13, and G15. The results of the generator
TABLE II
Generator Cluster Groups 39-Bus System
Cluster contingency
no(Gen.)
L
12
L
139
L
2829
1 G
1
G
1
G
1
2 G
2
,G
3
,G
4
,G
5
, G
6
,G
7
G
9
G
4
,G
5
,G
6
,G
7
3 G
9
G
4
,G
5
,G
6
,G
7
G
2
,G
3
4 G
10
G
8
,G
10
G
8
,G
10
5 G
8
G
2
,G
3
G
9
TABLE III
Generator Cluster Groups 68-Bus System
Cluster contingency
no(Gen.)
L
12
L
89
L
225
1 G
2
,G
3
,G
4
,G
5
,G
6
,G
7
G
4
,G
5
,G
6
,G
7
G
4
,G
5
,G
6
,G
7
2 G
10
,G
11
,G
12
,G
13
G
14
,G
15
,G
16
G
9
,G
10
3 G
9
G
2
,G
3
G
11
,G
12
,G
13
,G
14
,G
15
,G
16
4 G
2
,G
8
G
2
,G
8
G
2
,G
8
5 G
14
,G
15
,G
16
G
9
,G
10
,G
11
,G
12
,G
13
G
2
,G
3
cluster groups are shown in Table III. The lines with the high-
est PCA score vector length have been highlighted with bold
italic underlined letters in this table. Similarly, by applying
the above methodology, as discussed for the 39-bus system,
the input control signals selected for the 68-bus system are the
power ow in the lines L
3637
, L
12
, L
930
, L
5052
, L
936
, and
L
45
. The results of the cluster groups are shown in Table VI.
The lines with the highest PCA score vector length have been
highlighted with bold italic underlined letters in this table.
VI. Results for Control Signal Selection by
Using Pole Vectors
The most critical modes of oscillation under nominal pre-
fault operating conditions for the two systems studied are
given in Table I. A brief description of these systems is given
below.
A. New England 39-Bus System
The New England 39-bus power system, considered for the
study in this paper, is shown in Fig. 8. The line parameters,
generators and exciters data used in the system are taken from
[18]. This system consists of ten generators and each generator
is assumed to be provided with governor, AVR, and IEEE
ST1A type static exciter. Generators 3, 7, 8, and 9 are assumed
to be provided with CPSS.
B. New England and New York Interconnected 68-Bus Power
System
The New England and New York interconnected 68-bus
power system, considered in this paper, is shown in Fig. 9.
The data used are taken from [19]. This system consists of
16-generators. Each generator is assumed to be provided with
governors, and IEEE ST1A type static exciter. Generators 3,
7, 9, 13, and 16 are assumed to be provided with CPSS.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
6 IEEE SYSTEMS JOURNAL
TABLE IV
Line Power Cluster Groups for 39-Bus System
contingency
Cluster no
1 2 3 4 5
L
12
L
12
L
414
, L
56
, L
1624
, L
34
, L
45
, L
58
, L
139
, L
225
, L
1013
, L
611
, L
939
, L
1011
,
L
1727
, L
2324
, L
67
, L
78
, L
89
, L
1621
, L
2122
, L
2223
, L
1314
, L
1619
L
1415
, L
1516
, L
2627
, L
2628
, L
2629
,
L
2829
L
139
L
318
, L
1013
, L
2526
, L
225
, L
34
, L
58
, L
67
, L
78
, L
12
, L
414
, L
56
, L
23
, L
611
, L
939
, L
1011
,
L
1415
, L
1516
, L
45
, L
2627
, L
2628
, L
89
L
1314
, L
1619
, L
1621
, L
1617
, L
1624
, L
1727
,
L
1718
, L
2526
L
2629
L
2829
L
2122
, L
2223
L
2324
L
2829
L
611
, L
939
, L
225
, L
34
, L
45
, L
318
, L
1415
, L
414
, L
1624
, L
1727
, L
12
, L
139
, L
58
, L
56
,
L
1011
, L
2627
, L
1013
, L
1621
, L
2122
L
1516
, L
1617
, L
2324
L
67
, L
78
, L
89
, L
1314
,
L
2629
L
2223
, L
2526
, L
2628
L
1718
L
1619
Fig. 8. New England ten-machine power system showing selected input and
output signals with the proposed method.
The pole vector input and output indices have been calcu-
lated by using 2(a) and (b), as discussed in Section II. These
values have been normalized with respect to the highest index
value. In the case of the 39-bus power system, critical mode of
oscillation, with a frequency of 0.9223 Hz, as given in Table V,
has been considered to calculate the input and the output pole
vector index. The input and the output pole normalized vector
indices are shown in Fig. 10. From the results obtained with
the proposed signal selection approach, as given in Section V,
the ve-input and four-output signals have been selected. The
numbers of inputs/outputs were decided since the set of line
power ows, in 39-bus system taken as the input, got clustered
into ve groups, and the set of generator speed deviation, taken
as output to controller, got clustered into four groups. Only one
signal from each group has been taken as the input/output to
the controller. For the sake of fair comparison, the numbers of
input and output signals have been restricted to ve and four,
respectively, using the pole vector and geometric approaches.
Also, the line power ows selected as inputs and generators
as outputs are different with the three methods in the 39-
bus system (accordingly, in the 68-bus system, the number
of input and output signals have also been restricted to 6 and
6, respectively).
Fig. 9. New England and New York interconnected system, showing selected
input and output signals with the proposed method.
TABLE V
System Critical Modes of Oscillations
Test Systems
Critical
Eigenvalues
Damping Frequ-
Mode No (%) ency (Hz)
39-bus New 1 0.2383 5.7950i 4.11 0.9223
England 2 0.2436 3.7536i 6.48 0.5974
System 3 0.3481 2.0749i 16.54 0.3302
68-bus New 1 0.1824 3.6455i 4.99 0.5801
England 2 0.2731 4.1370i 6.58 0.6584
New York 3 0.3569 2.1748i 16.19 0.3461
The signal selection using a geometric approach is already
discussed in [6]. The results of the 39-bus system show that
generators G
7
and G
9
are contributing most to this mode of
oscillation, with index values of 1 and 0.98, followed by G
4
and G
5
, with index values of 0.369 and 0.45, respectively.
Hence, these four generator groups are selected to receive the
control output signals. Similarly, the signals P
89
and P
12
are
contributing most to the mode of oscillation with the highest
value of index 1, followed by the signals P
23
, P
1415
and
P
318
. Hence, these ve signals are selected as inputs to the
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
PADHY et al.: COHERENCY-BASED APPROACH FOR SIGNAL SELECTION 7
TABLE VI
Line Power Cluster Groups 68-Bus System
Contin- Cluster No
gency
1 2 3 4 5 6 7
L
12
L
45
, L
611
, L
89
,
L
1011
, L
1314
,
L
936
, L
127
L
1617
, L
1619
, L
1621
,
L
1920
, L
2122
L
23
, L
225
, L
2526
,
L
3436
, L
4748
, L
4048
,
L
3743
, L
4344
, L
4445
,
L
4551
, L
4252
, L
4041
L
34
, L
414
, L
56
,
L
1727
, L
2324
, L
2627
,
L
2628
, L
2629
, L
2829
,
L
3334
, L
3031
, L
3338
,
L
3945
L
58
, L
67
, L
78
,
L
1013
, L
1112
, L
1213
,
L
1415
, L
1516
, L
2223
,
L
3032
, L
131
, L
147
,
L
3944
, L
4142
L
318
, L
1624
, L
1718
,
L
3435
, L
3233
, L
3138
,
L
3846
, L
4649
, L
3545
,
L
5052
L
130
,
L
930
,
L
3637
L
89
L
12
, L
23
, L
225
,
L
1617
, L
2526
, L
3436
,
L
4748
, L
4048
, L
3743
,
L
4344
, L
4445
, L
4551
,
L
4252
, L
4041
L
414
, L
56
, L
1112
,
L
1213
, L
1621
, L
1727
,
L
2122
, L
2223
, L
2324
,
L
2627
, L
2628
, L
2629
,
L
2829
, L
3334
, L
3031
,
L
131
, L
3338
, L
147
,
L
3945
, L
4142
, L
127
L
45
, L
611
, L
1011
,
L
1314
, L
1619
,
L
1920
, L
936
L
130
, L
930
L
318
, L
1624
, L
1718
,
L
3435
, L
3233
, L
3032
,
L
3138
, L
3846
, L
4649
,
L
3545
, L
3944
, L
5052
L
34
, L
58
, L
67
,
L
78
, L
1013
, L
1415
,
L
1516
, L
3637

L
225
L
130
, L
23
, L
318
,
L
930
, L
1718
, L
1727
,
L
2526
, L
2627
, L
930
L
1415
, L
3637
L
58
, L
67
, L
78
,
L
1013
, L
1516
, L
1624
,
L
3435
, L
3233
, L
3032
,
L
3138
, L
3846
, L
4649
,
L
3545
, L
5052
L
45
, L
1112
, L
1213
,
L
1314
, L
2223
, L
2628
,
L
2629
, L
2829
, L
3334
,
L
131
, L
3338
, L
147
,
L
3944
, L
3945
, L
4142
,
L
127
L
414
, L
56
, L
611
,
L
89
, L
1011
, L
1619
,
L
1621
, L
1920
, L
2122
,
L
2324
, L
936
, L
3031
L
12
,L
34
, L
1617
,
L
3436
, L
4748
, L
4048
,
L
3743
, L
4344
, L
4445
,
L
4551
, L
4252
, L
4041

TABLE VII
Comparative Results for 39-Bus and 68-Bus Systems
Type Pole vector Geometric Proposed
39-bus New Generators G
4
, G
5
, G
7
, G
9
G
3
, G
7
, G
8
, G
9
G
3
, G
5
, G
9
, G
10
England System Power Flow In Lines L
89
, L
12
, L
23
, L
1415
, L
318
L
939
, L
1617
, L
23
, L
225
, L
45
L
1727
,L
1619
,L
939
L
225
, L
89
68-bus New EnglandNew Generators G
11
,G
12
,G
13
,G
14
,G
15
,G
16
G
3
, G
9
, G
10
, G
13
, G
14
, G
16
G
1
,G
3
,G
5
,G
9
,G
13
,G
15
York System Power Flow In Lines L
3637
,L
5052
,L
4041
,L
3436
,L
130
,L
936
L
12
, L
4252
, L
147
, L
3338
, L
2627
, L
930
L
3637
, L
12
, L
930
, L
5052
, L
936
, L
45
controller. The input and output pole vector indices for the 68-
bus New England and New-York interconnected power system
are shown in Fig. 11. The frequency 0.5801 Hz has been taken
to calculate these index values, which is the poorly damped
mode of oscillation present in the system, as shown in Table
V. The results show that the generator groups G
11
, G
12
, G
13
,
G
14
, G
15
, and G
16
have the largest input pole vector value.
Hence, these are selected to get voltage stabilizing signals
from the controller. Similarly, the output signals selected are
P
3637
, P
5052
, P
4041
, P
3436
, P
130
, and P
936
. All the input
signals selected in both the 39-bus and 68-bus systems are
observable as the corresponding index values are nonzero, as
seen in Figs. 10 and 11. The results obtained, as above with
the three methods, are summarized in Table VII.
VII. Validation of Results and Discussions
The proposed control signal selection scheme has been
evaluated by testing with small disturbances, such as variation
of small amounts of loads, and also for large disturbances,
such as three-phase fault conditions. The performance of
the proposed control scheme has been compared with a
geometric-based approach and pole vector direction based
signal selection approach. The results have been obtained with
wide-area robust control using mixed H2/H

output feedback
(WARDC) synthesis [6], [7], [20]. The proposed approach has
been implemented by using two robust control methods, one
based on mixed H2/H

output feedback, i.e., WARDC, and

the other by the wide-area TS fuzzy output feedback control
(WATSFDC) approach [6], [21]. These methods have been
tested under various test scenarios for the 39-bus and the 68-
bus systems. The results obtained for the two systems are given
the the following.
A. 39-Bus New England Power System
1) Case-1: Application of Three Phase Fault at Bus 8: For
the nonlinear simulation, a three-phase fault has been applied
at bus 8. The circuit breakers are assumed to clear the fault at
bus-8 after 150 ms and permanently open lines 7 and 8 after
150 ms, thereby changing the topology of the system. The
fault has been applied after 30 s from starting the nonlinear
simulation, as shown in Fig. 13. The results show that with
the proposed approach of the signal selection, the settling time
is less than 11 s, which is signicantly better than the other
selection methods and also with only the CPSS considered
in the system. It has been found that with the proposed
approach and the WARDC control, the rst peak overshoot
and undershoot is little higher, which has been reduced by
applying the WATSFDC approach.
2) Case 2: Line Outage: The line connecting buses 4 and
14 was suddenly opened 50 s after the start of the simulation,
which resulted in the change in topology and the operating
condition of the system. The results for the power deviation in
the line connecting buses 9 and 39 are shown in Fig. 12. It can
be seen that the response is more oscillatory without the wide-
area control. With the proposed method, the system settles to
steady state in 13 s, which is slightly better than the geometric
approach based signal selection approach and signicantly
improved as compared to that with the pole vector approach.
3) Case 3: Load Variation: The proposed control signal
approach is also tested for another critical case, i.e., for sudden
change of a large amount of load, which resulted in wide
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
8 IEEE SYSTEMS JOURNAL
Fig. 10. Normalized input and output pole vectors for a 39-bus system.
Fig. 11. Normalized input and output pole vectors for a 68-bus system.
variation in the system operating condition. The base case load
at bus 4 was 522 MW and 176MVR. To test the performance of
this controller, this load has been suddenly decreased to 40%
of the base value, as shown in Fig. 13. With the proposed
approach, it has been found that the speed deviation of G
7
shows a bit higher undershoot, but the system response settles
quickly to the steady state within 8 s, whereas with only CPSS,
it takes about 16 s to reach the steady state.
B. 68-Bus New England and New York Interconnected Power
System
1) Case 1: Application of Three Phase Fault: For the
nonlinear simulation, a three-phase fault has been applied at
bus 27 for 50 ms, after 30 s from starting of the nonlinear
simulation, as shown in Fig. 14. The results show that in the
proposed approach of the signal selection, the settling time
is less than 12 s, which shows signicant improvement as
compared to the other signal selection methods and also over
that with only the CPSS, which takes with about 20 s to reach
the steady state. It has been found that the application of
WATSFDC improves the result slightly as compared to that
with the WARDC control.
2) Case 2: Load Outage: The performance of the proposed
approach has also been tested for sudden reduction of large
amount of load at bus 1. The base case load at bus 1 was
252.7 MW and 118.56 MVR. To test the performance of the
controller, this load has been suddenly decreased to zero 50 s
after the start of the simulation. Fig. 14 shows the power
deviation in lines 930 due to this load outage. The results
demonstrate that with only the local control action, the power
oscillates up to around 30 s. However, with the proposed
approach and the geometric approach of the signal selection,
the oscillations settle in 10 s with an undershoot and overshoot
of quite low values.
VIII. Controller Performance Evaluation
The performance of the signal selection approaches has been
further evaluated by proposing an index, dened as
J
Index
=
N
g

j = 1
j = r
_
N
s

i=1

w
ij
w
ir

_
2
(8)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
PADHY et al.: COHERENCY-BASED APPROACH FOR SIGNAL SELECTION 9
Fig. 12. Line power deviation due line outage.
Fig. 13. Speed deviation of Gen 7 due to 3- fault and load variation.
TABLE VIII
Index Value for Various Test Cases
Test Systems Tests With CPSS Pole vectors Geometric
Proposed
WAR-DC WATS-FDC
3- fault 0.3418 0.200 0.1906 0.1810 0.0859
39-bus New Line outage 0.4892 0.3706 0.0474 0.0509 0.0418
England System Variation
0.4573 0.3372 0.0646 0.0613 0.0594
of load
68-bus New
3- fault 0.9238 0.0646 0.0004 0.0003 0.00034
England New
Load
0.3818 0.1954 0.200 0.1369 0.0852
York System
outage
3- fault 0.3418 0.200 0.1906 0.1810 0.0859
where N
g
represents the number of generators in the system,
N
s
is the total number of samples taken, w
ij
is the ith
sampled value of the jth generator rotor speed, and w
ir
is
the ith sampled value of the rotor speed of the reference
generator. For the 39-bus system, gen-1 has been taken as
the reference generator, and for the 68-bus system, the gen-16
has been taken as the reference generator. The performance
has been evaluated by normalizing the index in (8). The
normalized performance index has been dened with only
CPSS considered in the system (case 1) as follows:
J
normalized
Index
=
J
CPSS
Index
(J
CPSS
Index
+ J
Pole Vectors
Index
+ J
Geometric
Index
+
J
WARDC
Index
+ J
WATSFDC
Index
)
. (9)
For other cases of controllers, the performance index can
be dened by appropriately changing the numerator in (9).
Fig. 14. Speed deviation of Gen 7 and power deviation in lines 939 due to
3- fault and load outage.
The controller with the lowest index value provides better
performance and robustness toward the change in the operating
condition. The performance has been evaluated for various
test cases, as given in Table VIII. It is observed that in all
test cases, the index value is lowest with the proposed control
signal selection approach.
IX. Conclusion
A coherency-based wide-area signal selection approach for
damping interarea mode of oscillation was proposed. The
validity and performance of the proposed controller were
evaluated on the two test systems. The combined action of
both the CPSS and the wide-area controller provided better
performance than with only the local CPSS controllers. The
performance results, obtained for various test cases under
wide range of variation of the operating conditions, showed
that the proposed wide-area signal selection method exhibited
better performance as compared to the pole vector direction
and geometric based signal selection approaches, which was
further proved by the values of the proposed performance
index. It was also observed that the geometric approach gave
slightly better performance than the pole vector direction
based signal selection approach. Furthermore, it was found
that the TS fuzzy (WATSFDC) based controller provided
improved performance over the mixed H2/H

output feedback
(WARDC) based controller.
References
[1] P. Kundur, Power System Stability and Control. New York, NY, USA:
McGraw-Hill, 1994.
[2] J. De La, V. Centeno, J. S. Thorp, and A. G. Phadke, Synchronized
phasor measurement applications in power systems, IEEE Trans. Smart
Grid, vol. 1, no. 1, pp. 2027, Jun. 2010.
[3] B. Chaudhuri and B. C. Pal, Robust damping of multiple swing modes
employing global stabilizing signals with a TCSC, IEEE Trans. Power
Syst., vol. 19, no. 1, pp. 499506, Feb. 2004.
[4] W. Juanjuan, F. Chuang, and Z. Yao, Design of WAMS-based multiple
HVDC damping control system, IEEE Trans. Smart Grid, vol. 2, no.
2, pp. 363374, Jun. 2011.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
10 IEEE SYSTEMS JOURNAL
[5] B. P. Padhy, S. C. Srivastava, and N. K. Verma, A wide-area
continuous time model predictive controller, in Proc. 16th Na-
tional Power Syst. Conf., Osmania University, Hyderabad, India,
2010, pp. 272277.
[6] B. P. Padhy, S. C. Srivastava, and N. K. Verma, Robust wide-area
TS fuzzy output feedback controller for enhancement of stability in
multimachine power system, IEEE Syst. J., vol. 6, no. 3, pp. 426435,
Sep. 2012.
[7] Y. Zhang and A. Bose, Design of wide-area damping controllers for
interarea oscillations, IEEE Trans. Power Syst., vol. 23, no. 3, pp. 1136
1143, Aug. 2008.
[8] N. R. Chaudhuri, B. Chaudhuri, S. Ray, and R. Majumder, Wide-
area phasor power oscillation damping controller: A new approach to
handling time-varying signal latency, Proc. IET Gen. Transm. Distrib.,
vol. 4, no. 5, pp. 620630, 2010.
[9] A. Heniche and I. Kamwa, Assessment of two methods to
select wide-area signals for power system damping control,
IEEE Trans. Power Syst., vol. 23, no. 2, pp. 572581, May
2008.
[10] J. V. Milanovic and A. C. S. Duque, Identication of electrome-
chanical modes and placement of PSSs using relative gain ar-
ray, IEEE Trans. Power Syst., vol. 19, no. 1, pp. 410417, Feb.
2004.
[11] M. M. Farsangi, H. Nezamabadi-Pour, Y.-H. Song, and K. Y. Lee,
Placement of SVCs and selection of stabilizing signals in power
systems, IEEE Trans. Power Syst., vol. 22, no. 3, pp. 10611071, Aug.
2007.
[12] Y. Li, C. Rehtanz, S. Ruberg, L. Luo, and Y. Cao, Assessment and
choice of input signals for multiple HVDC and FACTS wide-area
damping controllers, IEEE Trans. Power Syst., vol. 27, no. 4, pp. 1969
1977, Nov. 2012.
[13] C. Chateld and A. J. Collins, Introduction to Multivariate Analysis.
London, U.K.: Chapman and Hall, 1980.
[14] T. Kohonen, The self-organizing map, Proc. IEEE, vol. 78, no. 9, pp.
14641480, Sep. 1990.
[15] K. Havre and S. Skogestad, Selection of variables for stabilizing control
using pole vectors, IEEE Trans. Automatic Control, vol. 48, no. 8, pp.
13931398, Aug. 2003.
[16] A. M. A. and A. M. Elabdalla, Geometric measures of modal control-
lability and observability of power system models, Electr. Power Syst.
Res., vol. 15, no. 2, pp. 147155, 1988.
[17] K. K. Anaparthi, B. Chaudhuri, N. F. Thornhill, and B. C. Pal, Co-
herency identication in power systems through principal component
analysis, IEEE Trans. Power Syst., vol. 20, no. 3, pp. 16581660, Aug.
2005.
[18] M. A. Pai, Energy Function Analysis for Power System Stability. Boston,
MA, USA: Kluwer Academic, 1989.
[19] G. Rogers, Power System Oscillations. Norwell, MA, USA: Kluwer
Academic, 2000.
[20] Matlab Users Guide, The Math Works, Inc., Natick, MA, USA, 2007.
[21] J. Yoneyama, M. Nishikawa, H. Katayama, and A. Ichikawa, Design
of output feedback controllers for Takagi-Sugeno fuzzy systems, Fuzzy
Sets Syst., vol. 121, no. 1, pp. 127148, 2001.
Bibhu Prasad Padhy received the B.E. degree in
electrical engineering from the University College of
Engineering, Burla, India, in 2006, and the M.Tech.
degree from the Indian Institute of Technology Roor-
kee, Roorkee, India, in 2009. He is currently pursu-
ing the Ph.D. degree at the Department of Electrical
Engineering, Indian Institute of Technology Kanpur,
Kanpur, India.
His current research interests include power sys-
tem dynamics and stability, wide area controller de-
sign, power system optimization, and AI applications
in power systems.
S. C. Srivastava (SM91) received the Ph.D. degree
in electrical engineering from the Indian Institute of
Technology Delhi, Delhi, India.
He is currently a Professor with the Department
of Electrical Engineering, Indian Institute of Tech-
nology Kanpur, Kanpur, India. His current research
interests include power system optimization, security
analysis, stability, power system restructuring, and
synchrophasor applications.
Dr. Srivastava is a Fellow of the Indian National
Academy of Engineering, the Institution of Engi-
neers, India, and the Institute of Electronics and Telecommunication Engi-
neers, India.
Nishchal K. Verma (M05) received the B.E. de-
gree in electrical engineering from the Faculty of
Engineering, Dayalbagh Educational Institute, Agra,
India, in 1996, the M.Tech. degree in electrical
engineering from the Indian Institute of Technology
Roorkee, Roorkee, India, in 2003, and the Ph.D.
degree in electrical engineering from the Indian In-
stitute of Techonology Delhi, Delhi, India, in 2007.
He is currently an Assistant Professor with the
Department of Electrical Engineering, Indian In-
stitute of Techonology Kanpur, Kanpur, India. His
current research interests include fuzzy systems, neural networks, data mining
tools, intelligent monitoring of systems, bioinformatics, video clip or image
sequence modeling, machine learning, and computational intelligence.