Anda di halaman 1dari 9

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2710134, IEEE
Transactions on Smart Grid
1

A New Adaptive Dependability-Security


Approach to Enhance Wide Area Back-up
Protection of Transmission System
Manas Kumar Jena, S.R.Samantaray, Senior Member, IEEE, B.K Panigrahi, Senior Member, IEEE

the protection logic based on the system condition.


AbstractThis paper presents a new methodology to The need of a reliable zone-3 operation of the distance relay
adaptively shift the dependability and security bias of the is discussed in [5]. Recent developments in wide area
transmission system protection logic depending on the system monitoring, control and protection technology have certainly
state. Contingencies such as generator outage, line outage and
enhanced the wide area situational awareness (WASA).
line fault can induce stressed system condition. Under stressed
conditions, protection logic should be more biased towards Further, the use of PMU information to enhance the
security. Phasor measurement unit (PMU) information is used at transmission system protection operation has gained
the system protection centre to estimate the system state. A data- momentum [6]-[10]. Synchrophasor assisted adaptive reach
mining model known as random forest (RF) is utilized to setting of the distance relays is suggested in [6]. A
accomplish the task of state assessment (SA). Two different classification approach using support vector machine (SVM)
protection logics are used to make the final relaying decision. The
to prevent relay mal-operation under power swing is reported
first protection logic is the existing distance relay and the second
protection logic is based on wide area information. Both the in [7]. In [8], positive sequence voltage and current signals are
protection logics are connected though logic gates to make the utilized to locate the faulted transmission line. Further, the
final relaying decision. The performance of the proposed WABP schemes for series compensated transmission line are
scheme is validated on the IEEE- 39 bus New England system suggested in [9] and [10].
and 246-bus Indian Northern Regional Power Grid (NRPG) The concept of adaptive dependability-security is reported
system. The test results indicate that the proposed algorithm can
for the very first time by Phadke et al. [11]. In [12], a decision
help in shifting the dependability-security bias of the protection
logic adaptively, which will help in mitigating cascaded outages tree (DT) model based adaptive dependability-security scheme
in the power transmission system. is reported. In this scheme, a set of three independent and
redundant relays are used to take the final relaying decision.
Index Terms Distance relay, line protection, phasor Similar attempt is also made in [13]. However, the core
measurement unit, random forest, variational mode algorithm behind all these relays remains same. That means all
decomposition (VMD), wide area back-up protection (WABP). of them work on the basic principle of distance relay. This is
where there is scope for improvement as far as the adaptive
I. INTRODUCTION
dependability-security based scheme is concerned. Most of the

M ajor power system blackouts in recent past are reported


to be caused due to protection logic mal-function [1]-
[5]. Majority of those mal-functions occur during stressed
relay mal-operations are related to zone-3 of distance relay.
Further, the zone-3 of distance relay mal-operate mainly due
to stressed conditions such as load encroachment, stable power
power system conditions. Planned line outage (maintenance), swing and voltage instability [5]-[10]. Thus, if one distance
generator outage, sudden line outage due to relay mal- relay of a particular line mal-operates during above mentioned
operation, line faults are some of the factors which make the stressed conditions, the other two redundant relays will also
system stressed. Further, loss of security during stressed mal-operate owing to their similar principle of operation.
condition is more dangerous than loss of dependability [5]. However, if the basic algorithm of the redundant independent
However, the existing protection logics are normally more relay is such that it is stable during stressed conditions, then
biased towards dependability. Thus, there is a need of new the objective of the adaptive dependability-security will be
algorithm which can shift the dependability-security bias of fulfilled. This is the main motivation behind the proposed
WABP scheme. The proposed scheme intends to use two
protection algorithms for back-up operation. The protection
Manas Kumar Jena and B.K. Panigrahi are with Department of Electrical
Engineering, Indian Institute of Technology Delhi, New Delhi-110016 India logic-1 (PL-1) is the conventional zone-3 distance relay. The
(e-mail: kumarmanas87@yahoo.com, bkpanigrahi@ee.iitd.ac.in). protection logic-2 (PL-2) is based on wide area information.
S.R. Samantaray is with the School of Electrical Sciences, Indian Institute of The remainder of the paper is organized as follows: Section
Technology, Bhubaneswar, Bhubaneswar-751013 India (e-mail:
srs@iitbbs.ac.in).
II discusses the detailed methodology of the proposed scheme.
Development of data-mining models is discussed in Section
III. Results analysis and conclusions are presented in Section

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2710134, IEEE
Transactions on Smart Grid
2

IV and Section V, respectively. in system state assessment. The details of the mining process
is explained in the next section. Thus, the next interesting
II. PROPOSED METHODOLOGY finding of this work is that the information from only
The system can be declared as stressed following any generator buses and boundary buses are sufficient to declare
event which may trigger relay mal-operation. The following the system state. In such situation, the number of PMUs
contingencies might make the system stress such as: (i) required will be very less. To summarize the state assessment
generator outage (ii) line outage (iii) line faults. Sometimes, process:
transmission lines remain out of service because of Power system is divided into different coherent
maintenance issues and this can be categorized as planned areas [14],[17],[20];
outage. The concept of relay margin (RM) is reported for the PMUs are assumed to be installed on generator
very first time in [15]. As shown in Fig.1, any contingency buses and boundary buses;
may reduce the RM of a distance relay. Decrease in RM Features which take part in state assessment are
increases the chance of relay mal-operation. Thus, during decided ( , , ) ;
stressed system condition, more bias should be given towards The features are used as input to a RF model (RF-
security aspect of the protection logic. As suggested in [12], 1);
situations in which the load flow converges can be classified Criteria for stressed and Normal system
as Normal. However, if the load flow does not converge, the condition are decided;
corresponding state is categorized as Stressed. In addition to
Development of learning sample;
this, large disturbances such as sudden generator outage, line
Training of RF-1 model followed by testing;
faults and line outage may also affect the RM of distance
relay. These situations may not create a non-convergence
case of load flow solution; however, such situations are also
kept in the stressed category, owing to their impact on RM
of a relay. In this study, following the contingency if RM of
any distance relay of the transmission system attains 1.3 times
reach setting of the zone-3 setting, then the corresponding
situation is declared as stressed condition. The threshold
1.3 is decided considering the trade-off between frequency
of shifting the bias towards security and the minimum time
required for coordinated PL-1 and PL-2 operation.
The next task is to decide the features which will take part in Fig. 1. Relay margin of a distance relay
predicting the system state. The investigation reveals that the
square of the positive sequence voltage magnitude of The next task is to develop algorithm for PL-2. Here, we
generator buses ( ), change in positive sequence voltage have used a recently introduced signal processing tool termed
phase angle of only generator buses
as VMD. The efficient mode decomposition feature of
and change in
VMD along with its robustness in noisy environment make it a
phase angle difference between any two boundary buses i
and j of any two areas ( ) can help in predicting the
suitable candidate for extraction of IMFs of the input signal
[22]. The objective of PL-2 is to correctly identify any power
system state reliably. This finding is also logically correct system event which behaves like a fault condition. Different
considering the fact that any disturbance pertaining to real
power flow will affect and ( ). Similarly, any
power system events such as generator outage, line outage,
line fault etc. often appear similar. This makes event
disturbance which involves change in reactive power flow, classification difficult if the raw PMU data is simply used.
will affect . The reason for considering square of the However, each event leaves its unique footprint on either the
positive sequence voltage is based on the fact that any voltage or frequency waveform of the system. The spectral
deviation from 1p.u. gets magnified and this helps the mining modes obtained through time-frequency resolution based
process. techniques can help in identifying different disturbances
Let us assume that the system under consideration is divided occurring in power system. Thus, the IMFs obtained through
into different coherent areas. The positive sequence voltage VMD help in deriving the features which are further utilized
phasors at the buses connected to the generators and at to distinguish no-fault situation from fault situation. Once the
boundary buses connecting two areas, are provided by the IMFs are obtained following the VMD based pre-processing,
PMUs. It is assumed that each area is coherent following a Hilbert Transform (HT) is applied to each IMF to extract the
contingency and the groups of coherent machines can be analytic signal. The features selected in the proposed study are
found out by applying the existing techniques [14], [17], [20]. the entropy (E) and standard deviation (S.D) of first three
Initially, authors used information from all buses of the IMFs of the input signal. It is to be noticed that the input
system. However, after the rigorous mining process, it was signal for the PL-2 is only the positive sequence voltage
found that very few PMU information are actually taking part magnitude of PMUs installed on the generator and boundary

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2710134, IEEE
Transactions on Smart Grid
3

buses. Similar to the state assessment process, the buses to be Thus, the proposed scheme ensures systems reliability.
considered for the PL-2 development, was concluded only
after the mining process. Another RF model (RF-2) is used to
predict the fault occurrence, anywhere in the transmission
system.
Figure 2 shows the flow chart of the proposed scheme. It is
inferred from the figure that there are two protection logics
which drives the back-up protection operation of a
transmission line (PL-1 & PL-2). PL-1 is the traditional zone-
3 distance relay and PL-2 is based on wide area information.
The positive sequence voltage magnitudes of the generator
and boundary buses are passed through VMD based pre-
processing tool. The first three IMFs extracted through VMD
are passed through a RF model, which helps in differentiating
fault condition from other power system disturbances.
However, if the system is in normal condition, the bias should
be more towards dependability. Thus, the state assessment
(SA) algorithm can assist in deciding whether the bias should
be more towards dependability or towards security. If the
SA logic detects that the system is in stressed condition,
then both PL-1 and PL-2 must be connected in such a way that
the overall bias of the protection scheme should be towards
security. In order to achieve this objective, the truth table is
drawn as shown in Table I. The following rules are followed
to build the truth table.
If output of SA is 0, it indicates normal operating
condition. Here, the overall bias of the back-up
protection scheme should be towards dependability.
Thus, the output function (F) should follow operation of
PL-1.
If output of SA is 1, it indicates stressed operating
condition. Here, the overall bias of the back-up
protection scheme should be towards security. In this Fig. 2. Conceptual block diagram of the proposed scheme
situation, in order to get a final trip signal, both the
protection logics PL-1 and PL-2 must provide their
output as 1.
Applying the concept of minterm to Table I, the
expression for the final output signal (F) can be
expressed as (1).
= + + (1)
Fig. 3 shows the k-map for the truth table. The final
expression for the output function is shown in (2).
= + (2)
where A, B and C indicate logical output signals from SA,
Fig. 3. K-map for the truth table
PL-1 and PL-2, respectively.

The gate level implementation of the output function


(F) is shown in Fig. 4.

The proposed scheme utilizes the available wide area


information to assist the existing relay in taking the reliable
back-up protection decision. If at all there is any
communication failure (PL-2 fails), then the existing zone-3
will take care of the back-up protection. Thus, the worst
condition of the proposed scheme is same as the present
Fig. 4. Gate level realization of the proposed scheme
practice i.e. zone-3 will take care of the back-up protection.

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2710134, IEEE
Transactions on Smart Grid
4

TABLE I
TRUTH TABLE
SA PL-1 PL-2 Output
(A) (B) (C) Function
(minterms, F)
N 0 0 0 0(m0)
O
R 0 0 1 0(m1)
M 0 1 0 1(m2)
A
0 1 1 1(m3)
L
S 1 0 0 0(m4)
T
R 1 0 1 0(m5)
E
S 1 1 0 0(m6)
S
E 1 1 1 1(m7)
D

III. DATA-MINING USING RANDOM FOREST


Fig. 5. Single line diagram of IEEE-39 Bus system
RFs are introduced by Leo Breiman [16]. RF is a grouping
of tree predictors. The prediction error of RF reduces as the Thus, in the proposed study, total 10 generator outage cases,
numbers of tree increases. Both regression and multiclass 46 line outage cases and 460 fault cases are simulated at each
classification problems are solved by RF. Recently many operating point of the system. The loads are varied from 90%
power system problems are solved through RF based to 200%. Total 2616 cases are generated out of which 70% of
prediction algorithm [17]-[19]. A RF is a combination of t the data sets are used to train the RF and rest 30% are used as
trees.
= ( ), ( ), . ( )
test file.
(3)
= ! , ! , . !" is a vector of n-dimensional
One of the important characteristics of the RF model is the
where correlation analysis. This analysis helps in deciding the
predictors.
#$ = % ( ), #$ = ( ), #$ = ( )& is the t outputs
optimal input features to be considered in the classification
process. A correlation analysis provides insights into the
of RF. The class predicted by majority of trees form the final
prediction output of RF #( . More details related to RF based
independence of the numeric input variables. Modeling often
assumes independence, and better models will result by using
data-mining approach can be found in [16]-[19]. independent input variables. In the proposed study, initially
112 input features were considered. However, after the
A. Generating Learning Sample correlation analysis, only 28 features ( , , ) are
The proposed scheme is initially validated on IEEE-39 bus retained for the classification process. Fig. 6 shows one of the
system. MATLAB 2015(b) is used for modeling the test examples of the degree of correlation among the input features
system. The single line diagram of the test system is shown in
considered in the proposed scheme (x1 to x10 represents of
generator buses, x70 to x75 along with x78, x79, x80, x81 are
Fig. 5. As shown in the figure, the test system consists of 10

of generator buses and the rest represents


generators and 46 transmission lines, 39 buses, 29 loads. The,
system is divided into 6 areas [17], [20], [21]. PMUs are across
assumed to be placed at all generator and boundary buses. In boundary buses). The hierarchical correlation dendrogram
order to generate learning data set, miscellaneous operating presents a view of the variables of the dataset showing those
points are created through load scaling. The system load is are highly correlated are identified. The length of the lines in
scaled methodically by successively increasing and decreasing the dendrogram provides a visual indication of the degree of
the load at each area of the 39- bus system. For each operating correlation. Shorter line lengths indicate more tightly
point, a contingency (generator outage, line outage, line fault) correlated variables. Once the highly correlated variables are
is simulated. This may lead to either convergence or non- identified, then the redundant variables are ignored for the
convergence of load flow solution. All non-convergence cases mining process.
are naturally categorized as stressed condition. However, Further, the classification accuracy depends on the quality
there are some cases which lead to load flow convergence of the features taking part in the classification process. One of
with maximum impact on RM of some critical relays of the the important characteristics of RF is that it investigates the
system. Those cases are also categorized as stressed impact of each input feature on the classification accuracy.
condition. This is useful to know which variable plays important role in

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2710134, IEEE
Transactions on Smart Grid
5

the classification process. Fig. 7 shows the top down introduction of VMD based mode extraction which assist in
importance of variables according to the accuracy loss when identifying a fault situation from other stressed conditions.

0.04
they are respectively removed from or included in the feature
OOB
set. It is observed from the figure that positive sequence 0
voltage angle difference between boundary buses (x61, x58, x39) 1

of the IEEE-39 bus system plays a vital role in the

0.03
classification process. This is the basis on which the optimal
number of features (information from generator bus and
boundary bus) are selected which take part in the SA process.

0.02
Error
As pointed out earlier, 30% of total data set are kept for
testing purpose. These data set are termed as out of bag
(OOB) data set. Fig.8 shows the error characteristics of the

0.01
RF-1 model. In the figure, x-axis represents the number of
trees and y-axis represents the error rate. It is concluded that
the OOB error stabilizes after around 15 trees.

0.00
x61 0 10 20 30 40 50
x55
x35 trees
x58
x48
Fig. 8. Convergence characteristics of OOB test cases
x39
x68
x66
x79
The three primary steps involved in VMD based pre-
x78
x75
processing are as follows: [22]
x80
x70
x81 (i) modes update
x72
x71 (ii) center frequency update
x2
x1 (iii) dual ascent update
x73
x7
x74
x6
(i) Modes update: In VMD, the Wiener Filtering is embedded
x10
x3
in order to update the mode. The process of updating the
x9
modes is accomplished in Fourier domain by tuning a filter to
the center frequency )*
"
.
x4
x8
x5

"- ())
+,*
$ " ())/2)
./()) 2* +, "- ()) 3* +, " ()) + (4
2.5 2.0 1.5 1.0 0.5 0.0

= ")
(4)
Fig. 6. Degree of correlation between the input features
1 + 28() )*

x61 x58 (ii) Center frequency update: The following equation is used
x39 x61
to update the center frequency.
= "- ())|
: )|+,* <)
x58 x66
x78 x78

)*
"-
= >= (5)
x68 x39
x48 x48

:> |+,*
"- ())| <)
x81 x68
x79 x79
x35 x35
x66 x55
x55 x71
x71 x70
x10 x81
x9 x72 (iii)Dual ascent update: The Lagrangian multiplier is updated
x80 x80
x72
x75
x75
x2
using the following equation.

4/"- = 4/" + @ A./ B +,* C (6)


x70 x10
"-
x2 x9
x74 x8
x5 x1
*
x1 x5
x7 x74
x73 x73
x8 x3
x4 x7
x6
x3
x4
x6 The process of signal decomposition continues till,

" "
B+,* +* /+,* < G (7)
-1 0 1 2 3 4 0 5 10 15 20 25
MeanDecreaseAccuracy MeanDecreaseGini "-
Rattle 2017-Jan-18 19:04:10 Manas

Fig. 7. Top-down importance of the variables according to the accuracy loss *


or misclassification rate reduction The detailed theoretical and mathematical background of
VMD can be obtained from [22].
The proposed scheme uses two RF model to assist the Let, I(J) be the Hilbert transform of IMF +(J), then the
performance of the existing distance relay. The modeling of analytic signal is computed by,
RF-2 is almost similar to that of RF-1.The difference is the

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2710134, IEEE
Transactions on Smart Grid
6

!(J) = +(J) + KI(J) = (J) expOK(J)Q (8) TABLE II


PERFORMANCE OF RF-2
= U(V)
where, I(J) = T :W= <@ . P is the Cauchy integral value.
Line faults Generator outage Line outage

S WV
Actual Predicted Actual Predicted Actual Predicted
IEEE- 1229 1218 6 6 26 26
39 bus
The instantaneous frequency is computed using the following Actual 1 1 1 1 - -
PMU
equation.
YU( )
data
YZ( )
1 +(J) Y +(J) Y I(J)
)(J) = (9)
2X + (J) + I (J)
IV. RESULT ANALYSIS AND DISCUSSION
Instantaneous amplitude is computed using the following The efficacy of the proposed scheme is further tested for
equation. various operating conditions which led to mal-operations of
(J) = \+ (J) + I (J) (10)
the conventional zone-3 based back-up protection logic. The
results are discussed below.
1) Performance during stressed conditions:
In [22], superiority of VMD over some of the recent signal The effectiveness of the proposed scheme is verified by
processing tools such as wavelet transform (WT), S-transform comparing it to the conventional blinder based schemes. Two
(ST) and empirical mode decomposition (EMD) in terms of concentric polygons blinders are used for power swing
decomposition quality, noise robustness, insensitivity to detection [26], [27] and load blinder [28] is used to block
sampling frequency is explored. This very property of zone-3 operation during load encroachment. The blinder
efficient mode extraction is used in this paper to assist settings detailed in [26], [27] and [28] are used for the test
distance relay in making a secure relaying decision during system. In case of concentric polygons blinders, the inner
stressed condition. Various events such as generator outage, blinder is placed outside the protection zone-3 characteristics.
line outage, line faults give rise to various spectral modes. As suggested in [26], the outer blinder settings are considered
These modes help in classifying different events occurring in to keep 5% security margin resistively away from the possible
the power system. The different modes corresponding to maximum load, and 20% margin reactively away from the
different disturbances can be quantified in terms of statistic- maximum mho relay reach. Settings of the load blinder are
based metrics. In order to achieve the same, the first three referred from [28]. The maximum slip frequency is considered
IMFs of positive sequence voltage signal are used. Further, to be 2 Hz and the minimum block duration is set to be 11ms
entropy (E) and standard deviation corresponding to each IMF [26].
are used as input to RF-2. The RF model is built using the Load encroachment is one of the possible causes behind
rattle software package [23]. The data collection and training relay mal-operations. In order to observe the performance of
process remains same as RF-1. Further, 1229 fault cases, 6 the proposed scheme during load encroachment, the following
generator outage cases and 26 line outage cases are used as steps are followed. Line 26-27 of IEEE-39 bus system is kept
unseen test file to the trained RF-2 model (IEEE-39 bus out of service. This is predicted as stressed condition by RF-
system). It was observed that RF-2 provides a high detection 1. Thus, the distance relays operating in the system are more
accuracy. (Table II). biased towards security. Now, the active and reactive powers
The efficacy of RF-2 is also tested for field data available of load at bus 26 are doubled. This caused dip in voltage and
from the PMUs installed at some strategic locations of the apparent impedance seen by the relay at bus-29 is shown
Northern Indian Grid. Training data is generated by modeling in Fig. 9. It is observed from the figure that the conventional
the Northern Regional Power Grid (NRPG) of Indian Power load blinder is unable to block the zone-3 operation.
system using Siemens commercial simulation software Similarly, power swing is simulated in the IEEE-39 bus
package named Power System Simulation for Engineering system by incepting a 3-phase fault on line 26-28 at 10.8 s
(PSS/E). The reduced NRPG system (220 kV and 400 kV followed by opening of the breaker at 10.88s. The
only) network consists of 246 buses, 376 branches corresponding R-X trajectory for the relay at bus-29 is shown
(lines/transformers) and 40 shunt reactors. The detailed line in Fig. 10. It is inferred from the figure that following the fault
data, bus data and the dynamic data required for the dynamic inception, the impedance locus enters the outer blinder.
modeling of the NRPG system are referred from [24]. However, because of instantaneous rate of change of
However, the testing is carried out on a completely new field impedance during fault situation, the blinder setting does not
data set. An instance of loss of generation is the loss of 2000 classify it as swing. Following removal of fault, the apparent
MW generator occurred at the Rihand Super Thermal Power impedance locus reenters the outer blinder at 11.4s followed
Station (STPS) complex on 1st June 2010 at 23:49 hrs. It was by inner blinder at 11.48s. Thus, this condition is incorrectly
a case of line to ground fault followed by generator outage. identified as an out-of-step phenomena. In both the cases,
Details of the above mentioned incident is reported in [25]. zone-3 at bus-29 mal-operates. The performance of the
Both the incidents are correctly predicted by RF-2. Table II proposed scheme is depicted in Table III. It is observed that
details the performance of RF-2 with respect to event RF-1 predicted the system as stressed condition. Thus, the
detection. back-up protection operation follows the logic presented in
Fig. 3. As RF-2 does not give any trip signal in both the

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2710134, IEEE
Transactions on Smart Grid
7

situations, thus the final trip signal is not generated by the


relay at bus-29.
The performance of the proposed scheme is extensively
tested in various operating conditions where distance relay
mal-operations are identified. All total, 2112 instances are
simulated (stressed) where mal-operation of at least one of the
zone-3 of distance relay (of any transmission line) takes place.

Reactance X (p.u.)
The test cases are simulated by initiating different (N-1)
contingencies (sudden loss of generation or transmission line)
and fault situations at various operating points. The loads are
varied from 90% to 200%. The objective is to induce zone-3
mal-operations due to load encroachment and stable swing
condition. Out of 2112 mal-operations, 1194 cases are due to
stable swing condition and the rest 918 cases are due to load
encroachment. It is observed that the proposed scheme could
able to avert 2092 cases of mal-operations. A new index
termed as Blocking efficiency (BE) is introduced in this
paper. BE is defined as follows.
`abcd efghijk al jidcm gcdanijcboaek cpijbiq
BE =
`abcd efghijk al jidcm gcdanijcboaek
(11) Fig. 10. Impact of power swing on relay at bus-29

TABLE III
PERFORMANCE OF PROPOSED SCHEME TO BLOCK RELAY AT BUS-29
Cases SA output Zone-3 at RF-2 output Final output
bus-29
Load 1 1 0 0
encroachment
Power swing 1 1 0 0

Generator 1 0 0 0
outage at bus-
38
Line outage 1 0 0 0
at 26-27

TABLE IV
OVERALL PERFORMANCE OF PROPOSED SCHEME
Stressed PL-1 PL-1 Mal-op. BE of the BE of
scenarios Mal-op. Mal-op. averted Blinder the
(2112) (without averted by PL-2 based proposed
blinders) with scheme scheme
blinders) (%) (%)
Stable swing 1194 606 1190 50.75 99.66
(1194)
Fig. 9. Impact of load encroachment on relay at bus-29 Load 918 513 902 55.88 98.257
Encroachment
We observed that PL-1 declares all 1194 stable swing (918)
scenarios as fault situations (without blinder operation)
whereas PL-2 declares 1190 cases as no-fault scenarios (4 2) Performance comparison between Data-mining techniques:
numbers of mal-operation). When concentric polygon blinders The performance assessment of the PL-2 developed using
are included in the distance relay modeling, only 606 cases of different data-mining tools such as neural network (NN),
zone-3 mal-operations are prevented. Similarly, out of 918 support vector machine (SVM), decision tree (DT) and
load encroachment cases, 902 cases are declared as no-fault random forest (RF) is depicted in Table V (NRPG-246 bus
conditions by PL-2 and only 513 cases of zone-3 mal- system). It is observed that the black box models such as NN
operations are avoided by load blinders. The main drawback and SVM show poor classification accuracy. In particular,
of blinder based scheme is to find the settings for blocking fault cases are mostly misclassified. Even though DT shows
during stable power swing and tripping during unstable power improved classification accuracy, however, it is very difficult
swings. It requires detailed stability studies of the system for for a single decision tree to capture the signature of each and
various contingency conditions [26]-[28]. The proposed every event accurately, particularly when lesser PMU
scheme provides much improved BE compared to the measurements are available. Even though DTs are more
conventional blinder based scheme. Table IV depicts the transparent, authors have opted for RF giving more priority to
detailed results. accuracy.

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2710134, IEEE
Transactions on Smart Grid
8

TABLE V 1) Category I: Unable to provide a bias towards security


PERFORMANCE COMPARISON WITH OTHER DATA-MINING TOOLS
Line faults Generator outage Line outage
when it is mostly desired. This circumstance characterizes
Act. Pre. CE Act. Pre. CE Act. Pre. CE the current protection practice. Therefore, this error does not
(%) (%) (%) go in detriment of any existing protection practice.
NN 10114 8011 79.2 35 29 82.8 106 89 83.9
SVM 10114 8125 80.3 35 31 88.5 106 92 86.7 2) Category II: Unable to provide bias towards
DT 10114 9214 91.1 35 33 94.2 106 98 92.4 dependability when it is mostly desired. A regular practice
RF 10114 10101 99.8 35 35 100 106 106 100 to enhance security is to implement a voting scheme in
Act.-actual, Pre.- predicted, CE- classification efficiency
which redundant relays continuously vote, irrespective of
3) Processing Time:
prevailing situations. Thus, under this type of error, the
The proposed scheme is developed to supervise the
scheme again reduces to current protection practices.
operation of zone-3 of distance relay which mostly operates
The worst possible situation does not deteriorate the
with an intentional time delay of 1-2 seconds to achieve the
performance of the prevailing protection logic and thus, the
coordination [5], [27], [29]. The proposed method is executed
proposed scheme provides a win-win situation. As a last
on a Core i7 CPU with 16 GB RAM running on windows 7.
resort, a solution suggested in [32] can be implemented
The average time required for the VMD based feature
where local event detection logic is applied at the substation
extraction and RF based classification is tabulated on Table
level itself to override the supervision of zone-3 of the
VI. The total average time consumed by PL-2 to classify an
distance relay by PL-2. In order to uphold dependability of
event is found out to be 20.11ms. However, the combined
overall protection logic during stressed condition, PL-1 can
fixed delay (assuming 1012 phasor measurements each of
be associated with a maximum wait timer, tw (tw must be
which is 4 bytes in length and has 10 input status channels,
less than the usual intentional time delay associated with a
which are 2 bytes in length) caused by processing,
zone-3 operation). That means during stressed condition,
multiplexing and transducers is estimated to be 75ms [8], [30]-
PL-1 will wait for a particular amount of time (tw) for the
[32]. The delay caused by fiber optics communication medium
arrival of PL-2 logical signal. If PL-2 logical signal does not
can be assumed to be around 25ms [8], [30]-[32]. Hence, the
arrive by that time, then PL-1 should be allowed to take the
total delay can be approximated to be around 100ms-130ms
decision alone after usual intentional time delay of 1000-
[8], [30][32]. As 1000-2000ms (considering 50Hz system) is
2000ms [5], [27],[29].
a common timer setting [5], [27], [29] for back-up protection
operation, hence, the cumulative response time of 120ms
V. CONCLUSIONS
150ms is acceptable. An intentional time delay will still be
required following the implementation of the proposed An adaptive dependability-security approach is suggested
scheme. for improving the performance of existing back-up protection
TABLE VI scheme. The main contribution of this paper is the ability to
AVERAGE PROCESSING TIME supervise zone-3 operation of the distance relay using limited
VMD RF-2 Total time PMU information. The correlation among different variables
For PL-2 processing helps us deciding the optimal number of features which take
Average 9.58 10.53 20.11
time(ms)
part in RF modeling. When the system is in normal condition,
then more bias is provided towards dependability. Similarly,
during stressed condition, more bias is given to security. The
4) Findings of this research work:
proposed scheme is extensively tested on IEEE-39 bus and
The important findings of the proposed research work are
NRPG-246 bus system. An incident of line to ground fault
summarized below. followed by loss of generation in the NRPG system is also
(a) The proposed scheme addresses two important issues of tested to observe the efficacy of RF-2. Comparison with other
todays complex power transmission system. These are (i) data-mining tools such as NN, SV, DT is also carried out. RF
improvement of wide area situational awareness (WASA) is considered mainly because of its highest classification
(ii) enhancement of back-up protection of transmission accuracy which is very much essential for secured relaying
lines. decision. Further, the proposed scheme is used for back-up
(b) The first part of the scheme, which is development of protection operation which inherently works with an
SA logic, can help in enhancing WASA. intentional time delay of around 1000-2000ms. Thus, the
(c) Use of PMU information from only generator buses and processing time of combined VMD and RF is also not an issue
boundary buses of a system is an important finding of this as far as back-up protection is concerned. The algorithm has
paper. performed reliably for more than 10,000 events drawn from
(d) Further, development of PL-2 is also helpful in assisting combination of actual and test systems. The findings of this
existing back-up protection relays in making a reliable research work can be utilized to improve the situational
relaying decision. awareness which is highly essential to prevent cascaded
(e) PL-2 is also helpful in efficient event detection. outages in todays modern power system.
(f) Adaptive dependability-security approach is suggested.
(g)The proposed adaptive dependability-security scheme is VI. ACKNOWLEDGEMENT
susceptible to two types of errors as follows. The authors would like to thank the authorities of the

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2710134, IEEE
Transactions on Smart Grid
9

Northern Regional Load Dispatch Centre (NRLDC) of IEEE Transactions on Smart Grid, vol. 1, no. 2, pp. 144-158, Sept.
2010.
Power Grid Corporation of India, for sharing the PMU [20] S. Yusof, G. Rogers, and R. Alden, "Slow coherency based network
information recorded in their system. Author M. K. Jena partitioning including load buses," IEEE Transactions on Power
gratefully acknowledges the financial and technical support Systems, vol. 8, pp. 1375-1382, 1993.
[21] IEEE. (2 april 2014). New England 10 Generators 39 Busses Test
received from Dr. Nilanjan Senroy of Indian Institute of
System. Available: http://sys.elec.kitami-it.ac.jp/ueda/demo/WebPF/39-
Technology Delhi, India. New-England.pdf
[22] K. Dragomiretskiy and D. Zosso, "Variational Mode Decomposition,"
VII. REFERENCES IEEE Transactions on Signal Processing, vol. 62, no. 3, pp. 531-544,
Feb.1, 2014.
[1] G. Andersson et al., "Causes of the 2003 major grid blackouts in North [23] Rattle (the R Analytical Tool to Learn Easily), by D.Williams, ver. May
America and Europe, and recommended means to improve system 2008. [Online]. Available: http://rattle.togaware.com/
dynamic performance," in IEEE Transactions on Power Systems, vol. [24] [online],NRPG data, http:// www.iitk.ac.in/ eeold/facilities/ Research_
20, no. 4, pp. 1922-1928, Nov. 2005. labs/ Power_System/ NRPG-DATA.pdf, accessed 16 Oct 2015.
[2] S. H. Horowitz and A. G. Phadke, "Blackouts and relaying [25] A. Prince, N. Senroy and R. Balasubramanian, "Targeted approach to
considerations - Relaying philosophies and the future of relay systems," apply masking signal-based empirical mode decomposition for mode
IEEE Power and Energy Magazine, vol. 4, no. 5, pp. 60-67, Sept.-Oct. identification from dynamic power system wide area measurement
2006. signal data," IET Generation, Transmission & Distribution, vol. 5, no.
[3] (2012). Report of the Enquiry Committee on Grid Disturbance in 10, pp. 1025-1032, October 2011.
Northern Region on 30 July 2012 and in Northern, Eastern and North [26] M. McDonald et al., IEEE PSRC Working Group D6, Power swing and
Eastern Region on 31 July 2012 New Delhi, India. [Online]. Available: out-of-step considerations on transmission lines, Jun. 2005.
http://powermin.nic.in/upload/pdf/GRID ENQ REP 16 8 12.pdf, [27] D. Kang and R. Gokaraju, "A New Method for Blocking Third-Zone
accessed Aug. 20, 2012 Distance Relays During Stable Power Swings," IEEE Transactions on
[4] V. Rampurkar, P. Pentayya, H. A. Mangalvedekar and F. Kazi, Power Delivery, vol. 31, no. 4, pp. 1836-1843, Aug. 2016.
"Cascading Failure Analysis for Indian Power Grid," in IEEE [28] [online],http://www.erlphase.com/downloads/application_notes/Perform
Transactions on Smart Grid, vol. 7, no. 4, pp. 1951-1960, July 2016. ing_Load_Encroachment_with_an_LPRO.pdf
[5] S. H. Horowitz and A. G. Phadke, "Third zone revisited," IEEE [29] IEEE PSRC Working Group D4, Application of overreaching distance
Transactions on Power Delivery, vol. 21, no. 1, pp. 23-29, Jan. 2006. relays, 2009.
[6] K. Seethalekshmi, S. N. Singh and S. C. Srivastava, "Synchrophasor [30] A. G. Phadke and J. S. Thorp, Communication needs for wide area
Assisted Adaptive Reach Setting of Distance Relays in Presence of measurement applications, presented at the 5th Critical Infrastruct. Int.
UPFC," IEEE Systems Journal, vol. 5, no. 3, pp. 396-405, Sept. 2011. Conf., Beijing, China, Sep. 2010.
[7] K. Seethalekshmi, S. N. Singh and S. C. Srivastava, "A Classification [31] I. Kamwa, L. Geoffroy, S. R. Samantaray and A. Jain," Synchrophasors
Approach Using Support Vector Machines to Prevent Distance Relay data analytics framework for power grid control and dynamic stability
Maloperation Under Power Swing and Voltage Instability," IEEE monitoring," IET Engg. And Technol. Reference, (early access).
Transactions on Power Delivery, vol. 27, no. 3, pp. 1124-1133, July [32] S. A. Lavand and S. A. Soman, "Predictive Analytic to Supervise Zone 1
2012. of Distance Relay Using Synchrophasors," IEEE Transactions on Power
[8] M. M. Eissa, M. E. Masoud and M. M. M. Elanwar, "A Novel Back Up Delivery, vol. 31, no. 4, pp. 1844-1854, Aug. 2016.
Wide Area Protection Technique for Power Transmission Grids Using
Phasor Measurement Unit," IEEE Transactions on Power Delivery, vol.
25, no. 1, pp. 270-278, Jan. 2010.
[9] Paresh Kumar Nayak, Ashok Kumar Pradhan and P. Bajpai, Wide- BIOGRAPHIES
area measurement-based back-up protection for power network with Manas Kumar Jena received the B.Tech. degree in
series compensation, IEEE Trans. Power Del., vol.29, no. 4, pp. 1970 electrical and electronics engineering from the Silicon
1977, Aug. 2014. Institute of Technology, Bhubaneswar, India, in 2011. He
[10] M. K. Jena, S. R. Samantaray, B. K. Panigrahi, "A New Wide-Area is currently working towards PhD degree in the
Backup Protection Scheme for Series-Compensated Transmission Department of Electrical Engineering at the Indian
System," IEEE Systems Journal , vol. PP, no.99, pp.1-11. Institute of Technology, Delhi, India.
[11] S. H. Horowitz, A. G. Phadke and J. S. Thorpe, "Adaptive transmission
system relaying," IEEE Transactions on Power Delivery, vol. 3, no. 4, S. R. Samantaray(M08SM10) received the B.Tech. in
pp. 1436-1445, Oct 1988. electrical engineering from UCE Burla, India, in 1999 and
[12] E. E. Bernabeu, J. S. Thorp and V. Centeno, "Methodology for a thePh.D. in power system engineering from the Department
Security/Dependability Adaptive Protection Scheme Based on Data of Electronics and Communication Engineering, National
Mining," IEEE Transactions on Power Delivery, vol. 27, no. 1, pp. 104- Institute of Technology, Rourkela, India, in 2007. Dr.
111, Jan. 2012. Samantaray holds the position of Associate Professor in the
[13] R. D. Quint and M. K. Thomas, "Implementation of an Adaptive Voting School of Electrical Sciences, Indian Institute of Technology Bhubaneswar,
Scheme using synchronized phasor measurements," Southeastcon, 2012 India.
Proceedings of IEEE, Orlando, FL, 2012, pp. 1-6. He serves as Editor of IEEE Transactions on Smart Grid, Associate Editor
[14] J. H. Chow, Power system coherency and model reduction: Springer, of IET, Generation, Transmission & Distribution.
2013.
[15] F. Dobraca, M.A. Pai, P.W. Sauer, Relay margin as a tool for dynamic Bijaya Ketan Panigrahi (SM06) received the Ph.D.
security analysis, Int. J. Electr. Power Energy Syst., vol. 12, no.4, pp. degree from Sambalpur University, Sambalpur, India, in
226-234, Oct. 1990. 2004. He is a Professor with the Department of Electrical
[16] L. Breiman and A. Cutler, "Random forests," Machine learning, vol. 45, Engineering, Indian Institute of Technology (IIT), New
pp. 5-32, 2001. Delhi, India. His research interests include digital signal
[17] H. Supreme, A. Dessaint, I. Kamwa A. Heniche-Oussedik, processing, power quality assessment, and application of
"Development of New Predictors Based on the Concept of Center of soft computing techniques to power system.
Power for Transient and Dynamic Instability Detection," IEEE He serves as Associate Editor of IEEE Systems Journal.
Transactions on Smart Grid , vol.PP, no.99, pp.1-1.
[18] I. Kamwa, S. R. Samantaray and G. Joos, "On the Accuracy Versus
Transparency Trade-Off of Data-Mining Models for Fast-Response
PMU-Based Catastrophe Predictors," IEEE Transactions on Smart Grid,
vol. 3, no. 1, pp. 152-161, March 2012.
[19] I. Kamwa, S. R. Samantaray and G. Joos, "Catastrophe Predictors From
Ensemble Decision-Tree Learning of Wide-Area Severity Indices,"

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Anda mungkin juga menyukai