International Journal of Innovative Technology and Exploring Engineering (IJITEE)

ISSN: 2278-3075 (Online), Volume-9 Issue-2, December 2019

Principle Component Analysis Based Data

Mining for Contingency Analysis on IEEE 30
Bus Power System
Lekshmi M, M S Nagaraj

The proposed work in this paper would gather the meter

Abstract: This Paper is an attempt to develop a Data Mining reading or the sensor data of power generated, power demand,
tool for the contingency of the power system. By mining the big power angles and the voltage firstly. Then the gathered data is
data in the power system and analyzing the early detection of the
contingency in the power system a larger cost cutting can be applied with the Fast Fourier Transform as a preprocessing
planned. As Mining would reduce the computational complexity step[4-5]. Then the preprocessed data is feature extracted
of the contingency analysis this attempt would lead to reduction in with Principle Component Analysis un order to get a feature
the hardware use. This paper uses Multiclass Relevance Vector of reduced dimension [6].
Machine(MCRVM) and Multiclass Support vector The feature thus extracted is used for training the MCRVM
machine(MCSVM) in order to mine the data which include the
and the MCSVM for the different kinds of contingencies in
voltage, power generated , power angles , power demand in
different lines of the power system. The Data mining would need a the power system. A tradeoff between both RVM and The
data transformation technique, which would reduce the SVM methods are carried out in the IEEE 30 bus system. The
dimensionality of the data introduced for mining. The idea developed in this paper is to avoid the load flow for
combination of Data cleansing and the Principal Component contingency analysis by getting the real time data from the
Analysis would act as the data transformation technique in this lines to be trained in the MCRVM and the MCSVM in order
paper. A Matlab based simulation is carried using the IEEE 30
bus system for the contingency analysis by incorporating the to get the contingency as a model to be used for further
loading risk assessment strategy using the Multiclass SVM and analysis for a system. The paper is organized as follows
RVM and the results are compared and the outputs are tabulated. Section II talks about the proposed method of online
Active power performance index and the reactive power contingency analysis, Section III discusses the
performance index are used in contingency analysis of the IEEE implementation details of the contingency analysis carried out
30 bus system thus used and the accuracy of classification and the
in the paper. Section IV would discuss the results and
speed of classification with the different methods and the
contingency rankings are found and displayed. discussion of the proposed work, and the conclusion would
follow with the references.
Keywords : Contingency Analysis, Principle Component
II. PROPOSED CONTINGENCY ANALYSIS-A
Analysis, Artificial Neural Network, Support Vector Machine.
CLASSIFCATION APPROACH
I. INTRODUCTION The implementation of the online contingency analysis of the
Smart Grid Implementation needs data from different sensors system for a 30 bus IEEE bus system is carried out with
and meters to be collected from the power system and these loading contingency like the light load, medium load and
data, which is collected, would be huge. The down time of any heavy loading condition to be assessed. The inputs like the
power system due to contingency would cause a loss to the voltage, voltage angle, real power generated and the reactive
power system economy. The Switzerland based Large power are introduced as the input which would be feature
Hadron Collider has developed a Big data experiment which extracted in order to develop a contingency analysis model
would conduct experiments that would deliver data from 150 using two different learning models of MCRVM and
million sensors which samples its output at the rate of 40 MCSVM. Performance measures that would decide the
million times per second. But only the 0.001% of the data is adaptability of any of these methods for contingency analysis
actually used for experiments carried out in the lab [1-3]. model is decided. The voltage, voltage angle, real power
generated and the reactive power from the IEEE 30 bus power
system under study thus obtained would be taken for feature
extraction for training the MCRVM and the MCSVM training
Revised Manuscript Received on December 30, 2019. algorithms for making a intelligent model for online
* Correspondence Author contingency analysis. The data cleansing analysis on the
Lekshmi M*, Research Scholar, Jain University, Asso. Prof. Dept of
EEE,AIT, Bangalore, India. Email:[email protected] signals is carried out and the results are applied with Principle
Dr.M S Nagaraj, HOD Dept. of EEE,BIET Davanagare, India. Email: Component Analysis in order to give a reduced size input to
[email protected] the MCRVM and the MCSVM in order to have a reduced
© The Authors. Published by Blue Eyes Intelligence Engineering and
computational complexity.
Sciences Publication (BEIESP). This is an open access article under the
CC-BY-NC-ND license http://creativecommons.org/licenses/by-nc-nd/4.0/

Retrieval Number: B7116129219/2019©BEIESP Published By:

DOI: 10.35940/ijitee.B7116.129219 Blue Eyes Intelligence Engineering
Journal Website: www.ijitee.org 878 & Sciences Publication
Principle Component Analysis Based Data Mining for Contingency Analysis on IEEE 30 Bus Power System

The PCA analysis is the projection method that would project The Training flow of the contingency ranking system for the
the complex system in to combination of the basis functions of IEEE 30 bus system is carried out as shown in the
Principle Components. Figure1.During the testing phase the outages are made to
After applying the data cleansing methods on the inputs occur in the buses randomly and the contingency severity
specified the PCA is applied on it in order to reduce the size of ranking is classified using the MCSVM and the MCRVM
the data. method.

III. CLASSIFICATION
The performance measures that have to be predicted for the
contingency analysis are the active power performance index
and the reactive power performance index. The calculations
of the performance indices can be described as defined in [7].
The orthogonal characteristics between the active and the
reactive power which is a decoupled measurements on the
power system can act as a parameter that would act as the
measure for the contingency prediction. The active power
performance index is given as ,
PI p = å w p L(PL / PL lim )2
a

where, PL power in line L. w p is the weight value for the line

power PL , PL lim is the real power limit in the line L. a is the
set of lines which has the power limit crossed .The reactive
power performance index is dependent on the voltage
variation in the buses that would cause the reactive power
injection in the lines. The formula for the reactive power
performance index is given as [7]
PI v = å wvi ( Vi -Vi lim ) / Vi lim
b

where Vi is the voltage magnitude at bus i, Vi lim is the

voltage magnitude limit in line i, wvi is the weight value for
the line voltage, b is the set of buses that crosses the limits.
The contingency ranking is carried out using the two methods
of MCSVM and MCRVM in this paper.

Fig. 2.MCSVM and MCRVM Testing

The testing phase of the contingency ranking classification is
carried out as given in Figure 2.
The IEEE 30 bus system is applied with the contingency
condition and for each condition the real and reactive power
performance index is calculated and the data is collected for
training. The voltage at each bus in the 30-bus system is taken
as the input and the FFT is calculated for the voltage wave and
the Principle Component Analysis using the Singular Value
Decomposition (SVD).
Fig. 1.MCSVM and MCRVM Training

Retrieval Number: B7116129219/2019©BEIESP Published By:

DOI: 10.35940/ijitee.B7116.129219 Blue Eyes Intelligence Engineering
Journal Website: www.ijitee.org 879 & Sciences Publication
 International Journal of Innovative Technology and Exploring Engineering (IJITEE)
ISSN: 2278-3075 (Online), Volume-9 Issue-2, December 2019

The output of the SVM and the RVM training is taken as the voltage profile of the 30 buses in per unit values while
composite performance index the PCA on the parameters of initializing is as given in Figure.4 as a plot of bus versus the
each bus is the input that is given for training.. The 30-bus voltage per unit value.
input of all the line parameters with the PCA extracted and the
corresponding ranking values as the output of the training
data are given for the multiclass training. The multiclass
training would correspond to the ranking of the bus in terms of
the bus number, which is more severe.

Figure 4.Voltage Profile of the IEEE 30 bus system

As the input for the training of the MCSVM and the MCRVM
algorithm the real power performance index and the reactive
power performance index using the Fast Decoupled method is
carries out to get the contingency ranking for the different
outages has to be tabulated. The contingency ranking values is
predicted by means of applying a thresholding values on both
the performance index values considered. The Table 1 depicts
the performance index values calculated from the Fast
Decoupled method, MCSVM and MCRVM on the IEEE 30
Fig. 3.IEEE 30 Bus System bus system. The MCSVM uses the Radial Basis Function for
training and the MCRVM uses the probabilistic method for
The complete topology of the IEEE 39 bus system is as given
training. The input is the vector of 60 inputs that is the PCA
in Figure 3.
extracted from the 30 buses for both the voltage and power
waveforms.
IV. RESULTS AND ISCUSSION
The ranking obtained from the Fast Decoupled method
Matlab Mfile based simulation is carried out on the IEEE 30
bus system as shown in the Figure 3. The IEEE 30 bus system
is a 41 line power system which has the generators in the
2,5,8, 11 and 13 buses and slack bus is in the bus 1. The

Table 1.Contingency Ranking

MCRVM
Fast Decoupled MCSVM
Line No. Line outage PI v PI p Predicted
Rank Predicted Rank
Rank

1 1-2 68.96 22.98 1 1 1

5 2-5 26.7 8.79 2 2 2

6 2-6 23.56 7.65 3 3 3

15 4-12 21.45 6.78 4 4 4

36 28-27 20.988 6.34 5 5 5

9 6-7 19.56 5.98 6 6 6

2 1-3 18.7 5.7 7 7 7

Retrieval Number: B7116129219/2019©BEIESP Published By:

DOI: 10.35940/ijitee.B7116.129219 Blue Eyes Intelligence Engineering
Journal Website: www.ijitee.org 880 & Sciences Publication
Principle Component Analysis Based Data Mining for Contingency Analysis on IEEE 30 Bus Power System

18 12-15 18.34 5.55 8 8 8

4 3-4 18.26 5.34 9 9 9

3 2-4 18.1 5.23 10 10 10

38 27-30 17.89 5.1 11 11 11

37 27-29 17.78 4.87 12 12 12

25 10-20 17.65 4.56 13 13 13

27 10-11 17.56 4.45 14 14 14

19 12-16 17.45 4.40 15 15 15

17 12-14 17.42 4.35 16 16 16

22 15-18 17.40 4.32 17 17 17

24 19-20 17.35 4.31 18 18 18

41 6-28 17.33 4.27 19 19 19

12 6-10 17.30 4.195 20 20 20

30 15-23 17.28 4.0976 21 21 21

39 29-30 17.089 3.899 22 22 22

35 25-27 16.986 3.565 23 23 23

31 22-24 16.898 3.232 24 24 24

26 10-17 16.656 3.123 25 25 25

28 10-22 16.367 3.099 26 26 26

21 16-17 16.233 2.998 27 27 27

40 8-28 16.008 2.877 28 28 28

14 9-10 15.8787 2.645 29 29 29

23 18-19 15.766 2.564 30 30 30

32 23-24 15.667 2.449 31 31 31

Retrieval Number: B7116129219/2019©BEIESP Published By:

DOI: 10.35940/ijitee.B7116.129219 Blue Eyes Intelligence Engineering
Journal Website: www.ijitee.org 881 & Sciences Publication
 International Journal of Innovative Technology and Exploring Engineering (IJITEE)
ISSN: 2278-3075 (Online), Volume-9 Issue-2, December 2019

20 14-15 15.0087 2.3354 32 32 32

The execution time for the contingency rank prediction using

the fast decoupled method ,MCSVM and MCRVM methods
are as given in Table-II. Dr. M. S. Nagaraj is heading Dept. of EEE, BIET,
Table- II: Comparison of execution time taken by Davanagere. India. His area of specialisation is AI
various methods techniques in power system. He has published papers in
various journals.
Methods Fast Decoupled methods MCSVM MCRVM

Execution Time 22.34 secs 12.34sec 11.45sec

s s

While automating the contingency ranking of the lines the

hardware needs and the algorithm reaction time are important
aspects of the implementation. It is observed the time taken
for prediction of the MCSVM and the MCRVM
implementation takes lesser time facilitating the chances of
obtaining the results quickly.

V. CONCLUSION
Matlab based implementation is carried out with the
traditional method of Fast decoupled method and the
proposed method of MCSVM and MCRVM. The results
obtained from the proposed method matches the output
obtained from the traditional method. The advantage of the
proposed method is that the execution time of the proposed
method is around 50% of the total time. This execution time
of the proposed method suggests getting a best alternative to
the traditional methods while online contingency ranking is a
requirement. The hardware implementation of this proposed
algorithm in the online contingency ranking algorithm would
be most feasible as the execution time is optimal.

REFERENCES
1. Martin, James, Strategic Information Planning Methodologies, 2nd ed,

Prentice-Hall, 1989, p. 24. ISBN 0-13-850538-1. 

2. White, Tom (10 May 2012). Hadoop: The Definitive Guide. O'Reilly

Media, p. 3. ISBN 978-1-4493-3877-0.
3. CERN-Brochure-2010-006-Eng. LHCBrochure, English CERN.
Retrieved 20 January 2013.
4. B. G. Jo and M. H. Sunwoo, "New continuous-flow mixed-radix
(CFMR) FFT Processor using novel in-place strategy," IEEE Trans.
Circuit. Syst. I: Regul. Pap., vol. 52, no.5, pp. 911-919, May. 2005. 

5. S. Khomfoi and L. M. Tolbert, “Fault diagnostic system for a multilevel
inverters using a neural network,” IEEE Trans. Power Electron., vol. 22,
no. 3, pp. 1062–1069, May 2007. 

6. W. Chenglin, H. Jing, W. Tianzhen, and C. Zhiguo, "Relative PCA with
applications of data compression and fault diagnosis," Acta. Autom.
Sin., pp. 1128-1139, 2008. 

7. Tarlochan S. Sidhu, “Contingency Screening for Steady-State Security
Analysis By Using FFT and Artificial Neural Networks “IEEE
TRANSACTIONS ON POWER SYSIBMS, VOL. IS. NO. I ,
FEDRUARY 2000

AUTHORS PROFILE

Lekshmi.M is working as an Asso. Professor in Dept

of EEE, AIT, Bangalore,India. She is aresearch scholar
at Jain University, Bangalore. Her research area is
contingency analysis in power system using AI
Techniques..

Retrieval Number: B7116129219/2019©BEIESP Published By:

DOI: 10.35940/ijitee.B7116.129219 Blue Eyes Intelligence Engineering
Journal Website: www.ijitee.org 882 & Sciences Publication