Fifteenth National Power Systems Conference (NPSC), IIT Bombay, December 2008
Understanding Power System Behavior Through
Mining Archived Operational Data
Sarasij Das, P S Nagendra Rao
Abstract—This paper is the outcome of an attempt in mining
recorded power system operational data in order to get new
insight to practical power system behavior. Data mining, in
general, is essentially Þnding new relations between data sets by
analyzing well known or recorded data. In this effort we make use
of the recorded data of the Southern regional grid of India. Some
interesting relations at the total system level between frequency,
total MW/MVAr generation and average system voltage have
been obtained. The aim of this work is to highlight the potential
of data mining for power system applications and also some of
the concerns that need to be addressed to make such efforts more
useful.
I. I NTRODUCTION
Advances in electronics, computer and information technology are fueling major changes in the area of power system instrumentation. More and more microprocessor based
digital instruments are replacing analog meters. Data logging
is becoming automatic and frequent. Vast quantities of data
generated by extensive deployments of digital instruments are
creating information pressure on Utilities. The legacy SCADA
based data management systems do not support management
of such huge data. The present practice is to store the acquired
data in SCADA for only a few months and then delete. In few
cases after removing from the SCADA system, these data are
stored in compact discs. At present the usefulness of historical
data is not fully explored. So, utilities do not give importance
to store such data efÞciently.
The traditional integrated power industry is going through a
deregulation process. The market principle is bound to force
competition between power utilities, which in turn demands
a higher focus on proÞt. To optimize system operation and
planning utilities need better decision-making processes that
depend on the availability of reliable system information. It
is expected that in this context historical data is going to be
a vital asset. In [1] some possible applications of historical
power system data is presented.
Apart from a business perspective, historical data is important
from another point of view also. Electric power system is
a very complex system. Most of the mathematical models
used for analyzing/predicting system behaviors are based on
several assumptions. The availability of detailed measurements
of power system parameters could provide an opportunity to
Sarasij Das obtained M.Sc(Engg) in Electrical Engineering from Indian
Institute of Science, Bangalore. He is currently with Power Research Development Consultant Pvt. Ltd. (e-mail: sarasijdas@gmail.com).
P S Nagendra Rao is with the Electrical Engineering Department, Indian
Institute of Science, Bangalore.(e-mail: nagendra@ee.iisc.ernet.in).
validate many of such models. The importance of data is being
recognized widely in the recent times.
Power system data management has been discussed in several
works[1][2][3]. Data warehousing technology is being proposed to meet the future requirement of power systems. In
[4] data mining as a feature of power system data warehouse
is mentioned. In [5] it has been mentioned that power systems
operation can be greatly improved through data analysis and/or
assimilation
In our work real system data is analyzed to Þnd interrelation
between several system parameters. This analysis can be
viewed as a small attempt of data mining. Data mining is
essentially an analysis of data sets in order to discover new
relations between various quantities which is not obvious from
the recorded data in its normal form.
For our investigation, data of Þves system parameters- voltage,
frequency, MW and VAr generation, system demand - of
the southern regional grid of India, collected from Southern
Regional Load Despatch Center has been used.
This paper is organized as follows. Section II outlines some
features of the Southern Regional Grid. In Section III a brief
description of the data set used is given. Section IV presents
the results of data analysis. The paper is concluded in Section
V.
II. S OUTHERN R EGIONAL G RID F EATURES
The southern regional grid of India covers an area of
6,51,000 Sq. km encompasses four states namely Andhra
Pradesh, Karnataka, Kerala, Tamilnadu and one Union Territory of Pondicherry. This region comprises of several central
and state owned generating stations, independent power producers, distribution companies and state transmission utilities.
In India, including SRLDC, there are Þve major regional
grids. After August 2006 four of the Þve regions excluding
the southern region are (synchronously) interconnected. The
Southern region is connected with the other regional grids only
in an asynchronous manner. Southern region is connected to
Western region through HVDC back to back at RamagundamChandrapur and to Eastern region at Jepore-Gazuwaka backto-back and point to point HVDC line between KolarTalcher.
The total installed capacity of southern region as in the
beginning of 2007 is about 37370 MW. Some other salient
features of the southern regional grid are [6] :
• Covers approximately 19% of the geographical area, 22%
of population and 29% of the installed capacity of the
country.
248
Fifteenth National Power Systems Conference (NPSC), IIT Bombay, December 2008
•
•
•
•
•
30-70% hydro-thermal mix
3300 MW wind generating plants
8000 MW capacity independent power producers.
2000 MW capacity HVDC Talcher-Kolar double crcuit
interconnection with the Eastern Region.
400/220 KV transmission system
III. DATA D ESCRIPTION
The control center at Bangalore of SRLDC is equipped
with a computerized load despatch and communication facilities. Around 320 Remote Terminal Units(RTU) are used for
real time system monitoring and grid management. Through
SCADA these RTUs communicate with the control center.
From the collected data only Þve system parameters- voltage,
MW generation, MVAr generation, frequency, system demand
- are made available to us for use in this work. These data
were in the form of Microsoft Excel Þles stored in Compact
Discs. Data, starting from Jan 2004 to June 2006, is collected.
The data logging interval is 1 minute for all parameters.
Following are some salient features of the data of the Þve
parameters are chosen.
channels, etc. The second class of outlier is the outcome of
some extraordinary events. In the context of power systems
faults, switching operations etc could be the cause for such
events. This type of outliers should be retained in the data
set. The third class of outlier comprises of extraordinary
observations for which there is no explanation. These outliers
must be retained to capture some characteristics of the system
(not known/explained). The fourth and Þnal class of outlier
are observations that fall within the range of each of the
variables but are unique in their combination of values across
the variables. In the data made available to us, it appears that
no attempt had been made to identify and replace outliers.
Investigated data set contains all the four types of outliers.
The outliers identiÞed in the data set are:
- Considerable change in value for in one or two consecutive instances
- Occurrence of 0 values for a considerable time period
- Occurrence of values not possible from system point of
view
IdentiÞable outliers are substituted with average value of
previous and next non-outlier values.
A. Voltage
Voltage data consists of measurements at the 26 buses of
the 400 KV grid. All voltages have been stored (in EXCEL
Þle) as integers with a least count of 1 KV. Among the 26 bus
voltages, the Kolar bus voltage remains constant at 400 KV
all the time.
B. Frequency
Frequency data consists of frequency of the region. Precision of frequency data is 0.01216 Hz.
C. MW Generation
MW generation data consists of information from 60 generation buses. The precision of stored (in EXCEL Þle) data is 1
MW. The 60 outputs at generation buses represent either unit
outputs or the total station output. Types of plants are hydro,
thermal or nuclear.
D. MVAr generation
IV. A NALYSIS
An overview of some features of the collected data set has
been given in the last section. In this section we investigate
the interrelation of some of the measured system variables.
In Figures 1 and 2 system frequency vs. average system
voltage is plotted for 06/06/2006 and 16/4/2006. By average
system voltage we mean average voltage of all 400 KV buses.
It is seen that the points are clustered around one of the
diagonals. It can be seen that to some extent higher voltages
correspond to higher frequency and lower voltage corresponds
to lower frequency. This correlation is relatively well deÞned
in Figure 2.
In Figures 3 and 4 total system demand vs. system frequency
is presented for 6/6/2006 and 16/4/2006. In Figures 3 and 4
by and large higher system demand corresponds to low system
frequency and at lower demands system frequency is high.
Figure 5 presents the total system demand vs. average system
MVAr generation data corresponds to 88 generation units.
Precision of stored(in EXCEL Þle) data is 1 MVAr. The 88
generation units include hydro, thermal and nuclear power
units. In this case the data corresponds to individual units in
all cases.
50.4
50.2
50
System frequency
49.8
E. System demand
System demand data consists of system demand of four
states and the total demand of the region. The precision of
stored(in EXCEL Þle) data is 1 MW.
One of the major challenges in handling real data is the issue
of the outliers. Outliers are basically records whose features
are distinct from the other records of the group. Outliers are
classiÞed into four classes. The Þrst class arises from data
entry error. For power systems data this type of error is
generated due to malfunction of SCADA or communication
OF PARAMETER INTERRELATIONS
49.6
49.4
49.2
49
48.8
48.6
405
410
415
420
System average voltage on 06/06/2006
Fig. 1.
System frequency vs. Average system voltage on 06/06/2006
voltage plot for 19/1/2006. Time instants corresponding to
249
Fifteenth National Power Systems Conference (NPSC), IIT Bombay, December 2008
system voltages. In Figure 10 average generator power factor
vs. total system demand is plotted for 19/1/2006. It can be seen
that as system demand increases in the morning the power
factor starts improving. After 6 a.m the power factor enters a
random ßuctuation zone but remains high in value. At night,
as system demand starts decreasing the power factor starts to
decrease.
In Figure 11 total system demand vs. average system voltage
50.6
50.4
50.2
System frequency
50
49.8
49.6
49.4
49.2
49
19,500
48.8
408
412
414
Average system voltage on 16/04/2006
416
418
420
19,000
18,500
System frequency vs. Average system voltage on 16/04/2006
Total system demand
Fig. 2.
410
4
2.1
x 10
Total system demand
2
18,000
17,500
17,000
16,500
1.9
16,000
1.8
15500
48.8
1.7
49
49.2
49.4
49.6
49.8
System frequency on 16 04 2006
50
50.2
50.4
50.6
1.6
Fig. 4.
1.5
48.6
Fig. 3.
48.8
49
49.2
49.4
49.6
System frequency on 06 06 2006
49.8
50
50.2
Total system demand vs. System frequency on 16/04/2006
50.4
21,000
Total system demand vs. System frequency on 06/06/2006
08:00
20:00
16:00
19:00
20,000
12:00
14:00
10:00
Total system demand
some data points are also indicated in the Þgure. It can be
seen that from 02:00 a.m, the average system voltage gradually
decreases as the system demand increases with time. After
6:20 a.m the plot shows random changes. But, after 7:00 p.m
till the end of the day the average system voltage increases
with decrease in total system demand. In Figure 6 the same
plot of Figure 5 is shown but without the random portion of
the graph. It can be seen that the plot takes two distinct paths
at the start and end of the day. For the same system demand
average system voltage is lower for the morning portion of
the day and higher during the night. This corresponds to a
′
Hysteresis′ type of variation.
In Figure 7 total system demand vs. total MVAr generation is
presented for the same day. It is interesting to see that this plot
also shows a ′ Hysteresis′ loop when the random portions are
excluded. The Figure 7 is similar to the Figure 5 with the only
difference being that in Figure 7, with time the plot moves anticlockwise where as in Figure 5 it moves clockwise. In Figure
8 the same Figure 7 is shown but with the random variation
region excluded.
In Figure 9 total MVAr generation vs. average system voltage
is plotted. From the Figure it can be seen that almost linear
relationship exists between total MVAr generation and average
system voltage. As average system voltage increases MVAr
injection drops and generators absorb MVAr at high average
18:00
19,000
18,000
18:20
06:00
18:10
22:00
17,000
16,000
23:59
00:00
05:00
15,000
02:00
04:00
14000
402
Fig. 5.
404
406
408
410
Average system voltage
412
414
416
418
Total system demand vs. Average system voltage on 19/01/2006
is plotted for 6/6/2006. In this case also random movement
is seen during the middle of the day while the remaining
part exhibits hysteresis type of behavior. We have seen this
hysteresis across several days taken from several months.
More careful investigations are necessary to identify speciÞc
performance patterns. What is evident is that there is some
interesting behavior seen in these plots. Further study could
help to understand this in a better way.
In Figure 12 total system MVAr generation vs. average system
voltage is presented for 6/6/2006. The relationship between
MVAr and average system voltage is almost linear (in an
average sense). As average system voltage increases large
250
Fifteenth National Power Systems Conference (NPSC), IIT Bombay, December 2008
21,000
21,000
20,000
20,000
19,000
19,000
Total system demand
Total system demand
Morning data points
Nigth data points
18,000
17,000
18,000
17,000
16,000
16,000
15,000
15,000
14000
404
14000
800
Night data points
Morning data points
406
408
410
412
Average system voltage
414
416
418
Fig. 6. Total system demand vs. Average system voltage plot (after removing
random portion) of on 19/01/2006
600
400
200
Total MVAR generation
0
200
400
Fig. 8. Total system demand vs. total MVAr generation plot (after removing
random portion)on 19/01/2006
21,000
800
08:00
20:00
600
16:00
20,000
14:00
10:00
12:00
Total MVAR generation
Total system demand (MW)
400
19,000
18:00
18,000
06:00
17,000
22:00
200
0
200
16,000
23:59
400
00:00
02:00
15,000
600
04:00
14000
800
Fig. 7.
600
400
200
0
Total MVAR generation
200
400
600
800
402
800
404
406
408
410
Average system voltage
412
414
416
418
Total system demand vs. total MVAr generation plot on 19/01/2006
Fig. 9. Total MVAr generation vs. average system voltage plot on 19/01/2006
MVAr is consumed by generators and as average system
voltage becomes low generators inject large MVAr into the
grid. Near linear relationship between the parameters is also
evident for other days also.
Till now we have discussed the inter-parameter relationship
considering the whole system. Parameter relationships are
also investigated for individual generation buses. In Figure
13 scatter plot of MVAr and voltage at different generator
buses are presented for the day 6/6/2006. It can be seen that
the MVAr vs. voltage scatter plot at RGM generator bus is
different from other plots. For each voltage at the bus two
distinct MVAr values are seen. Scatter plots (for all the 8
generating stations) indicate a near linear relationship between
bus voltage and MVAr output of the units. Figure 14 shows
the plot of average of MVAr generation at each voltage value
vs. bus voltage value of generating stations (corresponding to
Figure 13). The slope of the MVAr-voltage plot is different
for different generating stations in Figure 14. The nominal
voltage at VTS bus is 220 KV. As the voltage is increasing
the MVAr injection at VTS bus is also increasing. On the other
hand MAP and MTPS are injecting less MVAr with increasing
voltage. KAI, SHVT and SIM are absorbing large MVAr at the
higher voltage levels while injecting a small amount at lower
voltages. RTPS is absorbing smaller MVAr at higher voltage
as compared to the absorption at lower voltages. In Figure 15
plot of voltage vs. MVAr (averaged for each voltage) is shown
for the day 13/3/2006. In this case RGM is absorbing large
MVAr at the higher side of the voltage range while injecting
small MVAr at lower side of the voltage range. The nature
of variations at other generating stations except RTPS in the
Þgure remain same as on 6/6/2006. In Figure 15 RTPS is
injecting a small MVAr at high voltage while injecting large
MVAr at low voltage. In Figure 16 Voltage vs. VAr injection
(averaged for each voltage) is shown for the whole month of
March 2006. In this case also a near linear relationship can be
seen between VAr and voltage. For SHVT and MAP the plots
are not linear at the extremities. Actually the number of sample
points are very small in this range. In Figure 17 total system
demand (averaged for each frequency) vs. system frequency
is plotted for the month of December 2005. Except at higher
251
Fifteenth National Power Systems Conference (NPSC), IIT Bombay, December 2008
0
RGM
10:00
08:00
1
407
408
409
410
411
412
413
414
12:00
20:00
00:00
2
3
4
14000
Fig. 10.
KAI
02:00
15,000
16,000
17,000
18,000
Total system demand
19,000
20,000
230
231
213
214
215
216
217
218
219
220
234
235
236
237
238
239
240
241
404
406
408
410
412
406
408
410
412
414
416
200
0
229
100
221
402
60
80
404
400
232
0
200
233
21,000
229
0
200
212
200
22:00
23:59
228
MTPS
04:00
1
200
100
227
200
14:00
MAP
18:00
RTPS
VTS
0
0
500
400
40
415
16:00
SHVT
Power factor angle in degree
500
1000
406
300
06:00
500
SIM
2
230
231
232
234
235
236
237
50
0
219 220 221 222 223 224 225 226 227 228 229
242
Voltage (KV)
Total system demand vs. power factor plot on 19/01/2006
233
Voltage (KV)
Fig. 13. Scatter plot of Voltage vs. MVAr injection at different bus on
06/06/2006
21,000
08:00
19,000
14:00
SIM
407
408
18,000
17,000
04:00
412
413
414
415
228
229
230
231
400
40
100
80
404
213
214
215
216
217
218
219
220
KAI
02:00
410
415
0
50
100
233
420
Average system voltage
234
235
236
237
238
239
240
241
406
408
410
412
406
408
410
412
414
416
100
MAP
15000
405
404
200
0
229
100
221
50
402
60
232
0
200
212
00:00
16,000
411
MTPS
SHVT
23:59
06:00
410
150
100
227
18:00
409
200
VTS
16:00
22:00
0
200
600
406
12:00 10:00
200
400
RTPS
20,000
Total system demand
RGM
200
20:00
230
231
232
Total system demand vs. Average system voltage on 06/06/2006
234
235
236
237
50
0
219 220 221 222 223 224 225 226 227 228 229
242
Voltage (KV)
Fig. 11.
233
Voltage (KV)
Fig. 14. Plot of Voltage vs. VAr injection (Averaged for each voltage) at
different bus on 06/06/2006
1200
RGM
800
403
404
405
406
407
408
409
410
411
100
200
50
0
400
415
KAI
410
420
System average voltage on 06/06/2006
228
229
211
212
213
214
215
231
232
233
234
Voltage ( KV )
frequencies the graph appears to be nearly a quadratic. As
system demand increases system frequency decreases and vice
versa. At the higher end of frequencies the graph is irregular
231
216
217
235
236
237
403
404
405
406
407
408
409
410
228
229
230
231
232
233
234
235
500
0
227
200
218
0
100
230
Fig. 12. Total system MVAr generation vs. Average system voltage on
06/06/2006
230
0
200
210
100
600
227
MAP
200
226
MTPS
225
200
0
25
20
15
10
5
402
1000
RTPS
VTS
400
SHVT
System Var generation
600
800
405
0
500
402
100
0
100
200
300
403 404 405 406 407 408 409 410 411 412 413
SIM
500
1000
100
0
220
222
224
226
228
230
232
Voltage ( KV )
Fig. 15. Plot of Voltage vs. VAr injection (Averaged for each voltage) at
different bus on 13/03/2006
due to insufÞcient number of sample points. In Figure 18 the
same plot is shown for the month of April 2006. Here also
252
Fifteenth National Power Systems Conference (NPSC), IIT Bombay, December 2008
150
MAP
KAI
100
0
100
228
230
232
234
236
238
50
0
216 218 220 222 224 226 228 230 232 234 236
240
200
SHVT
400
MTPS
100
200
0
224
226
228
230
232
234
236
238
0
200
208
210
212
214
216
218
220
SIMH
200
0
200
400
395
400
405
410
Bus voltage for the month of March 2006
415
Fig. 16. Plot of Voltage vs. VAr injection (Averaged for each voltage) for
the month of March 2006
System demand averaged over each frequency for Dec 05
18,400
17,600
16,800
16,000
15,200
14,400
13,600
12,800
12000
48.5
49
49.5
50
50.5
51
System frequency
System demand averaged for each frequency over month of April 06
Fig. 17.
Plot of System demand averaged at each frequency vs.system
frequency for the month of December 2005
22,000
of India. By performing similar analysis on other systems the
similarities/differences in their behavior can be found. It must
also be pointed out that the primary aim of the present work
is to emphasize that new relations/insights can be obtained by
analyzing operational data.The results presented are incidental.
The present investigation had some constraints beyond our
control. For example, all the parameters contained outliers.
Erroneous outliers are to be identiÞed and eliminated for the
sake of meaningful analysis at the time of archiving. it is
difÞcult to do it later. We have identiÞed some of the erroneous
values and substituted them with reasonable values. With more
system information it would have been possible that more
outliers are identiÞed. Much work is needed to Þnd suitable
ways to identify and replace erroneous values at the time
of recording/archiving. Application of statistical methods for
outlier identiÞcation and replacement can be an interesting
issue for further research.
To get the real beneÞt from data analysis, utilities have to
focus on eliminating the limitations of the present data storage
practices. The limitations observed are :
1) The data set has many errors
2) The data is not complete
3) the data set acquisition/storage was not motivated by
data mining consideration
The aim of this investigation is to argue that if some of
these limitations are overcome (it can be in fact done fairly
easily), then mining such data could be extremely proÞtable
from the point of view of efÞcient planning and operation of
power systems.
ACKNOWLEDGMENT
The authors would like to thank Southern Regional Load
Despatch Center, Bangalore for providing system data for this
work.
R EFERENCES
21,000
20,000
19,000
18,000
17,000
16,000
15,000
48.5 48.6 48.7 48.8 48.9 49 49.1 49.2 49.3 49.4 49.5 49.6 49.7 49.8 49.9 50 50.1 50.2 50.3 50.4 50.5 50.6 50.7 50.8 50.9 51
System frequency
Fig. 18.
Plot of System demand averaged at each frequency vs.system
frequency for the month of April 2006
[1] M.Werner and U.Hermansson, ”Integrated Utility Data Warehousing A Prerequisite To Keep Up With Competetion On Electricity Markets”,
Power Systems Management and Control.2002.Fifth International Conference on(Conf.Publ.No.488), 17-19 April 2002, pp 130-135
[2] D Shi, Y Lee, X Duan, Q.H.Wu, ”Power systems data warehouse”, IEEE
Computer Applications in Power, July 2001, Vol. 14, No. 3, pp 49-55
[3] Lin Xu, ”Data modelling and processing in deregulated power systems”,
Ph.D Theses, Washington State University, May 2005
[4] Xiaofeng He, Gang Wang and Jiancang Zhao, ”Research on the SCADA
/EMS System Data Warehouse Technology”, IEEE/PES Transmission and
Distribution Conference and Exhibition: Asia and PaciÞc, Dalian, China,
2005, pp 1-6
[5] E.H.Abed, N.S.Namachchivaya, T.J.Overbye, M.A.Pai, P.W.Sauer and
A.Sussman, ”Data-Driven power systems Operations”, Lecture Notes In
Computer Science, 2006, NUMB 3993, pp 448-455
[6] http://www.srldc.org/Downloads/Srldc
the graph tends to show a similar nature except at the higher
frequency range.
V. C ONCLUSION
The simple analysis attempted in the work brings out several
interesting characteristics of the overall system behavior that
are not readily available anywhere for the souther regional grid
253