2020 IEEE International Conference on Power Systems Technology (POWERCON)

Analyses of Indian Power System Frequency

Aman Gautam Rahul Shukla Kamal Kishore
System Operation System Operation Information Technology
Power System Operation Corporation Power System Operation Corporation Power System Operation Corporation
Limited Limited Limited
[email protected] [email protected] [email protected]

Priyam Jain R K Porwal N Nallarasan

System Operation System Operation System Operation
Power System Operation Corporation Power System Operation Corporation Power System Operation Corporation
Limited Limited Limited
[email protected] [email protected] [email protected]

Abstract— Frequency is a vital indicator for the reliable infrastructure. The presence of high frequency synchronized
operation of power systems. The frequency of major power wide-area measurement systems at control centres is
systems around the world vary only slightly around the generating immense volume of power system data. Historical
nominal frequency. The variation is mostly on account of data from January 2015 to December 2019 from phasor
dynamically changing demands, changes in generation and measurement units of different stations in different time
variation of renewable generation with increasing penetration frames has been analysed in this paper. The frequency data
of renewable energy based generation. The various controls used in this paper has been first cleaned using bad data
have been put in place to control the variations in frequency. detection techniques and then particular station data has been
The quality of frequency is of importance to system operators
selected based on the maximum availability of good quality
as the various frequency control measures operate as per the
data. In further subsections analyses are done for average
settings identified by them. The variations in frequency affect
the dynamics of power system and require analysis of past data
frequency for each month of the year, average frequency for
to arrive at effective conclusions. The behavior of frequency in each day of the week, average frequency for each hour of the
different time scales ranging from minutes of an hour to day, average frequency for each time block of the day and
months of a year has been found to have a definite pattern. The average frequency for each minute of the hour.
analysis of pattern so derived can be a vital input in short, A. Impact of poor frequency
medium and long term operational planning. The Phasor
Measurement Units (PMU) help in providing high resolution The synchronous generation has been a major source of
accurate data from various locations in a power system. This generation in most of the power systems. The speed of
paper presents the analysis of Indian Power system frequency rotating parts in these sources is governed by frequency. A
of 5 years and derives important conclusions. very low frequency will cause low speed of generating unit
and consequently a load required to be trip to avoid damage
Keywords— Real time systems, Power systems, Frequency to turbine blades of generator. A high frequency will cause
estimation, Power system protection, Power system relaying, the problem of over speed and attraction between the
Voltage, Power system measurements, Control systems, magnetic field of rotor and stator will be lost causing the
Distortion measurement, Frequency measurement tripping of unit. A significant high speed may also damage
the parts of generating unit. The deviation from normal
I. Introduction frequency also affects the system impedances. The variation
The stability and control of power system frequency has in system impedances may affect the loadings and voltages
always been an important requirement of power system across the power system.
operation of all the grids across the world. The requirement
In real time operations during low frequency periods, the
of maintaining frequency within the desired range is getting
tripping of motors and generating units, loss of auxiliary
more and more challenging in the recent times. In the era of
drive motors as well as errors in electrical measurement
expanding power markets and integration of technical
based time instruments is observed. Similarly, the high
advancements in power sector, the system operator’s role is
frequency affects the operation of power system operation.
critical in maintaining the real time frequency. The large
The daily use equipment’s in our homes or offices are
scale integration of renewable generation and its intermittent
designed to operate at a standard frequency within tight
nature has further compounded the problem. The
tolerance, so it’s very important to keep the frequency of our
replacement of synchronous generation with rotating parts by
power supply stable. Also, the solar and wind power plants
non-rotating decoupled mass of renewable generation has
usually work in frequency following mode. The working of
presented a challenge to manage the system frequency with
PLL of renewable plants are very much dependant on grid
low inertia. The literature related to managing frequency via
frequency and hence, any deviations from the standard
primary, secondary and tertiary control has been discussed a
frequency affects the functioning of PLL.
lot in literature. However, the idea of utilising the historical
frequency data to assess the power system behaviour has B. Balance of power
rarely been found in the literature. Electric power systems The impact of significant deviation in frequency has been
are taking drastic measures in deployment of information and highlighted in previous paragraph. In normal scenario, the
communication technologies; several new measurement frequency does not deviate much from its nominal or
devices are installed in forms of advanced measurement reference value. As per electricity regulator India, user/state

978-1-7281-6350-5/20/$31.00 ©2020 IEEE

Authorized licensed use limited to: Universidad de Sonora. Downloaded on February 08,2024 at 15:23:29 UTC from IEEE Xplore. Restrictions apply.
electricity companies are not allowed to vary its load 0.2 % data are outliers. The PSEUDO code used for outlier
suddenly by more than 100 MW. Hence, there are not much detection is
variations in loads and the balance between generation and
demand is maintained which leads to the operation of system outlier ← [ ] //Empty array for outlier value
around the nominal frequency. This delicate balance is j ←0
disturbed whenever there is a significant and sudden change threshold ← 3
in either generation or load. Indian grid is among one of the mean← mean(data) //Mean value of data set
major synchronised power systems in the world which std← stdev(data) //Standard deviation of data set
operates at a nominal frequency of 50 Hz. The frequency
for (i← 0 to n) //Loops iterate no. of data points times
band has been kept as 49.90 Hz to 50.05 Hz by the electricity
z ← (data[i] – mean)/std
regulator of India. Significant market based reforms as well
as control actions have helped in maintaining the frequency if (absolute_value(z) > 3)
in the narrow frequency band. The improvement in outlier[j] ← data[i]
frequency profile of Indian grid over the years can be seen in j ← j+1
Fig 1. It can be observed that average frequency has moved print(outlier)
closer to 50 Hz and in initial months of 2015, maximum to
minimum spread of frequency was in range of 0.18 Hz, The gaps in the data, after removal of the outliers, was made
which is now being reduced to .06 Hz. equal to average of all frequencies from other stations at that
bad data instant only. After this, best frequency data for a
particular day is selected on the basis of least outliers,
maximum availability of phasor measurement unit, and
maximum status of good quality tags associated with data.
III. BIG DATA ANALYTICS
The digital revolution of last decade has resulted in
significant accumulation of large amounts of data. The
Internet of Things (IoT) concept resulted in increase of
measurement devices. The power sector has also seen an
unprecedented growth in this aspect and with availability of
high resolution accurate data made available by PMU,
various data mining techniques are being introduced.
Synchrophasor technology and systems use PMUs to
monitor electrical quantities (e.g. voltage, current phasors,
and frequency) at specific locations on an electric power
Figure 1: Average Frequency of Indian Power System system. PMUs estimate phasor values (typically 25 or more
measurements per second) using the measured
II. DATA SELECTION AND CLEANING voltage/currents and time-stamp these phasor values
(synchrophasor data) using the Global Positioning System
The five years’ complete data of frequency in sampling rate (GPS) signal as a reference clock for time alignment. The
of 10 seconds from phasor measurement units was collected resulting information gives transmission grid planners and
from the various regional sites. The next task was to clean operators a high-resolution view of power system states
the data and then select best frequency data of a particular throughout the grid in real time and provides data for post-
station from the data available more than 200 stations. analysis of various types of disturbances.
An outlier is a data point in a data set that is distant from
all other observations and lies outside the overall Indian power system started its PMU experience in 2012
with a pilot project with approx. 60 PMUs reporting at
distribution of data set. Due to these, data points cause
national control centre. The utility of PMU in system
various problems during our statistical analysis, forecasting
operation and after the event analysis provided the way for a
analysis, predictive modeling analysis etc. To avoid these full-fledged PMU project with PMU placement at
issues in analyzing the frequency trends and pattern, Z score approximately all the important nodes of Indian power
technique is used here for removing the outliers in system. The PMU data has satisfied the characteristics of Big
frequency data. Z-score technique assumes a Gaussian data. The three V’s of Big data i.e. volume, velocity, and
distribution of data. The outliers are the data points that are variety as defined in [2] is observed in data of PMU in Indian
in the tails of the distribution and far from mean. Z-score is Power system. The addition of two V’s in big data model has
calculated by below formula: increased the ambit of big data and five V’s have been
Z = (xi – µ)/ơ identified in many researches [3][4].
Where
The load curve of a typical day as well as frequency
xi is data point.
profile for the same day for Indian power system is given in
µ is mean of data set. Fig 2 and Fig 3. The close observation of the two figures will
Ơ is standard deviation of data set. help in understanding a lot about the power system size and
its operation. In further subsections, analyses are done for
When computing the Z-score of each sample on the data set average frequency for each month of the year, average
a threshold must be specified. Threshold can be 2, 2.5, 3 frequency for each day of the week, average frequency for
standard deviation from mean. In data set of frequency, we each hour of the day, average frequency for each time block
have considered 3 as threshold and it almost has detected of the day and average frequency for each minute of the

Authorized licensed use limited to: Universidad de Sonora. Downloaded on February 08,2024 at 15:23:29 UTC from IEEE Xplore. Restrictions apply.
hour. Average frequency is the sum of all data points strategies of unit commitment and planning of generation to
divided by number of data points. The average frequency match high ramp rate of loads may help in maintaining
indicates the central value of frequency in case when frequency close to 50 Hz in low frequency months.
frequency is normally distributed. The average frequency
does not provide information about outliers and thus may not
be a good indicator of performance; however, it is very
helpful in comparison of performance for equal durations.
The frequency being a nominal value with approximate
symmetry can be represented well by average values.
Standard deviation is also used in the paper wherever
average gives inconclusive results.

Figure 4: Average frequency of each month of the year for 60 consecutive

months

B. Average Frequency for each day of a week

The results from average frequency value over a large
time scale as that of a month validates the basic power
system frequency and demand relationship. The similar
analogy was tried to arrive at correlation between days of
week and average frequency over a day. The average
Figure 2: Load curve of a typical day
frequency for each day of a week for five consecutive years
is given below in figure 5. It is observed that the frequency
on Sunday remained very high in comparison to other days.
Frequency mostly remained high due to lean demand on
Sunday. During the week days, more variation is observed
in frequency and it remained lowest usually on Tuesdays. A
short term week ahead planning of unit commitment
considering the minimum up and down time of units can help
in maintaining sufficient spinning reserves and thus
improvement of frequency in weekdays. Also there is need to
consider the minimum spinning reserve requirement
considering the single largest unit outage contingency in the
system during week ahead hydro thermal schedule planning.
In demand pattern of the week, the demand keeps increasing
till the middle of the week and then reduces gradually in
second half. A proper demand forecasting can help in
maintaining sufficient units on bar.

Figure 3: Frequency profile from 0000 hrs to 0024 hours

A. Average Frequency for each month of year

The average frequency data of a month was calculated
for five consecutive years. The bar chart plot given in Figure
4 indicates the average frequency for a month for 60 months.
The average frequency for all these 60 months is 49.988 Hz.
The months where frequency was lower than 49.988 Hz can
be termed as under frequency months and those with
frequency more than 49.988 Hz can be termed as over
frequency month. The observation of plot indicates that
generally high demand summer season months are under
frequency months and lean demand winter season are over
frequency ones. The average frequency is usually low during
summer periods due to high demand and low amount of
spinning reserves. A change in long term and medium term
Figure 5 : Average Frequency for each day of a week

Authorized licensed use limited to: Universidad de Sonora. Downloaded on February 08,2024 at 15:23:29 UTC from IEEE Xplore. Restrictions apply.
C. Average Frequency for each hour of the day
The typical day load curve is given above in figure 2. It
can be observed that peak demand periods during morning
and evening time are generally associated with high ramp
rate in demand. To analyze this aspect, the average frequency
for each hour values of the day are plotted for a period of
five years and plot is given as Fig 6. It can be observed that
the maximum variation in frequency is during high ramp up
and down periods. The frequency is below nominal value for
periods associated with heavy ramp up periods and above
nominal values for periods when demand is ramping down.
The frequency is also expected to be above nominal value
during low demand night periods. The frequency is at its
highest in-between 13:00 to 14:00 hrs. The high ramping
periods of 1900 -2000 hours of evening peak are very much Figure 7: Standard deviation of hourly frequency
visible with less frequency values. D. Average Frequency for each minute of an hour
The results observed via plots of average frequency over
a longer time scale are the ones which are evident as per the
theory. The seasonal and intra-day variation in demand is the
major factors which decide the average frequency. It is of
much interest to a system operator to forecast the behavior of
frequency for smaller intervals say next five minutes. The
average frequency of an hour as given in previous section
decides the central values but it is very important to identify
the spread of frequency in an hour’s time frame. The average
frequency of a minute in an hour for five consecutive years is
given in figure 8. The inference from this plot of minute wise
frequency is given below:

• Frequency is on higher side at the start and end of an

hour. The statement can be justified on account of major
schedule changes of load at hourly boundaries. The
sudden large changes in schedules give rise to large
Figure 6: Average frequency of an hour variations in frequency. The hourly schedule changes are
not matched by loads owing to reduced ramping tendency
One interesting observation in the Fig 6 is that though the thereby resulting in high frequency at hourly boundaries.
average frequency appears at 50 Hz for the 1700-1800
• Frequency is at comparatively lesser values in the middle
hours, the hour is generally associated with change in
of an hour, this is generally due to balance in load
demand owing to lighting loads due to sunset. The
generation achieved after a major change in schedules.
frequency changes from over-frequency to under-frequency
• On close examination of the frequency plot, it can be
and sometimes large variations are visible in operation. The
observed that there are four terminal points within an hour
Standard Deviation, therefore, can be a better statistical
and four sections in the plot can be identified. The
parameter to assess the behavior of frequency at this
sections of plot are at periodicity of fifteen minutes and
particular instant. Standard deviation of a data set tells how
average frequency trend can be seen changing at every
measurements for a group are spread out from the reference
fifteen minutes i.e. from decreasing to increasing or vice
value (in this case it is reference frequency of 50 Hz). A low
versa. The schedules of generation are made for every
standard deviation means that most of the dataset is close to
fifteen minutes in Indian power system and the balance is
the 50 Hz. A high standard deviation means that the data set
achieved after certain duration.
is more spread out. The standard deviation for each hour of
the day was calculated for similar data set and the results are
shown as Fig 7. It can be observed that standard deviation
(SD) of frequency is maximum around 1700-1800 hours.
The high values of SD indicate that variations are more
pronounced in these hours. The low SD values during early
morning hours indicate that frequency variation is less in
this duration. Hydro and gas units having high ramp rates
need to be optimized for use during hours having high SD
values. Generally, day ahead load forecasting need to look
for all these issues. Also an Economic Dispatch with battery
storage system can help in calculation of battery storage
capacity requirement to meet high ramp rate of demand
during evening peak hours.
Figure 8 : Average frequency of a minute and its spread in an hour

Authorized licensed use limited to: Universidad de Sonora. Downloaded on February 08,2024 at 15:23:29 UTC from IEEE Xplore. Restrictions apply.
 Frequency displays a significant jump at the boundary of
each scheduling interval. It takes approximately 3-5 minutes
for the frequency to settle to a lower value. At the beginning
of a new scheduling time-block, the generation changes
nearly instantaneously and the complex dynamically varying
power system needs some time to relax to its new operational
state. The impact of scheduling interval on the variation in
frequency is detailed in [9].
E. Average Frequency for each minute of 15 minute time
block
The analysis of frequency variation within a time block
of scheduling which is fifteen minutes also gives further
insights into Indian power system. The average frequency of
each minute of a fifteen-minute time block was averaged for
a data of five calendar years and the trend of frequency Figure 10 : SD values for each minute of an hour
within a time block is given as Figure 9. It can be observed
that during initial minutes of each time block, frequency F. Histogram
remains above reference frequency of 50 Hz. The post initial
minutes observe a decline in frequency and by tenth minute Histogram is plotted to check the maximum sets of
of a time block, the frequency is at its minimum of frequency various ranges. It can be observed in figure 11 that
approximately 49.977 Hz. In the end of a time block, maximum sets are at 50 Hz frequency and no of samples
frequency again starts rising and reaches 49.995 Hz at the decreases as the frequency deviates from 50 Hz. In future the
end of time block. upcoming renewable generation accompanied with less
inertia will increase the number of samples outside 50 Hz.
Monitoring of frequency histogram can help in maintaining
sufficient inertia levels and take effective measures to the
short circuit level and system inertia.

Figure 9: Average frequency for each minute of a time block Figure 11 : Histogram of Five year Frequency data sets
The spread of frequency around the nominal frequency is not Though the risks of large frequency deviations are minimised
observed within a time block which indicates that various in large interconnected grids, the frequency control measures
frequency controls are actively controlling the variations in need to be suitably implemented to control the excursions. A
frequency. The variations are high at the boundary of each stochastic model which can factor the scheduling time block
time block which are observed in almost all major power has been discussed as improved model to represent power
grids of the world. The SD values of each minute of an hour system in [10]. The replacement of synchronous generation
interval averaged over a period of five years indicate a large with inverters having low inertia is certainly going to add to
variation at the boundary of hour. The trend of SD is given as the parameters in model of the power grid.
Fig 10. The large synchronized grid of Indian power system
with high inertia values is expected to contain large IV. CONCLUSION
deviations in frequency post any sudden imbalance of load or The frequency of Indian power system has been analyzed
generation. The Gaussian distribution, therefore, will fail to
using the high resolution PMU data of last five years. The
capture the behavior as long tails will be observed. This
analysis of this data provided better understanding of the
confirms the inability of Gaussian distribution to represent
the power system frequency variation. This aspect has been behavior of Indian power system over different time
confirmed in detail in [9] where frequency time series for a horizons and helped in identifying areas where further
range of power grids are analyzed. Tertiary and secondary improvements are required for better system operation. The
controls can help maintain frequency for short to medium analysis of month and day wise average frequency data
range deviations in frequency. In Indian power system, clearly brought out the need for optimizing annual unit
AGC is yet to implement in all the generating units. With its outage plan and effective long & short term generation
implementation short term frequency deviations can be planning including unit commitment. The day wise analysis
avoided and will help to move frequency near to 50 Hz. also highlighted the importance of planning and maintaining

Authorized licensed use limited to: Universidad de Sonora. Downloaded on February 08,2024 at 15:23:29 UTC from IEEE Xplore. Restrictions apply.
adequate spinning reserves in operational horizon. The [4] International Journal of Distributed Sensor Networks, Volume 2015,
Article ID 902982
effect of demand ramp on system frequency was also shown
[5] Big Data in Power Systems Leveraging grid optimization and wave
through the hourly frequency analysis for each day. The energy integration Nuno Amaro R&D Nester Centro de Investigação
requirement of high ramping resources during the high em energia REN – State Grid, S.A. Lisbon, Portugal
demand ramp period was derived from this analysis. The [email protected] João Murta Pina Centre of Technology
minute wise analysis of frequency data for each hour and and Systems (CTS) UNINOVA Caparica, Portugal [email protected]
time block (15-minute interval) showed the effect of [6] Wide-Area Measurement-Driven Approaches for Power System
Modeling and Analytics Hesen Liu University of Tennessee,
changes in schedule on frequency especially at the hourly Knoxville, [email protected]
and time block boundaries. The variation of frequency [7] A. G. Phadke, J. S. Thorp, M. G. Adamiak, "A new measurement
within a time block showed the importance of time blocks technique for tracking voltage phasors local system frequency and
with smaller intervals for scheduling purpose in order to rate of change of frequency", IEEE Trans. Power Apparat. Syst., vol.
PAS-102, pp. 1025-1033, May 1983.
achieve effective load-generation balance. The improvement
in frequency over a period of five years was also elaborated [8] Robustness against Disturbances in Power Systems under Frequency
Constraints Dongchan Lee, Liviu Aolaritei, Thanh Long Vu and
The effect of various parameters like inertia, frequency Konstantin Turitsyn
control, grid size along with the impact of large scale [9] Modern Aspects of Power System Frequency Stability and Control
renewable integration in near future on system frequency 1st Edition,Authors: Andrew Dixon
were discussed. The integration of large scale power [10] Non-Gaussian power grid frequency fluctuations characterized by
electronic devices and their interaction with system Lévy-stable laws and superstatistics Benjamin Schäfer , Christian
Beck, Kazuyuki Aihara, Dirk Witthaut and Marc Timme.
frequency may be a case for further study in this regard.
V. ACKNOWLEDGMENT
The authors gratefully acknowledge the contributions of
fellow colleagues and higher management of POSOCO for
support as well as the authorities over there to make
available the relevant documents, based on which it has
been possible to prepare the paper.
VI. REFERENCES
[1] Power systems big data analytics: An assessment of paradigm shift
barriers and prospects HosseinAkhavan-HejaziaHamedMohsenian-
Rad, Energy Reports Volume 4, November 2018, Pages 91-100.
[2] NASPI WHITE PAPER Data Mining Techniques and Tools for
Synchrophasor Data Prepared by NASPI Engineering Analysis Task
Team (EATT) January 2019
[3] Zhu Ting, Sheng Xiao, Qingquan Zhang, Yu Gu, Ping Yi, and
Yanhua Li, “Emergent Technologies in big Data Sensing: a survey”,

Authorized licensed use limited to: Universidad de Sonora. Downloaded on February 08,2024 at 15:23:29 UTC from IEEE Xplore. Restrictions apply.