where:
Considering these applications, the LES technique
that simultaneously measures the amplitude and x2 = (Af) V cOSe
frequency of a voltage or current was developed = v~ P
[1,2,3]. The performance of the technique at
x3 = v sine x4 = (Af) v sine
P P
x6 = (Af)2 V sine
P
89 IY 050-6 PWRD A paper recommended and a p p r o v e d all = sin(2nfotl) a12 = 2ntl cos(2nfotl)
by the IEEE Power System Instrumentation 6 Yensure-
inents Conmittee of the IEEE Power Engineering Society
a13 = cos(2nfotl) a14 = -2ntl sin(2nf0t1 )
f o r presentation at the IEEE/PES 1989 Winter Meeting,
New York, “Tew York, lanuary 29 - February 3, 1989.
Manuscript submitted August 2 9 , 1988; made available a15 = -2 (nt1)2 sin(2nfotl)
f o r printing January 4 , 1989.
aI6 = -2 ( ~ t cos(2nfOtl)
~ ) ~ Af f - fo
0885-8977/89/0700-1573$01.0001989 IEEE
1574
The left hand side of this equation is known when the SOURCES OF ERROR - FREQuENcl ESTIMATION
voltage is sampled at time tl. Since time is an
arbitrary quantity and tl can be assigned a value, the As described in the last section, frequency
estimates can be obtained using combinations of the
lla" coefficients in Equation 3 are known but the "x's" estimates of V cose and Afv cos8, and V sine and
are not. Since there are six unknowns in Equation 3, P P P
five more equations must be established before solving AfV sine. The frequency estimation technique uses
P
for the unknowns. If the voltage is sampled at non-recursive digital filters for estimating each
intervals of AT seconds, six consecutive samples component. This section investigates the reasons for
provide six equations in six unknowns. It was the proposed technique providing accurate measurements
observed during the development of the technique that in a limited range only.
frequency should be estimated using more than six
equations so that the results are reasonably god. One of the options for calculating frequency is to
Also, other formulations of the equations could have use sampled values of a voltage and the filter coeffi-
more than six unknowns. Equation 4 expresses m such cients which are the elements of the matrix, [A]'
equations in n unknowns in the matrix form [l]. defined in Equation 5. Figure l(a) shows the
frequency response, IHl(f)l, of the filter that
(4) estimates V sine and Figure l(b) shows the frequency
P
response, IHz(f)l, of the filter that estimates A N
[VI P
where:
[XI
is the vector of voltage measurements,
is the vector of unknowns from which the sine . These filters use two cycle data windows and
frequency can be estimated and 720 Hz sampling rate. The frequency deviation from
[A] is the coefficient matrix whose elements the nominal value estimated by the proposed technique
are known. is the ratio of the magnitudes of the frequency
responses shown in Figures l(b) and l(a) respectively.
For m n , the following equation provides the least This is mathematically expressed as
squares solution.
[XI = [AI' [VI (5)
where: [A]+ is the left pseudo-inverse of [A] and .,
is given by [ [AIT [A]1-l [AIT. where: denotes that the frequency deviation is an
estimate.
The factors that affect the suitability of the LES
technique are the size of the data window, the Dividing and multiplying the right hand side of
sampling frequency and the truncation of the Taylor Equation 8 by Af provides
series expansions of the sine and cosine terms. A
considerable freedom in selecting these parameters
exists that enables the final design to be suitable
for power system applications [l].
This equation shows that the estimated frequency, Af,
-
Equation 5 shows that multiplying the row elements is equal to the true deviation of the frequency when
of the pseudo-inverse matrix with the digitized IIH2(f)l/Af) is equal to IIHl(f)l).
voltage samples provides the elements of the [XI
vector. The elements of each row of the pseudo- 0
0
1
oO.OO 20.00 40.00 60.00 10.00 100.00 120.00
(7.a) FREOUENCY
Figure 2 shows the frequency responses of functions can, therefore, be replaced by two series of
IH2(f)l/Af and IHl(f)l versus system frequency. This terms. Each term of the series is a product of a
figure shows that the magnitudes of the frequency numerical value and a phase delay operator. If the
responses are nearly equal in the 58 to 62 Hz range. data window contains 2k samples and the time reference
The difference between the two frequency responses is at the middle of the data window, the function
increases as the frequency deviates farther from 60 Hl(f) is
Hz. The mismatch between the two frequency responses
is the cause of the errors in the estimated frequency Hl(f) = c-kej2nf(-k+O.5)AT + ( k-1)
'
-
ej2nf(-k+l.5)AT +
(11)
1. the estimated frequency deviations verses The LES algorithms designed for data sampled at
frequency curve has a minima at 32 HZ and a 720 Hz and two and three cycle data windows were used
maxima at 85 Hz. in these studies. (However, results of the algorithm
that uses two cycle data window only are presented.)
2. the estimated values of frequency deviation The values of frequency corresponding to frequency
are identical at different frequencies; for deviations in increments of 0.1 Hz were calculated.
example, estimated frequency deviation is -15 The information was stored in a look-up table. The
Hz when the frequency is 20 Hz and 4OHz. range in which the corrections can be applied was also
established for each design of the algorithm.
It is, therefore, essential that the proposed
technique be applied over a limited range. To avoid It was decided that the effectiveness of the
uncertainties, frequencies below 40 Hz and above 80 Hz technique be checked for frequencies in the 40 to 70
should not be estimated using this design of the LES Hz range. Since the frequency deviation estimates are
technique for measuring frequency. accurate in the 58 to 62 Hz range, the correction
procedure was applied in the 40-58 Hz and 62-70 HZ
While developing the LES algorithm used in this ranges only. The values of the frequency corre-
example, the sine and cosine terms were expressed as a sponding to the frequency deviations from -20 to +10
Taylor series in the neighborhood of f = 60 Hz. This Hz range were calculated using Equation 15 and the
version of the technique provides accurate estimates Newton's method. Some of the selected frequency
in a close proximity of 60 Hz and the extended version deviations and the corresponding true values of the
provides reasonable estimates in the 40 to 80 HZ frequency obtained in this manner are given in Table
range. However, if it is necessary to measure I. The third column is included in this table to show
frequency in the 10 to 80 HZ range, two versions of the error that would be in a frequency estimate if it
the LES algorithm should be used; one designed for a was not corrected using the procedure.
nominal frequency of 60 Hz and the other designed for
a nominal frequency of 25 Hz. Table I. The Calculated Values of the Frequency
Deviation and the Corresponding True Values of the
Frequency.
;d
Y
g Error in the
.d Estimated value True value estimated value
of the frequency of the frequency of the frequency
deviation (Hz) (Hz) deviation (Hz)
-14.8 39.89 -5.31
-14.7 40.18 -5.12
-14.6 40.46 -4.94
-14.5 40.73 -4.77
Hl(f) was defined as the frequency response of the The studies reported in this section used data
from simulated voltages for investigating the
function used for calculating V sine. Let H3(f) be effectiveness of the correction process. Voltages of
P frequencies in the 40 to 70 Hz range were simulated
the frequency response of the function for calculating
V cos& These frequency responses are not matched and sampled at 720 Hz in a software program. The
P frequency deviations were estimated using the sampled
over the entire range of the frequency. values and the LES algorithm. The estimates were then
improved using the look-up table. The errors observed
After using the procedure described in the last in the frequency estimates before and after improving
section the frequency is known; let it be f The
gains of the functions corresponding to the fhequency
. the estimates are shown in Figure 4. A study of this
figure shows that, before applying the corrections,
f H (f ) and H (f ) are known. To obtain the true the errors are large at off-nominal frequencies.
eh.;i&tei of the2 r&l and imaginary components of a However, the use of the proposed procedure reduces the
voltage phasor, divide the estimate of V sine by errors significantly. Also, the improvement is less
P when the frequency is in the neighborhood of 40 Hz;
IHl(fl)I and the estimate of V cose by IH3(fl) 1. the errors are up to 0.15 Hz in this region. This is
P
because the difference in the actual frequencies when
TESTING THE PROPOSED TECHNIQUE the estimated frequency changes by 0.1 Hz is much more
at 40 Hz than it is at 55 Hz.
A method for improving the accuracy of the
frequency estimates has been described in the last The measurements shown in Figure 4 were repeated
section. The effectiveness of the proposed method was using a look-up table that used frequency deviation
investigated using software generated data and experi- steps of 0.05 Hz. Figure 5 shows the results obtained
mental data collected from experiments conducted in in this manner. This figure shows that the measure-
the laboratory. This section describes some results ment errors at off-nominal frequencies are less than
from those studies and discusses the effectiveness o t the errors observed in Figure 4.
the method.
1577
The measurements shown in Figure 4 were also The data were obtained in the laboratory from a
repeated using linear interpolation for the inter- Model FDF dynamic frequency source manufactured by the
mediate values of the frequency deviations. Figure 6 Doble Engineering Company. The output of the source
shows that the errors observed are now insignificant can be adjusted from 0 to 140 volts and 40 to 79.99
over the entire 40 to 70 Hz range. Hz. The voltage was set at 110 V and frequency was
selected in the 40 and 65 Hz range. The voltage level
was reduced to 3 V peak value using an isolating
transformer and voltage divider combination. The
reduced level signals were applied to a k5 volts, 12
@ f o r e correction bit analog to digital converter. The digitized values
of the samples were recorded on a magnetic tape via a
PDP-11/60 computer. The voltage magnitude and
$1 0
After correction
frequency of the dynamic frequency source are accurate
within one and 0.001 percent respectively. The A/D
converter used has an accuracy of 0.025 percent in the
full range.
'9 \ , Figure 7 shows the frequency estimates of a 55 Hz
sighal obtained from the calculated frequency devia-
FREQUENCY (HZ) tions before and after applying the proposed proce-
Fig. 4 . The errors observed before and after dure. This figure shows that the mean error is
correcting the frequency estimates. approximately 0.1 Hz when the frequency is calculated
from the estimates of the frequency deviations.
0 \ However, when the proposed procedure is applied using
the look-up table, the mean error reduces signifi-
cantly. The fluctuations of the calculated frequency
about the mean value are due to noise in the input
data; these fluctuations remain practically unchanged.
?
4 After correction
I 7
'4;).00' 44100 ' a:OO ' 5!!W 5S.W W.00 M.W a.00
FREOUENCY
\ Before correction
=
El% After Correction
' 1
0 03
" P TIME *(1/720) SEC
0
9 Fig. 7. The frequency estimates of a 55 Hz signal
0
'4b.00' 44100 ' 48:OO ' YlOO ' if00 ' 60~00 ' 64100 ' WrOq
with and without correction.
FREQUENCY ( H Z ) The study was also repeated using the data of 40 Hz
signal. The mean value of frequency calculated using
Fig. 6. The errors observed before and after the LES technique was about 45.2 Hz - an average error
correcting the frequency estimates by linear of over five Hz. After the proposed technique was
interpolation. applied, the mean error reduced to 0.1 Hz. However,
the fluctuations of the frequency estimates about the
Testing with Experimental Data mean value increased substantially. The reason for
this increase is that the nature of the look-up table
In addition to the simulated data used in the is such that it amplifies noise.
studies reported in the last section, experimental
data obtained in the laboratory were used for testing The problem was alleviated by suppressing noise.
the effectiveness of the proposed technique. Data of Averaging, exponential smoothing, or low-pass filters
several off-nominal frequencies was used for this could be used for this purpose. The measurements of
purpose. The measurements were simulated in a the 40 Hz signal were repeated using the LES tech-
software program using integer arithmetic during nique. The estimates of the frequency deviations were
computations. In each case, the improved frequency improved by averaging 16 consecutive calculations.
estimates were obtained using a predetermined look-up The procedure proposed in this paper was then applied.
table. Results from the application of the technique Figure 8 shows the results obtained in this manner.
to measurements of frequency of 55 and 40 Hz signals This figure shows that the maximum measurement errors
are presented in this section. are of the order of 0.1 Hz.
1578
:I,,
d0 . 0 0 IM.OO
, ,
w.00
, I
40.00
,
T I M E +(1/7201 SEC
, ,
6w.w m.oo
, , ,
900.00
, ~
1o50.00
1950 to 1968 in System Operation, Design and Planning.
He joined, in 1968, the University of Saskatchewan
where he is presently Professor and Head of Electrical
Enuineerinu. His areas of interest are Power Svstem
Anilysis &d Power System Protection.
Dr. Sachdev is a Fellow of the Institution of
Engineers (India) and a Fellow of the Institution of
Fig. 8. Frequency estimates of a 40 Hz signal using Electrical Engineers, London (England). He is a
the LES technique and rejecting noise by Registered Professional Engineer in the Province of
averaging sixteen consecutive estimates. Saskatchewan.
CmcLuSIms
A method for extending the useful measurement
range of the Least Error Squares technique for
simultaneously measuring peak value of a voltage or
current and its frequency has been presented in this
paper. This method is useful in designing digital
meters and relays that need to measure system
parameters accurately over a large frequency range.
The proposed procedure is quite general in form
and can be used in conjunction with other digital
techniques for accurately measuring peak values and
frequency over a wide range of frequency.
REFERENCES
Sachdev, M. S. and Giray M. M., "A Least Error
Squares Technique for Determining Power System
Frequency", IEEE Transactions on Power Apparatus
and Systems, Vol. PAS-104, No. 2, February 1985,
pp. 435-444.
Sachdev, 1. S. and Giray M. M., "Measurement of
Local Frequency from Digitized Bus Voltage
Samples", Transactions of the Canadian Electric
Association, Engineering and Operation Division,
Vol. 23, Part 3, 1984, Paper No. 84-SP-149.
Sachdev, M. S. and Giray M. M., "Amplitude and
Frequency Measurements by Discrete Fourier
Transform and Least Error Squares Technique",
Transactions of the Canadian Electrical
Association, Engineering and Operation Division
Vol. 24, Part 3, 1985, Paper No. 85-SP-150.
Alexander, G. W., Corbin, S. L. and McNutt, N.
J., "Influences of Design and Operation Practices
on Excitation of Generator Step-up Transformer",
IEEE Transactions on Power Apparatus and Systems,
Vol. PAS-85, NO. 8, August 1966, pp. 901-909.
Gartner, G. and Birch. F. H., "Transformer
Overfluxing Protection", Report by Working Group
01 of Study Committee No. 3 4 , Electra, 1973,
CIGFE Reports, pp. 65-73.
Goff, L. E., "V/Hz Relay Protects Transformers
against Overexcitation", Electrical world, No.
168, September 1967, pp. 47-48.
BIOGRAPHIES
Mahmut M. Gira (S'83) was born in Ankara, Turkey
on December, 1949 .'
He received the B.Sc. degree in
Electrical Engineering from the Istanbul Technical
University, Istanbul, Turkey in 1973 and M.Sc. and
Ph.D. degrees from the University of Saskatchewan in
1979 and 1987 respectively.