A Smart Method Makes DFT More Precise for Power System Frequency

Estimation

Jun-Zhe Yang Chih-Wen Liu

Department of Electrical Engineering, National Taiwan University, Taipei, Taiwan

Abstract - A precise digital algorithm based on Discrete There have been many digital algorithms applied to
Fourier Transforms (DFT) to estimate the kequency of a estimating frequency during recent years, for example
sinusoid with harmonics in real-time is proposed. This Modified Zero Crossing Technique [2], Level Crossing
algorithm that we called the Smart Discrete Fourier Technique [3], Least Squares Error Technique [4-61,
Transforms (SDFT) smartly avoids the errors that arise Newton method [7], Kalman Filter [8-10], Prony Method
when frequency deviates from the fundamental frequency, [ll], and Discrete Fourier Transform (DFT) [12], etc. For
and keeps all the advantages of the DFT e.g., immune to real-time use, most of the aforementioned methods have
harmonics of fundamental Erequency, obtaining easily the trade-off between accuracy and speed [ 131. A smart digital
parameters of amplitude and phase, and even the recursive algorithm, namely Smart Discrete Fourier Transform
computing can be used in SDFT. These make the SDFT (SDFT) is presented and tries to meet the real-time use.
more accurate than conventional DFT based techniques. In SDFT has the advantages that it can obtain exact solution
addition, this method is recursive and very easy to when frequency deviates from hdamental frequency, its
implement, so it is very suitable for use in real-time. We speed is even faster than DFT, and it can get exact solution
offer the simulation results compared with conventional in the presence of harmonics.
DFT method and second-order Prony method to validate The organization of this paper is as follows: We describe
the claimed benefits of SDFT. in very detail the SDFT in section 11. DFT, Prony method
and SDFT are tested by four examples in section 111.
Keywords: Discrete Fourier Transforms (DFT), Frequency Finally, we give a conclusion in section IV.
estimation, phasor measurement.
11. The Proposed Digital Algorithm
I. Introduction This section presents the algorithm of the SDFT that
Frequency is one of the most important quantities in estimates the frequency and phasor from a voltage/current
power system operation because it can reflect the dynamic signal. Consider a sinusoidal input signal of frequency
energy balance between load and generating power. So 0 = 2nf given by:
frequency is always regarded as an index of the operating
practices, and utilities can know the system energy balance
situations by observing frequency variations. Frequency
may vary very fast in the transient events such that it is where X : the amplitude of the voltage/current signal,
difficult to track it accurately. In addition, there are many 4 :the phase angle of the voltage/current signal
devices, such as power electronic equipment and arc
fumaces, etc. generating lots of harmonics and noise in
Suppose that x ( t ) is sampled with a sampling rate (60*N)
modem power systems. It is therefore essential for utilities
to seek and develop a reliable method that can measure Hz waveform to produce the sample set { ~ ( k})
frequency in presence of harmonics and noise.
With the advent of the microprocessor, more and more k
x ( k ) = x cos(0 -+ 4)
microprocessor-based equipments have been extensively 60N
used in power systems. Using such equipment is known to
provide accurate, fast responding, economic, and flexible The signal x ( t ) is conventionally represented by a phasor
solutions to measurement problems [ 11. Therefore, all we (a complex number) X
have to do is to find the best algorithm and implement it.
X = XeJ* = x c o s +~ jXsinQ (3)

Then x ( t ) can be expressed as

where * denotes complex conjugate.

Moreover, the fundamental frequency (60Hz) component

0-7803-4403-0/98/$10.00 0 1998 IEEE 909

of DFT of { ~ ( k ) is given by

2nk
2 N-I
i, = - C x ( k + r ) e - ' y
N k=O

Combing Eq.(4) and Eq.(5) and taking frequency deviation

( w = 2n(60 + Af)) into consideration, we obtain:
Then Eq.(9) can be expressed as

So far the development of the algorithm of SDFT are the

same as the conventional DFT method. So the SDFT can
keep all advantages of DFT such as recursive computing
We rearrange Eq.(6) as the following manner. But in the DFT, it assumes that the frequency
deviation is small enough to be ignored, and it always
2 jz(l+z)rN-l considers 2, II! A , , so traditional DFT based methods
x, = -e N 60 C e j 2 n6 0XN k
N incur error in estimating frequency and phasor when
-.
k=O

(2+f)
frequency deviates from nominal frequency (60 Hz). If we
x -J2"(1+df)r
N-1 -~2n-*k want to get exact solution, we must take B, into
+-e 6o Ce
N k=O consideration. So we define

We use the following identity to simplify Eq.(7)

NB
sin -
N-l
And from Eq.(lO), we will find the following relations
,=O
sin -
2 Arcl = A, * a
B,,l = B, * a-'
Then Eq.(7) can be expressed as
Then

=A,, +B,+I=A, * a +B, *a-' (17)

i + 2 = 4 + 2 + Br+2=4+1* a + B,+I * a-' (18)

If we multiply ' a ' on both sides of Eq.( 17) and Eq.( 1S),
respectively, then we get

where e l = *60 N
, and 6 = -
2n(2 + -)Af
N
60 .?r+l * a = A, * a2 + B,
x r + 2 a = A,+I * aZ + B,+I
* *
(19)
(20)
Rearranging Eq.(9) further, we obtain
Subtracting Eq.( 13)ftom Eq.( 19) and subtracting Eq.( 17)
from Eq.(20), respectively, we obtain

2r+l* a - 2, = A, (aZ- 1)
* a - i,+,
ir+Z = Arrl(a' - 1)

Dividing Eq.(22) by Eq.(21), we get

Ifwedefine A, and B, as

9 10
Then rearrange Eq.(23) as follows sinusoidal waveform. Fig. 1 also shows the performance of
SDFT, method and conventional DFT method. It is
observed that conventional DFT method gives the wrong
frequency estimates. In Fig 2a, SDFT, is observed to
Solve W24) to obtain obtain the exact solution. When the SDFT and Prony
methods test the same signal as SDFT, but filtered by a
(2, + i r + z ) k J ( i r+ i r + -4i,+,*
2)2 Blackman window (window size = 16) for estimation, we
Cl= . Then from the
%+I find that the SDFT and Prony methods have similar
definition of 'a' in Eq.( 14), we can get the exact solution of performance. In Fig 2b, the same test signal without
the frequency. Blackman window tests SDFT and Prony, we find that
Prony is worser than SDFT in the presence of harmonic. In
60N Fig.3a, the frequency of test signal is changed linearly
f = 60 + Af = cos-' (Re( U)) * - (25)
2% from 60Hz to 63.5Hz during 1 second. From Fig.3b, we
can observe that the error of DFT increases with the
From Eq.(25), it is observed that SDFT can provide exact frequency deviation. However, the absolute error of SDFT
frequency using ?, ,?,+, and ?,+, in the absence of noise. is bounded below a small value. In FigAa, the frequency is
Moreover, we can estimate phasor after getting exact ' f' changed as sin wave and 3'd harmonic is also added in test
signal during 1 second. We can observe the errors of DFT
by the following equations:
and SDFT in Fig.4b, and the errors of Prony and SDFT, in
i"+,
*U-;,
FigAc. Although SDFT, can resist the effect of the 3rd
A, =- harmonic, the effect of frequency variation makes SDFT,
112 -1
get some small errors. Anyway, from Fig4 we can
conclude that SDFT-based algorithms (SDFT, SDFT,) are
better than DFT method and Prony method for frequency
estimation. By comparison of computation speed, Table 1
shows the AMD K6-200 CPU time of each method. There
are 960 data per second computed by each method (the test
Actually, if we assume that signal is the same as Fig.3a) without Blackman windows
x ( t ) = XI cos(wt + 4,) + X , cos(3wt + $,)
from the to estimate the frequency, while adding a Blackman
window will add 0.91 second to the computation. We find
beginning of development of the algorithm, we will derive
that SDFT is the fastest method in these computation,even
a polynomial equation similar to Eq.(24) that provide exact
faster than DFT, because SDFT counts frequency directly,
frequency in the presence of 3'd harmonic. We use SDFT,
but DFT has to count phase first and then use the phase
that has taken 3rdharmonic into consideration to distinguish
difference to count frequency. The faster speed of SDFT
ffom SDFT. Of course, any other integral order harmonic
over the Prony method is because recursion can be used in
still can be taken into consideration, for example SDFT,,
SDFT.
and SDFT,,, take 3rd,5" harmonic and 3rd,5Ih, 7 I h harmonic
into considerationrespectively.
Table 1 computation time
Although we can take all of the harmonics into
consideration, we still need digital filter to decay noise and IProny I DFT SDIT I SDIT~ SDFT,~SDIT~=
high order harmonics. The advantages of digital filter are time@) I 2.03 I 0.71 I 0.54 I 1.01 1 1.71 1 3.44
no voltage drop, no temperature drift, no noise addition,
and don't care any filter element feature. Besides these, IV. Conclusion
digital filter can be implemented in microprocessor-based In this paper we introduce the SDFT method and
equipment. These make us choose the digital filter to filter demonstrate its performance. SDFT both keeps the
noise and high order harmonics. There are many digital advantages of DFT and also deals with the cause of
filters that we can choose eg., hanning, hamming and frequency deviation errors, while taking integral harmonics
Blackman window. In simulations we will use Blackman into consideration. These aspects make SDFT a fast,
window for filtering. accurate and harmonic-resisting method. But we do not
suggest taking all the harmonics into consideration, since
111. Simulation Results that would require too much computation time.
Simulation results presented in this section were all Alternatively, using a smooth window to decay the high
simulated from Matlab and showed for a fair comparison order harmonics and just taking the low order harmonics
to DFT method and Prony method. In Fig.1, we showed into consideration will be more efficient and suitable for
that SDFT could obtain an exact solution identical to the power systems under real-time demands.
Prony method under frequency deviation in a pure

911
 60 I 0 1

60 30
~ , b y , . , SDFI - .t- ,SDFI>, ,- . ,
U M (=)

Fig1 Comparison of frequency estimates among DFT, Prony, SDFT, and Fig.2a Comparison of frequency estimates among Prony, SDFT, and SDFT,
SDFT,. (Test signal: x(t)=cos(wt), f-61Hz). with Blackman window. (Test signal: x(t)=cos(wt)+[Link](3wt),
F60.5Hz).

6011 ' ' ' ' ' ' " "OD5

lim (16)

Fig.2b Comparison of frequency estimates among Prony, SDFT, and SDFT, Fig.3a Frequency variation of test signal, x(t)=cos(wt), f-3.5*tHz
without Blackman window.
SDFI DFI

0
um 6"

Fig 3b Comparison of error of frequency estimates between DFT and SDFT Fig.4a Frequency variation of test signal, x(t)=cos(wt), M(HOSsin(2 K t)Hz.
,101 DFI SDFI i10= m SDFb
I " I ' ' ' ' I ' ' I I ' ' \ . ' ' I' ' ' I

UM(Uc)

Fig.4b Comparison of error of frequency estimates between DFT and SDFT F i g . 4 ~Comparison of error of frequency estimates between Prony and
SDFT,

91 2
 Reference Biographies
P J Moore, R D Carranza, A T Johns, “Model System Tests on a New Wen Liu was bom in Taiwan in 1964.
Numeric Method of Power System Frequency Measurement”, IEEE He received the B.S. degree in Electrical
Transactions on Power Delivery, Vol. 11, No. 2, April 1996, pp. 696- Engineering from National Taiwan
701 University in 1987, Ph.D. degree in electrical
Miroslav M. Begovic, Petar M. Djuric, Sean Dunlap, Arun G. Phadke, engineering from Come11 University in 1994.
“Frequency Tracking in Power Networks in the Presence of Since 1994, he has been with National
Harmonics”, IEEE Transactions on Power Delivery, Vol. 8, No. 2, Taiwan University, where he is associate
April 1993, pp. 480-486. professor of electrical engineering. His
[31 [Link], [Link], “A New Technique for Rapid Tracking of research interests include application of
Frequency Deviations Based on Level Crossings”, IEEE Transacfions numerical methods to power system, motor
on Power Apparatus and Systems, Vol. PAS-103, No.8, August 1984, control and GPS time transfer.
pp.2230-2236.
[41 I. KAMWA, R. GRONDIN, “Fast Adaptive Schemes for Tracking
Voltage Phasor and Local Frequency in Power Transmission and
Distribution Systems”, IEEE Transactions on Power Delivery, Vol. 7,
Jun-Zhe Ymg was bom at Tainan, Taiwan,
No. 2, April 1992, pp. 789-795.
in 1971. He received his B.S. degree in
M.S. Sachdev, M.M. Giray, “A Least Error Squares Technique For
electrical engineering from Tatung Institute
Determining Power System Frequency”, IEEE Transacfionson Power
of Technology in 1992 and M.S. degree fiom
Apparatus andSystems, Vol. PAS-104, No. 2, February 1985, pp. 437-
National Taiwan University in 1995. He is
443.
presently a graduate student in the electrical
M.M. Giray, [Link], “Off-Nominal Frequency Measurements %-

engineering department, National Taiwan

In Electric Power Systems”, IEEE Transactions on Power Delivery,
University, Taipei, Taiwan
Vol. 4, NO. 3, July 1989, pp. 1573-1578.
Vladimir V. Tenia, Milenko B. Djuric, Branko D. Kovacevic,
“Voltage Phasor and Local System Frequency Estimation Using
Newton Type Algorithm”, IEEE Transactionson Power Delivery, Vol.
9, NO. 3, July 1994, pp. 1368-1374
M.S. Sachdev, H.C. Wood, N.G. Johnson, “Kalman Filtering Applied
To Power System Measurements For Relaying”, IEEE Transactions.
on Power Apparatus and System, Vol. PAS-104, No. 12, December
1985, pp. 3565-3573.
Adly [Link], T. L. Daniel Hwang, ”Optimal Estimation of Voltage
Phasors and Frequency Deviation Using Linear and Non-Linear
Kalman Filter: Theory and Limitations”, IEEE Transactions o n
Power Apparatus and Syslems, Vol. 103, No. IO, October 1984, pp.
2943-2949.
Adly A. Girgis, William L. Peterson, “Adaptive Estimation of Power
System Frequency Deviation and its Rate of Change for Calculating
Sudden Power System Overloads”, IEEE Transactions on Power
Delivery, Vol. 5,No. 2, April 1990, pp. 585-594.
Tadeusz Lobos and Jacek Rezmer, “Real-Time Determination of
Power System Frequency”, IEEE Transactions on Instrumentation
and measurement,Vol. 46, No. 4, August 1997, pp. 877-88 1.
A. G. Phadke, J. S . Thorp, M. G. Adamiak, ”A New Measurement
Technique for Tracking Voltage Phasors, ‘Local System Frequency,
and Rate of Change of Frequency”, IEEE Transactions on Power
Apparatus andSystems, V01.102, No. 5, May 1983, pp. 1025-1038.
Ph. Denys, C. Counan, L. Hossenlopp, C. Holweck, ”Measurement Of
Voltage Phase For The French Future Deffence Plan Against Losses
Of Synchronism”, IEEE Transactionson Power Delivery, Vo1.7, No.1,
Jan 1992, pp. 62-69

913