You are on page 1of 6


2, APRIL 2000

A Precise Calculation of Power System
Frequency and Phasor
Jun-Zhe Yang and Chih-Wen Liu

Abstract—A series of precise digital algorithms based on the advantages that it can obtain exact solution in the presence
Discrete Fourier Transforms (DFT) to calculate the frequency and of harmonics and frequency deviation from nominal frequency.
phasor in real-time are proposed. These algorithms that we called The organization of this paper is as follows: We describe basic
the Smart Discrete Fourier Transforms (SDFT) family not only
keep all of the advantages of DFT but also smartly take frequency principle of SDFT in section II. DFT, Prony method and SDFT
deviation, and harmonics into consideration. These make the are tested by four examples in section III. Finally, we give a
SDFT family more accurate than the other methods. Besides, conclusion in section IV.
SDFT family is recursive and very easy to implement, so it is very
suitable for use in real-time. We provide the simulation results
compared with conventional DFT method and second-order Prony
method to validate the claimed benefits of SDFT. This section presents the algorithm of the basic SDFT that es-
Index Terms—Discrete Fourier Transforms (DFT), Frequency timates the frequency and phasor from a voltage/current signal.
estimation, phasor measurement. Consider a sinusoidal input signal of frequency ! = 2f as fol-
I. INTRODUCTION x(t) = X cos(!t + ) (1)

F REQUENCY and phasor are the most important quanti-
ties in power system operation because they can reflect
the whole power system situation. Frequency can show the dy-
X : the amplitude of the voltage/current signal,
: the phase angle of the voltage/current signal
Suppose that x(t) is sampled with a sampling rate (603 N ) Hz
namic energy balance between load and generating power, while
phasor can constitute the state of system. So frequency and
waveform to produce the sample set fx(k)g
phasor are regarded as indices for the operating power systems  k

k = 0; 1; 2; 1 1 1 ; N 0 1:
in practice.
x(k) = X cos ! +
However, utilities have difficulty in calculating those quanti- 60N
ties precisely. There are many devices, such as power electronic (2)
equipment and arc furnaces, etc. generating lots of harmonics The signal x(t) is conventionally represented by a phasor (a
and noise in modem power systems. It is therefore essential for complex number) x
utilities to seek and develop a reliable method that can measure x = Xej = X cos  + jX sin : (3)
frequency and phasor in presence of harmonics and noise.
With the advent of the microprocessor, more and more Then x(t) can be expressed as
microprocessor-based equipments have been extensively used xej!t + x3 e0j!t
in power systems. Using such equipment is known to provide x(t) = (4)
accurate, fast responding, economic, and flexible solutions to where 3 denotes complex conjugate.
measurement problems [1]. Therefore, all we have to do is to Moreover, the fundamental frequency (60Hz) component of
find the best algorithm and implement it. There have been many DFT of fx(k)g is given by
digital algorithms applied to calculating frequency or phasor N 01
2 X
x(k + r)e0j N :
during recent years, for example Modified Zero Crossing 2k
x^r = (5)
Technique [2], Level Crossing Technique [3], Least Squares N k=0
Error Technique [4]–[6], Newton method [7], Kalman Filter
Combing Eq. (4) and Eq. (5) and taking frequency deviation
[8], [9], Prony Method [10], and Discrete Fourier Transform
(DFT) [11], etc. For real-time use, most of the aforementioned
( = 2 (60 + 1 ))
!  f into consideration, at last, we obtain:

x sin 2
methods have trade-off between accuracy and speed [12]. N 1
^r = N ej 60N (1f (2r+N01)+120r)

Unlike other methods, a series of precise digital algorithms, x
namely Smart Discrete Fourier Transform (SDFT) family, are sin 2
presented and tries to meet the real-time use. SDFT family has
N 2
x3 sin 2
+N e0j 60N (1f (2r+N01)+120(r+N01))

sin 22
Manuscript received February 4, 1999.
The authors are with the Department of Electrical Engineering, National
Taiwan University, Taipei, Taiwan.
Publisher Item Identifier S 0885-8977(00)03461-0. (6)
0885–8977/00$10.00 © 2000 IEEE

Authorized licensed use limited to: National Taiwan University. Downloaded on March 9, 2009 at 04:14 from IEEE Xplore. Restrictions apply.

we will find the following rela. x^r+1 and x^r+2 in the presence of frequency Br = 2 e 60N : (8) phasor using x N sin 2 deviation. (12). and Following similar steps developed previously. we can get: ^ xr  Ar . 2 : the phase angle. and 2 = : Ar = x^r+1a230a 10 x^r (19) N  1f  If we define Ar and Br as N 3 sin 60N sin 2 N 1 X = abs(Ar ) 3  1f  (20) Ar = x ej 60N (1f (2r +N 01)+120r)  (7) sin 60 N sin 21  = angle(Ar ) 0 60N 2 (1f 2 (N 0 1)): (21)  x3 sin N22 0j  (1f (2r+N 01)+120(r+N 01)) It is observed that SDFT can provide exact frequency and ^r . . (9). and after some algebraic manipulations we obtain: ^ = xr e ^ 3 a2 0 (^xr + x^r+2) 3 a + x^r+1 = 0: xr+1 (17) N sin 21 Solve Eq. the first half development of the algorithm of ( ) = X1 cos(!t + 1) + X2 cos(m!t + 2) x t (22) SDFT is the same as the conventional DFT method. Eq. we can estimate phasor after getting exact “f ” by the where  1  2 2 + 60 f following equations: 1 = 2601Nf . In the conventional DFT.YANG AND LIU: A PRECISE CALCULATION OF POWER SYSTEM FREQUENCY AND PHASOR 495 Moreover. (17) to obtain q (^xr + x^r+2) 6 (^xr + x^r+2)2 0 4^x2r+1 x31 sin N22 0j(=60N )(1f (2r+N 01)+120(r+N 01)) a= + e 2^xr+1 : N sin 22 Then from the definition of “a” in Eq.  2 1f )r NX (1+ 01 j 2 1f k r = arctan(imag(^xr )=real(^xr )) (10) ^r = x x j N e N 60 e 60N r 0 r01 k=0 f = 60 + 2 2 60N: (11) 3 2  1 f N 01 X Conventional DFT methods incur error in estimating fre. (7) and Eq. we take harmonics into consideration. Assume a sinu- Then Eq. So the where SDFT can keep all advantages of DFT such as recursive X1 . in the SDFT we take Br into 1 (2 + 60 ) f consideration. (15) and x1 Eq. + N e N 60 )r x 0j (1+ k=0 quency and phasor when frequency deviates from nominal frequency (60Hz). computing manner. Restrictions apply. it assumes that 1 . (6) can be expressed as soidal signal of frequency ! =2 f with mth harmonic given ^ = Ar + Br : xr (9) by: Actually. we can get the exact x2 sin N23 + solution of the frequency f = 60 + 1f = cos01(Re(a)) 3 602N : (18) N sin 23 Authorized licensed use limited to: National Taiwan University. X2 : the amplitude. Therefore. (8). Next. + Nx ej N (1+ 60 )mr k=0 m1f tions Ar+1 = Ar 3 a (13) (m 0 1 + 60 ) j 2 Br+1 = Br 3 a01: (14) 2e N k 2 1f )mr NX x3 0j (1+ 01 + N e N 60 Then ^ xr+1 = Ar+1 + Br+1 = Ar 3 a + Br 3 a01 (15) k=0 m1f (01 0 m + 60 ) ^ = Ar+2 + Br+2 = Ar+1 3 a + Br+1 3 a01 0j 2 k xr+2 2e N : (23) = Ar 3 a2 + Br 3 a02: (16) Then sin N21 j(=60N )(1f (2r+N 01)+120r) There are three unknown variables in Eq. Downloaded on March 9. (16). 2009 at 04:14 from IEEE Xplore. However. So we define 0j 2 k 2e N a = ej( 60N (21f +120)): (12) 2 1f NX 01 And from Eq. the frequency deviation is small enough to be ignored.

we need ^ ^ five equations to solve this problem. Downloaded on March 9. we used these four cases to compare the performance of these methods. . 15. 5th . Restrictions apply. xr+1 . nonintegral harmonics also can be developed. hence. harmonics and noise. Hence. ^ ^ ^ xr+3 and xr+4 . Test signal: v (t) = cos(!t). any = Ar 3 a + Br 3 a01 + Cr 3 am + Dr 3 a0m (26) other integral order harmonic can be taken into consideration too. we can get exact frequency and phasor in the presence of one harmonic. 5th harmonic and 3rd . SIMULATION RESULTS x32 sin N24 Most of the frequency estimation methods are concerned with Dr = sin 24 the performances of the four cases: frequency deviation. 1. sampling frequency 960 Hz. to denote that 3rd ^ xr+1 = Ar+1 + Br+1 + Cr+1 + Dr+1 harmonic has been taken into consideration. We use SDFT. where x2 sin N23 Similarly. 2. xr+2 .5 Hz. NO. simulated frequency 59.496 IEEE TRANSACTIONS ON POWER DELIVERY. 7th harmonic into consideration respectively. Sim- 2 e0j (=60N )(1f (2mr+N 01)+60(2mr+mN 0m+ N 01)) ulation results presented in this section were all simulated from Authorized licensed use limited to: National Taiwan University. Of course. for example: SDFT35 and SDFT357 take 3rd . 2 e0j (=60N )(m1f (2r+N 01)+60(2mr+mN 0m0N +1)) III. VOL. APRIL 2000 Fig. fre- N quency variation. We Cr = use SDFTn to denote that nonintegral harmonics has been N sin 23 taken into consideration. (25). 2009 at 04:14 from IEEE Xplore. 2 e0j (=60N )(m1f (2r+N 01)+60(2mr+mN 0m +N 01)) and  m1f  (24) 2 0m 0 1 + 60 4 = N : ^ = Ar + Br + Cr + Dr xr (25) There are five unknown variables in Eq. 2 ej (=60N )(m1f (2r +N 01)+60(2mr+mN 0m0N +1))  1  3 sin N24 2 m 0 1 + 60 m f + x2 3 = N sin 24 N . So using xr .

5 Hz during 1 second. sampling frequency 960 Hz. Test signal: v (t) = cos(!t). 1(a). we showed that SDFT family and the Prony frequency and phasor. 2. simulated frequency variation form 59. By comparison of computation speed. frequency deviation in a pure sinusoidal waveform. simulated frequency variation like sin wave during 1 second. Downloaded on March 9. Restrictions apply. 3. sampling frequency 960 Hz. 1(d) shows the AMD K6-200 CPU time of each method. .5 Hz to 60. It is clear Prony method. 2009 at 04:14 from IEEE Xplore. that conventional DFT method gives the wrong answer in In Fig.YANG AND LIU: A PRECISE CALCULATION OF POWER SYSTEM FREQUENCY AND PHASOR 497 Fig. Matlab and shown for a fair comparison to DFT method and show the performance of conventional DFT method. We also There are 960 data per second computed by each method. Test signal: v (t) = cos(!t). Fig. method could obtain an exact frequency calculation under Fig. Authorized licensed use limited to: National Taiwan University.

simulated frequency 60 Hz. simulated frequency 60. sampling frequency 960 Hz. Another test in Fig. Downloaded on March 9. . The frequency of test signal in Fig. Authorized licensed use limited to: National Taiwan University. variation. In Fig. 15. sampling frequency 960 Hz.5 Hz to 60. 4. Fig. 2 and Fig. Test signal: v (t) = cos(!t). Test signal: v (t) = cos(!t) + 0:05 cos(3!t) + 0:02 cos(5!t) + O:O1 cos(7!t).498 IEEE TRANSACTIONS ON POWER DELIVERY. 2 is varied linearly It is observed that the errors of conventional DFT method are form 59.5 Hz during 1 second. VOL. we put the emphasis on the frequency to make the frequency change like a sinusoid: f =60+ sin(2t). APRIL 2000 Fig. NO.05 Hz. 3 is larger than SDFT family in terms of frequency and phasor. 5. Restrictions apply. 3. 2009 at 04:14 from IEEE Xplore. 2. with 1% noise.

” IEEE Transactions on Power Delivery. Counan. M. Kovacevic. no.S. pp. and A. pp. Girgis and William L. J. and Ph. vol. May 1983. PAS-103. no. T. [7] Vladimir V. Petar M. Milenko B. REFERENCES Taipei. “Kalman Filtering Ap- In Fig. their performance. G. 9. H. Downloaded on March 9. is better than SDFT. Holweck. 1997. Local System Frequency. R. Although pp. He received his B. Terzija. Djuric.” IEEE Transactions on Power Apparatus and Systems. 102. 696–701. His other research interests in- on Power Apparatus and Systems. vol. Prony method can change its window size. M. Sachdev and M. but Of Voltage Phase For The French Future Deffence Plan Against Losses Of Synchronism. IV. Apr. Sachdev. no. 1985. 4.05Hz and added 3rd . no. trical Engineering from National Taiwan University in 1987. PAS-104. Srinivasan.S. pp. Adamiak. vol. [12] Ph. 1992. Feb. it must “Voltage Phasor and Local System Frequency Estimation Using Newton take noise into consideration. 1368–1374. “Frequency Tracking in Power Networks in the Presence of Since 1994. PAS-104. He is a member of the IEEE and serves as 480–486. SDFT family both keeps the advantages of [11] A. a reviewer for IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS. He received the B. Taiwan. no. 2.” IEEE Transactions on Power De- phasor in this case. We pp. Restrictions apply. and C. 4. no. 3. Djuric. “A New Measurement DFT and also deals with the difficulty of frequency deviation er. 1. Determining Power System Frequency. operation. Hossenlopp. Aug. Sean Dunlap. vol. It is shown that SDFT is better than DFT and Dec. 62–69. Apr. harmonic-resisting methods. 877–881. pp. where he is associate Harmonics. SDFT35n and SDFT357n get the exact frequency and ments in Electric Power Systems. “Fast Adaptive Schemes for Tracking 5th and 7th harmonics into test signal in Fig. “Adaptive Estimation of Power SDFT357n is better than SDFT. and Branko D. Technique for Tracking Voltage Phasors. pp. if SDFT family can be speeded up by Jun-Zhe Yang was born in Taiwan in 1971. while taking harmonics. GRONDIN. . degree from National Taiwan University in 1995. “A Least Error Squares Technique for better performance. C. National Taiwan University. pects make SDFT family accurate. July 1994. 1990. and the rest of SDFT family. no. 789–795. S. Thorp. We know that if a method can be used in real world. we recommend SDFT family. [1] P. [9] Adly A. and Arun G. Nguyen and K. 1992. Carranza. PART I. 46. His [3] C. if precise answer is essential. D. pp. Jan. and Systems. KAMWA and R. 5. Peterson. and M. Denys. 5. Apr. SDFT has better performance than DFT. Fig. we didn’t show the performance of Prony Method.” IEEE Transactions on Instrumentation and mea- In this paper we introduce the SDFT family and demonstrate surement. 3565–3573. Phadke. vol. pp. and we add one percent of white noise into signal. 2. and N.YANG AND LIU: A PRECISE CALCULATION OF POWER SYSTEM FREQUENCY AND PHASOR 499 We designed the test frequency to be 60. Begovic. G. size in this test. S. vol. no. J.” IEEE Transactions on Power Delivery. S. no.” IEEE Transactions monitoring. Johns. degree in Elec- 1996. Chih-Wen Liu was born in Taiwan in 1964. DFT is still a good choice. “Measurement If only rough answer is required. expect SDFTn .D. no. Authorized licensed use limited to: National Taiwan University. 12. July 1989. The frequency of test signal in Type Algorithm. [6] M. [4] I. 1025–1038. pp.S.S. it still can’t have [5] M. 7. vol. 585–594. vol. problems. Giray and M. 2009 at 04:14 from IEEE Xplore. Phadke.” IEEE Transactions since the performance of Prony method depends on window on Power Apparatus and System. 3. we Voltage Phasor and Local Frequency in Power Transmission and Distri- bution Systems. pp. T. vol. and Rate of Change of Frequency. “Real-Time Determination of Power System Frequency. He is presently a graduate student in the electrical engineering department. 7.” IEEE Transactions on Power Delivery. plied to Power System Measurements for Relaying. [8] M. Wood. “Model System Tests on a New Numeric Method of Power System Frequency Measurement. System Frequency Deviation and its Rate of Change for Calculating Sudden Power System Overloads. In Fig. S. degrees in electrical engineering from Cornell University in 1992 and 1994. These as. 1985. “Off-Nominal Frequency Measure- SDFT357 . 2. 1984. 1993. Moore. electrical engineering from Tatung Institute of Technology in 1992 and M. noise into consideration. Johnson.” IEEE Transactions on Power Apparatus rors. L. he has been with National Taiwan University. Apr. no. 8. “A New Technique for Rapid Tracking main research area is in application of computer technology to power system of Frequency Deviations Based on Level Crossings. Giray. 4a. vol. 5 is 60Hz.” IEEE Transactions on Power De- livery. livery. pp. professor of electrical engineering. 4. 2. 1573–1578. 11.” IEEE Transactions on Power Delivery. 2. protection and control. CONCLUSION [10] Tadusz Lobos and Jacek Rezmer. clude GPS time transfer and chaotic dynamics and their application to system Aug. G. vol.” IEEE Transactions on Power Delivery. 5. vol. [2] Miroslav M. no. 437–443. 2230–2236. degree in the advanced computing architecture. and M. find that Prony method is very sensitive to harmonic. believe that SDFT family has a great potential to replace con- ventional DFT method for power system frequency and phasor calculation in the future. C. Sachdev. 8.