You are on page 1of 11

www.ietdl.

org
Published in IET Generation, Transmission & Distribution Received on 18th June 2009 Revised on 12th October 2009 doi: 10.1049/iet-gtd.2009.0320

ISSN 1751-8687

Dynamic phasor and frequency estimator for phasor measurement units


R.K. Mai1 Z.Y. He1 L. Fu1 W. He1 Z.Q. Bo2
Department of Electrical Engineering, Southwest Jiaotong University, Sichuan Province 610031, Peoples Republic of China The Automation & Information Systems, Products R&D, AREVA T&D UK Ltd, Stafford ST17 4LX, UK E-mail: Mairk@live.com
2 1

Abstract: Synchrophasors and frequency estimations play an increasingly important role in power systems. Discrete Fourier transform (DFT) may introduce errors into phasor and frequency estimations under dynamic conditions, such as power oscillation. A dynamic phasor and frequency estimator for phasor measurement units (PMUs) is proposed to improve accuracy by considering dynamic characteristics of power systems expressed as Taylor derivatives. Firstly, phasor estimations of sequence components of difdent data windows are attained via DFT. Secondly Taylor derivatives, expressed by adjacent phasors at different data windows, are employed to improve accuracy of estimations by reassigning estimations with the knowledge of dynamic characteristics. Finally, dynamic characteristics of positive sequence components are applied to estimate the fundamental frequency with less delay compared to a state-of-the-art algorithm, phasor-based synchronised frequency measurement (PSFM). To verify the performance, a set of digital dynamic tests and a power oscillation model using PSCAD/EMTDC are presented. The simulation results show that proposed algorithm can achieve highly accurate estimations of phasor and frequency over a wide range of dynamic conditions. Even though a minor increase in calculation burden is required, this technique provides accurate phasor and frequency estimations without changing the legacy structure of PMU devices. Additionally, it has been found that proposed algorithm really meets the needs of online applications.

Nomenclature
AT AH A kAk Re (A) Im(A) diag(A) arg (A)

transpose of A conjugate transpose of A conjugate of A Euclidean norm of A real component of A imaginary component of A diagonal matrix formed from the vector A angle of A

important parameters to indicate system state. With these estimations from phasor measurement units (PMUs) around power grids, not only the capability for tracking system dynamic phenomena in real time is available, but also high-level applications such as state estimate [1, 2], wide area control systems [3] and adaptive relaying schemes [4] can be achieved. These allow power systems to operate in a safe manner and can help avoid slipping into a serious failure mode. In general, as power systems often operate close to stability limits, any operations or faults may lead power systems into transient and abnormal conditions, such as power oscillation. The performance of phasor and frequency estimations of PMUs under dynamic conditions plays an important role in many applications, as it is precondition for high-level applications to make a correct decision on how to take actions to stabilise power 73

Introduction

Phasor and frequency estimations of fundamental components of three-phase signals are two of the most IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 1, pp. 73 83 doi: 10.1049/iet-gtd.2009.0320

& The Institution of Engineering and Technology 2009

www.ietdl.org
systems. The waveforms of supplied signals may not be a standard cosine under transient and abnormal conditions. The estimation from discrete Fourier transform (DFT), which is widely implemented in commercial PMUs as its good harmonic rejection, is collapsed because of spectrum leakage phenomenon. Hence, accurate dynamic phasor and frequency estimator (DPFE) are necessary for PMUs and how to improve dynamic estimation performance has raised a few researchers interest [5 7]. It is not easy to implement innite impulse response (IIR) lter or smart lter, such as Kalman lter and notch lter, for PMU applications because of its phase delay uncertainty, even though they have the advantage of fast response to dynamic signal and less computation burden [8, 9]. A nite impulse response (FIR) lter is usually chosen as a phasor estimator of commercial PMUs because of its constant phase delay. Methods [10 12] based on frequency modulation are proposed to eliminate error caused by leakage effect and grid effect. However, their performance may deteriorate during power oscillation because both frequency and amplitude of signal are oscillating. Although dynamic phasor estimates (DPE) [13], using least-square method to calculate a dynamic phasor, improves the accuracy of phasor measurement under dynamic conditions, it requires more than three times computation of traditional DFT for a second-order model and it is sensitive to harmonics. Frequency is a vital parameter in power systems as it reects the balance between generation and load. Phasorbased synchronised frequency measurement (PSFM) [14] has recently been proposed to improve frequency estimation accuracy via adjusting the reference time. Since its frequency estimation is calculated by two phasors which have the same time distance to reference time, this approach needs more computation burden to calculate two more phasors at arbitrary time or has a time delay by using the phasor estimations at previous and next report time. That is a shortcoming for applications that have limited processor bandwidth or require a fast response. To overcome aforementioned problems, a DPFE is proposed to estimate the phasors and frequency of supplied signals during dynamic conditions. The organisation of the paper is as follows. Three-phase dynamic signal model and phasor estimation from DFT are introduced, along with the reason why DFT cannot give accurate estimation under dynamic conditions, in Section 3. The sequence model, derived from three-phase dynamic signal model, is presented in Section 4. The proposed algorithm is described in detail in Section 5. Section 6 summarises some presentations of the representative results obtained in the simulation study. Finally, the conclusions are drawn in Section 7. 74

2 Phase signal model and DFT estimate


2.1 Dynamic signal model
In power systems, phase quantities are not standard cosine functions with xed frequency and amplitude but are timevarying functions. Besides, as it is not likely to have fundamental frequency of power systems excurse far away for its nominal condition, it is reasonable to express threephase phasors as combinations of low-frequency p bandlimited phasors and a constant rotation vector 2ejv0 n , respectively, in discrete form as follows 2 0 3 3 Pa (n) p l a(n) jv n Pabc (n) 4 Pb (n) 5 24 l2 b(n) 5e 0 Pc (n) l1 c(n) 2

(1)

where Pa (n), Pb (n), Pc (n) and a(n), b(n), c(n) are three-phase phasors and low-frequency band-limited phasors at the nth sample. v0 2p=N is the discrete angular velocity of rotation vector. N is the number of samples per cycle. A phase rotation operator l ej2=3p is dened to rotate phasor vectors forward by 1208. The nth three-phase samples are expressed using real component of Pabc (n) as follows 3 l0 a(n) p 6 7 Sabc (n) 2Re[Pabc (n)] 4 l2 b(n) 5ejv0 n l1 c(n) 2 0 3 l a(n) 6 7 4 l2 b(n) 5 ejv0 n l1 c(n) Since we know that 1 Re[Pabc (n)] {Pabc (n) [Pabc (n)] } 2 Because the state of power systems changes slowly normally, low-frequency band-limited phasors can be accurately approximated by Taylor series within a short period of time by (3)
K X1 k0

(2)

x(n)

k!

x(k) nk Dx

(3)

where x [ {a, b, c} is one of three phases. K is the order of dynamic model. Dx is Taylor error of x phase. x(k) is the kth derivative of x(0). The higher the K is, the more accurately the dynamic characteristic can be described, while it needs more computation. IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 1, pp. 73 83 doi: 10.1049/iet-gtd.2009.0320

& The Institution of Engineering and Technology 2009

www.ietdl.org
By substituting (3) into (2) and ignoring Dx , the dynamic signal model is given by ! 3 K K X 1 (k) k P 1 (k) k jv0 n 0 jv0 n a ne a n e l 6l 7 k! 6 k0 k! 7 k0 6 7 ! 6 7 K X 1 (k) k K 6 2 P 1 (k) k jv n 7 Sabc (n) 6 l b n e 0 l1 b n ejv0 n 7 6 7 k! 6 k0 k! 7 k0 6 7 ! 6 P 7 K K 1 X 1 (k) k 4 1 (k) k jv0 n 2 jv0 n 5 c ne c n e l l k! k0 k! k0 2
0

3 Sequence component signal model


For most of the time, a power system is at or near a balanced state, even under dynamic conditions. This means A, B and C have nearly the same value. If this information, neglected by three-phase signal model, can be used properly, it is helpful for improving the accuracy of phasor and frequency estimations. Therefore symmetrical component method is employed to obtain information out of three phases and we have S012 (n) LSabc (n) where
0 1 4 l0 L l 3 l0

(4)

(7)

2.2 DFT estimator


Now we can calculate the phasors at reference point with DFT as follows ~ P abc (m)
Mm X nMm

l0 l1 l2

3 l0 l2 5 l1

h(n M m)Sabc (n)ejv0 n

3 2 0 3 l0 A lA 6 2 7 6 2 7 4 l B 5EK ,0,m,M 4 l B 5 EK ,2,m,M l1 C l1 C where A [a , . . . , a


(0) (K )

The estimations of sequence phasors at mR with data model (7) are given by (8) (5) ~ P 012 (m)
Mm X nMm

h(n M m)S012 (n)ejv0 n h(n M m)Sabc (n)ejv0 n

L ]

Mm X nMm

B [b(0) , . . . , b(K ) ] C [c (0) , . . . , c (K ) ] EK ,q,m,M [e(0, q, m, M), . . . , e(K , q, m, M)]T X 1 Mm e(k, q, m, M) h(n M m)nk ejqv0 n k! nMm where M (L 1)=2 and L is the length of data window. m mC mR is the sample distance between the centre of data window (mC ) and reference point (mR ). The coefcient of window function is dened as follows & h(n) = 0, 0, 0 n L1 n , 0, n ! L (6)

~ LP abc (m) T H m a b EK ,0,m,M m b a EK ,2,m,M

(8)

where we dene the zero, positive and negative sequence phasors as follows, respectively 1 m (l0 A l2 B l1 C)T 3 1 a (l0 A l0 B l0 C)T 3 1 b (l0 A l1 B l2 C)T 3 (9) (10) (11)

~ Estimation P abc (m) is a function against derivatives of phasors at reference point and h(n). For a standard cosine signal and synchronous sampling, the low-frequency band-limited phasors are constant and all its derivatives are zero except the zero derivative. Therefore the estimation is equal to theoretical value as long as e(0, 2, m, M) 0. Under dynamic conditions, errors arise by the combined effects of signals derivatives and h(n). The larger the derivatives are, the bigger the error will be for the same h(n). IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 1, pp. 73 83 doi: 10.1049/iet-gtd.2009.0320

4 Dynamic phasor and frequency estimator


Even though we can obtain a balanced three-phase system most of time, it is important to develop an estimator that can attain accurate estimations under unbalanced condition.

4.1 Zero sequence phasor estimator


The zero sequence component may not be equal to zero when an earth fault exists, that is m = 0. Taking zero sequence 75

& The Institution of Engineering and Technology 2009

www.ietdl.org
phasor estimation out of (8), we have ~ P 0 (m) mT EK ,0,m,M mH EK ,2,m,M (12) time correctly

m(tr ) e jv0 nr a(tr ) e jv0 nr b(tr ) e jv0 nr

K X1

k! k0
K X1

m(k) nk r a(k) nk r b(k) nk r


(19)

For K order system, it needs (K 1) equations to form a set of formulas to solve m out. After the derivation of Appendix 1, it is possible to have  m (E0 )1 P 0 (13)

k! k0
K X1

k! k0

The kth derivative of estimation of zero sequence phasor is given by

where nr (tr tw )fs is the sample point distance between the time of centre of data window tw and an absolute report time tr . fs is the sampling rate. Via an inverse transform, the phasors at tr can be expressed as follows [ Pa (tr ) Pb (tr ) Pc (tr ) ]T p 1 2L [ m(tr )

m(k) m(k) jm(k 1 K )


where m(k) is the kth element of vectorm.

(14)

a(tr ) b(tr ) ]T
(20)

4.2 Positive and negative phasor estimator


After taking positive and negative sequence phasors out of (8) and with some further manipulation, we have ( ~ P 1 (m) aT EK ,0,m,M bH EK ,2,m,M ~ [P 2 (m)] aT (EK ,2,m,M ) bH (EK ,0,m,M )

4.4 Dynamic frequency estimator


In this section, systems frequency and its rate of change are estimated through positive sequence component derivatives. First of all, dynamic positive sequence phasor is redened as an FM/AM model

(15)

a(n) euhn1=21n ej(wrn1=2sn

(21)

After the derivation of Appendix 2, the estimations of positive and negative sequence phasors are given by (16)

g (E12 )1 P 12

(16)

where u, h and 1 are the parameters to determine the amplitude of a(n) whereas w, r and s are the initial angle, frequency and rate of change of frequency, respectively. The rst and second derivatives of (21) at the centre of data window are given by
2 2 d[a(n)] [h 1n j(r ns)]euhn1=21n ej(wrn1=2sn ) jn0 dn

We have the kth derivatives of estimations of positive and negative sequence phasors

a(k) g(k) b(k) [g(k 1 K )]


where g(k) is the kth element of vector g.

(17) (18)

(h jr)a(0)

(22)

d2 [a(n)] dn2

4.3 Estimation at report time


The sequence phasors at the time of centre of current data window are available, but they are not the phasors to be sent as synchrophasors. Generally speaking, the time at the centre of the data window is decided by sampling time which has no knowledge of absolute time while the report time (absolute time) is derived from a GPS satellite receiver. Therefore these two times are not guaranteed to be coincident. More procedures should be carried out to express the estimations, given by (19), at an absolute report 76

 2 2 [h 1n j(r ns)]2 euhn1=21n ej(wrn1=2sn )    uhn1=21n2 j(wrn1=2sn2 )  (1 js)e e (h jr) a(0) (1 js)a(0)
2

n0

(23)

Combining (22) and (23), we have & ' d[a(0)] =a(0) r Im dn ( !2 ) d2 [a(0)] d[a(0)] =a(0) s Im =a(0) dn2 dn (24)

(25)

& The Institution of Engineering and Technology 2009

IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 1, pp. 73 83 doi: 10.1049/iet-gtd.2009.0320

www.ietdl.org
The frequency at report time can be described as fest f0 (r snr )fs =(2p) (26) Step 1: If the coefcient of FIR lter is changed, coefcient tables for (E0 )1 and (E12 )1 should be regenerated accordingly. Step 2: Three-phase phasors at current data window are calculated using (5). Step 3: Sequence phasor estimations are obtained via (8) and the modication to let previous six phasors have the same reference point is done by (28). Step 4: The reassigned sequence phasors and their derivatives are calculated by (14), (17) and (18). Step 5: The phasor estimations at the report time are calculated by (19) and the phasor of each phase is given by (20). Step 6: The frequency at the report time is attained by (26) and then go back to step 1. is valid to be applied for real-time application and the steps taken are shown below:

where f0 fs =N is the fundamental frequency estimation calculated according to fs , which is a variable when frequency tracking technique is employed.

4.5 Practical application of DPFE


The computation burden of signal processing algorithms is another important factor that should be considered, as there is a limited processing bandwidth available for applications. In order to reduce computation burden, it is necessary to make a few assumptions. As described above, because the fundamental frequency of power systems varies slowly, it is reasonable to assume that the sampling frequency fs remains as a constant within the analysed window even though frequency tracking is applied fs (L 1)=[t(M) t( M)] (27)

From most of situations we have investigated (oscillation frequency is less than 2.5 Hz), the second-order model (K 2) is accurate enough for slow oscillation conditions for PMU applications. To reduce computation burden, previous two phasors of centre of data windows are reused. Some modications are required to let them have the same reference point ~ P 012 (m1 ) ejv0 m1 P012 (m1 ) ~ P 012 (m2 ) ejv0 m2 P012 (m2 ) (28)

Performance analysis

P012 (m1 ) and P012 (m2 ) are previous two phasors with their own centre of data window as reference point. m1 and m2 are the sample distances away from the centre of current data window. As M, K and h(n) are decided before hand, (E0 )1 and (E12 )1 can be calculated off-line for different values of m1 , m2 and stored into a look-up table. Applying these assumptions to the algorithm can dramatically have computation burden within an acceptable level as shown in Table 1. Compared to the traditional DFT algorithm, it only needs a xed more operations despite of the length of data window and it is easily handled by a DSP processor. Therefore this new algorithm

In order to provide a comprehensive performance verication of DPFE, DFT and DPE are employed to be as comparative phasor estimators. The frequency estimators derived from DPE, PSFM and the phasor-based frequency estimator (PBFE) used in commercial PMUs are used as comparative frequency estimators. Two sets of simulation tests are performed, a digital signal model test and a PSCAD/ EMTDC test. In the rst set, theoretical value of phasors and frequency can be derived easily, so total vector error (TVE) [15] and frequency estimation errors are employed to verify the performance of proposed algorithm under various conditions. Another set is carried out by an oscillation model in EMTDC. Both frequency tracking technique [16] at N 48 samples per cycle (as some commercial products do) and xed sampling rate fs 2400 Hz) for a 50 Hz system are employed among the tests. Rectangle window function is applied to two-cycle data windows, namely L 96 and h(n) 1/L. All the simulation results presented in this section are simulated by Matlab and PSCAD.

5.1 Digital signal model test


The performance of proposed algorithm is veried in presence of harmonics and under a few dynamic conditions.

Table 1 Comparison of computation burden between DPFE and DFT Add DPFE DFT difference 3L 18 3(L 2 1) 21 Subtract Multiply Divide Look up 22 2 20 3L 27 3L 4 23 3 2 1 3L 18 3L 2 16

IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 1, pp. 73 83 doi: 10.1049/iet-gtd.2009.0320

77

& The Institution of Engineering and Technology 2009

www.ietdl.org
5.1.1 Harmonic test: The requirement of harmonic
rejection is indispensable for power systems. A static simulation test was performed to evaluate the performance of harmonic rejection and the parameters of signals are shown in Table 2. TVEs of DFT and DPFE are within 0.45%, but that of DPE is up to 1.22% as shown in Fig. 1.

5.1.2 Power system oscillation: Because power oscillation may reduce the available transmission capacity and even lead to cascading outages, it is critical to have accurate phasor estimations to achieve correct judgments to bring the state of power system back to nominal condition as soon as possible. Let the low-frequency band-limited phasor be described mathematically by (29)
& a(t) cos (2ptfpo ) j0:5, j0:5 0 t 1 otherwise (29)

Figure 2 Real component, imaginary component and magnitude, angle of A-phase low-frequency band-limited phasor

The rst derivative of phasor can be used to derive signals instant frequency considered as reference frequency 1 d{Arg[a(t)ej2pfcrt t ]} 2p dt

b(t) a(t)ej4=3p c(t) a(t)e


j2=3p

fref

(30)

where fpo 1 Hz is the power oscillation frequency. Real component, imaginary component and magnitude, angle of a(t) as a function against time are shown in Fig. 2.

where fcrt 50 Hz is the fundament frequency of the constant rotation vector. Phasor estimation accuracy of algorithms is depicted in Fig. 3a. TVEs of DPFE and DPE stay in a limited range during the whole dynamic process as they have the dynamic characteristic expressed properly, but that of DFT reaches the highest point at 250 and 750 ms since the value of the rst derivative of supplied signals is the largest at the moment. From Fig. 3b, although the performance of DPE is better than that of DPFE, the TVEs are so small that it cannot tell the difference when harmonic exists. The maximum TVEs of DPFE and DPE are 0.06824 and 0.00224% during oscillation except at the discontinuous point. TVEs jump at the beginning and the end of the dynamic process because neither algorithms can express the discontinuities correctly. Frequency estimation errors are shown in Fig. 3c. Performance of DPFE and DPE is similar and their frequency estimation errors are very close to zero. Although the performance of frequency estimation can be improved by carefully selected assigned point, PSFM will introduce a relative large estimation error around the inexion of theoretical frequency (250 and 750 ms). The maximum and mean frequency estimation errors are shown in Table 3.

Table 2 A-phase parameters for harmonic rejection simulation study Frequency, Hz Magnitude, pu Angle, deg 1 2 3 4 5 30 50 105 155 255 0.01 1.0 0.01 0.01 0.01 220 0 20 40 60

Figure 1 Performance comparison in presence of harmonics among DPFE, DFT and DPE 78

5.1.3 Multiple modes oscillation: In real systems, the supplied signal is complicated but can be approximately represented as a sum of damped sinusoids to reect electromechanical characteristics. Besides, so much
IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 1, pp. 73 83 doi: 10.1049/iet-gtd.2009.0320

& The Institution of Engineering and Technology 2009

www.ietdl.org
Table 4 Parameters of A-phase low-frequency band-limited phasor i 1 2 3 4 shown in Table 4
I X i1

xi , pu
4 1.5 10 1

li

fi , Hz ui , deg 210 290 0 135

20.2 21.5 22.0 21 0.0 22.0 0 0.5

a(t)

xi eli t ej(2pfi tui )

(31)

where xi , li , fi and ui are the amplitude, damping factor, frequency and initial phase of the ith component, respectively; I is the number of components. In this case, a frequency tracking technique is applied. A-phase waveform and theoretical frequency are shown in Fig. 4. There are quite large magnitude, angle errors and TVE in the estimations of DFT compared to that of DPE and DPFE, as shown in Fig. 5. Because both of DPE and DPFE can properly express the dynamic characteristic of supplied signals, their estimation errors can be limited to a low level during the whole dynamic condition. As the

Figure 3 TVEs and frequency errors of different algorithms under power oscillation Table 3 Maximum, mean and variance of frequency errors under power oscillation Frequency estimation error Maximum, Hz Mean, Hz PSFM PBFE DPE DPFE 0.05046 0.17756 0.00326 0.00571 0.00937 0.07962 0.00104 0.00192 Variance 0.00023 0.00906 1.96892e-06 6.83631e-06

research [17, 18] has been done by using this model to estimate parameters of power systems successfully. Therefore it is necessary to verify the performance of the proposed algorithm with this model, and its parameters are IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 1, pp. 73 83 doi: 10.1049/iet-gtd.2009.0320

Figure 4 A-phase waveform and theoretical frequency under multiple modes oscillation model 79

& The Institution of Engineering and Technology 2009

www.ietdl.org

Figure 6 Frequency Errors of different algorithms under multiple modes oscillation

5.2 PSCAD/EMTDC test


In order to demonstrate the effectiveness of proposed algorithm used for PMUs application under dynamic conditions, an extensive simulation including power system oscillation and fault condition was performed with the help of PSCAD/EMTDC. The conguration of a 230 V power system is shown in Fig. 7. During oscillation, an A-phase earth fault happened at 100 ms and it was cleared at 300 ms. The waveform of phase A is shown in Fig. 8. In order to simulate eld data, white noise (SNR 60 dB), second harmonic (0.5%) and third harmonic (0.3%) were added into the waveforms. In order to have intuitive comparison between magnitude p estimations and supplied signals, it is good to have 2 times

Figure 5 Magnitude errors, angle errors and TVEs of different algorithms under multiple modes oscillation model dynamic characteristic decreases, estimations become more accurate as time goes on for all algorithms. Table 5 shows the maximum errors of different algorithms from 0 to 500 ms. Frequency estimations performance of different algorithms is pretty good except that of PBFE as shown in Fig. 6 and the maximum frequency estimation errors are: 0.00310 Hz (PSFM), 0.04640 Hz (PBFE), 0.00248 Hz (DPE) and 0.00292 Hz (DPFE), respectively. Table 5 Maximum errors of different algorithms with multiple modes oscillation model Angle error, deg Magnitude error, pu DFT DPE DPFE 0.31973 0.02122 0.03111 0.05787 0.00119 0.00299 TVE, % 0.67079 0.05448 0.06673 Figure 8 A-phase waveform under phase earth condition IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 1, pp. 73 83 doi: 10.1049/iet-gtd.2009.0320

Figure 7 One-line diagram of a test system for dynamic test

80

& The Institution of Engineering and Technology 2009

www.ietdl.org

Figure 10 Frequency estimations at the beginning of fault as shown in Fig. 10. The estimation of DPFE is coincident with that of PSFM.

Acknowledgment

The authors gratefully acknowledge the support of the Chinese National Science Fund (No. 50877068), and Projects of Education Ministry Planning on New Century Excellent Technologist (No. NCET-06-2008).

Conclusion

An algorithm for PMUs to estimate the phasor and frequency of supplied signals under dynamic conditions is proposed by this paper. The proposed algorithm can obtain dynamic characteristics of power systems in the form of Taylor derivatives. By the benet of the knowledge of dynamic characteristics, the accuracy of phasor and frequency estimations can be dramatically improved. Two comprehensive sets of simulation results under dynamic condition tests have shown that although minor computation burden has been added to DSP, it does not affect PMU applications and it has great advantages over other algorithms discussed above: 1. The proposed algorithm just needs slightly more computation burden than DFT, whereas DPE needs more than three times computation burden of traditional DFT for a second-order model. 2. Magnitude frequency response is not changed signicantly by DPFE, which means we can ignore it. At the same time, that of DPE is changed signicantly to improve the performance under dynamic conditions. In other words, DPE may be sensitive to some harmonics. 3. Considering the dynamic characteristic by DPE and DPFE, the accuracy of the dynamic phasor estimation can be improved dramatically, while errors arise using DFT under dynamic conditions. 81

Figure 9 Magnitude estimations during the fault condition

of magnitude and supplied signals expressed together as shown in Fig. 9. During the fault condition, although all the estimations are affected by dynamic characteristic and harmonics, the performance of DPFE is much better than the others, as the estimation of DFT is mainly distorted by the dynamic characteristic and that of DPE is sensitive to harmonic, as shown in Fig. 9a. The estimation of DFT has the largest ripple during a fast dynamic condition as shown in Fig. 9b. When the dynamic slows down, the harmonic leads the estimation of DPE to display a large ripple as shown in Fig. 9c. There is a overshoot in the frequency estimation of DPE because of signals discontinuity which contains a huge number of harmonics around 100 ms, whereas the frequency estimations of the others appear relatively smooth IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 1, pp. 73 83 doi: 10.1049/iet-gtd.2009.0320

& The Institution of Engineering and Technology 2009

www.ietdl.org
4. Accurate frequency estimation can be attained by DPFE without waiting for the next phasor estimation as PSFM does. Also, the proposed algorithm is being applied in a relay. [13] DE LA SERNA J.A.D.: Dynamic phasor estimates for power system oscillations, IEEE Trans. Instrum. Meas., 2007, 56, (5), pp. 1648 1657 [14] DAWEI F., CENTENO V. : Phasor-based synchronized frequency measurement in power systems, IEEE Trans. Power Deliv., 2007, 22, (4), pp. 2010 2016 [15] IEEE standard for synchrophasors for power systems. IEEE Std C37.118-2005 (Revision of IEEE Std 1344 1995), 2006, pp. 7 8 [16] BEGOVIC M.M., DJURIC P.M., DUNLAP S., PHADKE A.G.A.P.A.G.: Frequency tracking in power networks in the presence of harmonics, IEEE Trans. Power Deliv., 1993, 8, (2), pp. 480486 [17] TAWFIK M.M., MORCOS M.M.: On the use of Prony method to locate faults in loop systems by utilizing modal parameters of fault current, IEEE Trans. Power Deliv., 2005, 20, (1), pp. 532 534 [18] FAIZ J., LOTFI-FARD S., SHAHRI S.H.: Prony-based optimal Bayes fault classication of overcurrent protection, IEEE Trans. Power Deliv., 2007, 22, (3), pp. 1326 1334

References

[1] RICE M. , HEYDT G. : Power systems state estimation accuracy enhancement through the use of PMU measurements. PES TD 2005/2006, 2006, pp. 161 165 [2] HONGXIAN W., GIRI J.: PMU impact on state estimation reliability for improved grid security. Transmission and Distribution Conf. and Exhibition, 2005/2006 IEEE PES, pp. 1349 1351 [3] MILOSEVIC B., BEGOVIC M.: Voltage-stability protection and control using a wide-area network of phasor measurements, IEEE Trans. Power Syst., 2003, 18, (1), pp. 121127 [4] BRAHMA S.M., GIRGIS A.A. : Development of adaptive protection scheme for distribution systems with high penetration of distributed generation, IEEE Trans. Power Deliv., 2004, 19, (1), pp. 56 63 [5] DE LA O SERNA J.A.: Dynamic phasor estimates for power system oscillations and transient detection. Power Engineering Society General Meeting, 2006, vol. 7, pp. 1 7 [6] STENBAKKEN G., ZHOU M.: Dynamic phasor measurement unit test system. Power Engineering Society General Meeting, 2007, pp. 1 8 [7] MUNOZ A.T. , DE LA O SERNA J.A. : Shanks method for dynamic phasor estimation, IEEE Trans. Instrum. Meas., 2008, 57, (4), pp. 813 819 [8] MACIEJ N., ADAM S.: A simple way of increasing estimation accuracy of generalized adaptive notch lters, IEEE Signal Process. Lett., 2007, 14, (3), pp. 217 220 [9] NIEDZWIECKI M., SOBOCINSKI A.: Generalized adaptive notch smoothers for real-valued signals and systems, IEEE Trans. Signal Process. (see also IEEE Trans. Acoust. Speech Signal Process.), 2008, 56, (1), pp. 125 133 [10] SO H.C., CHING P.C.: Adaptive algorithm for direct frequency estimation, IEE Proc. Radar Sonar Navig., 2004, 151, (6), pp. 359 364 [11] ROSNES E., VAHLIN A.: Frequency estimation of a single complex sinusoid using a generalized Kay estimator, IEEE Trans. Commun., 2006, 54, (3), pp. 407 415 [12] HOU M.: Amplitude and frequency estimator of a sinusoid, IEEE Trans. Autom. Control, 2005, 50, (6), pp. 855 858 82

Appendix 1
m mr jmi
(32)

In this appendix a detail derivation is given for (13). First of all, we will like to rewrite m as

where mr and mi are the real and imaginary components of m, respectively. As there are (K 1) unknown parameters in vector m, we will employ (K 1) phasor estimations of different data windows to form a set of formulas, as follows 8 > P 0 (m0 ) mT EK ,0,m0 ,M mH EK ,2,m0 ,M > ~ < . . . > > :~ T P 0 (mK ) m EK ,0,mK ,M mH EK ,2,mK ,M

(33)

By substituting (32) into (33) and rearranging formulas, we have 8 ~ > Re[P 0 (m0 )] Re(EK ,0,m ,M EK ,2,m ,M )T mr > > 0 0 > > > > Im( EK ,0,m0 ,M EK ,2,m0 ,M )T mi > > > > > . > . > > . > > T >Re[P (m )] Re(E ~0 K > > K ,0,mK ,M EK ,2,mK ,M ) mr > > > < Im( EK ,0,mK ,M EK ,2,mK ,M )T mi (34) > Im[P 0 (m0 )] Im(EK ,0,m ,M EK ,2,m ,M )T mr ~ > > 0 0 > > > > Re(EK ,0,m0 ,M EK ,2,m0 ,M )T mi > > > > > . > . > > . > > T >Im[P (m )] Im(E ~0 K > > K ,0,mK ,M EK ,2,mK ,M ) mr > > > : Re(E E )T m
K ,0,mK ,M K ,2,mK ,M i

& The Institution of Engineering and Technology 2009

IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 1, pp. 73 83 doi: 10.1049/iet-gtd.2009.0320

www.ietdl.org
By redening vectors, we have P 0 E0 m (35) By redening vector, we have P 12 E12 g (38)

~ ~ ~ where P 0 [Re[P 0 (m0 )], ..., Re [P 0 (mK )], Im[P 0 (m0 )], ... , (0) (0) (K ) ~ 0 (mK )]]T Im[P m [mr , ... , mr , mi , ..., m(K ) ]T (see i equation at the bottom of the page) As long as jE0 j = 0, the solution is given by

~ ~ ~ ~ where P 12 [P 1 (m0 ), [P 2 (m0 )] , . . . , P 1 (mK ), [P 2 (mK )] ]T , T H T g [a , b ] 2 6 H 6 (E 6 K ,2,m0 ,M ) 6 . 6 . 6 . 6 6 (EK ,0,m ,M )T 4 K (EK ,2,mK ,M )H (EK ,0,m0 ,M )T 7 (EK ,0,m0 ,M )H 7 7 7 . 7 . 7 . 7 (EK ,2,mK ,M )T 7 5 (EK ,2,m0 ,M )T 3

m (E0 )1 P 0

(36)

E12

10

Appendix 2

(EK ,0,mK ,M )H

In this appendix a detailed derivation is given for (16). As there are 2(K 1) unknown parameters in vectors a and b, we will employ 2(K 1) phasor estimations of different data windows to form a set of formulas, as follows ~ P 1 (m0 ) (EK ,0,m0 ,M )T a (EK ,2,m0 ,M )T b ~ [P 2 (m0 )] (EK ,2,m0 ,M )H a (EK ,0,m0 ,M )H b . . . ~ P 1 (mK ) (EK ,0,mK ,M )T a (EK ,2,mK ,M )T b ~ [P 2 (mK )] (EK ,2,mK ,M )H a (EK ,0,mK ,M )H b (37)

As long as jE12 j = 0, the solution is given by

g (E12 )1 P 12

(39)

Re(EK ,0,m0 ,M EK ,2,m0 ,M )T 6 6 . . 6 . 6 T 6 Re(E K ,0,mK ,M EK ,2,mK ,M ) 6 E0 6 6 Im(EK ,0,m ,M EK ,2,m ,M )T 6 0 0 6 . 6 . 4 . Im(EK ,0,mK ,M EK ,2,mK ,M )T IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 1, pp. 73 83 doi: 10.1049/iet-gtd.2009.0320

3 Im( EK ,0,m0 ,M EK ,2,m0 ,M )T 7 7 . . 7 . 7 Im( EK ,0,mK ,M EK ,2,mK ,M )T 7 7 T 7 7 Re(EK ,0,m0 ,M EK ,2,m0 ,M ) 7 7 . 7 . 5 . Re(EK ,0,mK ,M EK ,2,mK ,M )T 83

& The Institution of Engineering and Technology 2009

You might also like