Professional Documents
Culture Documents
6, DECEMBER 2017
Abstract—This paper proposes a new fast algorithm to estimate networks (ANNs) [1]–[8]. The DFT is widely used for phasor
the fundamental phasor of fault current signals, based on discrete estimation because of its simplicity, high accuracy and speed [9].
Fourier transform (DFT). Since fault currents include decaying dc The DFT can filter out harmonics and estimate the magni-
component (ddc), DFT-based algorithms have inaccuracy in phasor
estimation. The proposed algorithm consists of four steps. First, a tude and phase angle of a periodic signal [1]–[4]. However,
new auxiliary signal is introduced based on a simple high-frequency the presence of non-periodic decaying dc component (ddc)
modulation of the fault current. Then, DFT of the fault current and in fault currents causes large error (up to 15%) in estimated
sample-summation of both fault current and auxiliary signal are magnitude [10], [11].
calculated, for each one-cycle-length data stream. Next, DFT esti- Research has been performed in the open literature to reduce
mation error due to the ddc is calculated. Finally, by removing this
error, accurate fundamental phasor is obtained. This paper uses the DFT phasor estimation error in presence of ddc [12], [13].
an innovative and accurate formulation based on notations of sig- There are two main groups for this work:
nal processing literature. To validate the proposed algorithm, sev- In the first group, the ddc is first filtered out, and then the
eral computer-simulated signals and an electromagnetic transient DFT is applied on the remaining signal containing the funda-
program-generated signal are employed. The proposed algorithm mental and other harmonics to estimate the fundamental phasor
is compared with recent phasor estimation methods, using wide
variety of test signals. Standard indices of rise time, settling time, [10], [14]–[16]. Implementation of mimic filter is one of the con-
and percentage overshoot are used for comparison. These compar- ventional solutions for filtering the ddc. However, this high-pass
isons show that the proposed algorithm has robust performance filter amplifies harmonics and inter-harmonics, which causes os-
in off-nominal frequency condition, and presence of harmonics, cillatory error in estimated magnitude [10]. In [14], a modified
noise and multiple ddc components. Moreover, it provides faster notch filter is performed to filter out the ddc. References [15],
convergence speed and lower computation burden.
[16] introduce an auxiliary signal containing reduced ddc, based
Index Terms—Damping factor, decaying dc component, discrete on fault current. Then, the DFT is applied to the auxiliary sig-
Fourier transform, high-frequency modulation, numerical relays,
phasor estimation. nal to obtain its fundamental component. Finally, fundamental
current phasor is estimated based on fundamental phasor of
I. INTRODUCTION the auxiliary signal. The estimated magnitude of this method
severely oscillates, especially in case of small ddc time con-
OMPUTER based protection has attracted many
C researches because of its applications in wide area moni-
toring, protection and control (WAMPAC). In this type of protec-
stants.
The second group estimates the ddc, subtracts it from fault
current signal, and applies DFT to estimate the fundamental
tion, accurate estimation of voltage and current phasors plays an
phasor [17]–[29]. The value of ddc is estimated in [17]–[19] by
important role. Therefore, several researches investigate various
decomposing samples into odd and even strings and perform-
algorithms for phasor estimation. The presented algorithms for
ing DFT on each of them. These methods suffer from heavy
this purpose are: discrete Fourier transform (DFT), least square
computational burden. In [20], ddc parameters are obtained by
error (LSE), cosine filters, discrete wavelet transform (DWT),
two partial sums. However, this algorithm cannot completely
Kalman filters and intelligent techniques such as artificial neural
filter out all of the harmonic components in fundamental phasor
Manuscript received June 23, 2016; revised September 16, 2016; accepted
estimation. In [21], the ddc value is estimated using compli-
November 4, 2016. Date of publication November 16, 2016; date of current cated equations, based on DFT of three consecutive strings of
version October 16, 2017. Paper no. TPWRD-00781-2016. the sample data. This method takes one cycle plus two ad-
B. Jafarpisheh is with the Department of Electrical Engineering, Faculty
of Engineering, University of Isfahan, Isfahan 81746-73441, Iran (e-mails:
ditional samples to estimate the fundamental phasor. Another
b.jafarpisheh@eng.ui.ac.ir, babak_jafarpisheh@yahoo.com). DFT-based method described in [22] suffers from heavy com-
S. M. Madani is with the Department of Electrical Engineering, Faculty putational burden. It requires computing DFT twice for each
of Engineering, University of Isfahan, Isfahan 81746-73441, Iran (e-mails:
m.madani@eng.ui.ac.ir, madani104@yahoo.com).
string of data. References [23], [24] presented another DFT-
S. Mohammad Shahrtash is with the Department of Electrical Engineering, based method which takes one cycle plus two additional sam-
Center of Excellence for Power System Automation and Operation, Iran Uni- ples. A long duration method for phasor estimation that takes
versity of Science and Technology, Tehran 16844, Iran (e-mail: shahrtash@
iust.ac.ir).
one and half cycle is presented in [25]. In [26], ddc parameters
Color versions of one or more of the figures in this paper are available online are estimated by integration from fault current signal. However,
at http://ieeexplore.ieee.org. due to linear approximation of ddc, it has poor performance for
Digital Object Identifier 10.1109/TPWRD.2016.2629762
small values of ddc time constant. In continuation of [26], other
0885-8977 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
JAFARPISHEH et al.: NEW DFT-BASED PHASOR ESTIMATION ALGORITHM USING HIGH-FREQUENCY MODULATION 2417
DFT-based methods are presented in [27]–[29] for estimation can be rewritten as shown in (2).
of fundamental and sub-synchronous components. Heavy ma-
Δ
trix and logarithmic computations are the main disadvantages i[n] = i(nΔt) = iddc [n] + iac [n]
of these methods.
where :
For most of the mentioned algorithms, the performance of
phasor estimation is acceptable either for small or large ddc iddc [n] = I0 e−n Δ t/τ
time constants. Outside of their specific range, their estimated
N /2−1
phasor has oscillatory error. 2πk
This paper presents an algorithm to suppress the ddc for wide iac [n] = |Ik | cos n + θk (2)
N
k =1
range of ddc time constant, avoiding heavy computations. An
innovative, accurate formulation is applied using notations of where N is the number of samples per cycle and Δt is the
signal processing literature [1], [2]. The proposed algorithm sampling interval. It is assumed that harmonic numbers higher
consists of four steps: (1) A new auxiliary signal is intro- than (N/2 − 1) are eliminated by a low-pass filter, to avoid the
duced which is easily generated from high-frequency modu- aliasing phenomenon.
lation of the fault current signal, (2) One-cycle-length strings According to (2), the fault current signal is divided into two
from both fault current and auxiliary signals are acquired. The terms. The first term, which is denoted by iddc [n], is the ddc
sample-summation of the both strings and the DFT of the fault value of fault current signal. The second periodic term, which is
current data stream are calculated, (3) The DFT error due to shown by iac [n], contains the fundamental and other harmonic
the ddc is calculated, (4) The calculated error is subtracted components.
from DFT estimation, which results in accurate fundamental The ddc is generated by any fault or disturbances in power
phasor. systems. It can also be generated by current transformers (CTs)
To validate the proposed algorithm, several computer- [17], [18]. The ddc time constant varies from 0.5 to 5 cycles,
simulated signals and an Electromagnetic Transient Program depends on the system configuration, fault location and fault
(EMTP)-generated signal are employed. The proposed algo- resistance [10]. Since ddc is non-periodic, its presence causes
rithm is compared with recent phasor estimation methods, us- significant error in DFT-based estimated-phasor.
ing wide variety of test signals including different values for The estimated phasor of the fundamental component (Iˆ1 ) by
ddc magnitude and time constant. These comparisons show DFT is:
that the proposed algorithm has robust performance in offnom-
N −1
inal frequency condition, and presence of harmonics, noise Δ 2 2π
Iˆ1 = Iˆ1 < 0 , N −1 > = i[n]e−j N n
and multiple ddc components in fault current signals. It also N n =0
provides faster speed of convergence and lower computation
burden. = I1 + ΔI1ddc< 0,N −1>
The rest of this paper is organized as follows. In Section II, where :
the proposed algorithm is completely described within three
subsections of: a brief review on DFT and ddc; presenting the I1 = |I1 | ej θ 1 (3)
proposed algorithm; and the flowchart of the proposed algo-
rithm. In Section III, the performance of the proposed algorithm where Iˆ1 is the fundamental phasor estimated by DFT using
for both computer-simulated and EMTP-generated signals is N-length string of data stream, in which 0 and N−1 denote
evaluated. Finally in Section IV, some concluding remarks are the first and last sample number of the string. I1 represents
discussed. the exact fundamental phasor, and ΔI1ddc denotes the error in
the estimated fundamental phasor caused by the ddc, which can
II. PROPOSED ALGORITHM be obtained as follows:
A. A Brief Review on DFT and ddc
E = e−Δ t/τ
Δ
where I0 and τ are the magnitude and time constant of the Equation (4) is obtained using geometric series. The damping
ddc, k is the harmonic number, P is the maximum harmonic factor E depends on the sampling interval Δt and ddc time
number, |Ik | and θk are the magnitude and phase angle of the constant τ . ΔΘ is the phase difference between two consecutive
k-th harmonic. The discretization form of the fault current signal samples and can be precalculated.
2418 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 32, NO. 6, DECEMBER 2017
n =0 n =0 N /2−1
1 − ( − 1)N ej 2π k
= Ik (13)
Since the ac part of the fault current iac [n] is periodic, its 1 + ej k Δ θ
k =1
sample-summation over one cycle is equal to zero:
⎡ ⎤ It is clear that ej 2π k = 1. Since N is an even number,
N
−1 N
−1 N/2−1 ( − 1)N = 1. Thus, (13) is rewritten as:
⎣ 2πk
iac [n] = |Ik | cos n + θk ⎦ = 0 (7)
N N /2−1
n =0 n =0 k =1 1−1
S̃y ac = Ik =0 (14)
1 + ej k Δ θ
So, iddc [n] is the only remaining term of the fault current k =1
sample-summation Si . So, high-frequency modulation of iddc [n] is the only remain-
N −1 N −1
ing term of auxiliary signal sample-summation Sy .
1 − EN
Si = iddc [n] = I0 E n = I0 (8) N
−1 N
−1
n =0 n =0
1−E Sy = (−1)n iddc [n] = (−1)n I0 E n
n =0 n =0
Similarly, the sample-summation of auxiliary signal Sy is
N N
defined as: 1 − ( − 1) E 1 − EN
= I0 = I0 (15)
N −1 N −1 1 − ( − 1)E 1 + E
(−1)n i[n]
Δ
Sy = Sy < 0 , N −1 > = y[n] = As mentioned previously, Si denotes the summation of N
n =0 n =0 consecutive samples of fault current data stream. Actually, Si
N
−1 acts as a low-pass filter for fault current. Since the auxiliary-
= ( − 1)n [iddc [n] + iac [n]] (9) signal harmonic-components are obtained by mirroring the fault
n =0 current harmonic-components around the mid-frequency, the
frequency response of Sy can also be obtained by mirroring
It will be proved that the second term in (9) which is called
the Si frequency response around the mid-frequency (harmonic
Sy ac is equal to zero:
number of N/4). Therefore, Sy acts as a high-pass filter. Figure 1
N
−1 depicts normalized magnitude of the frequency responses for Si
Sy ac (−1)n iac [n] = 0 (10) and Sy , for N = 16 and fundamental frequency of 60 Hz. Since
n =0 it is assumed that harmonic numbers higher than (N/2 − 1) are
eliminated by an anti-aliasing filter, the frequency responses are
To prove (10), the phasor representation of Sy ac is used:
depicted for maximum frequency of 420 Hz.
⎡ ⎤
N −1 N /2−1 Dividing (8) by (15), results in:
S̃y ac (−1)n ⎣ |Ik | ej 2π k n /N ej θ k ⎦ Si 1 + E
n =0
= (16)
k =1 Sy 1−E
where : Finally, the damping factor is easily obtained as:
Si − Sy
Sy ac = Re S̃y ac (11) E= (17)
Si + Sy
JAFARPISHEH et al.: NEW DFT-BASED PHASOR ESTIMATION ALGORITHM USING HIGH-FREQUENCY MODULATION 2419
TABLE I
PERFORMANCE INDICES FOR THE CASE OF 90° FAULT INCEPTION ANGLE
Performance Indices
Fig. 3. Basic signal with ddc time constant of 200 (ms) and fault inception TABLE II
angle of zero. AVERAGE OF THE INDICES FOR THE EIGHTEEN CASES
Magnitude
Fig. 4. Estimated fundamental phasor for the worst case of basic signals.
Fig. 7. Fundamental phasor estimation for fault current signal with multiple
ddc components. Fig. 9. Fundamental phasor estimation for fault current signal including har-
monics and white random noise.
Fig. 8. Fault current signal including harmonics and white random noise.
Fig. 10. Fundamental phasor estimation for fault current signal in offnominal
where, the fault inception angle ΘF and first ddc time constant frequency condition.
(τp ) are assumed to be zero and 5 cycles, as the worst condition.
The time constant of the second ddc τs is set to 20 cycles [17].
The estimated fundamental phasors are depicted versus time, D. Test for Offnominal Frequency
in Fig. 7. Compared to the recent mentioned methods, the pro- During transient condition, the fundamental frequency
posed algorithm has better performance, in presence of multiple slightly deviates from its nominal value, which is called offnom-
ddc components (Fig. 7). inal frequency condition. This deviation causes oscillatory and
steady state error in estimated phasor [29]. To verify the per-
formance of the proposed algorithm, the test signal in previous
C. Test for Harmonics and Noise subsection is used by changing its power frequency to 59.5 Hz.
In this subsection, the performance of the proposed algorithm The estimated phasors are depicted versus time in Fig. 10.
is evaluated in presence of harmonics and white random noise As shown in Fig. 10, in offnominal frequency condi-
[17], [24]. As shown in (23), the fault inception angle and ddc tion, the proposed algorithm estimates the fundamental
time constant are assumed to be zero and 5 cycles, which is phasor faster and more accurate than the other men-
the worst condition. The signal-to-noise ratio (SNR) for white tioned methods. This is because Si and Sy have approxi-
random noise is set to 20 dB. mately the same gain for frequencies around fundamental
frequency of 60 Hz (Fig. 1).
N /2−1
|I1 | 2πk
−n Δ t/τ
i3 [n] = |I1 | e − cos n + kθ1 E. Test for EMTP-Generated Signal
k N
k =1
At this subsection, the proposed algorithm is evaluated, using
+ White Random Noise (23) typical electric signals that take place on electric power systems.
An EMTP-generated test signal corresponding to a single-
The fault current and estimated phasors are depicted versus phase fault at the mid-point of a 103.47 km long, 220-kV trans-
time in Figs. 8 and 9, respectively. mission line is used to evaluate the performance of the proposed
As seen in Fig. 9, the proposed algorithm has better perfor- algorithm [23]. The fault current is shown in Fig. 11. Figure 12
mance in fundamental phasor estimation than the other three depicts the estimated magnitude and phase angle of the funda-
methods, in presence of harmonics and noise. This is because, mental component versus time for the proposed, S-CharmDFT
as shown in Fig. 1, harmonic components and noise with max- [16], modified DFT [17] and Prony-DFT [28] algorithms. For
imum frequency of 420 Hz (harmonic number of N/2-1), are more clarity, Fig. 13 depicts the estimated phasors versus time,
severely suppressed by Si and Sy in the proposed algorithm. from one cycle after the fault inception time.
2422 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 32, NO. 6, DECEMBER 2017
TABLE III
COMPARISON OF THE COMPUTATION BURDEN AND NUMBER OF SAMPLES
REQUIRED FOR PHASOR ESTIMATION ALGORITHMS
[26] 2 61 80 N+ 1
[24] 2 61 48 N+ 1
[21] – 182 140 N+ 2
[20] – 151 64 N
[23] – 148 50 N+ 2
[30] – 48 59 N+ 1
[17], [18] – 37 95 N
[31] – 35 94 N+ 2
[16] – 34 76 N+ 1
Fig. 11. EMTP-generated fault current signal. [22] – 30 65 N
Proposed Algorithm – 15 65 N
IV. CONCLUSION
This paper presents a new DFT-based phasor estimation al-
gorithm, which contains four steps. First, a new auxiliary signal
is introduced based on a simple high-frequency modulation of
the fault current signal. Then, sample-summation of N-length
strings from fault current and auxiliary signals are calculated.
DFT of the fault current string is also calculated. Next, the DFT
error caused by the ddc is calculated. Finally, the fundamental
phasor component is estimated by eliminating the error value.
The performance of the proposed algorithm is compared with
Fig. 13. Estimated phasor from one cycle after the fault inception time for three recent phasor estimation methods, using various types
EMTP-generated fault current signal. of computer-simulated signals and an EMTP-generated signal.
Several computer-simulated signals are employed, considering:
As seen in Figs. 12 and 13, all of the three performance different values for ddc magnitude and time constant, offnom-
indices of: rise time, settling time and percentage overshoot of inal frequency condition, and presence of harmonics, noise
the proposed algorithm are significantly smaller than the other and multiple ddc components. Three performance indices of
phasor estimation methods. rise time, settling time and percentage overshoot are used for
comparison.
F. Computation Burden Compared to the mentioned recent methods, advantages of
For evaluation of the computation burden, total numbers of the proposed algorithm are:
logarithmic, exponentional and basic-mathematical operations 1) Faster convergence speed,
are calculated, for the proposed and several recent phasor esti- 2) Less dependency to ddc parameters,
mation methods of [16]–[18], [20], [21], [23], [24], [26], [30], 3) Robust performance in presence of multiple ddc compo-
[31]. It is assumed that some parameters are precalculated and nents in fault current signal,
used in the implementation of the methods. Table III also shows 4) Robust performance in presence of harmonics and noise,
the required number of samples per data window for each phasor 5) Better performance in offnominal frequency condition,
estimation method. The number of samples per cycle is assumed 6) Less computation burden.
to be 16. In this case, each (fast) DFT computation consists of The low demanding property of the proposed algorithm
10 multiplications and 30 summations [4]. makes it applicable for fast protection purposes.
JAFARPISHEH et al.: NEW DFT-BASED PHASOR ESTIMATION ALGORITHM USING HIGH-FREQUENCY MODULATION 2423