Professional Documents
Culture Documents
IEEE - A New Half-Cycle Phasor Estimation Algorithm
IEEE - A New Half-Cycle Phasor Estimation Algorithm
Abstract—Phasor estimation algorithms for protective relaying fundamental frequency signal by filtering all other unwanted
are required to filter out unwanted components from the input signals. The time constant and amplitude of the decaying dc
signals and retain only the components of interest. The compo- are unknown and are associated with the fault resistance, fault
nents to be removed include harmonics and the decaying-exponen-
tial transient (dc offset) component. They affect the accuracy and position and the fault beginning time. The time constant of the
the speed of convergence of the phasor estimation algorithms to a decay is generally determined by the X/R ratio (inductive reac-
great extent. This paper presents a new technique, which effectively tance to the resistance ratio) of the system. For high resistance
removes the harmonics and the decaying dc component present earth faults the decay rates are very high, sometimes less than
in the input signals, within half a cycle of the power system fre- half the power system frequency cycle. Decaying dc is a non-
quency. This is achieved by means of a simple computational pro-
cedure using three off-line look-up tables. The proposed algorithm periodic signal and its frequency spectrum encompasses all fre-
has been tested for a wide variety of signals to assess its perfor- quencies. The decaying dc component therefore, seriously af-
mance. The performance is also compared with the two most pop- fects the accuracy and convergence of digital filter algorithms
ular half-cycle phasor estimation algorithms; the half-cycle least such as Fourier, cosine, Walsh, Kalman and least-error-squares
error squares algorithms and the mimic plus half-cycle Fourier (LES) filters [1]–[6]. Due to this, the estimated phasors using
algorithm. The test results show that the proposed relaying algo-
rithm has a faster convergence and better accuracy compared to these algorithms contain errors if decaying dc is present in the
these previously proposed algorithms. The results also indicate that input signal.
the proposed algorithm converges to its final value within half a Phasor estimation algorithms can be divided into two cate-
cycle of the power system frequency as compared to the other two
algorithms, which take more than half a cycle to converge, when a gories: the full cycle algorithms and the half-cycle algorithms.
decaying dc component is present in the input. The full cycle algorithms typically take one power system cycle
Index Terms—DFT, digital filtering, least error squares algo- to estimate the voltage or current phasor. These algorithms may
rithm, mimic filters. not be suitable for use in EHV/UHV system applications where
the decisions have to be taken in less than one cycle. The most
popular of the full cycle algorithms are the DFT algorithm and
I. INTRODUCTION the LES algorithm. Since the response speed is of paramount
importance, half-cycle algorithms or short window algorithms
T HE RIGID interconnectivity existing among modern
power systems make them highly unstable when, faults
are not cleared rapidly. If a fault in an important transmission
have been proposed whose convergence speed to the final value
are faster than the full cycle algorithms. The two such important
line is not identified and removed at the shortest possible time, half-cycle algorithms are the digital mimic filter [1] plus half-
it might lead to a widespread damage in the power system. cycle DFT and half-cycle LES algorithm. These half-cycle al-
The job of the protective relays is to identify and isolate the gorithms suffer from some major disadvantages. The half-cycle
faults in power system networks. The unpredictable nature of DFT filters cannot eliminate even numbered harmonics and the
the power system signals during faults, make their extraction a decaying dc present during fault conditions, hence affecting the
challenging job. In order to prevent the damages from spreading accuracy and convergence. This leads to a large amount of error
to the healthy parts of the power system, the protective relaying in the estimated phasor of the current or voltage signal if those
algorithms need to detect the faults within sub cycles of the harmonics are present in the input signal. To make these filters
power system frequency. This makes the challenge even more insensitive to the decaying dc component, the input signals are
difficult, since, lesser data is provided to a digital filter to extract pre-processed by a digital mimic filter before being processed
the desired signal frequency. by the half-cycle DFT filter. Mimic filter removes the decaying
For most relaying algorithms, the digital filters extract the dc if the time constant of the decaying dc is known before hand,
fundamental frequency signal. This needs to be done quickly which is usually not the case in a real power system. In addition
and accurately. The fault signal is normally composed of fun- to that, the mimic filter is susceptible to noise. In the half-cycle
damental frequency component, harmonics and a decaying dc least error squares technique, the decaying dc component is in-
component. The task of the digital filters is to extract only the cluded in its model. The decaying dc component can be mathe-
matically expressed by a Taylor series expansion. In the LES
Manuscript received November 24, 2003; revised January 3, 2004. Paper no.
technique, the first two terms of the Taylor series expansion
TPWRD-00590-2003. are used to model the decaying dc component. The least error
T. S. Sidhu and V. Balamourougan are with the University of Western On- squares technique is then applied to estimate the fundamental
tario, London, ON N6B 5A9, Canada (e-mail: sidhu@eng.uwo.ca; bvinayag@
uwo.ca).
frequency phasor and other harmonics, provided they are mod-
Xudong Zhang was with University of Western Ontario, London, ON N6B eled in the technique.
5A9, Canada, and is now with the Central China Electric Power Group Com-
pany, Wuhan, Hubei, China. Fig. 1 shows the frequency analysis of three decaying dc com-
Digital Object Identifier 10.1109/TPWRD.2004.834677 ponents with different time constants. It can be seen that the
0885-8977/$20.00 © 2005 IEEE
1300 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 2, APRIL 2005
in this look-up table. This ratio is fixed for a given time con-
stant, irrespective of the initial magnitude of the decaying dc.
This ratio can be expressed as,
(2)
(3)
where, is the ratio at a given time constant tuned to the fundamental and the other set is tuned to the
is the output of the cosine filter of filter Set I, which computes mth harmonic.
the imaginary part of the phasor defined at time constant • From the output of the orthogonal filter Set II, the ratio
is the output of the sine filter of filter Set II, of the real to imaginary part is determined during each
which computes the real part of the phasor defined at time sampling interval.
constant . The ratios and are also fixed for a given time • The time constant , corresponding to this ratio is obtained
constant, irrespective of the initial magnitude of the decaying from the look-up Table I.
dc. It is important to note that the orthogonal filters of filter • The ratios and corresponding to this time constant
Set II contain only the effect of decaying dc in their outputs. are obtained from the look-up Tables II and III, respec-
Once the output of these filters are known, the effect of the tively.
decaying dc on the outputs of the filter Set I can be determined • The real part of the decaying dc component can be calcu-
from (3) and (4) with the help of the three look-up tables. lated now from (3), since, the ratio and are
The next step is to use the look-up tables and perform error known for the time constant .
correction on the output of filter Set I for obtaining the correct • The imaginary part of the decaying dc component simi-
phasor estimate. larly can be calculated now from (4), since, the ratio
and are known for the time constant .
C. Error Correction • The error correction is done, by subtracting the real and
The error correction procedure to estimate the correct phasor imaginary component values of the decaying dc obtained
is shown schematically in Fig. 2. The stepwise procedure is from the above step, from the respective real and imagi-
given below: nary parts of the outputs of the orthogonal filter Set I. This
now leaves only the real and imaginary component of the
• The low pass anti aliasing filter first filters the input signal,
fundamental frequency phasor.
• The filtered signal, is digitized and applied to This way the decaying dc effect can be eliminated completely
the two sets of the orthogonal filters. One set of filters is from the estimated phasors irrespective of the time constant of
1302 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 2, APRIL 2005
D. Computations
The technique needs the following computations
1) Design of two sets of orthogonal filters
2) Creation of three look-up tables
3) Process the input signal using the two filters
4) Determine the time constant and the correction factors
from the look-up table.
5) Perform error correction using the corrections obtained
6) Estimate the phasor
Fig. 3. Frequency response of the orthogonal filters of filter Set I.
The first two computations can be performed off-line and the
others can be performed on-line. With modern microprocessor
technology, the above-mentioned computations will not intro- III. IMPLEMENTATION
duce significant computational burden. The results prove that
the algorithm has a better performance in the time and frequency The nominal frequency is taken as 60 Hz. The sampling fre-
domain. The details of which are given in Section IV. quency of the input signal is assumed to be 3840 Hz. The data
The orthogonal filter Sets I and II can be any conventional window size is taken as half a cycle, which is 32 samples for
digital filters e.g., LES, DFT, Walsh functions, etc. In the next the sampling frequency of 3840 Hz. The low pass anti-aliasing
section the proposed technique is implemented using least error filter is chosen to be a 3rd order Butterworth filter with a cut-off
square filters. frequency at 180 Hz. A very low cut-off frequency is taken here
so that the digital filters can provide a better noise suppression
capability.
E. Parameter Selection
The sampling frequency of the input signal, the cut-off fre- A. Orthogonal LES Filters I and II
quency of the low pass anti-aliasing filter and the center fre-
The orthogonal filter Sets, I and II, explained in the previous
quency of the orthogonal filter Set II are three very important
section are designed using the least error square principle. The
parameters whose selection criterion is important for this algo-
orthogonal filters of filter Set I are least error square filters de-
rithm.
signed to extract the fundamental frequency and eliminate up to
A higher sampling rate allows for a higher cut-off frequency the sixth harmonic. Conventionally, the decaying dc component
of the low-pass anti-aliasing filter and a better response. A would have been included in the least error squares model. How-
higher cut-off frequency for the low pass anti-aliasing filter ever, only the fundamental and the harmonics to be eliminated
means more harmonics are present in the signal available to the are included in the model. The frequency responses of these fil-
digital filters. If the orthogonal digital filters are designed using ters are shown in Fig. 3. These filters have a better frequency
LES technique, then the filters should be designed to eliminate response at higher harmonics than the conventional half-cycle
these harmonics. The frequency response of these filters at LES filters where the decaying dc is modeled.
off-nominal frequencies becomes worse if more components Orthogonal filters of filter Set II are also least error square
are modeled in the design of the LES digital filters. For a higher filters. They are tuned at 10th harmonic . The 10th
cut-off frequency for the low pass filter, the digital LES filters harmonic is chosen because the designed low pass anti-aliasing
provide poor noise suppression capability. For a better noise filter completely eliminates 10th harmonic and the harmonics
rejection capability a lower cut-off frequency is needed for the above it effectively. This filter is designed to eliminate all
low pass anti-aliasing filter which means the response speed harmonics except the decaying dc. It should be noted that the
is compromised. It is important to realize that the low pass low pass anti-aliasing filter has already eliminated the 10th har-
anti-aliasing filter is an essential element for all digital relaying monic. Therefore, the output would be only due to a decaying
algorithms. dc component. The frequency response of these filters is shown
The selection of the center frequency for the orthogonal dig- in Fig. 4.
ital filter Set II depends on the cut-off frequency of the low
pass anti-aliasing filter. The center frequency should be chosen
B. Look-Up Tables I, II, and III
well above the cut-off frequency of the low pass anti-aliasing
filter. The important criterion is that the center frequency signal In order to build the look-up Table I, the orthogonal filters of
should not be present in the filtered input signal. A higher center filter Set II are given an input of a decaying dc signal of different
frequency allows a selection of higher cut-off frequency for the time constants. The ratios of the real part to imaginary part of
low pass filter and better response speed. these filters for each time constant are stored in this look-up
In a way these parameters are interdependent and their selec- table. This ratio is fixed for a given time constant, irrespective of
tion depends on the compromise the designer is willing to make the initial magnitude of the decaying dc. The plot of this ratio,
for a particular application. for different time constant is shown in Fig. 5.
SIDHU et al.: A NEW HALF-CYCLE PHASOR ESTIMATION ALGORITHM 1303
(5)
(a)
(b)
Fig. 8. (a) Output of the half-cycle LES and proposed algorithm when the
data window is completely filled with fault data. (b) Output of the mimic plus
half-cycle DFT and proposed algorithms when the data window is completely
filled with fault data.
A. Frequency Response
REFERENCES
[1] G. Benumouyal, “Removal of DC-offset in current waveforms using
digital mimic filtering,” IEEE Trans. Power Del., vol. 10, no. 2, pp.
621–630, Apr. 1995.
[2] A. G. Phadke, T. Hlibka, and M. Ibrahim, “A digital computer system
or EHV substations: Analysis and field tests,” IEEE Trans. Power App.
Syst., vol. PAS-95, no. 1, pp. 290–301, Jan./Feb. 1976.
[3] A. G. Phadke and J. S. Thorp, Computer Relaying for Power Systems,
Hertfordshire, U.K.: Research Studies, 1988.
[4] M. S. Sachdev and M. A. Baribeau, “A digital computer relay for
impedance protection of transmission lines,” in Proc. Spring Meeting
Engineering Operating Division Canadian Electrical Association,
Vancouver, BC, Canada, Mar. 1979, pp. 1–14.
[5] M. S. Sachdev, IEEE Tutorial Course, Microprocessor Relays and
Protection Systems: IEEE Service Center, 1988, Publication no.
88EH0269-1-PWR. Coordinator.
[6] M. S. Sachdev, H. C. Wood, and N. G. Johnson, “Kalman filtering ap-
(a)
plied to power system measurements for relaying,” IEEE Trans. Piwer
App. Syst., vol. PAS-104, no. 12, pp. 3565–3573, Dec. 1985.
[7] B. P. Lathi, Signal Processing and Liner Systems. New York: Oxford
Univ. Press, 2001.