Вы находитесь на странице: 1из 11

1136 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 15, NO.

4, OCTOBER 2000

An Adaptive PMU Based Fault Detection/Location


Technique for Transmission Lines—Part II: PMU
Implementation and Performance Evaluation
Joe-Air Jiang, Ying-Hong Lin, Jun-Zhe Yang, Tong-Ming Too, and Chih-Wen Liu, Member, IEEE

Abstract—Part I of this paper set sets forth theory and algo- current phasors to determine the impedance to the fault loca-
rithms for adaptive fault detection/location technique, which is tion, and both suffers from errors mentioned in many papers
based on Phasor Measurement Unit (PMU). This paper is Part II [10]–[13]. For example, the one-terminal data method needs to
of this paper set. A new timing device named “Global Synchronism
Clock Generator, GSCG” including its hardware and software de- make some assumptions for ground condition; the two-terminal
sign is described in this paper. Experimental results show that the impedance based method usually needs accurate and synchro-
synchronized error of rising edge between the two GSCGs clock is nized measurements for extracting the phasors. These problems
well within 1 ps when the clock frequency is below 2.499 MHz. The are completely solved by the proposed adaptive technique pre-
measurement results between Chung-Jeng and Chang-Te 161 kV sented in the paper.
substations of Taiwan Power company by PMU equipped with
GSCG is presented and the accuracy for estimating parameters In Part I, theory and algorithms of an adaptive transmission
of line is verified. The new developed DFT based method (termed line fault detection/location technique are presented. The adap-
as Smart Discrete Fourier Transform, SDFT) and line parameter tive fault detection/location technique uses synchronized fun-
estimation algorithm are combined with PMU configuration to damental voltage and current phasors at both ends of transmis-
form the adaptive fault detector/locator system. Simulation results sion line, and of course, requires an extremely accurate common
have shown that SDFT method can extract exact phasors in the
presence of frequency deviation and harmonics. The parameter timing reference. In the past, it was difficult to precisely syn-
estimation algorithm can also trace exact parameters very well. chronize the sampling at the ends of transmission lines due to
The SDFT method and parameter estimation algorithm can the lack of a common timing reference. At an early stage, it
achieve accuracies of up to 99.999% and 99.99%, respectively. usually adopts broadcast signal as timing reference, or utilizes
The EMTP is used to simulate a 345 kV transmission line of telecommunication technique to transmit remote data to the con-
Taipower System. Results have shown that the proposed technique
yields correct results independent of fault types and is insensitive trol center. However, these introduce new problems i.e. there
to the variation of source impedance, fault impedance, and line will exist multi-path effects or it needs to precisely determine
loading. The accuracy of fault location estimation achieved can be the delay of communication channel and to correctly compen-
up to 99.9% for many simulated cases. The proposed technique sate for this delay. Now, using the accurate timing signal pro-
will be very suitable for implementation in an integrated digital vided by Global Positioning System (GPS) as common time
protection and control system for transmission substations.
base for measuring instruments located at both ends of line, we
Index Terms—Discrete Fourier transforms (DFT), fault detec- can highly promote the accuracy of synchronized measurements
tion/location index, phasor measurement unit (PMU). and reduce the cost of equipment greatly. Because of low fre-
quency of the timing signal of GPS, it can not be used as sam-
I. INTRODUCTION pling signal directly. This means that a timing device is needed
to do frequency multiplication/division task. We will present the

F AULT detection/location on transmission lines is a very


well-known problem that has been studied for a long time.
An accurate fault detection/location technique is of special im-
hardware and software design of such a timing device and inves-
tigate performance of this device through field-test in the paper.
The proposed PMU based fault detection/location technique
portance in improving power system reliability including re-
incorporated with line parameter estimation forms an adaptive
laying, analysis for line inspection, and routine maintenance.
technique. The adaptive fault detection/location technique can
In the past, the most common approach was to model trans-
cope with various factors associated with the accuracy of fault
mission lines using phasors [1]–[9]. This approach is referred
detector/locator mentioned above. Its performance will be eval-
to as impedance based measurement technique, and is classi-
uated with respect to various factors such as fault resistance,
fied to two methods. The earlier developed one is one-terminal
source impedance variation, line loading, fault incidence angle,
data method, and the other is the currently more prevalent one
etc. The paper is organized into four sections, Section I is the
so-called two-terminal method. These methods use voltage and
introduction. In Section II, we first briefly describe the con-
figuration of PMU, and then present the design concept of a
Manuscript received February 5, 1999; revised February 1, 2000. This new timing device (GSCG). The performance of PMU has been
work was supported by the National Science Council under the Grant NSC verified in field-test at 161 kV substations of Taiwan Power
88-2612-E-002-005. System (Taipower System) and the results are also presented
The authors are with the Department of Electrical Engineering, National
Taiwan University, Taipei, Taiwan. in this section. The performance of the proposed adaptive fault
Publisher Item Identifier S 0885-8977(00)10302-4. detection/location algorithms will be evaluated with respect to
0885–8977/00$10.00 © 2000 IEEE
JIANG et al.: AN ADAPTIVE PMU BASED FAULT DETECTION/LOCATION TECHNIQUE FOR TRANSMISSION LINES—PART II 1137

currents are digitized by A/D converter, and then fetched into


microprocessor. The microprocessor determines the phasor
according to built-in recursive SDFT program, and the time
marker information from GPS is assigned to the phasors as their
identifying tag. These time-marked quantities are transmitted
to remote site through the communication channel.

B. GSCG Configuration
In order to reduce settling time, frequency jitter, and steadys-
tate errors, we design a new timing device named “Global Syn-
chronism Clock Generator (GSCG)” whose timing error could
remain within that of satellite’s receiver. The block diagram
and actual implementation of GSCG are shown in Fig. 1(b) and
(c), respectively. The vital core of GSCG is a microprocessor
based phase locked loop (PLL). The PLL consists of time-error
counter, microprocessor, digital-adjustable oscillator, and fre-
quency divider. The time-error counter compares the difference
between 1 Hz signals generated by GPS and frequency divider
(divided by ). Then, the error quantity is sent to micropro-
cessor, and is processed with the time marked data provided by
GPS receiver herein. Microprocessor uses a P–I type (propor-
tional and integral) control program to correct the error quantity
in sampling clock. Concurrently, microprocessor will transmit
the clock signal to another frequency divider (divided by ),
and from there the accurately synchronized clock output goes
to two A/D converters. The digital-adjustable oscillator is con-
trolled by voltage provided by the microprocessor. Its output
frequency is a function of the control voltage. The important
merits of PLL utilized in GSCG are that it adopts close-loop
feedback control scheme, and uses a gain-programmable oscil-
lator. Hence, the new designed GSCG is a completely digitized
timing device, its speed is very fast and is quiet suitable for the
applications of computer relaying. Besides, the very low cost for
constructing the GSCG is another one of several merits of our
Fig. 1. The configuration of phasor measurement unit based on GPS and its
actual implementation. design, for example, the crystal oscillator utilized in our GSCG
costs merely 32 NT$ (about 1 US$). Moreover, another im-
various facts mentioned above through the EMTP generated portant advantage of GSCG is its ability to automatically com-
data. Simulation results will be presented Section III. Finally, pensate unsynchronization error by software built in micropro-
Section IV is the conclusions. cessor. From our investigations, the frequency drift of GSCG
can be controlled within 0.1 PPM and the error of rising edge
II. IMPLEMENTATION OF PHASOR MEASUREMENT UNIT of two GSCGs clock is able to remain within 1 s (equiv-
alent to 0.0126 phase difference at 60 Hz). These facts certify
A. PMU Configuration the synchronism between two PMUs when they are used to mea-
As mentioned in Part I, the phasor approach is quite attractive sure phasors at different ends of transmission line.
as long as all of the algorithm assumptions are met and phasor
quantities are estimated accurately. It is widely recognized that C. Laboratory Test
two-terminal based fault detectors/locators are subject to errors In order to demonstrate the accuracy of PMU without the ef-
coming from unsynchronized sampling clock. Use of a timing fects of CTs and CCVTs, we have performed lots of laboratory
signal from GPS can greatly reduce or eliminate such errors. tests. Selected results will be presented in this paper. At first,
In this paper, we proposed an extremely accurate con- we input a sinusoidal signal to the A/D converter
figuration for measuring phasors by virtue of Phasor Mea- terminals of two different PMUs directly. The computed am-
surement Units (PMU) technique, whose configuration is plitudes and phases of the test signal by two PMUs are shown
illustrated in Fig. 1(a). The GPS receiver provides 7 pps (one in Fig. 2(a) and (b), respectively. The aim of the upper half of
pulse-per-second) signal and a time marker, which are sent to Fig. 2(a) is to exhibit the ability for tracing variation trend mea-
GSCG and microprocessor respectively. The 1 pps signal will sured by both PMUs. The data are output per one-cycle. It is
be utilized to process the frequency multiplication/division task clearly seen that both measured results are traced very well.
in the GSCG, and then GSCG provides synchronized sampling From Fig. 2, it can be seen that the difference between the two
pulses to the A/D converter. The measured 3 voltages and computed amplitudes remains within 0.03%, and the computed
1138 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 15, NO. 4, OCTOBER 2000

Fig. 3. Comparison between theoretical values and field-measurements


for c-phase current in Chung-Jeng and Chang-Te substations at GPS TIME
4/22/1998 02:57:00.

Fig. 3 simultaneously for comparison. Fig. 3(a) only shows


the -phase current per unit value of transmission line, where
Fig. 2. Performance evaluation of PMUs in laboratory test.
we can observe that there are only about 2.5% difference
between measured values and theoretical values at Chung-Jeng
phase errors can also remain within 0.02 . These facts indi- substation. The measured and theoretical phase angle values of
cate that the new designed PMUs can indeed achieve relatively -phase of transmission line are shown in Fig. 3(b). The max-
high accuracy for measuring phasors in the absence of CTs and imum computed phase difference between the measured data
CCVTs. and theoretical data is only 0.47 . Besides, the amplitude and
D. Field Test in 161 kV Substations phase measured at both substations also agree with theoretical
values very well. This fact thereby verifies the synchronism
The difference between laboratory test and field test in between these two PMUs. It is worthy to note that there are
161 kV substations is that the results of field test will include a lot of different cases tested in our studies and all provide
the errors involved in CTs and CCVTs. At the GPS system time extremely accurate results.
4/22/1998 02:57:00, the performance of PMU-GSCG has been
verified with the field measurements which were performed III. PERFORMANCE EVALUATION
at a single 161 kV transmission line between Chung-Jeng and
Chang-Te substations of Taipower system. The distributed A. Simulation Example
-model parameters of this transmission line are shown as The proposed adaptive fault detection/location technique was
followings: tested using data obtained from the EMTP simulator. Different
types of fault, different fault resistances, different synchroniza-
tion errors, various power flow conditions, and inception angles
[p.u.] of the fault occurrence were considered. For these situations,
a 345 kV, 100 kilometers transposed transmission line is sim-
MVA BASE MVA ulated with distributed parameters model that is shown in the
followings:
V BASE kV
1) Transmission Line Model: Zero-sequence parameters:
The data measured at Chang-Te substation (receiving end)
km
was substituted into the -model transmission matrix of the
line to calculate the theoretical values of voltage and current H/km
at Chung-Jeng (sending end). These three data are plotted in F/km
JIANG et al.: AN ADAPTIVE PMU BASED FAULT DETECTION/LOCATION TECHNIQUE FOR TRANSMISSION LINES—PART II 1139

Positive-sequence parameters:
/km
H/km
F/km
*Conductance is neglected in the simulation.
Both ends of the line are replaced by Thevenin’s equiva-
lent impedance with an angle different than that of the line
impedance, and are shown in the followings:
Sending End:

line-to-line
Receiving End:

line-to-line
In our simulation cases, except for investigating the effects
of unsynchronized measurements and power flow capacity
variations, the phasor quantities for voltage and current at re-
ceiving bus were intentionally shifted ten degrees (same value
as Ref. [5]). The total time of simulation is (m-sec),
and the data sampled at sampling rate of 1.92 kHz (32 times
60 Hz). The moving data window contains thirty-two measured
data per cycle, and which are initiated from the simulation start. Fig. 4. Test signal: v(t) = cos(wt): simulated frequency = 59.5 Hz;
sampling frequency = 960 Hz.
We first conduct performance evaluation for SDFT algorithm
and transmission parameter estimation algorithm, and then
fault detection/location index. The error of using the proposed
fault location index, , is expressed as a percentage of the total
line length, i.e.

error EstimatedTotal
Location Actual Location
Line Length
(1)

B. Performance of SDFT Method


In order to demonstrate that the performance of the SDFT
method for extracting phasors of fundamental frequency is su-
perior to that of conventional DFT method, we use EMTP to
generate the data for comparison. In order to completely reflect
actual system operating situations, Gaussian-type noise (pro-
posed by A. G. Phadke et al. [14]) will be added to simulated
data. First, we show that SDFT could obtain an exact frequency
calculation under frequency deviation in a pure sinusoidal wave-
form. The simulated frequency is set at 59.5 Hz. The conven-
tional DFT and the SDFT are utilized to extract phasor and fre-
quency. The extract phasor, frequency, and phase error are illus- Fig. 5. Comparison between conventional DFT method and SDFT method for
trated in Fig. 4(a)–(c) respectively. From Fig. 4, it is observed extracting phasor and true system frequency under a-phase ground fault.
that the conventional DFT method gives the worser phasor and
frequency estimates. However, the SDFT method could give a The fault location was set to be . Simulated frequency
stable phasor measurement and an exact frequency identical to was intentionally chosen to be 60.5 Hz. From Fig. 5, it is ob-
the setting value, i.e. 59.5 Hz. Fig. 5 shows an another test case. served that the conventional DFT gives the worse phasor and
In this test case, we use EMTP to simulate -phase ground fault. frequency estimates. However, the SDFT method could give a
1140 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 15, NO. 4, OCTOBER 2000

Fig. 6. The ability of the proposed estimation algorithm to trace line parameter
variations. Fig. 7. Detecting the occurrence of single-phase ground fault whose fault
index is D = 0:8.

stable phasor measurement and an exact frequency identical to


the setting value in EMTP, i.e., 60.5 Hz. D. Performance Evaluation for Adaptive Fault
Detection/Location Index
C. Performance of Line Parameter Estimation Algorithm
1) The Performance of Fault Detection: It is observed that
From many researches in the field of fault location at earlier absolute values of and [(6) in Part I] will all be held at zero
stage, uncertainty about the line parameters is one of the most before the occurrence of a fault. This can be proved by substi-
severe factors that could impact on the fault location accuracy. tuting the measured data ( , ) and ( , ) into the formulae
In practice, the line parameters will vary not only with the en- of and , respectively. Since such measured components all
vironmental conditions but also with the system operation sit- satisfy the transmission line equation, i.e., (2) (in Part I), after
uations. It is difficult to precisely presage what will happen to algebraic manipulation, we can prove that and are indeed
the line parameters with the transmission line in service. A. T. identical to zero, and hence the algorithm will give an indefinite
Johns et al. [6], [7] have adopted the method that varied the line value of fault location index through (5) (in Part I). However, as
impedance matrix and about the true values for the actual soon as the post-fault measured data is input into the algorithm
line to evaluate the effect of line parameter uncertainty on the through the moving data window, the computed absolute values
fault location accuracy. D. J. Lawrence et al. [8], [9] assess the of and abruptly deviate from zero, and hence the fault lo-
sensitivity of their algorithm about this topic. The method uti- cation index also quickly converges to some value. After ap-
lized by those authors is to vary the self-impedance and mutual proximately the first two or three cycles of fault, the computed
impedance in the EMTP. Both of these methods will give indica- absolute values of and would stabilize, and they would
tive and satisfactory results, but those authors have not proposed gradually stabilize to a constant gap between each other. While
a method to get rid of such a parameter uncertainty problem. the computed absolute values for and will take two or three
In this paper, we take characteristic impedance and propaga- cycles to converge to stable values, the fault location index
tion constant of line as line parameters, whose results are illus- (for most of fault-types) takes much less time to achieve a stable
trated in Fig. 6(a) and (b), to demonstrate the performance of value. These facts imply that , , and location index could
the proposed approach. In this test, we will adopt 10% varia- be employed as detection index of fault.
tion of line parameters about the true values as the quantities of Many fault events have been tested with respect to different
uncertainty. From Fig. 6(a) and (b), it can be clearly seen that fault-types, different fault resistance, and various fault locations
estimated parameters will trace exact parameters very well. The in our simulation. Fig. 7 shows a typical fault detection re-
errors of both parameters remain well within 0.01%. This pa- sponse for single-phase-to-earth fault involving a 1- fault re-
rameter estimation algorithm has been demonstrated with field sistance to ground in -phase. The fault time and fault location
test at 161 kV substation of Taipower system, and the results are are set at 37.47 milliseconds and 80 kilometers away from the
very good. receiving end, respectively. From Fig. 7, it is observed that
JIANG et al.: AN ADAPTIVE PMU BASED FAULT DETECTION/LOCATION TECHNIQUE FOR TRANSMISSION LINES—PART II 1141

TABLE I
STATISTICAL RESULTS OF THE RELIABILITY EVALUATION OF THE PROPOSED
RELAYING UNDER VARIOUS PERMANENT INTERNAL FAULTS

TABLE II
SIMULATION RESULTS OF THE SECURITY EVALUATION OF THE FAULT
DETECTOR UNDER VARIOUS RERMANENT EXTERNAL FAULTS

Fig. 8. Detecting the occurrence of b–c phase ground fault whose fault index
and fault resistance are D = 0:8 and R = 10 k
, respectively.

and were identical to zero before the fault event, and hence
the fault index will be an indefinite value. When the measured
post-fault data were fetched in the algorithm, the computed
and values quickly deviated from zero, finally reaching stable
values (in this case, the absolute values ,
). The location index abruptly converged
to 0.8. As mentioned above, the fault detection performance of
the proposed algorithm is well suited for any type of fault events.
Another simulation case is shown in Fig. 8. This case is a –
phase ground fault case, whose fault resistance and fault loca-
tion are set at 10 k and 80 km, respectively. While this case
is a high fault resistance case, Fig. 8 clearly shows that the pro-
posed index once again provides excellent performance for fault
detection. The security performance of the fault detector has also been
A lot of cases of all possible faults on the protected line were examined with respect to various external fault locations and
simulated. Using the generated data, the reliability, security and intermittent disturbance (for example, switching of power-com-
responsibility of the fault detector were also checked. With re- pensated capacitors). For permanent external faults, the security
spect to five different thresholds of index , for example, we of the fault detector achieved can be up to 100% for these simu-
have performed 756 tests of internal fault events to evaluate the lated cases, whose results are summarized in Table II. Using the
reliability of the fault detector. The results are summarized in EMTP generated data, the responsibility of the fault detector
Table I. The fault detector is regarded as failures to trip when it was also checked. From the simulation results, it is observed
takes over the specified decision-limited periods. There are three that the tripping decision time of the fault detector is relative
specified decision-limited periods, i.e. 1-cycle, 1.2-cycle, and insensitive to the fault type and fault locations. For the most of
1.5 cycle, are chosen as restraint periods for the simulation tests. the tested cases, the maximum time for tripping decision takes
In these tests, it is clearly seen that larger threshold of index only about one-third cycle even under extreme proximal fault
is chosen, the more reliability of the fault detector possesses. events. From our investigations, it has been found that the fault
When the threshold of is set at 0.0285, for example, the relia- detector indeed provides an excellent performance in the fault
bility of the fault detector can achieve even high to 100.0% and detection.
98.41% with respect to 1.5-cycle and 1-cycle decision-limited Besides, the implementation cost for the proposed PMU, in-
periods, respectively. From the shown table, it is observed that cluding Pentium® II-300 PC, DSP card, Data Acquisition Card
the relaying scheme can indeed achieve an extremely high level (PCL-818HG), 12-bit A/D converter, GPS Receiver, and other
of reliability. necessary devices, is approximately $3125. This means that the
1142 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 15, NO. 4, OCTOBER 2000

Fig. 9. Effects of fault type and fault location on index accuracy. Fig. 10. Index accuracy of a-phase ground fault under different resistance
situations.

cost effectiveness can be achieved. The proposed technique will


thus be very suitable for implementation in an integrated digital
protection and control system for transmission substations.
2) Effects of Fault Type and Fault Location: In order to
demonstrate the performance of the proposed fault location
index, we have generated a number of test cases using an
EMTP. This includes all possible types of fault. Selected results
are presented in terms of error in the fault location estimates,
and are shown in Fig. 9. The errors between the estimated
location and actual location are plotted by percentage on the
-axis of the figure; all distances are measured from the
receiving substation. An error of 0.1% indicates that the Fig. 11. Index accuracy for a–c phase ground fault (d = 0:2 p.u.) and a–b–c
phase ground fault (D = 0:5 p.u.) on different resistance.
index determines the fault to be 0.1% of the total line length
closer to the receiving substation than the actual location.
As Fig. 9 shows, the errors of the proposed index are well double-line ground fault and triple-line ground fault cases
within 0.4% for most of fault types and fault locations, which shown in Fig. 11, where the maximum error is well below
means that the accuracy of this index is essentially indepen- 0.2% for both cases. From our investigations, it is clearly
dent of fault type and fault location. In general, the accuracy observed that errors will tend to increase slightly as faults
of fault events associated with ground faults is better than that move closer to both ends of line or fault resistance increases to
of line-to-line short faults (i.e., ungrounded faults). Moreover, it very large value for some fault events. However, those errors
can be seen from Fig. 9 that two peak errors (about 0.417% and are so small that will not affect the accuracy of this index. All
0.306%) of – line short faults occurred at 40 and 60 km, re- of these facts mentioned above imply that the proposed fault
spectively. The same situation is in a three-line short fault, and locator gives an inherently accurate estimation of fault position
these two larger errors are about 0.04% and 0.12% located in that is almost independent of the fault resistance. This is a very
the mentioned positions. Although the proposed index displays significant advantage over conventional approaches.
the phenomenon mentioned above, the maximum error of the 4) Effects of Two-Terminal Timing Accuracy: To investigate
index still remains well within 0.2% for most fault events. the impact of synchronization error on fault location accuracy,
3) Effects of Fault Resistance: Fault resistance is one of the authors have utilized EMTP to perform a lot of simula-
major factors that affect the accuracy of the impedance-based tion examples with respect to different fault types and various
one-terminal data fault locator methods. In general, the errors fault locations. In these test cases, an error phase shift is added
of such methods are proportional to the fault resistance. Even to the phasors from receiving end and compared with that of
small fault resistance (1 ) can cause a significant arc voltage sending one. These errors range from 10 to 10 degree (approx-
drop that can produce large fault location errors for remote imately equivalent to 0.463 milliseconds of sampling time
faults [9]. The unknown resistance comes from the arc resis- error). Fig. 12(a) and (b), respectively show the errors of single-
tance, the resistance in any additional paths to ground, and the phase ground fault and double-line short fault due to the data
errors in ground resistivity modeling. A typical fault resistance measured at both ends of transmission line asynchronously. In
is seldom over hundreds of ohms. our investigations, it is observed that the synchronization error
Figs. 10 and 11 respectively illustrate the impacts of involved in ungrounded fault cases is much more than that of
single-line ground fault and multiple-line ground fault events other fault events. When synchronization error is 10 degree, for
with respect to different fault resistance on fault location example, it will cause up to 4.76% (minus sign has been ne-
estimates. It is evidently from Fig. 10 that the proposed index glected) error in the fault location accuracy under the -phase
is relatively insensitive to magnitudes of fault resistance. Even ground fault cases. Under those ungrounded fault events, e.g.
in the extreme case of 10 k fault resistance (this could be – line short, we find that the effect of timing error is more
regarded as a high impedance fault case), the maximum error severe on fault location accuracy. These facts are illustrated in
is 0.1% approximately. Another results could be found in the Fig. 12(b) and when synchronization error is set at 10 , the
JIANG et al.: AN ADAPTIVE PMU BASED FAULT DETECTION/LOCATION TECHNIQUE FOR TRANSMISSION LINES—PART II 1143

Fig. 13. Effects of line parameter uncertainty on fault location index.

Fig. 12. Effects of two-terminal timing accuracy for various fault types.

error increased up to 5.66% under the case of fault location


index (p.u.).
From the papers [6], [7] presented previously, for example,
those authors usually adopted a one-cycle moving data window
to estimate the phasor quantities. Suppose that the window
length is thirty-two samples per cycle. This means that if the
moving data window mismatches only one sample, then it
exhibits 11.25 degrees phase error between the data measured
from both ends of line. From Fig. 12, such a synchronization
error will possibly cause a severe error in the fault location
accuracy.
The synchronization problem can be solved by use of a timing
signal such as GPS. In the proposed technique, we have de-
signed an extremely accurate timing device (GSCG), which is
equipped in PMU and can be utilized to precisely provide a Fig. 14. Performance evaluation of parameter estimation algorithm.
timing clock signal synchronized with the timing signal issued
by GPS satellite. As mentioned in Section II, the timing error
between two GSCGs clock is able to remain within 1 s -phase ground fault: resistance up, capacitance and induc-
(equivalent to 0.0216 degree), and its performance has also been tance up.
demonstrated in field test very well (see Fig. 3 shown). In our in- -phase ground fault: resistance up, capacitance and induc-
vestigations, it is seen that the synchronization error in the data tance down.
measured by the PMU based on GSCG can remain well within The fault location in both simulated cases is chosen at 40 km
1 degree for most of test cases. This fact guarantees that the lo- for comparison. Fig. 14(a) and (b) respectively show the sim-
cation accuracy of the proposed index will be hardly affected by ulation results of both cases described above, from which we
the synchronization error. find that there indeed exists an error in fault location accuracy
5) Effects of Line Parameter Uncertainty: Fig. 13 shows the due to parameter uncertainty. In order to eliminate the error in-
error associated with the variation of line parameters for -phase volved with such uncertainty described previously, the proposed
ground fault. In this test case, the variations of various line line parameter estimation serves the purpose. As Fig. 14 shows,
model parameters in EMTP are all from 20% to 20%. The when we use the parameter estimation algorithm to correct such
errors of fault location are computed in terms of -mode under parameter uncertainties, the results are almost unaffected by the
this test, whose results are shown in Fig. 13. It is found that the mentioned problems.
impact of line parameter uncertainty on the remote-end fault 6) Effects of System Noise and Frequency Fluctua-
case (i.e. 80-km) is minimum. However, the error of 30-km tion: System noise and frequency fluctuation that originated
faulted case will be up to 6.7% approximately. Usually, line from system unbalance, measurement hardware error, and line
parameters do not vary consistently, or rather they would change loading variation, etc. will affect the fault detection/location
vagariously. Thus, the simulations were further run assuming accuracy. In order to verify the performance of the SDFT
the following conditions: method applied to fault detection/location, we have studied a
1144 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 15, NO. 4, OCTOBER 2000

Fig. 15. Comparison between DFT and SDFT for a-phase ground fault under
frequency-floating and system noise-added conditions.

Fig. 17. Effects of line loading variations on measuring accuracy.

between system operating conditions at the occurrence of fault.


It can be clearly seen that for the source impedance considered,
Fig. 16. The influence of source impedance on fault location index.
the technique gives an accurate result of fault location. When
lot of fault events. We will present selected results among those source impedance variations increase, the errors of the proposed
cases as followings. technique will hardly be affected, and the maximum error can
Consider the -phase ground fault case, whose result is shown remain well within 0.07%.
in Fig. 15. In this case, the simulated system frequency and fault 8) Effects of Line Loading: The system operation conditions
location are set to be 60.5 Hz and 60 kilometers away from frequently vary with time. This will therefore cause the magni-
receiving end of line, respectively. In this test, Gaussian-type tude and phase angle of equivalent source at both ends of line to
noise has also been added into the raw data generated by EMTP. also change with time. To investigate the impact of line loading
The measuring accuracy of the SDFT method will remain within on the proposed fault locator accuracy, we follow the ways pro-
0.5%. It is once again found that the performance of the SDFT posed by D. J. Lawrence et al. [9] to estimate the fault location
method is superior to that of conventional DFT method. error. We utilized EMTP to run three power flow cases that are
7) Effects of Source Impedance: In order to demonstrate representatives of typical system operating conditions:
that the proposed algorithm is hardly affected by source Case 1) light load, magnitude of equivalent source at re-
impedance variation, we will follow the way proposed by the ceiving end downs 50%, and receiving end phase
D. J. Lawrence et al. [9] to evaluate the performance of the pro- angle remains unchanged.
posed technique. First, let source impedance be held constant Case 2) heavy load, magnitude of equivalent source at re-
to evaluate the sensitivity of the technique versus various fault ceiving end increases 50%, and receiving end phase
locations. Next, we changed fault location every ten kilometers angle remains unchanged.
for various fault types, and then investigated the sensitivity of Case 3) heavy load, magnitude and phase angle of equivalent
proposed technique with respect to different source impedance. source at receiving are increased by 50% and 30
All of these results were put together in Fig. 16 for comparison. degree, respectively.
-phase ground fault is selected as simulation example for While these simulation cases do not coincide with the prac-
this test. Fig. 16(a) and (b) show the results of the zero and tical situations exactly in power systems, they are approximated
-mode, respectively. In this case, both the self and mutual sufficiently to permit investigation of line loading conditions.
impedance of the remote equivalent source are varied by a Fig. 17 presents the impact of line loading variation on the
factor of five to simulate the impact of severe mismatches fault locator accuracy. The – – line short fault has been
JIANG et al.: AN ADAPTIVE PMU BASED FAULT DETECTION/LOCATION TECHNIQUE FOR TRANSMISSION LINES—PART II 1145

from 0 to 360 degrees, respectively. The fault incidence angles


of fault event refer to the phase angle of -phase. In -phase
ground fault, the fault resistance is set to be 1 . The fault lo-
cation of the simulated case is set at . The results evi-
dently show that the proposed technique maintain an extremely
high degree of accuracy which is almost independent of the fault
incidence angle. The error of -phase ground fault will remain
well with 0.06% for zero mode and about 0.1% error for
-mode. These facts once again demonstrate that the proposed
technique can indeed provide an excellent performance for fault
location.

Fig. 18. Effects of fault incidence angle on fault location accuracy under
different fault events.
IV. CONCLUSION
The adaptive PMU based approach to accurately detecting/lo-
cating faults on an EHV/UHV transmission line is presented.
selected as the fault event in this case. For simplicity, we will
Fault location index as well as computed factors and
consider herein only one fault location, i.e., (p.u.).
can serve as fault detector and locator simultaneously. The pro-
Fig. 17(a) shows the detection performance of the proposed
posed index does not need any assumption, so it is a very robust
technique under various line loading conditions described
index. The proposed SDFT algorithm is also a recursive, fast,
above. The results of Cases 1–3 are put in Fig. 17(a) for
and stable algorithm that can be applied in the field of computer
comparison. It is observed that the technique can provide an
relaying. The parameter estimation algorithm is also proposed
excellent and stable performance for detecting fault occurrence
to solve the uncertainty of line parameters. This algorithm can
under different line loading conditions. Fig. 17(b) shows the
use the measured data on both ends of line to on-line correct
sensitivity of the proposed technique with respect to these test
line-parameter variations due to environmental change and var-
cases. For the fault events considered, it can be clearly seen that
ious system operating situations. The performance of parameter
the proposed technique will give extremely accurate results.
estimation algorithm has been verified with the data generated
This means that each curve of the errors of the technique versus
by EMTP. Simulation results show that this algorithm can trace
sampling step falls approximately within 0.2% from 2 cycles
parameter variations very well. Hardware errors, system noises,
after occurrence of fault. The proposed technique is least
harmonics, and system frequency fluctuation problem have been
sensitive to the effects of pre-fault loading. In our laboratory,
considered in this paper. A special filtering technique so-called
the authors have run many simulation cases with respect to
the SDFT method has been developed to solve such problems
different fault events. From such investigations, it has been
mentioned above. The performance of the SDFT method for
found that the proposed technique indeed provides an excellent
extracting true system frequency and fundamental phasors has
performance in the fault detection and location.
been verified. The accuracy of both SDFT method and param-
9) Effects of Fault Incidence Angle: As we known, when
eter estimation algorithm achieved can be up to 99.999% and
faults occur near voltage maximum and near voltage zero, the
99.99%, respectively. The accuracy of fault location estima-
higher frequency transients and DC offsets will take place in a
tion achieved can be up to 99.9% for many simulated cases. A
transmission line; and thereby will cause large difficulty in the
high performance phasor measurement unit (PMU) is designed
fault location method. This means that the burden of filtering
and implemented in the paper. Combining robust fault detec-
will increase largely and might need a more delicate method to
tion/location index, parameter estimation algorithm, the SDFT
reject such transients and to extract true fundamental phasors
method, and the well-designed PMU, the proposed technique
utilized in fault location algorithm. In practice, the fault inci-
will be an adaptive, high performance, and low-cost fault detec-
dence angle, however, cannot be predicted in advance. H. Y. Li,
tion/location technique.
et al. [13] have utilized one-cycle Fourier window in their pro-
tection algorithm to investigate the effects of the fault incidence
angle. They found that the protection algorithm may be unable ACKNOWLEDGMENT
to recognize the fault under the condition that post-fault and pre- The authors are thankful to C.-S. Chen for his editing assis-
fault information exist simultaneously in the sampling window. tance. Also they are thankful to Dr. J.-S. Yang at Power Research
In our technique, however, we adopt the moving data window to Institute of Taiwan Power Company for his technical advice.
sample data, and hence the proposed algorithm is more insensi-
tive to the problem mentioned above. This fact has been demon-
REFERENCES
strated clearly in previous presentations. For example, the fault
detection performance of our technique is hardly affected. [1] T. Takagi, Y. Yamakoshi, J. Baba, K. Uemura, and T. Sakaguchi, “A
new algorithm of an accurate fault location for EHV/UHV transmission
To demonstrate this performance of the proposed technique, lines—Part I: Fourier transformation method,” IEEE Trans. on Power
we have performed many simulation tests: Fig. 18 shows the se- Apparatus and Systems, vol. PAS-100, no. 3, pp. 1316–1323, Mar. 1981.
lected results among those test cases. In this case, the -phase [2] , “A new algorithm of an accurate fault location for EHV/UHV
transmission lines—Part II: Laplace transform method,” IEEE Trans. on
ground fault is chosen as illustrative case. This fault event is as- Power Apparatus and Systems, vol. PAS-101, no. 3, pp. 564–573, Mar.
sumed to occur at point on waves for every 45 degrees spacing 1982.
1146 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 15, NO. 4, OCTOBER 2000

[3] M. Kezunović, J. Mrkić, and B. Perunicić, “An accurate fault location Ying-Hong Lin was born in Taipei, Taiwan, in 1970.
algorithm using synchronized sampling,” Electric Power Systems Re- He received his B.S. degree in electrical engineering
search, vol. 29, pp. 161–169, 1994. from Taiwan University of Technology in 1995 and
[4] A. A. Girgis, D. G. Hart, and W. L. Peterson, “A new fault location tech- is presently working on M.S. degree from National
nique for two-and three-terminal lines,” IEEE Trans. on Power Delivery, Taiwan University, Taipei, Taiwan. His interested
vol. 7, no. 1, pp. 98–107, Jan. 1992. researches are the application of GPS and PMU in
[5] D. Novosel, D. G. Hart, E. Udren, and J. Garitty, “Unsynchronized two- power system.
terminal fault location estimation,” IEEE Trans. on Power Delivery, vol.
11, no. 1, pp. 130–137, January 1996.
[6] A. T. Johns and S. Jamali, “Accurate fault location technique for power
transmission lines,” IEE Proceedings, pt. C, vol. 137, no. 6, pp. 395–402,
Nov. 1990.
[7] P. K. Aggarwal, D. V. Coury, A. T. Johns, and A. Kalam, “A practical
approach to accurate fault location on extra high voltage teed feeders,”
IEEE Trans. on Power Delivery, vol. 8, no. 3, pp. 874–883, July 1992. Jun-Zhe Yang was born in Tainan, Taiwan, in 1971.
[8] D. J. Lawrence, L. Z. Cabeza, and L. T. Hochberg, “Development of an He received his B.S. degree in electrical engineering
advanced transmission line fault location system—Part I: Input trans- from Tatung Institute of Technology in 1992 and
ducer analysis and requirements,” IEEE Trans. on Power Delivery, vol. M.S. degree from National Taiwan University in
7, no. 4, pp. 1963–1971, Oct. 1992. 1995. He is presently a graduate student in the
[9] , “Development of an advanced transmission line fault location electrical engineering department, National Taiwan
system—Part II: Algorithm development and simulation,” IEEE Trans. University, Taipei, Taiwan.
on Power Delivery, vol. 7, no. 4, pp. 1972–1981, Oct. 1992.
[10] B. Lian and M. M. A. Salama, “An overview of digital Fault Location Al-
gorithms for power transmission lines using transient wavefomts,” Elec-
tric Power Systems Research, vol. 29, pp. 17–25, 1994.
[11] Power Engineering Education Committee and Power System Relaying
Committee of the IEEE Power Engineering Society, “Advancements in
microprocessor based protection and communication,” in IEEE Tutorial Tong-Ming Too was born in Yuan-Lin, Taiwan, in
Course, 1997, pp. 63–72. 1971. He received his M.S. degree in electrical engi-
[12] A. G. Phadke and J. S. Thorp, Computer Relaying for Power Systems: neering from National Taiwan University in 1998 and
John Wiley & Sons, 1988. is presently working on Mass Transportation Bureau
[13] H. Y. Li, E. P. Southern, P. A. Crossley, S. Potts, S. D. A. Pickering, B. in Taipei, Taiwan. His interested researches are the
R. J. Caunce, and G. C. Weller, “A new type of differential feeder pro- measurement technique in power system.
tection relay using the global positioning system for data syndironiza-
tion,” IEEE Trans. on Power Delivery, vol. 12, no. 3, pp. 1090–1097,
July 1997.
[14] A. G. Phadke, J. S. Thorp, and K. J. Karimi, “State estimation with
phasor measurements,” IEEE Trans. on Power Systems, vol. PWRS-1,
no. 1, pp. 233–241, Feb. 1986.

Chih-Wen Liu was born in Taiwan in 1964. He


received the B.S. degree in electrical engineering
Joe-Air Jiang was born in Tainei, Taiwan, in 1963. from National Taiwan University in 1987, and M.S.
He graduated from National Taipei University of and Ph.D. degrees in electrical engineering from
Technology in 1983 and received M. S. degree Cornell University in 1992 and 1994. Since 1994,
in electrical engineering from National Taiwan he has been with National Taiwan University, where
University, Taipei, Taiwan in 1990. Since 1990, he is Associate Professor of electrical engineering.
he has been with Private Kuan-Wu Institute of He is a member of the IEEE and serves as a
Technology and Commerce, where he is Instructor reviewer for IEEE TRANSCATIONS ON CIRCUITS
of electrical engineering. He is currently working on AND SYSTEMS—PART I. His main research area is in
a Ph.D. research project in power system protection application of computer technology to power system
and control, also at National Taiwan University. His monitoring, operation, protection and control. His other research interests
area of interest is computer relaying and bio-effects include GPS time transfer and chaotic dynamics and their application to system
of EM-wave. problems.

Вам также может понравиться