Professional Documents
Culture Documents
(45 Akim) Girgis1992 PDF
(45 Akim) Girgis1992 PDF
[i] E]
Equation 4 can be rewritten as
= D or Y = MD
j = a,b,c
where
Defining xi and x2 to be D and e j', equation (14) can be rewritten as
Yj = Vjl - VS + L ZjiIa
i=cr,b,C
Yj=Mlj X I + M2j ~ 2 M3j
+ X I ~2 (15)
or
Mj = zji (Iil +liz) (7)
Vjl
kb.c Yj=M&) , where Yj=-vfl
j = 4b.c
Equations 14 or 15 represent three complex measurements in two
Equation 5 represents three complex equations or six real equations in
unknowns. One unknown is real (xi), while the other unknown is
one unknown, D.
The solution for D can be then obtained using the least-squares complex e$. Starting with initial conditions xi,, and x20, the values of
estimates as xi and x2 can be updated using least squares state estimation as
D = (M+M)-' M+ Y (8)
where M' is the conjugatetransposeof M.
where H is the Jacobian matrix (dYjldxi). and
It should be noted that this procedure is independent of fault type or
fault resistance. AY = Yk - Yk-1
Using the computed value of D, the fault boundary conditions This iterative process may be terminated when the test for convergence
V F h and IF& can be obtained as follows is satisfied. That is when the differencein the estimated values of D in
- DWabcl
VFdc = V ~ M (9) successiveiterations is less than a prespecified value, 0.05 percent for
example.
= Iob~l+Iabc2 (10) The elements of the Jacobian matrix can be then obtained as
Using the voltage and currentvectors at the fault, the fault type may also
be identified.
Each of equations (23)and (24) represent three complex equations in
one unknown (real) or six r d equations in one unknown, A solution
for D can then obtained fiom (23)as
where
j = 4b.c
It is assumedhere that the process starts with kl.
To initialize the estimation process, initial values for for D and
-
D = (McA4)-1M+ Yl
+D -t
L1
L,-D
L2-
(3)
ciamnceded. Iftheprobabilitydiseibutionoffaultlocationisknown,
the mean value for fault location can be used for D and a value of zero
can be used for 6. A reasonable initial value for D may be 0.5L. Also
a value of zero for both D and 6 can be used as initial values. The
technique thenproceeds as shown above. SYSTEM 1 SYSTEM2
L3
2. ThreeTenninalLines
a.
SYSiEM 3
2-b
Ll L2-
SYSTEM 1
1abcg.i
F3
Li = 30 miles
L.2 = 10 miles
L3 = 50 miles
SYSTEM 3
24
Figure 2. Possible Fault Locations in a Three-TerminalLine
Similarly, for a fault at F3, assuming equal line impedances per mile,
I:[; =E]
equation (24) still holds, but equation (23) becomes
Li
Thus,it can be easily shown that for equal line impedances per mile,
(28)
-
Considerthe voltage and currentphasors at bus ( 1 ) to be reference. v,
The voltage and currentsmeasuredat bus (2) have synchronizationerror only satisfied if D is replaced by Li.
of -61 and the voltages and currents measured at bus (3) have Forequallineimpedancespermile,solvingequations(34)and(36)
synchronizationerror of -62. Thus; independently with the results denoted by D and D' for the distance to
the fault leads to the followingresults:
6)
1. Fault on L1 (at F i )
VFdc + D L l h c l
Vab~i= (29) D < L1 (from equation 34) and
D' c L1 (from equation 36)
2. Fault on La (at F2)
D > L1 from (34) and
D' =L1 from (36)
3. Fault on L3 (at F3)
D=Li
D' > L1
Using the above procedure the fault location can be obtained in any of
the threesections.
. .
for Fault
Equation(32) represents three simultaneous non-linear algebriacequa- To solve equation (34) for D or equation (36) for D' initial values
*,
tions functionsof (D. e e"). Rearranging equation (32) results in are required. One method is to have a flat starting as:
D = O , e*=ej%=lLO
Vi1 and -.
This leads to the computed values of - Vi1
vn '13
The vector [D. e*, e&] can be then iteratively estimated as
2
vfl
= M1P + M2j e + M3j ea +M4P e* + M5,D eib (34)
where Hlk-1 is the Jacobianmatrix computed at step k-1.
j = 4b.c
102
The elements of the Jacobian mahix Hl&i(ij] can be obtained as 0.02453+ 0.01451+ 0.01417+
&&Xj wherefi is the right hand side of k
! and xj isjth element of the j 0.08418 j 0.03650 j 0.02982
VP 0.01982+ 0.01451+
unknown vector. Thus, zobc= j 0.08346 j 0.03650
I
0.02453+
(Mla + M4a e is, + M5a e 4
(M2,+M4a D) ( M a + M5a D) - j 0.08418
(M1b+M4bea + M 5 b e 9 (M?b+M4bD) (hf3b+M5bD)
(Mlc + M4, e is,+ e '"L>
(M& +M4, D) (MC + M5c D) The voltage and currentvectors (in per unit) for a single-line-to-ground
fault on phase A at 5 miles from bus 1, obtained from the fault analysis
program, are as follows:
(38)
unnptedwithvaluesofD, e & , eh atstepk-1.
Similarly using the Jacobian matrix H2 and equation (34) leads to vabd =
0.41116- jO.44443
-0.46270- j 0.824981,
-0.45694+ j 0.91119
[
4.30996- j 5.42317
Iki = 0.21988- j 0.405431
0.18471- j 0.35920
0.54412- j 0.37226 3.31969- j 3.69772
vabc2=
-O.42367+j 0.92512 -O.18468+j 0.35919
These voltage and current vectors were used to compute the Y vector
and the M matrix shown in equations 6 and 7. The fault location was
then computed using equation 8 to be 5.0 miles from bus 1.
1
0.57184- j 0.35254 3.31856- j 3.77992 Values computed using Jacobian matrix Hi
-O.41765+j 0.92604 -O.15328+j 0.30786
The algorithm computed D and D' to be 32.0 and 30.0 miles, respec-
tively. Since D is greater thanLl(30) andD' is equal toL1, the algorithm
determined that the fault is on L1 at 2 miles from the tap.
Case (iii): Fault on L3 (at F3). The third example shown here is
for a single-line-to-groundfault at 5 miles from bus 3. The voltage and Values computed using Jacobian matrix H2
current phasors (in pa.) are as follows:
I Iteration
1 Fault
1 I
1
0.59508- j 0.34504 Number Location ,-j61
Miles
(0')
-O.41034+j 0.93042
1 29.8397 0.97551-j 0.21992 0.98480-j 0.17364
0.56327- j 0.36271
II 1 II 1
3 30.oooO 2.19600- j 0.38848 0.98480-j 0.17364
-O.41766+j 0.92713
0.41116- j 0.44443 Case (iii): Fault on L3 (at F3) at 5 miles from bus 3
Values computed using Jacobian matrix Hi
-O.45694+j 0.91119
The values of D and D' were computed as 30.0 and 75.0 miles. I t e I Fault e;-
Number Location e-lzl
respectively. Since the value of D is equal to L1 and the value of D' is (D)Miles
greater than 151, the fault is determined to be on L3 (at F3) at 75 miles 30.0310 0.96592- ' 0.25881 0.00270- '0.10797
from bus 1 or 45 miles from the tap.
29.9999 0.96592- ' 0.25881 0.00250- ' 0.10791
4. Three-Terminalh e with Unsynchronized Data 29.9999 0.96592- ' 0.25881 0.72052- ' 0.17513
Case (i): Fault on L1 (at F1) at 3 miles from bus 1. SinceDisequaltoL1 andD'is greaterthanL1inthefinaliteration,
the algorithm correctly determined that the fault is in L3 (at F3) at 45
Values computed using Jacobian matrix H1 miles from the tap.
Fault Location
In testing the algorithm for a two-terminalline with synchronized
data, the maximum error in computing fault location was less than 0.05
percent. When the data at bus 2 were shifted by ten degrees, the
104
2
1
The fault location technique was then tested using data obtained Values computed using Jacobianmatrix H2
from the EMTp software. A three-phase load flow analysis program
was run first to obtain the prefault voltage conditions. The Eh4TF'
program was then used to obtain the samples of currents and voltages
at each terminalfollowingthe faultoccurrence. The program used a flat
line configuration, with a spacing of 12 feet between conductors, to -14.9196
computethe line parameters per mile, L,C, and R matrices. The series
impedanceand shunt admittancematrices, at 60 Hz, per mile are shown -14.9184
m the Appendix. Although different line impedames were used in the (iii) Fault at F3 75 miles from bus 1,61= 15". & = 20". Rf= 0.2 p.u.
three-terminal cases, D and D' were calculatedas before. An estimation
program was then used to obtain the phasor quantities of the 60/50 Hz Values computed using Jacobian matrix Hi
voltage and currentwaveforms. Thesephasor quantitieswere then used
in the fault location program. Different fault resistances, different
synchronizationerrors and load flow conditionswere considered. Some
examplesthatshowedarelativelyhighpercen~eenorareshownhere.
Two-Terminal Line
Data Synchronized
Values computed using Jacobian matrix H2
Fault h t i m Miles
Exact I camplted I Error%
1 71.2763 -20.5873
2 74.8017 -20.2452
Iteration FaultLocation Error% -&(degrees) Error% 3 74.9805 -0.026 -20.2431 1.215
Number
1 4.957 -11.9412 CONCLUSIONS
2 4.9984 -0.032 -11.9812 -0.157 Using the three-phasevoltage andcurrentphasorsfor fault location
techniques is certainly advantageous when data are available at line
terminals. This concept is not affected by fault resistance. The techxi-
que is general for any type of fault. No fault classificationis required.
When the technique was tested using a steady state fault analysis
program, the errors were less than 0.05 percent. However, when the
techniquewas tested using the E m generated data, the error in most
cases was less than 1percent. The high range of the percentageerrors
was at close in faults and low fault resistance. Many factors contribute
to these errors. These factors are: inaccuracy in the three-phase load
flow, inaccuracyin the line model in the EMTP program and any error
in the estimation of the phasor quantities. Also, even though the
properties that allow idenMicationof the faulted section are based on
the assumption that all line impedances per mile are the same, the
algorithm still performed well when relatively close impedanceswere
Values computed using Jacobianmatrix HI Used.
Research in this area will continue using EMTP generated and
actual recorded data to account for inaccuracy in line parameters and
mutually coupled lines.
I 1
2
I
I
3.0607
3.0557
I
I
I
I
-5.3228
-5.2714
I
I I
I
105
ACKNOWLEDGEMENT For line L2
The authors acknowledge the support of the National Science 0.143 1354
Foundation and Clemson University Electric Power Research Associa- +j 1.309240
tion (CUEPRA). The authors would like to thank Mr. William Quain- 0.09313005 0.1431354
&= ohms/mile
tan= for preparing the data using the Eh4TP. +j 0.6285601 +j 1.309240
0.09311420 0.09313005 0.1431354
REFERENCES +j 0.5444561 +j 0.6385601 +j 1.309240
[l] T. Tagaki, et al., "Development of New Type Fault Locator using
the One-Terminal Voltage and Current Data, "
I Power A
IQ
1982, pp. 2892-2898.
V Vol. PAS-101, NO.8, August
1.032294I 120.1970 -1
0.965000 1-5.6810"
0.982000 1-125.4520
0.985000 I 114.2790
The prefault voltages for the three-terminalline are
P.U.
[
VI = 1.030939 1-119.8747
1.030939I 120.1253 I [
P.U.
1.028642 10.1266
V2= 1.034807 1-119.91153 P.U.
1.034893 1120.1185 1
-1
0.965000 1-5.6810
[
Fault Locations from Digital Impedance Relay Measurements,"
' V3 = 0.982000 1-125.4520 P.U.
, Vol. PWRD-3, No. 1,
January 1988, pp. 121-129. 0.985000 / 114.2790
[6] E. Schweitzer, "Evaluation and Development of Transmission The base values were 138 KV and 100 MVA.
Line Fault Locating Techniques Which ' 2 Sinusoidal Steady-
-
State Information," paper presented to the 9th Annual Western Bglv A. Gig& (SM '81) received the B.S. (with distinction first
Protective Relay Conference, Spokane, WA, October 1982. class honors)and the M.S .degreesin ElectricalEngineering from Assuit
University, Egypt, in 1967 and 1973. respectively. He received the
[7] V. Cook, "Fundamental Aspects of Fault Location Algorithms
used in Distance Protection," m,
Proce Vol. 133, Pt. C, Ph.D. degree in Electrical Engineering from Iowa State University in
1981. From 1967 to 1976, he was an Instructor in the Electrical
No. 6, September 1986, pp. 359-368.
Engineering Department, Assuit University. From 1976 to 1981, he
[8] A. 0. Ibe and B. J. Cory, "Fault Location Algorithm for Multi-
Phase Power Lines," IF.E Procee Vol. 134, Pt. C, No. 1, was teaching and perforrning research in computer relaying of power
January 1987, pp. 43-50 systems at Iowa State University. From 1981 to 1985, he was Assistant
Professor of Electrical Engineeringat North Carolina State University.
[9] I. S.Thorp,A. G. Phadke, and S.H. Horowitz, "Some Applications
of Phasor Measurements to Adaptive Protection,"IEEF.Transac- He is currently Professor of Electrical and Computer Engineering at
on Power Svstems. Vol. PWRS-3, No. 2, May 1988, pp. Clemson University, Clemson. SC. Dr.Girgis has published numerous
79 1-798. technical papers and holds four U.S. patents. He is the recipient of the
[lo] E. B. Makram and Adly A. Girgis, "A Generalized Computer 1989 McQueen Quattlebaum-Faculty outstanding achievementaward
Techniquefor the Development of Three-phaseImpedanceMatrix and the 1990 Edison Electric Institute Power Engineexing Education
for Unbalanced Power System." mctric Power Svstem Reseadl award. His present research interests are real-time computer applica-
a Vol. 15, 1988, pp. 41-50. tions in power system control, instrumentation,and protection. signal-
processing. and Kalman filtering applications. Dr.Girgis is a member
of Phi Kappa Phi, Sigma Xi, and is a registered ProfessionalEngineer.
APPENDIX
The series impedance and shunt admittancematrices at 60 Hz for DavidG. W w a s b o m i n 1962inUnion,SC. In 1985. hereceived
the system used are as follows for the three-terminalline. the B.S. degree (summa cum laude) in Mathematics and Physics from
For lines L1 and L3 Wofford College in Spartanburg, SC. He received the M.S.degree in
Electrical Engineeringin 1987 from Clemson University in Clemson,
Z& =
0.2088354
+j 1.358336
0.0931313 0.2088354
Oh/mile
1I SC. In September of 1991, he completed the Ph.D. programinElectrical
and Computer Engineering at Clemson University. His area of interest
is power system protection and control. Mr. Hart is a Phi Beta Kappa
+ i 0.6447627 + i 1.358336 and a studentmember of IEEE.
I+
0.-0931191 Ob31313 0.2088354
j 0.5606579 +j 0.6447627 + j 1.3583361 (M '82) received the B.S. degree in Electrical
Engineering from GeorgiaTech in Atlanta, GA, in 1982. From 1982
j 4.986234
j 0.9532975
-j 0.4042021
j 5.135725
-j 0.9532975 j 4.986234 1 pmhos/m'le to 1986 he worked in the System Operations Department at Carolina
Power and Light Company. In 1987, he received the B.S. degree in
Applied Physics from Georgia Tech. In 1988, he received the M.S.
degree in Electrical and Computer Engineeringfrom Clemson Univer-
sity in Clemson, SC. He is presently with Ferranti InternationalCon-
trols Corporation, Sugar Land, TX. Mr. Peterson is a registered
ProfessionalEngineer in North Carolina and Georgia.
106
Discussion deduce the faulted section. Could the authors comment on what would be
behaviour of the variables D and D in comparison to L, for an external
Prof. A. T. Johns, (Power and Energy Research Centre, The City fault, particularly for faults just outside any of the three sections, L,, L,
University, Northampton Square, London EClV OHB, England, U.K.): I and L,.
would like to thank the authors for a most interesting paper on an area of
increasing importance and I would like to offer the following comments:
me method proposed neglects the effect of shunt capacitance. My M. S. Sachdev (Power System Research Group, University of Sas-
experience as related in Reference 1 given below, is that such neglect can katchewan, Saskatoon, Sask., Canada.): The paper presents a method to
be justified only if the faulted current is relatively high, as in the case of compute fault locations in two- and three-terminal transmission lines. The
solid or near solid faults. However, fault location is of importance technique uses transmission line models and voltages and currents ob-
operationally under contingency fault conditions, e.g., high impedance served at the terminals of the lines. The discusser offers the following
faults involving vegetation, where in many cases, no significant visual comments on the paper.
evidence exists. Under such conditions the shunt capacitance current The fault location technique described in the paper, uses transmis-
component can attain a much more significant proportion of the total sion line models which consist of series impedances only. It is well
current than is the case for solid faults and in consequence, significant known that long and high-voltage transmission lines have significant
errors can arise if shunt capacitance is neglected. Can the authors’ capacitances and, therefore, it is not appropriate to represent these
algorithm be extended to include the effect of shunt capacitance and have lines by their series impedances only. Substantial errors in the
they any results to offer which show how accurately the method might be estimates of the fault locations can result if the line capacitances are
under high resistance fault conditions? neglected.
In many situations, parallel lines run in close proximity, e.g., a The accuracy of estimates provided by the proposed technique will
double-circuit line. On the face of things, the method proposed will be also depend on the accuracy of the line parameters which change due
affected by signal components induced from parallel circuits. Could the to changes in the environment, e.g., temperature. Even the line
authors give some indication of the effect of parallel circuits on the length is not known with great accuracy.
accuracy of the algorithm?
It would appear that in common with other methods relying on phasor
values, the overall accuracy attainable is critically dependent on being able
to extract accurate phasor measurands from information that contains
transient components. Experience shows that this is not always particu- ADLY A. GIRGIS, DAVID G. HART (Clemson University,
larly easy to achieve with very high accuracy, particularly in situations Clemson S.C.), AND WILLIAM L. PETERSON (Ferranti
where high speed fault clearance provides only a small window of heavily Control, Sugarland Tx). The authors thank the discussers for their
corrupted post fault information. Hardware errors also conspire to reduce interestingcomments and questions.
the level of overall accuracy attainable. Can the authors provide the The discussers raised concerns about the effects of line capacitance.
benefit of any experience they may have in overcoming such problems and Ignoring the line capacitance would affect the calculation of the fault
perhaps also give an indication of the order of overall accuracy they location when a single equation is used to determine the fault
expect to achieve when the cumulative effect of all sources of error is location. However, the error is minimizeddue to the redundancy of
included in the computation? the fault location equations. For example. in the weterminal case,
as illustrated by equation 5; the unknown distance to the fault is
References obtained from the solution of three complex equations or six real
equations. This leads to a least squares estimate which reduces the
[I] A. T. Johns & S. Jamali: “Accurate Fault Location Technique for effect of enws due to ignoring the l i e capacitance. That is because
Power Transmission Lines,” IEE Proceedings, Part C, No. 6, Nov. each of the six equations can be solved separately to obtain the
1990, pp. 395-402. distance to the fault. The result would include an error due to the
line capacitance, if the line is too long. However, when all six
equations are used to obtain the least square estimate, the error is
R. K. Aggarwal and D. Coury, (Power and Energy Systems Research greatly reduced. Similar arguments apply to the two-terminal
Group, School of Electronic and Electrical Engineering, University of unsynchronized case (equation 13) and the synchronized three-
Bath, Claverton Down, Bath, BA2 7AY, England, U.K.): Developments terminal case (equations 23 and 24). In the three-terminal
of techniques for very accurately locating faults on high voltage transmis- unsynchronized case, there are two sets of equations. Each set
sion systems is becoming increasingly important and, in this respect, the represents six equations in the unknowns D (distance to the fault)
authors have presented one such technique which we found quite interest-
ing. Our first comment is regarding the degree of inaccuracy in the and 6 (the synchronization correction angle). This also leads to a
convergence of the iterative method employed for solving for the fault least squares estimate of the fault location. The algorithm in this
distance via the Jacobian matrix. It would be interesting to know from the paper was tested using data obtained from the EMTP program. The
authors whether they attained the same degree of accuracy for all types of lines were modeled using line series impedance and shunt
different system and fault conditions. admittance matrices. However, the total error was less than 1% in
The method developed has been tested using power frequency voltage most of the cases tested. This total error includes the effect of line
and current phasors from very simple steady-state models of the high capacitance. However, in very long transmission lines, the effect of
voltage system and the fault algorithm totally ignores the line shunt the line capacitance may be greater.
capacitance. Whilst the latter assumption could be somewhat justified for Dr. M.S. Sachdev raised an important issue involving the
relatively short lines of the type considered by the authors, however, the inaccuracy in line parameters and line length. These inaccurate
line shunt capacitance can become quite significant as the line length goes parameters will certainly affect any fault location algorithm. These
up and this in turn could cause quite significant errors in the actual inaccuracies, as well as others, may be most severe in algorithms
location of the fault, using the proposed algorithm. Furthermore, although which solve for the fault location using one equation with one
the authors have also presented a limited study using more realistic voltage unknown. Their effect here is minimized by use of redundant
and current waveforms attained from the EMTP, however, there is no equatlons.
mention of the effect of CVTs, CTs, transducer/recording-equipment In response to other remarks by Dr. A.T. Johns: By use of
interfaces and in particular, the filtering technique used for extracting the measurement data at two or more buses, the fault resistance is
fundamental components of voltage and current phasors from the fault eliminated from the fault location algorithm. In principle, the
generated noisy waveforms. Could the authors comment on how the algorithm becomes independent of the fault boundary conditions.
accuracy of their technique would be affected in a more practical situation However, special consideration may be given to very high
and also upto what maximum line lengths could the high accuracy be impedance fault cases, where the line charging current is more
sustained. dominant than the fault current. Parallel lines are not considered in
Their extension of the technique for locating faults on three-terminal this study. As before, the redundancy of equations may provide
lines is also very interesting, in particular the method adopted for identify- some minimization of error due to parallel lines. However, if the
ing the faulted section of the system. It essentially hinges upon deriving lines are strongly coupled and the measurements from the other lines
values for the two variables D and D’ and then devising a logic check in are available, the parallel line would have to be modeled in the
comparison with line length L ,, the length of the first section, in order to algorithm. The voltage and current phasor estimates were obtained
I
107
using a Kalman filtering algorithm. Properly defined, Kalman error, roundoff error, and other sources of error could actually be
filteringis the optimal linear estimator and is designed for estimation modeled in the Kalman filtering algorithm used to obtain the phasor
in noisy transient environments. Thus, the Kalman filter will yield estimates, and should not adversely affect the results. Redundancy
the best possible estimates in the small window available. As the will also limit the effects due to these errors. The line length may
Kalman filter is a state-variable timedomain technique, it is possible become a factor when the lines are so long that line capacitance is a
to include measurement and other errors in the estimation process. major factor. In this case, it may be included in the algorithm. Last,
Thus, the filter may compensate for the time window and other in this study it is assumed that the fault is internal and there is no
sources of measurement error. need to apply the algorithm for an external fault. However, if the
In response to the other questions raised by R.K. Aggarawal: The fault is external, the voltage at the tie bus would be unique using the
algorithm was tested using approximately forty cases of three-phase, voltage and current at any of the terminals as shown by equations 23
line-to-line, double line-to-ground, and single-line-to-groundfaults and 24. Therefore,the value of D and D would be L1.
and performed equally well in these cases. Thus, fault type or
location does not seem to affect the convergence of the solution
using the Jacobian matrix. As mentioned before, measurement Manuscript received August 2 3 , 1991.