You are on page 1of 8

1

Two-Terminal Fault Location on Unsymmetrical


Transmission Lines
Roberto Schulze, Member, IEEE and Peter Schegner, Senior Member, IEEE

Abstract—The outage of a line due to a fault can be expensive, Today most of the substations or process control centers do
hence the fault has to be cleared as fast as possible. Digital protec- not fulfil all of these conditions. Especially the influence
tion relays consist of fault locators based on several methods. Due of the third point (cf. above) on the fault location is not
to the availability of highly accurate GPS based time references
the investigation of two-terminal methods combined with signal investigated yet. In [2] a traveling wave technique with a
analysis tools makes sense in the first place. practical application has been proposed, whereas a sampling
The aim of this paper is to develop a two-terminal fault location frequency of 5 MHz and GPS based time servers (time delay
algorithm based on two kinds of modeling the samples of a between two devices ≤ 100 ns) are used. A disadvantage of
fault record. Derived from the coupled π-sections line model this approach is, that only the first impulse is detectable (due
an estimation equation is developed. This equation contains time
derivatives of the measured currents. To overcome this problem to the damping of the line) and hence can be used for locating
a frequency and a time domain approach are presented. The the fault. An other approach has been published in [3], which
derivation of the voltage drops from both stations to the fault is based on the telegraph equation in the case of a symmetrical
location (either in the frequency or in the time domain) allows line. This method yields very exact fault locations, but it does
the calculation of the fault location via the least squares method. not consider line unbalances and line losses in the case of
A simulation model serves for testing the algorithm. Different
scenarios e. g. several fault locations, fault resistances and ”long” transmission lines (cf. [3]). An easier two-terminal
fault types allow a detailed investigation. Furthermore identified method containing a single impedance has been investigated
i. e. uncertain line parameters are used to calculate the fault in [4]. This fault location algorithm uses time-varying phasors.
location as well as the influence of inexact time synchronisation Due to the kind of calculating time-varying phasors the fault
is investigated. location contains errors. Furthermore this approach assumes
Index Terms—signal modeling, prony’s method, time-varying symmetrical transmission lines.
phasors, transmission line model, two-terminal fault location. Since the approach of this paper is based on identified pa-
rameters of a line model consisting of coupled unsymmetrical
I. I NTRODUCTION π-sections according to [5], [6], at first the considered line
model is presented. In the following section III two approaches
ASICALLY fault location techniques are divided into
B one- and two-terminal methods. It is well known, that
one-ended impedance-based techniques are sensitive to the
for modeling the fault segment of a fault record are introduced.
Based on the line model a fault location algorithm in the
frequency domain and a further one in the time domain are
unknown fault resistance and hence the calculated fault lo-
developed in section IV. Finally test results and a sensitivity
cation is inaccurate. Due to the continuous improvement of
analysis are presented in section V.
the time synchronisation (e. g. via longwave time signal
DCF77 in Germany or via GPS time synchronisation) of
digital protection relays or digital fault recorders two-terminal II. T RANSMISSION L INE M ODEL
approaches become more and more importance. As described The goal of this paper is the development of a fault
in [1] numerous methods exist for the two-terminal approach. location algorithm which allows the inclusion of modeled
Basically they can be divided into line unbalances as has been done in [5], [6]. Hence the
• Traveling wave methods
• Methods using synchronized phasors . Δv a
A iAa iλa iBa B
The first approach theoretically guarantees very exact fault
Z aa
locations if all of the following conditions hold: v Aa Y ac , Y ab Z ab iλb+Z ac iλc Y ac , Y ab v Ba
Δv b
1) Accurately time synchronised fault recorders iAb iλb iBb
2) A sufficient high sampling frequency
3) An accurate transformation of the high dynamic primary v Ab Y bc Z bb Z i +Z i Y bc v Bb
ab λa bc λc
Δv c
voltage and current signals via the voltage and current iAc iλc iBc
transformers into secondary values Z cc
v Ac v Bc
4) The propagation speed of the traveling wave on a specific Z ac iλa+Z bc iλb
Y aa Y bb , Y cc Y aa Y bb , Y cc
line must be known
R. Schulze and P. Schegner are with the Institute of Electrical Power Sys-
tems and High Voltage Engineering, TU Dresden, 01062 Dresden, Germany,
e-mail: roberto.schulze@tu-dresden.de Fig. 1. Model of an unsymmetrical transmission line in natural components

978-1-4244-6551-4/10/$26.00 ©2010 IEEE


2

model displayed in Fig. 1 need to be introduced. As can (Rkl + j ω Lkl ) I λl + Lkl İ λl + (2)
  
be seen, this model allows different resistive and inductive =:Z kl
couplings between the phases. Currents and voltages have
(Rkm + j ω Lkm ) I λm + Lkm İ λm ,
been introduced as complex signals, which is convenient for   
the frequency domain approach in section III-A. The model =:Z km
in Fig. 1 can as well be modified in order to describe the
system behavior in time domain. Since the currents through whereas
the admittances are small compared to iλ (especially in the
k ∈ {a, b, c}
fault case) and due to the impossibility of calculating them in
the case of a faulty line, they are omitted. l ∈ {a, b, c} \ {k}
It is assumed that all impedances are known, e. g. via a m = {a, b, c} \ {k, l} .
parameter identification as result of an analysis of fault records
The advantage of this approach is the easy calculation of the
with time synchronously recorded data [5].
time derivative of the slow moving phasor. This can be done
by using the ordinary difference quotient. A further advantage
III. M ODELING S AMPLED DATA IN T IME AND is the robustness to the influence of noise. But since this
F REQUENCY D OMAIN procedure contains a low pass filtering, the fault signals are
Obviously this method requires the fault segment of a fault not mapped correctly into the frequency domain. Furthermore
record. In this paper it is assumed that the range of the fault transients occur after filtering since the number of samples
within the record is exactly known. The challenge now is to inside the fault segment is relatively small. These transients
calculate the voltage drops from both stations A and B to have a significant influence on the estimated fault location as
the fault location. Therefore it is necessary to calculate time can be seen in section V.
derivatives of the measured currents. In order to perform this
it is convenient either to transform the measurements into B. Time Domain Approach – Prony’s Method
the frequency domain or to identify the parameters of an
appropriate signal model. To overcome the problems remarked in the previous section
the samples x[n] within the fault segment shall be modeled via
an appropriate signal model by applying the prony method [9].
A. Frequency Domain Approach – Time Varying Phasors Since a linear fault impedance as well as a linear line model
Transforming the measurements into the frequency domain are assumed, the following signal model of order K based on
means in this case to remove the fundamental oscillation ω the sampling time T is valid.
according to the demodulation technique [7], [8]. The result
x̃[n] = x− + x∼ [n] + Δx[n]
is the so-called time-varying phasor X(t) which is related to
K

the time signal x(t) as follows.
= x− + hk zkn + Δx[n] , (3)
x(t) = Re {x(t)} = Re {X(t) exp(j ω t)} (1) k=1

The absolute value |X(t)| is the envelope of the time signal with hk = xk ej ϕk and zk = e(αk +j βk )T . As can be seen,
as can be seen in Fig. 2. The approach (1) applied in order to the model (3) consists of a constant x− and a time-varying
direct and alternating part x∼ [n] as well as measurement
4 noise Δx[n]. An extended method based on the original prony
concept [10] yields the optimal values for the parameters
x(t)
(xk , ϕk , αk , βk ). The model (3) corresponds directly to the
±|X(t)|
x(t), ±|X(t)| →

2 real-valued, symbolically existent signal model


M

0 x̂(t) = x0 + xm exp (t/τm ) cos(ωm t + ϕm ) , (4)
m=1
−2 whose time derivative is easy to determine automatically
within an algorithm. The model order is typically M = 3
−4 (corresponds to K = 5), which allows two different damped
0 0.05 0.1 0.15 0.2 0.25 frequency components and one single damping. In Fig. 8 the
t/s → high accuracy in modeling the voltage and current samples
will be demonstrated.
Fig. 2. Typical signal of a fault record and its envelope

calculate the voltage drop along the line (cf. Fig. 1) results in IV. FAULT L OCATION A LGORITHM
the following equation The next step in developing the fault location algorithm is
to modify the line model Fig. 1. The objective is to formulate
ΔV k = (Rkk + j ω Lkk ) I λk + Lkk İ λk + the voltage drop from station A and B to the fault location
  
=:Z kk lAf resp. lBf . As can be seen in Fig. 3 an arbitrary fault is
3

   
lAf (Rab iAb lAf (Rac iAc lBf (Rab iBb lBf (Rac iBc
 d  d  d  d
 +Lab dt iAb ) +Lac dt iAc ) +Lab dt iBb ) +Lac dt iBc )
iAa lAf Raa lAf Laa 
lBf Raa lBf Laa iBa

vAa vBa
   
lAf (Rab iAa lAf (Rbc iAc RF a lBf (Rab iBa lBf (Rbc iBc
 d  d  d  d
 +Lab dt iAa ) +Lbc dt iAc ) +Lab dt iBa ) +Lbc dt iBc )
iAb lAf Rbb lAf Lbb 
lBf Rbb lBf Lbb iBb

vAb vBb
   
lAf (Rac iAa lAf (Rbc iAb RF b lBf (Rac iBa lBf (Rbc iBb
 +Lac dt
d
iAa ) +Lbc dt
d
iAb ) +Lac dt
d
iBa ) +Lbc dt
d
iBb )
iAc lAf Rcc lAf Lcc 
lBf Rcc lBf Lcc iBc

vAc vBc
RF c

Station A RF
Station B

lAf fault location lBf

Fig. 3. Line and fault model used for deriving the fault location algorithm

allowed. Based on this model the following equations can be The cumulation of N measurement sets results in eq. (10)
developed
  Ξ = Ψ lAf (10)
d ⎛ ⎞ ⎛ ⎞
v A − v B =l R iB + L iB + (5) ξ1 ψ1
dt ⎜ .. ⎟ ⎜ .. ⎟
  ⎝ . ⎠ = ⎝ . ⎠ lAf ,
d
lAf R (iA − iB ) + L (iA − iB ) , ξN ψN
dt
which yields the least squares solution
whereas l is the known line length and lAf is the fault distance
from station A. v A and v B are the measurable phase to ground ΨT Ξ
voltage vectors at both stations: lAf = . (11)
ΨT Ψ

T
T In the case of using the signal model (4) eqns. (9)-(11) can be
v A = vAa vAb vAc v B = vBa vBb vBc (6)
applied directly. While using the time-varying phasor approach
and iA and iB the measurable phase current vectors, resp. from section III-A eqns. (9) and (11) need to be modified as
follows.

T
T If complex-valued signals describe the system behavior
iA = iAa iAb iAc iB = iBa iBb iBc (7)
the regularities shown in eq. (2) have to be applied to the
In terms of a short spelling the explicit denotation of time estimation eq. (9) which yields
dependence has been omitted in eqns. (6) and (7). The per-
unit-length line parameter matrices are ξ k = ψ k lAf (12)
⎛  ⎞ ⎛  ⎞
d
Raa Rab  
Rac Laa Lab Lac v A,k − v B,k − l R + j ω L iB,k + L iB,k =


 dt 
R =⎝ Rab 
Rbb  ⎠
Rbc L =⎝ Lab Lbb Lbc ⎠ . (8) 

d

 
Rac Rbc Rcc 
Lac Lbc Lcc R + j ω L iA,k − iB,k + L iA,k − iB,k lAf .
dt
The unknown fault distance lAf in eq. (5) can be calculated by The vectors v A,k , v B,k , iA,k and iB,k of course contain the
applying the ordinary least squares method to the estimation time-varying phasors, e. g.
equation at time tk = k T
T
iA,k = I Aa,k I Ab,k I Ac,k (13)
ξ k = ψ k lAf (9)
  Analogous to eq. (10) the fault location in the complex case
d
v A,k − v B,k − l R iB,k + L iB,k = is obtained with the least squares solution
dt
 
  d ΨH Ξ
R (iA,k − iB,k ) + L (iA,k − iB,k ) lAf . lAf = , (14)
dt ΨH Ψ
4

where ΨH denotes the conjugate transpose. After producing a phase ’a’ to ground fault at xF = 12.5 km
The terms expressing the time derivatives in eqns. (9),(12) (this is equivalent to 5 % of l) the currents displayed in Fig. 5
are exactly written as follows: occur in phase ’a’. It can be seen, that the fault is mainly
 supplied by grid 1.
d d 
iA,k = iA (t) (15)
dt dt t=k T
10
iAa
V. S IMULATION R ESULTS AND S ENSITIVITY OF THE
5 iBa
A LGORITHM

i(t)/kA →
After deriving an algorithm for calculating the fault location
via two kinds of signal preprocessing it shall be tested on 0
a certain simulation model. The model Fig. 4 has been
implemented in Matlab/Simulink using the SimPowerSystems- −5
Toolbox [11] in order to produce fault records with a sample
time of T = 0.5 ms. At both stations A and B the current and −10
voltage signals are ”recorded” in a realistic way by modifying 0 0.05 0.1 0.15 0.2 0.25
them with a 16-bit ADC model. This model contains the t/s →
impact of the quantization as well as an error induced by
the inaccuracy of the three low-order bits, which is normally Fig. 5. Simulated fault currents of phase ’a’ at stations A and B
distributed. As can be seen in Fig. 4 the faulty line connects the
As described in section IV the first step of the algorithm
two grids 1 and 2 in the 380 kV level. During normal operation
is either to transform the records into frequency domain (cf.
section III-A) or to determine the parameters of a signal
Station A Station B model (cf. section III-B). The result of the time-varying phasor
v A , iA v B , iB
approach is similar to the plot of ±|X(t)| in Fig. 2. As was
RAf , RBf , mentioned in section III-A transients occur in the run of the
LAf , C Af LBf , C Bf time-varying phasor.
grid 1 grid 2 In order to minimize their influence, a window is stepped
RF a,b,c through the fault segment (cf. Fig. 6). A reliable fault location
can be obtained by determining the direct part of the fault
xF = lAf locations x̂F (Δtf i ) calculated by analyzing different window
0 l
positions. The time difference Δtf i in Fig. 6 expresses the
Fig. 4. Simulation model for testing the fault location algorithm distance of the first sample of the window to the fault inception
time. In general the tilde  ˜ denotes nominal values and the
mode a load flow of PA = 500 MW and QA ≈ 164 Mvar (this ˆ
hat  characterizes estimated values in the following.
corresponds to cos ϕ = 0.95) from station A to B exists. Both
grids have a short circuit level of 5 Gvar and a X/R-ratio of
10. The nominal per-unit-length parameters of the line are
⎛ ⎞ 240
x̂F (Δtf i )/km →

58.221 47.104 47.063



R̃ = ⎝47.104 58.224 47.104⎠ mΩ/km (16)
47.063 47.104 58.221 235
⎛ ⎞
1.6876 0.8652 0.7267

L̃ = ⎝0.8652 1.6876 0.8652⎠ mH/km (17) x̂F,DT
0.7267 0.8652 1.6876 x̂F,P M
⎛ ⎞ 230 x̃F
11.305 −2.446 −0.820

C̃ = ⎝−2.446 11.775 −2.446⎠ nF/km . (18) 0 10 20 30 40 50
−0.820 −2.446 11.305 Δtf i /ms →
These parameters arise from a line parameter computation Fig. 6. Calculated fault location x̂F via DT and PM of a fault phase ’a’ to
based on certain conductor characteristics as well as a given ’b’,nominal fault location x̃F = 237.5 km, RF a = RF b = 100 Ω, nominal
 
tower geometry [11]. The line is assumed to be untransposed, line parameters R̃ , L̃
i. e. unsymmetrical. The tower geometry is shown in the
Appendix (Fig. 9 and table VII). While using the signal model approach it is important to
The two parts of the faulty line (each modeled as three discard a few samples direct after the fault inception, since
coupled π-sections, cf. Fig. 1) consist of the parameters switching transients lead to inexact signal model parameters as
 
RAf = lAf R̃ , RBf = lBf R̃ etc. with a total line length of can be seen e. g. in the case of modeling vAi (t), i ∈ {a, b, c}
in Fig. 7. Furthermore it is recognisable that the model order
l = 250 km . of K = 3 is too low. E. g. the model for vAa at least needs
5

60 of the phase i to ground voltage at station A, I AR is the


Δ vAa residual current phasor at A, I Ai is the current phasor of the
40 Δ vAb faulty phase and kE = (Z 0 − Z 1 )/(3 Z 1 ). It is obvious that
Δ vA (t)/kV →

Δ vAc Z 0 and Z 1 are the zero-sequence and positive-sequence line


20 impedance respectively. It is well known, that the result of this
method is influenced by the fault resistance and the load flow.
0 In this example only a single phase to ground fault shall be
investigated.
−20
In order to apply this simple method the unsymmetrical
−40 line introduced in eqns. (16)-(18) has to be investigated in
0 20 40 60 80 symmetrical components
t/ms →   

Z 120 = T −1
120 R̃ + j ω L̃ T 120 , (20)
Fig. 7. Absolute error Δ vAi (t), i ∈ {a, b, c} of the signal models by
using model orders of Ki = 3 and analysis windows with 156 samples; each whereas T 120 is the transformation matrix. Since the off-
starting at the fault inception time diagonal entries in eqns. (16)-(18) are not equal, Z 120 is not
a diagonal matrix, i. e. the component systems are coupled. In
order to calculate the fault location with eq. (19) it is assumed,
a further harmonic, i. e. K = 5. By using a model order that the off-diagonal entries in (20) are zero, hence
of K = 5 and discarding 10 samples direct after the fault 

inception time the measurements are modeled exactly (up to Z 1 Z 2 Z 0 := diag(Z 120 ) (21)
the noise induced by the implemented ADC-model) which is
is defined.
displayed in Fig. 8.
To apply eq. (19) the phasors of voltages and currents have
to be calculated. This is done by using the Phadke-Ibrahim
40 algorithm [12], which can be used in distance protection
Δ vAa algorithms.
20 Δ vAb Table I shows the relative errors (22) of phase ’a’ to ground
Δ vA (t)/V →

Δ vAc faults while varying fault resistance RF a and fault location x̃F .
0 x̂F − x̃F
δ x̂F /% = 100 . (22)
l
−20 The left column of table I contains the fault resistances RF a .
The fault distance x̃F /% means
−40 x̃F
10 20 30 40 50 60 70 80 ,
x̃F /% = 100 (23)
l
t/ms →
beginning at station A. As can be seen, the fault location
Fig. 8. Absolute error Δ vAi (t), i ∈ {a, b, c} of the signal models by
using model orders of Ki = 5 and discarding 10 samples direct after the TABLE I
fault inception time R ELATIVE ERRORS δ x̂F A /% AND δ x̂F B /% – PHASE ’a’ TO GROUND
FAULT, USING ZERO - SEQUENCE AND POSITIVE - SEQUENCE LINE
IMPEDANCES
After describing these two approaches the results using a
classical method are presented in the following. After that x̃F /% 5 30 50 70 95
simulation results of different experiments are discussed. δ x̂FA/FB A B A B A B A B A B
RF a /Ω
3.8 0.1 -2.8 0.3 0.0 0.8 0.0 1.6 -0.2 2.5 -0.4
A. Results of a Classical Fault Location Method 10.0 0.2 / 0.1 -3.8 0.2 -2.2 0.0 -1.6 -2.4 -1.2
In this section some results based on a conventional fault 50.0 1.4 / -0.4 / -2.4 / -5.9 / / -7.2
location method are presented as reference for the new method. 100.0 3.7 / 0.6 / -2.7 / -7.7 / / /
Up to now fault locators are based on the line impedances
in symmetrical components, i. e. a not perfectly transposed
transmission line is assumed to be symmetrical. According to x̂F has been calculated from both terminals A and B. Only
[1] in the case of a phase i, (i ∈ {a, b, c}) to ground fault for low resistance faults (RF a = 3.8 Ω) acceptable results
eq. (19) is used. are obtained. For the relative errors then δ x̂F ∈ [−5%, 5%]
  holds. Entries denoted with ”/” contain relative errors δ x̂F ∈
/
V
Im I +kAi I [−10%, 10%]. These results are unaceptable. Furthermore it
Ai E AR
x̂F A = (19) can be seen, that from station A the fault locations can be
Im {Z 1 }
calculated more exactly. This is due to the fact, that the fault
x̂F A is the calculated distance from station A to the fault is mainly supplied by grid 1, cf. Fig. 4 (a load flow from A
location by using the measurements from A. V Ai is the phasor to B is existing) and Fig. 5.
6

Due to the inclusion of the measurements from both ter- been identified by applying the procedure published in [5].
minals and the complete resistance and inductance matrices it This identification algorithm is based on the line model in
is expectable, that the results with the proposed method are Fig. 1 and synchronously sampled voltage and current signals
more exact. In the following sections simulation results are at the terminating stations of a transmission line. In order
presented. to identify the line parameters it is necessary to analyse the
transient part of the fault segment within a fault record. So
B. Results Based on Nominal Line Parameters called time-varying phasors [7] are used to model the fault in
At first the algorithm is tested by providing the nominal the frequency domain. By using the LS-method and assuming
line parameters R̃ and L̃ (cf. eqns. (16,17)). For different fault that the self impedances are equal the matrices (25) and (26)
types, fault locations and fault resistances simulations by using have been identified.
⎛ ⎞
the demodulation technique (DT) and the prony method (PM) 56.621 46.289 47.060

have been done. The tables II-III contain the relative errors R̂ = ⎝46.289 56.621 45.089⎠ mΩ/km (25)
δ x̂F /‰ of the calculated fault locations x̂F related to the line 47.060 45.089 56.621
⎛ ⎞
length l according to 1.6809 0.8631 0.7189

x̂F − x̃F L̂ = ⎝0.8631 1.6809 0.8577⎠ mH/km . (26)
δ x̂F /‰ = 1000 . (24) 0.7189 0.8577 1.6809
l
The left column of each table contains the fault resistances For that identification a phase ’a’ to ground fault has been
RF i . In the case of a three phase fault i ∈ {a, b, c}. Addition- used, which led to the errors in (27), (28). As can be seen,
ally it is remarked that three phase to ground faults consists the couplings between the phases not drained by the short
of an additional fault resistance RF = 40 Ω (cf. Fig. 3). The circuit current have been identified worse compared to those
fault distance x̃F /% is defined in (23) and is related to station parameters modeling the coupling to the faulty phase ’a’.
A. ⎛ ⎞
 2.7 1.7 0.0
δ R̂
TABLE II = − ⎝1.7 2.7 4.2⎠ (27)
R ELATIVE ERROR δ x̂F /‰ – PHASE ’a’ TO GROUND FAULT, USING %
  0.0 4.2 2.7
NOMINAL LINE PARAMETERS R̃ , L̃ ⎛ ⎞
 0.4 0.2 1.0
δ L̂
x̃F /% 5 30 50 70 95 = − ⎝0.2 0.4 0.9⎠ (28)
%
RF a /Ω DT PM DT PM DT PM DT PM DT PM 1.0 0.9 0.4
3.8 1.6 0.0 -0.1 0.0 -0.7 0.0 -1.7 0.0 -2.0 0.0
After performing the algorithm described in the previous
10.0 1.7 0.0 -0.1 0.0 -1.6 0.0 -2.4 0.0 -3.3 0.0
sections the results in tables IV-V have been obtained. Again
50.0 1.5 0.0 -0.2 0.0 -1.3 0.0 -2.3 0.0 -3.1 0.0
the relative errors are quite small (< 1%).
100.0 1.4 0.0 -0.3 0.0 -0.7 0.0 -2.3 0.0 -2.9 0.0
TABLE IV
R ELATIVE ERROR δ x̂F /‰ – PHASE ’a’ TO GROUND FAULT, USING
 
ESTIMATED LINE PARAMETERS R̂ , L̂
TABLE III
R ELATIVE ERROR δ x̂F /‰ – PHASE ’abc’ TO GROUND FAULT, USING
  x̃F /% 5 30 50 70 95
NOMINAL LINE PARAMETERS R̃ , L̃
RF a /Ω DT PM DT PM DT PM DT PM DT PM
x̃F /% 5 30 50 70 95 3.8 1.0 -0.6 -0.6 -0.6 -1.1 -0.4 -2.0 -0.2 -1.7 0.1
RF i /Ω DT PM DT PM DT PM DT PM DT PM 10.0 1.0 -0.6 -0.7 -0.6 -2.1 -0.4 -2.7 -0.2 -3.2 0.1
3.8 2.4 0.0 -1.0 0.0 -4.1 0.0 -6.7 0.0 -8.8 0.0 50.0 0.8 -0.7 -0.8 -0.6 -1.9 -0.5 -2.7 -0.4 -3.1 0.0
10.0 1.1 0.0 -1.5 0.0 -4.5 0.0 -6.4 0.0 -8.0 0.0 100.0 0.3 -0.7 -1.0 -0.7 -1.3 -0.6 -2.7 -0.4 -3.1 -0.2
50.0 -0.7 0.1 -8.8 0.0 -4.1 0.0 -6.7 0.0 -8.4 0.0
100.0 -2.2 0.1 -0.3 0.1 -4.5 0.0 -7.6 0.0 -9.7 -0.1
TABLE V
R ELATIVE ERROR δ x̂F /‰ – PHASE ’abc’ TO GROUND FAULT, USING
 
While studying the results in tables II-III it strikes that the ESTIMATED LINE PARAMETERS R̂ , L̂
results obtained via the prony method correspond nearly to the
reference values x̃F despite the fact that the samples have been x̃F /% 5 30 50 70 95
modified by the ADC model. In general the results are very RF i /Ω DT PM DT PM DT PM DT PM DT PM
good (the entries in the tables are δ x̂F /‰!) By examining the 3.8 2.3 -0.1 -0.8 0.2 -3.9 0.3 -6.3 0.4 -8.3 0.5
tables it is not possible to derive a rule describing the accuracy 10.0 1.1 0.0 -1.4 0.2 -4.2 0.3 -6.0 0.4 -7.6 0.5
of x̂F in dependence of the fault resistance. 50.0 -0.7 0.1 -8.1 0.3 -3.7 0.4 -6.2 0.5 -7.7 0.6
100.0 -2.1 0.2 0.4 0.4 -4.0 0.5 -6.9 0.7 -8.9 0.7
C. Results Based on Identified Line Parameters
In the second step estimated line parameters shall be used By analyzing the results it is obvious, that the fault location
in order to analyse the robustness of the algorithm to uncer- algorithm based on the prony method yield more accurate fault
tainties of the line parameters. The parameters (16), (17) have locations as well in the case of uncertain parameters where
7

no relative error is > 1%. In general it can be summarized, δ x̂F < 1%. A further simulation of asynchronous recorded
that the fault location of single phase to ground faults can be fault data shows aceptable results (relative errors δ x̂F < 10%)
determined more accurate than others. for low resistance faults and time delays ≤ 1 ms.
Due to the increasing number of installed GPS based time
D. Influence of Asynchronously Recorded Data references within substations this method is applicable in the
future. Since the fault records are analysed offline the higher
The proposed algorithm is based on time synchronously
computational burden of the more accurate prony method is
recorded samples. Since every GPS based time reference has
no exclusion criterion of this method.
an error, the algorithm shall be tested by providing artificial
generated asynchronous test measurements. This was done by
shifting the measurement sets of one station by n = 1, 2 A PPENDIX
samples. The time delay can be calculated with T OWER G EOMETRY AND C ONDUCTOR C HARACTERISTICS

ΔtAB = tA − tB = n T . (29)
Using a sample time of T = 0.5 ms results in the time delays
ΔtAB1 = 0.5 ms and ΔtAB2 = 1.0 ms. As was concluded in d
section V-C the approach based on the prony method leads to
more accurate fault locations. For that reason the investigations
are referred to that method.
The results of this experiment are entered in table VI. DB bundle of
In order to indicate solely the influence of time delays the subconductors
nominal parameters R̃, L̃ were used. Since the results do not
h DC
TABLE VI T
I NFLUENCE OF A TIME DELAY ΔtAB DUE TO ASYNCHRONOUSLY
RECORDED SAMPLES ON THE FAULT LOCATION x̂F – RELATIVE ERROR
 
δ x̂F /%, x̃F = 125 km, USING NOMINAL LINE PARAMETERS R̃ , L̃

fault type ’a’-GND ’ab’ ’ab’-GND ’abc’-GND


ΔtAB /ms 0.5 1.0 0.5 1.0 0.5 1.0 0.5 1.0
RF i /Ω
ρE
3.8 0.6 1.2 1.6 3.5 1.8 3.7 0.8 1.8
10.0 0.8 1.7 3.1 6.5 3.0 6.2 2.3 4.7 Fig. 9. Tower geometry and conductor characteristics corresponding to the
50.0 2.4 5.1 12.8 26.0 9.3 18.7 11.5 23.4 line parameters (16)-(18) in section V
100.0 4.7 9.7 24.7 49.0 14.3 28.5 22.8 45.7

TABLE VII
depend significantly on the fault location, only x̃F = 125 km G EOMETRY DATA OF THE T OWER AND C ONDUCTOR D IMENSIONS
was used to show the influence. While regarding table VI it
Quantity Symbol Value
strikes, that the relative errors obtained with ΔtAB1 in the
height h/m 20.7
case of low resistance faults (≤ 10 Ω) are ≤ 5 %. In the case
bundle distance d/m 12.8
of ΔtAB2 the relative errors are < 10 %. For high resistance
bundle diameter DB /cm 64.66
faults the relative errors only in the case of single phase faults
conductor diameter DC /cm 3.56
are < 10 % for both investigated time delays.
ratio of hollow cylinder T /DC 0.375
conductor DC resistance RC /Ω km−1 0.043
VI. C ONCLUSION ground resistivity ρE /Ωm 100
A new approach for calculating fault locations based on
synchronized samples has been presented. After introducing
the assumed line model two kinds of modeling the samples
within the fault segment are described. The computational ACKNOWLEDGMENT
burden of the frequency domain approach is relatively low.
The authors would like to thank the Deutsche Forschungs-
The prony method which is the key part of the time domain
gemeinschaft (German Research Foundation) for supporting
approach is a more complex method and hence the com-
this research.
putation is more sophisticated. A derivation of the voltage
drops from the two terminals to the fault location allows
the determination of the fault location by applying the least R EFERENCES
squares method. The obtained simulation results indicate a [1] IEEE Guide for Determining Fault Location on AC Transmission and
very high accuracy of this method especially by using the Distribution Lines, IEEE PES Std. C37.114™ -2004, 2005.
[2] S. Zimath, C. Dutra, C. Seibel, M. A. Ramos, and J. S. Filho, “Traveling
time domain approach. A simulation with identified, uncertain wave fault locators – application and results,” white Paper REASON
line parameters yields as well good results with relative errors Tecnologia S.A.
8

[3] M. Kezunović, J. Mrkić, and B. Peruničić, “An accurate fault location


algorithm using synchronized sampling,” Electric Power Systems Re-
search, vol. 29, no. 3, pp. 161–169, 1994.
[4] R. Živanović and C. Cairns, “An application of the pmu technology
in fault location: Preliminary study,” in Southern African Universities
Power Engineering Conference SAUPEC’96, Johannesburg, Republic of
South Africa, January 1996.
[5] R. Schulze and P. Schegner, “Parameter identification of unsymmetrical
transmission lines,” in IEEE PES PowerTech. Bucharest, Romania:
IEEE PES PowerTech, 28 June - 2 July 2009.
[6] R. Schulze, P. Schegner, and P. Stachel, “Parameter identification of
unsymmetrical transmission lines using accurately re-synchronised fault
records,” in IEEE PES. Calgary, Canada: IEEE PES, 26-30 July 2009.
[7] V. Venkatasubramanian, “Dynamic analysis of the general large power
system using time-varying phasors,” International Journal on Electric
Power and Energy Systems, pp. 365–376, 1994.
[8] S. Haykin, Communication Systems. John Wiley & Sons, Inc., 2001.
[9] S. L. J. Marple, Digital spectral analysis with applications, ser. Prentice-
Hall Signal Processing Series, A. V. Oppenheim, Ed. Englewood Cliffs,
NJ, Prentice-Hall, Inc., 1987.
[10] R. Živanović, P. Schegner, O. Seifert, and G. Pilz, “Identification of the
resonant-grounded system parameters by evaluating fault measurement
records,” IEEE Transactions on Power Delivery, vol. 19, no. 3, pp. 1085–
1090, July 2004.
[11] Matlab – SimPowerSystems™4 Reference, Hydro-Qubec, Transnergie
Technologies.
[12] A. G. Phadke, T. Hlibka, and M. Ibrahim, “A digital computer system for
ehv substations: Analysis and field tests,” IEEE Transactions on Power
Apparatus and Systems, vol. 95, no. 1, pp. 291–301, Jan. 1976.

Peter Schegner received his Dipl.-Ing. degree in


electrical power engineering from the Darmstadt
University of Technology, Germany, in 1982 and his
Dr.-Ing. degree from Saarland University, Germany
in 1989.
Currently, he is Professor at the Institute of Electri-
cal Power Systems and High Voltage Engineering,
TU Dresden, Germany.

Roberto Schulze received his Dipl.-Ing. degree in


automation and control from the TU Dresden, Ger-
many in 2005. From 2005 until 2007 he was working
in the field of electrical drives at the Technische
Universität Bergakademie Freiberg. Since 2008 he
has been a Ph.D. candidate at TU Dresden. His
research interests include analysis of fault records
and signal theoretic basics.

You might also like