Electrical Power and Energy Systems 63 (2014) 779–786

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Fault location on multi-terminal DC systems using synchronized current

measurements
Sadegh Azizi ⇑, Saeed Afsharnia, Majid Sanaye-Pasand
The School of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, Iran

a r t i c l e i n f o a b s t r a c t

Article history: Multi-terminal dc transmission systems are suitable substitutes for their ac counterparts, once the asso-
Received 28 September 2013 ciated costs are justified. This paper proposes a new traveling wave-based fault location method for
Received in revised form 15 June 2014 multi-terminal dc (MTDC) systems. In doing so, the fault section and its included lines are specified using
Accepted 18 June 2014
a straightforward algorithm. Then, an evaluation index is developed to determine the faulted line
Available online 17 July 2014
amongst all included ones in the fault section. Finally, a system of linear equations taking into account
the fault inception time, the fault location and the first arrival times at different detector locations are
Keywords:
constructed and solved. The method merely needs the first current surge arrival times, thus eliminating
Fault location
Multi-terminal dc (MTDC) system
the practical difficulties in relation with identifying other traveling wave features. Besides, the fault is
Traveling wave easily located despite the complexity of the network topology, i.e., the number of its meshes and radial
Wavelet transform (WT) lines. To examine the effectiveness of the proposed method, a large number of simulation studies are con-
ducted on a complicated MTDC system. Obtained results verify the high accuracy, noise immunity and
fault impedance robustness of the proposed method.
Ó 2014 Elsevier Ltd. All rights reserved.

Introduction due to a fault on a transmission line propagate all over the network
through the conductors. Accordingly, it is possible to identity the
To date, various aspects and potentials of multi-terminal dc fault inception time and its location by recording and analyzing
(MTDC) systems in a variety of applications have been addressed the high frequency traveling waves and their features such as mag-
in many research works [1–6]. Underground cables may be pre- nitude, polarity and time intervals between the arriving waves [9].
ferred to overhead lines especially in urban areas of large cities Despite different traveling wave features, the fault location can
due to environmental and safety reasons [1]. Nonetheless, ac oper- be pinpointed just by using the traveling wave arrival times at
ation of underground cables is often thought of as a major chal- some dispersed locations. The wavelet transform (WT) has been
lenge because of the associated high capacitance. Besides, since widely utilized so far as a useful tool for detecting the abrupt
the submarine ac cables are not able to exchange a significant changes in the input signals [8,9]. Various types of traveling
amount of active power, high voltage direct current cables are usu- wave-based methods may be deployed for fault location on a
ally preferred for wind farm applications [6]. Meeting the expected two-terminal line. A series of methods uses the subsequent arriv-
flexibility and reliability requirements, MTDC transmission is a ing surges at one terminal [10,11]. Some other methods use the
suitable choice for large offshore wind farms [7]. Cost, reliability, time difference between the first arriving waves at both terminals
protection and a series of various technical criteria should be taken [12,13]. Furthermore, it is possible to makes use of time delays
into consideration to choose an appropriate topology for a MTDC between modal components identified at one terminal to deter-
system [7]. mine the fault location [14]. Algorithms using one end measure-
Although the important task of separating the faulted zone is ments do not require a communication link. Hence, they may be
assigned to the protection system, fault location is also required adopted if the difficulties associated to reflected waves discrimina-
to fix the problem and restore the normal operating condition of tion; especially in the case of close-in faults can be well handled.
the system [8]. The initiated voltage and current traveling waves The traveling wave-based fault location in multi-terminal sys-
tems is a difficult task because of the variations of shortest paths
⇑ Corresponding author. Tel.: +98 912 4070714; fax: +98 21 88778690. to different detectors as a result of changes in the fault location.
E-mail addresses: [email protected] (S. Azizi), [email protected] (S. Afsharnia), Authors in [15–17] have proposed a piecewise linear optimization
[email protected] (M. Sanaye-Pasand). method for wide area fault location using a limited number of

http://dx.doi.org/10.1016/j.ijepes.2014.06.040
0142-0615/Ó 2014 Elsevier Ltd. All rights reserved.
780 S. Azizi et al. / Electrical Power and Energy Systems 63 (2014) 779–786

Nomenclature

Aki difference between traveling times from the fault point n number of detector-equipped terminals
and terminal i, to terminal k Si set of detector-equipped terminals whose first arriving
EIi,j evaluation index of line (i, j) surges have traversed node i
FS(i,j) binary variable whose value of zero implies line (i, j) is t0 fault inception time
not faulted tm
i first surge arrival time at detector i
i, j indices of line terminal Ti,j surge traveling time between two terminals i and j
k, p indices of detector-equipped terminal Vi,j surge velocity through line (i, j)
L set of all network lines

traveling wave detectors across the power system. Ref. [18] pre- tm
i ¼ t 0 þ t f ;i ; ð1Þ
sents a novel method based on the electromagnetic time-reversal tm
j ¼ t 0 þ t f ;j ; ð2Þ
theory for locating faults in power networks. A combined
impedance- and traveling wave-based method suitable for multi- where tf,i and tf,j are the times a surge needs to travel the direct
terminal ac transmission lines has been presented in [19]. An algo- paths from the fault point f to buses i and j, respectively. In addition,
rithm for fault locations on the star-connected MTDC transmission t0 is the fault inception time. Assume the time it takes for a surge to
lines was proposed in [20]. Nevertheless, this topology is not com- travel the whole length of the transmission line is equal to Ti,j. Tak-
mon because of its low level of reliability [7]. ing into account the extra equation
This paper proposes a new method for fault location on
MTDC systems using the measurements taken at a limited num-
ber of locations. Reducing the computational burden, a straight- T i;j ¼ t f ;i þ tf ;j ; ð3Þ
forward technique is utilized to limit the number of possibly
would result in a system of equations for three unknowns, solving
faulted lines as much as possible. For these lines, a low memory
of which gives tf,i, tf,j and t0. If vi,j is the propagation velocity of
demanding index is evaluated to specify the actual faulted line.
the traveling surge, it can be concluded that the fault has occurred
Eventually, a corresponding system of linear equations relating
at tf,i  vi,j ahead of terminal i.
the first arrival times of traveling waves at different detectors
to the fault location is constructed. Solution of this system of
equations corresponds to an accurate estimation to the fault Proposed fault location methodology
location. Since there is no need to the other traveling wave fea-
tures, the practical issues regarding their identification would be In this section, the proposed fault location methodology is
discarded. The case studies well reflect the method comprehen- explained in details. First, a technique is proposed to reduce the
siveness and its capability to pinpoint the fault location for com- computational burden by reducing the number of possibly faulted
plex MTDC topologies. lines as much as possible. Then, an algorithm is proposed to deter-
The rest sections of the paper are organized as follows. In ‘Trav- mine the actual faulted line from those included in the fault sec-
eling wave-based fault location’, the theoretical foundation of trav- tion. Eventually, the time equations regarding the first arrival
eling wave-based fault location using two-end synchronized times at different detectors are constructed and solved to pinpoint
measurements is explained. The proposed fault location algorithm the fault location.
is systematically set forth in ‘Proposed fault location methodology’. For the sake of convenience, hereinafter, the variables in upper-
To validate the proposed method, simulations results are provided case letters represent those constants whose values are only
and discussed in ‘Simulation results’. Finally, the paper is con- dependent on the network topology. Besides, the variables whose
cluded in ‘Conclusion’. values are also dependent on the fault location (in addition to
the network topology) would be shown in lowercase letters.

Traveling wave-based fault location

Determining the fault section
High sampling rate recording of transient signals during faults
is not a difficult task considering recent developments in trans- The presented approach in the previous section is suitable for
ducer technology [21]. For example, the electronic current trans- fault location on a single transmission line whose both ends are
ducers based on Rogowski coil, having a wide-band frequency equipped with traveling wave detectors. In practice, however, the
response, are commonly used for traveling wave-based applica- problem may be more complicated because of the limited number
tions [22,23]. On the other hand, by time-stamping the captured of measurement devices available for fault location. It should be
samples at a resolution in the order of microseconds, it would be pointed out that the fault location problem becomes more compli-
possible to use the remote locations data to accurately pinpoint cated for a MTDC system including several meshes and radial lines.
the fault [12,21]. This can be easily achieved using the widely As opposed to a single line, the shortest path to each detector, and
spread global positioning system (GPS) signals [21].
Fig. 1 demonstrates a single line connecting two terminals i and
Ti, j
j. Assume both of these terminals are equipped with high sampling
rate measurement devices which are synchronized thanks to the
GPS signals. The first arriving surge at each detector is the one hav- f
ing traveled the shortest path from the fault point f to that detec- i T f ,i Tf , j j
tor. Therefore, if the surge arrival times at terminals i and j are
detected and denoted by tm m
i and t j , two equations can be derived Fig. 1. Traveling wave-based fault location using two-end synchronized
as: measurements.
 S. Azizi et al. / Electrical Power and Energy Systems 63 (2014) 779–786 781

the lines included in this path vary with respect to the fault loca-
Start with all lines in
tion in the MTDC system. l=0
the fault section
To clarify the problem, suppose a fault has occurred on a trans-
mission line, say line (i, j). As the initiated traveling wave propa-
gates toward terminals i and j, the first arriving surge at any l=l+1
detector has certainly traversed either of these terminals on its
path. Consider the faulted transmission network shown in Fig. 2.
Accordingly, the surge arrival time identified by a detector s=0 Yes l ≤ N L?
installed at terminal k can be derived from:
  s=s+1
tm
k ¼ min T i;k þ t f ;i þ t 0 ; T j;k þ t f ;j þ t 0 ; ð4Þ

No
where Ti,k and Tj,k are times it take for a surge to travel the shortest
paths from terminal k to terminals i and j, respectively.
No s ≤ N DP ?
Suppose n traveling wave detectors are installed at the same
number of terminals across the network. Without loss of general-

Yes
ity, assume the detector-equipped terminals are terminals 1, Exclude line l from
2, . . . , n and the sequence of first arrival times at different detec- the fault section
tors are as t m m m
1 < t 2 <    < t n . Let l be the index of the line connect-
ing two terminals i and j. Besides, let s represent a pair containing
indices of two detector-equipped terminals p and q so that t m m
q < tp .
Constraint (5)
Now assume the following topology related inequalities hold true Yes holds for line l and No Declare the
for line l and terminals p and q: detector-pair s? fault section

T i;q > T i;p
: ð5Þ
T j;q > T j;p
Fig. 3. The flowchart of the proposed algorithm for fault section restriction.
The above relations express that the shortest path from any
arbitrary point on line l to detector p is shorter than the one to
detector q. For faults on line l, therefore, it is not possible to detect T j;k  T i;k þ t f ;j þ t0 6 Aki 6 t f ;i þ t0 : ð7Þ
the first arriving surge at detector q before a surge has arrived at It will be proved as a lemma in Appendix A that Tj,k  Ti,k would take
detector p. Otherwise, it can be concluded that the fault has been
its smallest value if the shortest path from terminal i to terminal k
located on a line other than line l. This property is exploited in this traverses terminal j. In this case, Tj,k  Ti,k = Ti,j. Accordingly, it is
paper to exclude the lines having no possibility to be faulted. Fig. 3
possible to alter the lower boundary of Eq. (7) to Ti,j + tf,j + t0 and
illustrates the flowchart of the proposed algorithm for fault section hence, this inequality can be simplified to:
restriction. In this flowchart, NL and NDP denote the number of lines
and distinguishable detector-pairs, respectively. As can be seen, if t f ;i þ t0 6 Aki 6 t f ;i þ t0 : ð8Þ
Eq. (5) holds for line l and any detector-pair, that line is excluded
Using the same procedure, it can be concluded that:
from the fault section. This procedure is pursued for all lines to
restrict the fault section as much as possible. t f ;j þ t0 6 Akj 6 t f ;j þ t0 : ð9Þ

Faulted line identification Aki takes its possible maximum, i.e., the upper boundary of Eq. (8),
for the detectors whose first arriving surges have traversed terminal
Let define the variable Aki as the difference between the travel- i. The same status holds true for Akj and the detectors whose first
ing wave arrival time at detector k, and the time it takes for a surge arriving surges have traversed terminal j in their paths. If the asso-
to travel the distance between terminals i and k. Hence, ciated maximums are respectively denoted by Amax i and Amax
j , then
Amax
i ¼ t f ;i þ t 0 and Amax
j ¼ t f; j þ t 0 . Additionally, it follows from the
Aki ¼ t m
k  T i;k : ð6Þ addition of two inequality Eqs. (8) and (9) that:
Assume a fault has occurred on line (i, j). With regard to Eq. (4), t0  T i;j 6 Aki þ Akj 6 t0 þ T i;j : ð10Þ
Aki would be equal to either of tf ;i þ t0 or T j;k  T i;k þ tf ;j þ t0 . How-
ever, the second value is only taken if the path traversing terminal Let define the range of a set of data as the difference between the
j is shorter than the one traversing terminal i, and therefore its largest and smallest values. Thus, for the faulted line (i, j), the range
value is smaller than the first one. Thus, it follows that: of Aki þ Akj would be less than or equal to 2Ti,j according to Eq. (10).
This property can be exploited to further restrict the fault section.
Depending on which end of the line the first arriving surge has
Ti, j traversed, either of the following expressions would take the value
i j
of zero:
f 8  max 
t f ,i tf, j < tm
k  Ai þ T i;k
 : ð11Þ
max
: tm
k  Aj þ T j;k
Remaining Network
Therefore, the following evaluation index is theoretically equal to
zero, and in practice would take a very small value, for the faulted
line:
Ti, k k T j, k
X   max    

Fig. 2. Wide-area traveling wave-based fault location using remote terminals EIi;j ¼ min tm
k  Ai þ T i;k ; t m max
k  Aj þ T j;k  : ð12Þ
measurements. 8k
782 S. Azizi et al. / Electrical Power and Energy Systems 63 (2014) 779–786

As will be shown in the simulation section, this evaluation index is

capable of determining the faulted line, if it is calculated for all sus- Start
pected lines in the fault section.

Calculate the traveling time

Time equations regarding the first arriving surges
between all terminal pairs
The faulted line would be easily identified as explained in the
previous subsection. Furthermore, the term in Eq. (11) having a Calculate traveling wave arrival times
smaller absolute value (ideally zero) implies which end of the at every detector using the WT
faulted line has been traversed by the first arriving surge in its path
to detector k. Let Si and Sj be sets respectively containing the indi-
Specify the fault section
ces of detectors whose arriving surges traversed terminals i and j.
Now, it is possible to extract the traveling time equations from
the fault point to the various detectors. The respective system of Determine the faulted line
equations is as follows: considering evaluation index (12)

tm
k ¼ T i;k þ t f ;i þ t 0 ; 8 k 2 Si
: ð13Þ Use the LES method to solve (14)
tm
k ¼ T j;k þ t f ;j þ t 0 ; 8 k 2 Sj and find the fault location
Taking into account that tf,j = Ti,j  tf,i, the above system of equations
can be rewritten in a matrix form as:
End
 T
M ¼ H  t0 t f ;i : ð14Þ
Fig. 4. The flowchart of the proposed fault location algorithm.
This is an over-determined system of linear equations which
can be solved for its two unknowns using the least error squares
(LES) method. Thus, the fault location and its inception time would Simulation results
be accurately obtained.
For fault location using the proposed method, two independent To investigate the applicability of the proposed fault location
equations are required for the system of Eq. (14) to be solvable. method, it has been tested on a number of MTDC transmission net-
This condition can be checked for every line by investigating two works already studied in literature. The accuracies of the obtained
faults at the beginning and the end of that line. Accordingly, for results justify that the method can be utilized for fault location in
any given set of traveling wave detectors, it is possible to deter- any MTDC system. In this section, a complex MTDC test system
mine unobservable parts of the MTDC system from the fault loca- containing a ring, two meshes, and three radial lines is considered
tion point of view. In this paper, the measurements’ set is chosen to demonstrate the capability of the proposed method. Numerous
so as to make the whole system observable for fault location. fault simulations are conducted on this test system to scrutinize
the accuracy and robustness of the obtained results versus differ-
Proposed fault location algorithm ent factors. In the first part, it is supposed that a fault on line (g,
h) occurs at t = 0.1 ms. Then the whole procedure on how to carry
The system of Eq. (14) would result in a precise fault location out the fault location based on the proposed algorithm is explained
provided that the traveling wave arrival times at various detectors for this specific fault case. In the next part, the accuracy of the
are identified. Being capable of simultaneous time and frequency results with respect to the noise level, fault impedance and sam-
domain analysis, WT has been extensively exploited for signal pro- pling frequency is examined.
cessing applications in power systems [8]. Since the fault location The studied MTDC test system and its graph representation are
calculations are performed offline, the continuous wavelet trans- shown in Fig. 5(a and b), respectively. In addition, the MTDC test
form (CWT) is preferred to accurately detect the abrupt changes system specifications are provided in Appendix B. The traveling
in the measured signals [20]. wave detectors are installed at the converter stations a, b, c, d
To calculate parameters of the system of Eq. (13), the shortest and e on the MTDC test system. Traveling time distance between
path and as a result, the time distance between all terminal-pairs, various terminal pairs of this test system is shown in Table 1.
i.e., Ti,j’s, should be calculated. This would be easily achieved by PSCAD/EMTDC software [26] is deployed to conduct the simula-
using the effective algorithm proposed in [24, 25]. Now, the pro- tions. Moreover, the detailed model of the HVDC converters and
posed fault location algorithm can be described by several stages the frequency-dependent model of cables are utilized. It should
as follows: be noted that the proposed method can be easily applied to any
MTDC system having an arbitrary combination of overhead, cable
I. The first arrival times at different detector locations across and mixed dc lines. The surge propagation velocity in the dc cable
the MTDC system are derived using the signal processing lines is 145 m/ls. In order to generate precise signals, simulations
method of WT. time step is set to 0.1 ls. Then, the current signals at detector loca-
II. The fault section is distinguished by excluding irrelevant tions are sampled with a rate of 1 MHz. To detect the traveling
lines using the flowchart shown in Fig. 3. wave arrival times CWT coefficients with the scale of 32 as well
III. The evaluation index Eq. (12) is calculated for all remaining as the ‘‘Haar’’ mother wavelet is taken into account as reported
lines in the fault section. The line is approved as the faulted in [20].
one whose associated value is smallest amongst all.
IV. The fault location would be accurately determined by solv-
ing system of Eq. (14) regarding the specified faulted line. Fault location procedure for a specific fault case

Flowchart of the proposed fault location algorithm is shown in It is assumed that a fault has occurred at 2.5 km ahead of termi-
Fig. 4. nal g on line (g, h). With respect to the network topology, it take
 S. Azizi et al. / Electrical Power and Energy Systems 63 (2014) 779–786 783

20 km 20 km g
b

20 km 5 km 20 km 5 km d
h

70 km 50 km 50 km
30 km 70 km 30 km

c e
40 km

40 km

40 km
40 km 30 km

30 km a p

(a) (b)
Fig. 5. (a) Studied MTDC test system. (b) Associated graph representation.

Table 1
Traveling time distance between various terminal-pairs.

Ti,j (ls) Terminal (j)

a p g b h d c e
Terminal (i) a 0 206.897 689.655 827.586 689.655 724.138 482.759 758.621
p 206.897 0 482.759 620.690 482.759 517.241 275.862 551.724
g 689.655 482.759 0 137.931 137.931 172.414 344.828 482.759
b 827.586 620.690 137.931 0 275.862 310.345 482.759 620.690
h 689.655 482.759 137.931 275.862 0 34.4828 206.897 344.828
d 724.138 517.241 172.414 310.345 34.4828 0 241.379 379.310
c 482.759 275.862 344.828 482.759 206.897 241.379 0 275.862
e 758.621 551.724 482.759 620.690 344.828 379.310 275.862 0

t = 806.9, 255.17, 427.59, 255.17, and 565.52 ls for the fault initi- (p, g), (g, b) and (g, h) remain as lines on which the fault may have
ated traveling surges to be reach terminals a, b, c, d and e, respec- occurred. Accordingly, the search space size decreases from 9 lines
tively. According to these values, it is expected that to 3 lines.
tm m m m m
b ¼ t d < t c < t e < t a regardless of the fault inception time (i.e., In the next step, Aki is calculated for all remaining line terminals
t = 0.1 ms). To demonstrate how to specify the fault section using in the fault section and various detectors, i.e., all i’s and k’s. Table 3
the proposed algorithm, this particular fault case is chosen so that calculates these variables for four terminals p, g, b and h which are
the distances between the fault point to detectors b and d are included in the fault section. Based on constraint Eq. (11), if a line
equal. To construct the fault section using algorithm Eq. (5), the (i, j) is the faulted one, Aki þ Akj should lie within an interval with
arriving time sequences between all detector-pairs are taken into the length of less than or equal to 2  Ti,j. Table 4 gives the value
consideration (first column of Table 2). For every arriving time of Aki þ Akj for all included lines in the fault section with respect
sequence, a number of lines are specified that have no fault possi- to different detector locations. As shown, line (g, b) can be rejected
bility and therefore, should be excluded from the fault section (sec- from the fault section in this stage, because the variation range of
ond column of Table 2). Logically, in the cases which the arrival Akg þ Akb are larger than 2  Tg,b, i.e., 344.83 > 275.86.
times of two different detectors are very close or the same, e.g., Table 5 gives the evaluation index Eq. (12) for two lines (p, g)
tm m
b and t d , the associated arrival sequence should be ignored due and (g, h) included in the fault section. As can be seen, the evalua-
to the possible measurement inaccuracies. If the listed lines in tion index for faulted line (g, h) is equal to zero which is much
Table 2 is eliminated from all the network lines, only three lines smaller than that of line (p, g). Although in practice, the evaluation
index may take a very small positive value, the conducted simula-
tions show the index reliably discriminate the faulted line amongst
Table 2 the suspected lines included in the fault section. Based on the third
Fault section construction with respect to traveling waves arriving sequence at
different locations.

Traveling wave arriving sequence Excluded lines Table 3

tm < tm
a (a, p) Differences between traveling times from the fault point and terminals included in
b
tm < tm (a, p) the fault section to various detectors.
d a
tm < tm (a, p)
c a
Aki (ls) Detector location (k)
tm
e < tm
a (a, p)
tm < tm (a, p), (h, d), (h, c), (h, e), (c, e), (p, c) a b c d e
b c
tm
b < tm
e (a, p), (c, e), (p, c) Line terminal (i) p 600 365.52 262.07 151.72 13.79
tm
d < tm
c (a, p), (c, e), (p, c) g 117.24 117.24 82.76 82.76 82.76
tm
d < tm
e – b 20.69 255.17 55.1724 55.17 55.17
tm
c < tm
e – h 117.24 20.69 220.69 220.69 220.69
784 S. Azizi et al. / Electrical Power and Energy Systems 63 (2014) 779–786

Table 4
Calculated and permissible range of Aki þ Akj for lines included in the fault section.

Aki þ Akj (ls) Detector location (k) Calculated range of Aki þ Akj (ls) Permissible range of Aki þ Akj (ls)
a b c d e
Line (i,j) (p, g) 717.24 248.28 179.31 234.48 96.55 965.517 965.517
(g, b) 96.55 372.41 27.59 27.59 27.59 344.828 275.862
(g, h) 96.55 234.48 165.51 165.51 165.52 206.897 275.862

Table 5
Determining the faulted line using the proposed evaluation index.

Line Arguments of min function in Eq. (12) Detector location (k) Evaluation index (ls)
a b c d e
  
(p, g) m max  0 965.51 862.07 448.28 586.21 103.44
tk  Ap þ T p;k 
  
m max  0 0 34.48 34.48 34.48
tk  Ag þ T g;k 
  
(g, h) m max  0 0 34.48 34.48 34.48 0
tk  Ag þ T g;k 
   
m max  103.45 241.38 0 0 0
tk  Ah þ T h;k 

Table 6
Fault location results for several fault points on line (p,g) of the MTDC test system.

Fault distance from terminal p (km) Detector location Fault inception No. of the included EIa (ls) Fault location
time (ls) lines in FS error (%)
a b c d e
Surge arrival time (ls)
10 Expected 1275.8 1551.7 1344.8 1586.2 1620.6 1000 1 2.41 5.9  104
Detected 1279 1555 1349 1590 1624 1003.4
35 Expected 1448.2 1379.3 1517.2 1413.7 1724.1 1000 1 4.34 0.051
Detected 1453 1385 1522 1418 1727 1004.5
50 Expected 1551.7 1275.8 1482.7 1310.3 1620.6 1000 1 3.27 0.045
Detected 1556 1279 1486 1315 1625 1004.06
a
Evaluation index value of the faulted line.

Now, the faulted line and the shortest paths from the fault point
Table 7 to various detectors have been specified. Consequently, the system
Noise influence on the fault location accuracy. of Eq. (13) for this fault case is constructed as follows:
SNR (db) Fault distance from terminal p (km) 8
>
> 806:9 ¼ 689:66 þ t f ;i þ t0
0 15 35 50 65 >
>
>
Fault location error (%) < 255:17 ¼ 137:93 þ tf ;i þ t 0
>
Without noise 0.059 0.047 0.056 0.062 0.061 255:17 ¼ 34:48 þ 137:93  t f ;i þ t0 : ð15Þ
>
>
60 0.059 0.056 0.058 0.059 0.061 >
> 427:59 ¼ 206:9 þ 137:93  t f ;i þ t 0
50 0.064 0.056 0.058 0.073 0.061 >
>
:
40 0.064 0.056 0.059 0.073 0.058 565:52 ¼ 344:83 þ 137:93  tf ;i þ t 0
This system of equation can be solved using the LES method,
where the solution would be t0 = 100 ls and tg,h = 17.24 ls. Thus,
if the time distance between the fault point to terminal p is multi-
Table 8 plied by the surge velocity, the fault location would be derived as
Effect of sampling frequency on the fault location accuracy. 2.5 km.
Sampling frequency Fault location error (%)
Avg. Max.
Fault location results
5 MHz 0.0421 0.1046
1 MHz 0.0955 0.1568
500 kHz 0.2815 0.6262 A large number of fault cases are simulated on the MTDC test
250 KHz 0.5792 1.0879 system. Various locations on different transmission lines are exam-
ined as fault locations. Moreover, the fault impedance is altered
from 0 to 100 X. In order to identify the first surge arrival times,
the current measurements at different detector locations are
and forth rows of Table 5, it is possible to determine that the first post-processed using the WT. To calculate the percentage error,
arriving surge has traversed either terminal g or terminal h on its the absolute difference between the actual and estimated fault
path to different detectors. Consequently, the first arriving surges locations are divided by the respective faulted line length. Next,
at detectors a and b have traversed terminal g and those detected the maximum and average percentage errors are calculated for
at terminals c, d and e have traversed terminal h, respectively. all simulated cases. A total of more than 1000 cases are simulated,
 S. Azizi et al. / Electrical Power and Energy Systems 63 (2014) 779–786 785

Table 9
Fault location accuracy for high impedance faults.

SNR (db) Error (%) Fault impedance (X)

0 25 50 100
60 Avg. 0.085 0.085 0.085 0.086
Max. 0.124 0.124 0.124 0.124
40 Avg. 0.085 0.085 0.086 0.087
Max. 0.124 0.124 0.125 0.129

and average and maximum percentage errors of 0.082% and 0.187% paper. In doing so, the fault section is distinguished by excluding
are respectively obtained. lines having no possibility of being faulted. Then, the faulted line
To provide more details, three fault cases at the beginning, the is specified by evaluating a simple index for all included lines in
middle and end of line (p, g) are shown in Table 6. The fault incep- the fault section. Eventually, the fault location and inception time
tion time and location are obtained according to the calculated would be accurately determined, by solving the corresponding sys-
surge arrival times. As can be seen, the proposed method accu- tem of equations.
rately pinpoints the fault location regardless of the complexity of The proposed fault location algorithm just requires the first
the MTDC system. surge arrival time at different detector locations which is an easily
extractable feature. Therefore, the practical difficulties associated
Noise influence on the estimation accuracy with detecting subsequent arriving surges and extracting their
other features are totally eliminated. Although the proposed algo-
To examine the noise influence on the accuracy of the fault loca- rithm exploits a limited number of traveling wave detectors
tion results, the current signal is contaminated with different levels installed at converter stations, it pinpoints the fault location inde-
of noise. Due to noise existence, the WT coefficient takes a nonzero pendent of the complexity of the network topology, i.e., the num-
value at the instances not corresponding to the traveling wave ber of its rings, meshes and radial lines. On the other hand, the
arrivals. However, simulation results demonstrate that for a wide method demonstrates high noise immunity and robustness against
range of noise intensities, the WT coefficients of noise are not as fault impedance.
significant as those corresponding to the traveling wave arrivals.
Therefore, it would be helpful to use a threshold for eliminating Appendix A. Supplementary material
the noise associated WT coefficients [13]. For five different fault
cases on line (p, g), the generated traveling waves are contami- Supplementary data associated with this article can be found,
nated with 60, 50 and 40 db white Gaussian noise. Table 7 summa- in the online version, at http://dx.doi.org/10.1016/j.ijepes.2014.
rizes the fault location results. The estimations accuracies confirm 06.040.
that the first arrival times can be properly identified by defining an
appropriate threshold to reject the unwanted WT coefficients.
References

Sampling frequency effect [1] Jiang H, Ekstrom A. Multiterminal HVDC systems in urban areas of large cities.
IEEE Trans Power Del 1998;13(4):1278–84.
[2] Lianga J, Gomis-Bellmuntb O, Ekanayakea J, Jenkinsa N, Anc W. A multi-
In the proposed method, the outputs of more than one terminal HVDC transmission system for offshore wind farms with induction
measurement devices are utilized using LES technique for fault generators. Int Elect Power Energy Syst 2012;43(1):54–62.
location. Accordingly, the average error of the fault location esti- [3] Salomonsson D, Soder L, Sannino A. Protection of low-voltage dc microgrids.
IEEE Trans Power Del 2009;24(3):1045–53.
mations is much lower than its possible maximum. Evidently,
[4] Yang J, Fletcher JE, O’Reilly J, Adam GP, Fan S. Protection scheme design for
the measurement device sampling rate still imposes a limitation meshed VSC-HVDC transmission systems of large-scale wind farms. In: Proc
on the fault location accuracy. In this part, the sampling rate is 9th IET international conf on AC and DC power transmission, Glasgow, UK;
altered from 250 kHz to 5 MHz and the obtained results are 2010.
[5] Ciezki JG, Ashton RW. Selection and stability issues associated with a navy
reported in Table 8. As can be seen, the accuracy of the algorithm shipboard dc zonal electric distribution system. IEEE Trans Power Del
decreases while the sampling frequency is reduced. Besides, the 2000;15(2):665–9.
noise effect is negligible in all simulated cases. [6] Fletcher SDA, Norman PJ, Galloway SJ, Burt GM. Determination of protection
system requirements for dc unmanned aerial vehicle electrical power
networks for enhanced capability and survivability. IET Elect Syst Trans
High impedance faults 2011;1(4):137–47.
[7] Gomis-Bellmunt O, Liang J, Ekanayake J, King R, Jenkins N. Topologies of
multiterminal HVDC-VSC transmission for large offshore wind farms. Elect
A distinguished superiority of the proposed fault location Power Syst Res 2011;81(2):271–81.
method is its robustness against fault impedance. To verify this [8] Galijasevic Z, Abur A. Fault area estimation via intelligent processing of fault
issue, a number of simulated fault cases on line (c, h) with different induced transients. IEEE Trans Power Syst 2003;18(4):1241–7.
[9] Sharafi A, Sanaye-Pasand M, Jafarian P. Ultra-high-speed protection of parallel
amount of fault resistance are summarized in Table 9. The maxi- transmission lines using current travelling waves. IET Gener Transm Distrib
mum and average fault location errors in the presence of 60 and 2011;5(6):656–66.
40 db white Gaussian noise are tabulated in the table. It can be [10] Eisa AA, Ramar K. Accurate one-end fault location for overhead transmission
lines in interconnected power systems. Int Elect Power Energy Syst
observed that as far as the fault impedance lies within a practical 2010;32(5):383–9.
range, the fault locations is accurately identified even in the pres- [11] Spoor D, Zhu JG. Improved single-ended traveling-wave fault location
ence of noise. algorithm based on experience with conventional substation transducers.
IEEE Trans Power Del 2006;21(3):1714–20.
[12] Dewe MB, Sankar S, Arrillaga J. The application of satellite time references to
Conclusion HVDC fault location. IEEE Trans Power Del 1993;8(3):1295–302.
[13] Jafarian P, Sanaye-Pasand M. A traveling-wave-based protection technique
using wavelet/PCA analysis. IEEE Trans Power Del 2010;25(2):588–99.
A new traveling wave-based fault location method suitable for [14] Abur A, Magnago FH. Use of time delays between modal components in
wide-area fault location in MTDC systems was proposed in this wavelet based fault location. Int Elect Power Energy Syst 2011;22(6):397–403.
786 S. Azizi et al. / Electrical Power and Energy Systems 63 (2014) 779–786

[15] Korkali M, Lev-Ari H, Abur A. Traveling-wave-based fault location technique [21] Saha MM, Izykowski JJ, Rosolowski E. Fault location on power networks. 1st
for transmission grids via wide-area synchronized voltage measurements. IEEE ed. London: Springer; 2010.
Trans Power Syst 2012;27(2):1003–11. [22] Practical aspects of Rogowski coil applications to relaying. In: IEEE PSRC
[16] Korkali, Abur A. Optimal deployment of wide-area synchronized special report, IEEE Power Engineering Society; 2010.
measurements for fault-location observability. IEEE Trans Power Syst [23] Pathirana KPAN, Rajapakse AD, Nanayakkara OMKK, Wachal R. Detecting fault
2013;28(1):482–9. generated surges in DC line of VSC HVDC schemes for travelling wave based
[17] Korkali M, Abur A. Robust fault location using least-absolute-value estimator. fault location. In: 2012 CIGRÉ Canada conference, Canada; 2012.
IEEE Trans Power Syst 2013;28(4):4384–92. [24] Deo N. Graph theory with applications to engineering and computer science.
[18] Razzaghi R, Lugrin G, Manesh H, Romero C, Paolone M, Rachidi F. An efficient 1st ed. NJ: Prentice Hall: Englewood Cliffs; 1974.
method based on the electromagnetic time reversal to locate faults in power [25] Dantzig GB. All shortest routes in a graph. In: Proc int symposium, Rome, Italy;
networks. IEEE Trans Power Del 2013;28(3):1663–72. 1966.
[19] Ngu EE, Ramar K. A combined impedance and traveling wave based fault [26] PSCAD/EMTDC, Ver. 4.2 Manitoba HVDC Research Centre; 2005.
location method for multi-terminal transmission lines. Int Elect Power Energy
Syst 2011;33(10):1767–75.
[20] Nanayakkara OMKK, Rajapakse AD, Wachal R. Traveling wave based line fault
location in star connected multi-terminal HVDC systems. IEEE Trans Power Del
2012;27(4):2286–94.