Electric Power Systems Research 130 (2016) 78–88

Contents lists available at ScienceDirect

Electric Power Systems Research

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

A differential protection technique for multi-terminal HVDC

Ahmed E.B. Abu-Elanien a,∗ , Ahmed A. Elserougi a,c , Ayman S. Abdel-Khalik a,c ,
Ahmed M. Massoud a,b , Shehab Ahmed c
a
Electrical Engineering Department, Faculty of Engineering, Alexandria University, Alexandria, Egypt
b
College of Engineering, Qatar University, Doha, Qatar
c
Texas A&M University at Qatar, Doha, Qatar

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

Article history: This paper presents a technique for differential protection of Multi-terminal High Voltage Direct Current
Received 7 February 2015 (MTHVDC) transmission lines. The proposed technique depends on the electrical current data at both ends
Received in revised form 29 May 2015 of each line section. Discrete Wavelet Transform (DWT) is used to detect DC faults as well as filter out
Accepted 26 August 2015
the high frequency transients superimposed on the current signals. An operating signal and a restraining
Available online 12 September 2015
signal are used in this technique to discriminate between internal faults and external faults through their
ratio in each section of the MTHVDC. The operating and restraining signals depend on energy contents
Keywords:
of the de-noised current signals at both ends of each line section. MTHVDC modeling and relay design
Multi-terminal HVDC
Protection
are carried out in the MATLAB environment. The results demonstrate the high reliability of the proposed
Wavelet transform relay in the zonal protection of MTHVDC.
Wind farm © 2015 Elsevier B.V. All rights reserved.

1. Introduction
Voltage source Converter (VSC)-HVDC installations are promis-
ing power transmission methods for transmission of bulk power
over long distances. The deployment of offshore wind power gener-
ation is expected to contribute a major portion of the world’s future
energy portfolio due to the expected depletion of fossil fuels along
with the climate changes introduced by CO2 emissions. Offshore
wind power generates a large amount of power that can be col-
lected and transmitted to different onshore grids via HVDC systems.
Different wind farms can be ultimately interconnected to multi-
ple grids by means of multi-terminal HVDC (MTHVDC). Moreover,
MTHVDC enables the transmission of high power levels, which can
contribute to the realization of the future super grids such as the
European super grid [1–3].
Many research articles discuss the protection of a two-terminal
HVDC system such as the research reported in [1,4–8]. However,
these methods are suitable only for two-terminal HVDC systems.
Some of these methods cannot be extended to MTHVDC while oth-
ers have not been tested for operation in MTHVDC. Generally, the
design of protection systems for MTHVDC grids is considered in its
early stages. Limited research has tackled the protection problems
of MTHVDC [9–14].
∗ Corresponding author. Tel.: +20 1129857562.
E-mail address: ahmed.abuelanien@alexu.edu.eg (A.E.B. Abu-Elanien).
The research carried out in [9] has presented a thorough discus-
sion concerning the DC-fault current and voltage waveform shapes,
but it did not present a complete protection scheme for MTHVDC
systems. The employment of circuit breakers (CBs) on the AC side
of the converter for protection of MTHVDC was introduced in [10].
In case of faults inside the DC grid, all CBs on the AC sides of all
terminals isolate the DC grid, then the faulted section is identified
by means of a handshaking method [10]. The faulted section is iso-
lated and the DC grid is restored afterwards. This method avoids
using DC CBs, however, it takes a long time for isolation, fault zone
identification, and system restoration. Moreover, the MTHVDC sys-
tem is completely shut down unnecessarily. Alternatively, the use
of DC CBs is now recommended to avoid unnecessary shutdown of
healthy DC lines, isolate the fault quickly, and increase the DC sys-
tem reliability [2,3,9,11,12,15]. The research relevant to DC CBs is
promising and several architectures of DC CBs have been developed
in literature [2,3,9,11,12,15]. With proper design of a protective
relay, the DC CBs can isolate the faulted section quickly without
the need to trip the whole DC system.
Yang et al. simulated the DC faults in MTHVDC including wind
farms in [11] and proposed a distance protection scheme in [12] to
be used in MTHVDC. However, this method did not address large
systems with a high fault resistance. The largest rate of change of
current is taken as a discrimination feature to identify the faulty line
in [13]. However, it depends on system topology, where in some
DC fault cases, the healthy line may have a larger rate of change of
fault current. Moreover, this method avoids the deployment of DC

http://dx.doi.org/10.1016/j.epsr.2015.08.021
0378-7796/© 2015 Elsevier B.V. All rights reserved.
 A.E.B. Abu-Elanien et al. / Electric Power Systems Research 130 (2016) 78–88 79

CB; therefore, it takes a relatively long time for isolation. The work Table 1
MTHVDC parameters.
done in [14] presents three criteria to identify the faulted cables
in meshed MTHVDC systems. These criteria are voltage Wavelet Section Resistance Inductance Capacitance
coefficients, current Wavelet coefficients, and voltage derivative (/km) (␮H/km) (nf/km)
and magnitude. Nevertheless, the authors did not consider the non- Overhead line 1 0.028 553 20.2
zero resistance fault case. Moreover, the proposed method depends Overhead line 2 0.028 553 20.2
on the initial change of voltage and current Wavelet coefficients, Undersea cable 0.007 200 226

which may be confusing to the relay especially during the initial

transient period of the fault. Furthermore, a combination between section 2, and section 3 are 200 km, 300 km, and 60 km, respec-
overhead lines and underground cables is not considered in this tively.
paper, which is the existing case in practical systems. A relay is placed at each end of each section with a sampling
The proposed protection algorithm in this paper presents a frequency of 100 kHz, which corresponds to a sampling time of
differential protection method for protection of MTHVDC with 10 ␮s. A SIMULINK/MATLAB model is built to simulate the whole
different configurations of DC sections (overhead lines and under- system. 3L-VSCs are used throughout the whole system. The con-
ground cables). The method depends on calculating an energy verters’ switching frequency is set to 1 kHz, a switching frequency
index for the de-noised versions of positive and negative pole of (1–2 kHz) is recommended for VSC in HVDC applications [18].
currents at both ends of a line to discriminate between inter- The DC-link capacitor of each converter is 1 mF. The grid side
nal and external faults. Pole to ground, pole to pole, and pole to converters’ controllers control the DC-link voltages based on the
pole to ground faults are simulated to test the proposed method. well-known droop control [19], and the wind-turbine side con-
Various fault resistances were simulated to prove the reliability verter controllers are set to active and reactive power (PQ) control
of the proposed method in low resistance and high resistance mode [20]. The droop gains are adjusted based on line impedances
faults. The paper is organized as follows. Section 2 presents the as given in [19]. The MTHVDC parameters are shown in Table 1.
MTHVDC model used to test the protection algorithm. Section
3 introduces the main concept of the proposed method. Protec- 3. Concepts of the proposed algorithm
tion functions and the design steps of the proposed algorithm
are explained in Section 4. Section 5 presents the simulation In MTHVDC, when a fault occurs in any section, all converters
results, and finally, Section 6 concludes the work presented in this contribute to the fault current; VSC control is able to block the oper-
paper. ation of the IGBT switches. DC fault in a line section fed from a
VSC can be represented by three equivalent circuits representing
three operational stages as shown in Fig. 2 where L and R are the
2. Power system model line equivalent inductance and resistance from VSC to fault posi-
tion. In order to analyze the short circuit current, the short circuit is
The most common voltage source-based HVDC converters are decomposed into 3 different stages of operation: Converter DC-link
two-level (2L-VSC), three-level (3L-VSC), and Modular Multilevel capacitor discharge stage, equivalent inductor discharge stage, and
Converters (MMC). The MMC can be classified into half-bridge MMC AC grid feeding stage.
(HBMMC), full-bridge MMC (FBMMC), and clamped double sub- In the capacitor discharge stage shown in Fig. 2(a), the converter
module based MMC (CDSM-MMC). The latter two types of MMC DC-link capacitor voltage can be expressed as follows, assuming
are out of the focus of this work, as they do not need DC CBs due underdamped performance [21]
to their current suppression capability during DC side faults, but
operate with lower efficiency compared to the half-bridge MMC in vC (t0 ) ωo   iL (t0 ) −ıt
vC = e−ıt sin ωt + ˇ − e sin (ωt) (1)
conjunction with a DC CB [16]. It has to be noted that 2L-VSC, 3L- ω ωC
VSC, and HBMMC have the same nature during DC side faults, and and the inductor current is [21]
they require DC CB for clearing the DC fault current. So this work
dvC iL (t0 ) ωo −ıt   vC (t0 ) −ıt
targets these types of converters, whose behavior during a DC side iL = C =− e sin ωt − ˇ + e sin (ωt) (2)
dt ω ωL
fault will be identical. The presented results in this work use the
3L-VSC Neutral point clamped (NPC). Most of the worldwide VSC- where t0 is the fault inception time, ı = R/2L, ω =
    2   
HVDC projects are based on conventional 2L-VSC and 3L-VSC NPC 1/LC − R/2L , ωo = ı2 + ω2 , and ˇ = arctan ω/ı .
[17].
This response continues until the full discharge of the converter
A three-terminal bipolar HVDC model is considered for this
DC-link capacitor [21]. Solving (1), the time at which the capacitor
study. The three-terminal system contains three sections. Section
voltage drops to zero can be found.
1 is connected to AC grid 1 via a 3L-VSC. Section 2 is connected
to AC grid 2 via a 3L-VSC, and section 3 is connected to a 3L-VSC ( − )
t1 = t0 + (3)
fed from an offshore wind farm. Sections 1 and 2 are overhead ω
bipolar HVDC lines and section 3 is a bipolar undersea cable con- where  can be calculated as follows [21]
necting the offshore wind farm with the overhead lines. The system    
configuration is shown in Fig. 1. This model is suitable for imple- vC (t0 ) ωo C sin ˇ
mentation in the European super grid, in which, the MTHVDC is
 = arctan   (4)
vC (t0 ) ωo C cos ˇ − iL (t0 )
used to transmit electrical power from an offshore wind farm to
two different AC grids. Nevertheless, the proposed algorithm can be For positive pole to ground fault 50 km from BB3, the values of L,
typically implemented on other MTHVDC configurations including R, and C according to Table 1 are 0.01 H, 0.35 , and 1 mF. The values
meshed configurations. However, meshed networks need compli- of iL (t0 ), and vC (t0 ) are 2000, and 250 kV. For this fault, values of ω,
cated control systems which are outside the scope of this research. ı, ωo , ˇ, and  are 315.74, 17.5, 316.22, 1.515, and 1.5406, respec-
The bipolar HVDC system, shown in Fig. 1, is operating at a nominal tively. As a result, t1 = t0 + 5.071 ms. Fig. 3(a) shows the positive pole
voltage of ±250 kV. The rated power of the wind farm is 1000 MVA current of converter 3 during positive pole to ground fault.
at a unity power factor. The nominal power flow is considered from At t = t1 , the second stage starts as shown in Fig. 2(b). The circuit
the wind farm toward the two AC grids. The lengths of section 1, inductor starts to discharge through the VSC freewheeling diodes.
80 A.E.B. Abu-Elanien et al. / Electric Power Systems Research 130 (2016) 78–88

Fig. 1. MTHVDC model.

Fig. 2. DC fault equivalent circuit (a) converter DC-link capacitor discharge stage. (b) Equivalent inductor discharge stage. (c) AC grid feeding stage.

Freewheeling diodes behave like an uncontrolled rectifier bridge
that feeds the DC fault [21] after inhibiting the gate pulses of the
self-commutated semiconductor devices (e.g. IGBTs).
The inductor current has an initial value of iL (t1 ). The inductor
current during this stage is
iL = iL (t1 ) e−(R/L)t
Fig. 3(a) shows the positive pole current of converter 3 during
positive pole to ground fault located at 50 km from BB3. It is clear
that the first stage ends 5.071 ms after fault inception. After 5.071
from fault inception, the converter DC-link capacitor voltage drops
to nearly zero as shown in Fig. 3(b). After the capacitor discharge
stage, inductor starts to discharge as shown in Fig. 3(a). The current
starts to decay exponentially at this point of time. The diodes start to
conduct at inductor discharge stage and the diode current increases
abruptly to high values.
The diode current in this stage will be one third of the inductor
current. At the beginning of the inductor discharge stage, the diode
current change abruptly, which may damage the diodes. To avoid
damaging the diodes during the faults, a thyristor can be connected
(5) across the IGBT to share the fault current with its freewheeling
diode during DC side faults, as thyristors are able to withstand
high surge currents [22–24]. This will help in reducing the over-
current stresses on the freewheeling diodes pending the clearing
of the fault. This can be realized by turning the thyristor on when
a DC side fault is initiated. This configuration provides additional
time for the protection system to identify the faulted zone without
damaging the VSC freewheeling diodes. The proposed algorithm is

4
x 10
Positive pole current, A

10

DC link Inductor
capacitor discharge
5 discharge
stage
stage

0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
(a)
5
x 10
Positive pole voltage, V

-1
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
(b) time, s

Fig. 3. Positive pole to ground fault at 50 km from BB3. (a) Positive pole current at converter 3. (b) Positive pole voltage at converter 3.
 A.E.B. Abu-Elanien et al. / Electric Power Systems Research 130 (2016) 78–88 81

4 4
x 10 x 10
2 2

R12 negative
R11 negative

current, A
current, A
1 1

0 0
fault inception
-1 -1
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
(a) (d)
4 4
x 10 x 10
2 2

R22 negative
R21 negative

current, A
current, A

1 1

0 0

-1 -1
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5

(b) (e)
4 4
x 10 x 10
5 4

R132 negative
R31 negative
current, A

current, A
0 2
-5
0
-10
-2
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5

(c) (f) Time, s

Fig. 4. Negative pole to ground fault at 30 km from BB3. (a) Negative pole current of R11 . (b) Negative pole current of R21 . (c) Negative pole current of R31 . (d) Negative pole
current of R12 . (e) Negative pole current of R22 . (f) Negative pole current of R32.

designed for VSCs such as the 2L-VSC, 3L-VSC, and HBMMC with this
thyristor-based built-in bypassing technique [22–24], particularly
in the case of freewheeling diodes with limited surge current han-
dling capabilities. Advancements in the semiconductor industry
have made free-wheeling diodes with higher surge current with-
standing capabilities possible [25], which provides additional time
for the protection system to achieve the task without damaging the
converter freewheeling diodes even with no bypass thyristors.
In the AC grid feeding stage shown in Fig. 2(c), the capacitor
and inductor are under forced response from AC side. The inductor
current in this case will be the summation of the diodes currents in
the positive direction. The inductor current in this stage is
iL = iAC
where iAC is the summation of the three phases current when the
current is larger than zero [21]. The protection system operates
usually in the first or second stage of the fault. The fault should not
left until reaching the third stage because it may cause damage to
the DC system.
For a fault in the DC system, the fault is fed from all sources in the
system. Accordingly, the sending end and receiving end currents of
the same pole in the faulted section will be different in magnitude.
The magnitude of the fault current at each end depends on the DC
system topology, the inductance to resistance ratio of each line,
the total impedance of each line, converter DC-link capacitor sizes,
and short circuit capacity of each AC grid side. For healthy sections,
the sending end and receiving end currents of the same pole will
have nearly the same current during a fault. Part of the high fre-
quency components of the current is shunted to ground during a
fault condition due to line/cable stray capacitances.
Fig. 4 shows the negative pole currents of relays R11 , R12 , R21 , R22 ,
R31 , and R32 , for a bolted negative pole to ground fault at 30 km from
BB3. The negative pole currents of both relays at the two ends of a
healthy section are nearly the same. This is obvious from Fig. 4(a),
(b), (d) and (e), where the negative pole current of R11 is nearly the
same as that of R12 ; and the negative pole current of R21 is nearly the
same as that of R22 . On the other hand, the negative pole currents
of the two ends’ relays of section 3 are different. The degree of
dissimilarity depends on system topology, fault location, etc.
The similarity of the sending end and receiving end currents is
not limited to only the healthy sections, but also the healthy poles
in the faulted sections as shown in Fig. 5. The positive pole currents
of relays R31 and R32 are nearly the same as depicted in Fig. 5(c) and
(f) although there is a positive pole to ground fault in section 3.
The aforementioned facts are used in the design of the differen-
tial relay. However, any fault generates wide band high frequency
current transients which propagate along the transmission line to
both ends [26–29]. After reaching the line end, part of these tran-
(6)
sients is reflected back and the remaining is transmitted to other
sections of the transmission system. Very high frequency transients
are affected by the system capacitances. Part of these high fre-
quency transients is shunted to ground. Therefore, even in case
of a healthy section, the sending end and receiving end currents
will not be exactly the same. The small difference will be due to
the high frequency noise. Thus, removing this high frequency noise
will allow similar sending and receiving end currents for healthy
sections.
DWT is a powerful tool for analysis of non-stationary signals,
such as short circuit currents, due to its ability to extract frequency
bands with keeping time information. Theory and more details of
DWT can be found in [30]. DWT has the advantage of analyzing
the signal frequency spectrum without losing the time information.
Fourier transform breaks the signal into scaled sinusoids; likewise,
Wavelet transform (WT) breaks up a signal into a scaled and dilated
version of a mother wavelet, thus keeping time information. There
are many types of mother wavelets such as Haar, Daubichies (db),
Coiflets (coif), and Symmlets (sym) wavelets. The choice of mother
is very important for the prompt collection of transient data. More
details about using DWT in analysis of power system transients can
be found in [31].
82 A.E.B. Abu-Elanien et al. / Electric Power Systems Research 130 (2016) 78–88

2000 2000

R11 positive

R12 positive
current, A

current, A
1000 1000

0 0
fault inception
-1000 -1000
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5

(a) (d)
2000 2000
R21 positive

R22 positive
current, A

current, A
1000 1000

0 0

-1000 -1000
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5

(b) (e)
4000 4000
R31 positive

R32 positive
current, A

current, A
2000 2000

0 0

-2000 -2000
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5

(c) (f) Time, s

Fig. 5. Negative pole to ground fault at 30 km from BB3. (a) Positive pole current of R11 . (b) Positive pole current of R21 . (c) Positive pole current of R31 . (d) Positive pole
current of R12 . (e) Positive pole current of R22 . (f) Positive pole current of R32.

The output of DWT is an approximation (A) and a number of
details (d). The approximation covers the low frequency band,
while the details cover the higher frequency bands. The approx-
imation can be considered the basic de-noised signal. The noise is
included in the transform details. Daubechies mother wavelet fam-
ily (db mother wavelet family) is compactly supported orthonormal
family of wavelets introduced by Ingrid Daubechies. It consists of
10 mother wavelets suitable for discrete wavelet analysis. The used
mother wavelet in this algorithm is Daubechies 4 (db4) because
it is suitable for short and fast transients, such as fault generated
transients [30].
Fig. 6 depicts a negative pole to ground fault 260 km from BB2.
Fig. 6(a), (f), and (k) shows positive pole and negative pole currents
(Ip and In ) at relays R11 , R21 , and R31 , respectively. Fig. 6(b), (g), and
(l) shows that the approximations are very close to their respective
original current signals.
The proposed differential relay contains two main signals,
namely, operating signal and restraining signal. The operating
signal tends to operate the relay for internal faults, while the
restraining signal tends to restrain the operation of the relay in
case of external faults. The operating and restraining signals are
designed to depend on the energy contained in the current signals.
For each pole, two energy indices are calculated for the current at
each line end to form the operating and restraining signals as in the
following equation:

ti +delay
Ep11 = A2p11 (k) t
k=ti
where Ep11 is the energy index for the positive pole at end 1 of
section 1 using an approximation of positive pole current at end 1
of section 1 (Ip11 ); Ap11 (k) is the DWT approximation of the pos-
itive pole current at end 1 of section 1 at time k; k is an index
representing the sample time; ti is the fault inception time; delay
is an interval of time used to calculate the energy index; and t is
the sampling time. The energy index for the positive pole at end 2
(Ep12 ), the energy index for the negative pole at end 1 (En11 ), and
the energy index for the negative pole at end 2 (En12 ) are also calcu-
lated using the same approach. Similarly, the energy indices can be
calculated for the remaining two sections. For internal faults, the
energy indices of the faulted pole (poles) at both ends of the line
will be dissimilar. On the contrary, for external faults, the energy
indices of the faulted pole (poles) at both ends of the line will be
nearly similar. Hence, the operating signal and restraining signals
for the positive pole are given by the following equations:
Sop1 = Ep11 − Ep12 (8)
Srp1 = Ep11 + Ep12 (9)
where Sop1 is the operating signal for the positive pole at section 1
and Srp1 is the restraining signal for the positive pole at section 1.
In a similar way, the operating signal for the negative pole (Son1 )
and the restraining signal for the negative pole (Srn1 ) can be calcu-
lated for section 1. Moreover, the operating and restraining signals
for the other sections can be calculated in the same way. Fig. 6(d),
(i), and (n) shows the negative pole operating signals at relays R11 ,
R21 , and R31 . The light colored (red) curves show the operating sig-
nals calculated using the original current signals. The black curves
show the operating signals calculated using the DWT approxima-
tion. Fig. 6(e), (j), and (o) shows the negative pole restraining signals
(7) at relays R11 , R21 , and R31 with the same coloring order like oper-
ating signals. It can be shown that the operating signals of the
un-faulted sections using the DWT approximations are lower than
those using original signal, which means better discrimination of
faults and vice versa for faulted sections. Moreover, the usage of
DWT approximation improves the restraining signals too. Note that
the operating signal will be multiplied by a constant K as will be
discussed in (10).
The operating signal and restraining signal are calculated to
ensure a decision for the relay, i.e. trip or no trip. The rip signal
will be given if the operating signal is dominant, or no trip signal
 A.E.B. Abu-Elanien et al. / Electric Power Systems Research 130 (2016) 78–88 83

4 4 4
x 10 x 10

Current, A
2 x 10
2 5
0 0 0
-2 -2 -5
5 10 15 20 5 10 15 20 5 10 15 20
(a) (f) (k)
4 4 4

Approximation
x 10 x 10 x 10
2 2 5
0 0 0
-2 -2 -5
5 10 15 20 5 10 15 20 5 10 15 20
(b) (g) (l)
100 1000 200
d1

0 0 0
-100 -1000 -200
5 10 15 20 5 10 15 20 5 10 15 20
(c) (h) (m)
4 7 5
x 10 x 10 x 10
Operating

10 4 4
signal

5 2 2
0 0 0
5 10 15 20 5 10 15 20 5 10 15 20
(d) (i) (n)
7 7 7
x 10 x 10 x 10
Restraining

2 5 4
signal

1 2
0 0 0
5 10 15 20 5 10 15 20 5 10 15 20
(e) (j) (o) time (ms)

Fig. 6. Negative pole to ground fault at 260 km from BB2. (a) Positive and negative poles’ currents at R11 . (b) DWT approximations of positive and negative poles’ currents at
R11 . (c) Detail 1 coefficients for negative pole current at R11 . (d) Negative pole operating signal at R11 .(e) Negative pole restraining signal at R11 . (f) Positive and negative poles’
currents at R21 . (g) DWT approximations of positive and negative poles’ currents at R21 . (h) Detail 1 coefficients for negative pole current at R21 . (i) Negative pole operating
signal at R21 .(j) Negative pole restraining signal at R21 . (k) Positive and negative poles’ currents at R31 . (l) DWT approximations of positive and negative poles’ currents at
R31 . (m) Detail 1 coefficients for negative pole current at R31 . (n) Negative pole operating signal at R31 . (o) Negative pole restraining signal at R31 . (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)

is given if the restraining signal is dominant. The trip decision is
taken based on trip ratio (10);
KSop1
Rp1 =
Srp1
where Rp1 is the trip ratio for positive pole at section 1; K is a
constant.
In the same way, the trip ratio for the negative pole at section
1 (Rn1 ), and ratios of the other sections can be calculated. In case
of external faults, the operating signal will be very small, while the
restraining signal will be much higher than the operating signal. As
a result, the trip ratio for the positive and negative poles will be very
small. In case of internal faults, the operating signal will be much
higher than that in case of external faults. The trip ratio in this case
will not be as small as the case of external faults. The constant K
is a constant used to adjust the trip ratio to be larger than unity in
internal faults, and less than unity in the case of external faults.
4. Algorithm design
In order to design any protection system, the protection
functions should be defined. The proposed protection algorithm
functions are: (i) fast fault detection; (ii) internal and external fault
discrimination; (iii) faulty pole classification; and (iv) identification
of fault location.
4.1. Fault detection
For fast fault detection, a fast measurement technique should
be used. The technique should have different characteristics
pre-fault and post fault. The high frequency transients during faults
are used as a fast and reliable measure to detect faults. The high
frequency transients in a certain frequency band have zero value
(10) during normal conditions; whereas, they attain high values dur-
ing fault conditions. Moreover, the smooth change of load current
generates minimal high frequency transients which can be ignored.
The most important issue is to keep the time information together
with frequency information of the signal after analysis. This time
information allows for detection of fault inception time if the high
frequency transients take a higher value than zero. Fig. 6(c), (h) and
(m) shows clearly the abrupt change of detail 1 (d1) coefficients
for the negative poles upon fault occurrence. This property can be
used to detect any fault in the system. It should be noticed also
that d1 coefficients of the negative and positive poles’ currents at
relays R11 , R21 , and R31 rise abruptly upon fault occurrence. More-
over, d1 coefficients of the negative and positive poles’ currents at
relays R12 , R22 , and R32 rise abruptly upon fault occurrence. There-
fore, the fault will be recognized at both ends of the three sections
using this criterion; which is, a fault is detected if d1 coefficients
for the negative pole or the positive pole increase over 0.4 A. The
fault is detected in a time frame of one sample (10 ␮s) based on
simulations.
4.2. Identification of fault zone
After fault detection, the relay should take a decision about
the zone of the fault. For internal faults, the relay sends a trip
signal to the DC CBs to isolate the faulty section. The converter
control system will adjust the DC power flow accordingly. The
differentiation between internal faults and external faults is done
84 A.E.B. Abu-Elanien et al. / Electric Power Systems Research 130 (2016) 78–88

Positive pole
2000 800 2000

current , A
1000 600 1000

0 400 0
0 0.1 0.2 0.3 0 0.1 0.2 0.3 0 0.1 0.2 0.3
(a) (g) (m)
4
Negative pole

x 10
1 2000 0
current , A

In11 In12
0 0 -5000

-1 -2000 -10000
0 0.1 0.2 0.3 0 0.1 0.2 0.3 0 0.1 0.2 0.3
(b) (h) (n)
4 4 5
x 10
energy indices

x 10 x 10
Positive pole

10 10 4

5 5 2

0 0 0
0 0.1 0.2 0.3 0 0.1 0.2 0.3 0 0.1 0.2 0.3
(c) (i) (o)
6 4 6
x 10 x 10 x 10
energy indices

2 10 2
Negative pole

1 Son1Srn1 5 1

0 0 0
0 0.1 0.2 0.3 0 0.1 0.2 0.3 0 0.1 0.2 0.3
(d) (j) (p)

0.1 0.1 0.04

Positive pole

0.05 0.05 0.02

ratios

0 0 0
0 0.1 0.2 0.3 0 0.1 0.2 0.3 0 0.1 0.2 0.3
(e) (k) (q)

3 1.5 3
Negative pole

0.75 1.5
ratios

0 0 0
0 0.1 0.2 0.3 0 0.1 0.2 0.3 0 0.1 0.2 0.3
(f) (l) (r) Time, s

Fig. 7. Relay output signals for a negative pole to ground fault 195 km from BB1 with 50  fault resistance. (a) Positive pole currents for both ends of section 1. (b) Negative
pole currents for both ends of section 1. (c) Positive pole energy indices for section 1. (d) Negative pole energy indices for section 1. (e) Positive pole ratio for section 1. (f)
Negative pole ratio for section 1. (g) Positive pole currents for both ends of section 2. (h) Negative pole currents for both ends of section 2. (i) Positive pole energy indices for
section 2. (j) Negative pole energy indices for section 2. (k) Positive pole ratio for section 2. (l) Negative pole ratio for section 2. (m) Positive pole currents for both ends of
section 3. (n) Negative pole currents for both ends of section 3. (o) Positive pole energy indices for section 3. (p) Negative pole energy indices for section 3. (q) Positive pole
ratio for section 3. (r) Negative pole ratio for section 3.

using the differential principle explained in Section 2 of this
paper. Upon fault detection, the energy index for both positive and
negative poles will be calculated using DWT approximation. Pos-
itive and Negative poles’ current signals at both ends are already
analyzed using DWT until the 5th level in order to detect any fault.
Detail 1 (d1) to detail 5 (d5) cover a frequency band from 1.5 to
50 kHz. The 5th level approximation is used to calculate the energy
indices.
The energy calculation is repeated at each sample after fault
inception to formulate an energy index curve for each pole at
each line end. This process will continue from fault inception
until 1000 samples after fault detection (i.e. 10 ms). The relay at
each side of the line stores this energy data. After delay com-
pletion (1000 samples), each relay will send these 1000 energy
samples to the other end of each pole. For example, R11 will send
1000 energy samples for both the positive pole and the nega-
tive poles at end 1 to R12 , and R12 will send the 1000 energy
samples for both the positive pole and the negative poles at end
2 to R11 . This process is done in all sections of the MTHVDC system.
This communication process takes some time (e.g. line-of-sight
microwave radio which is a direct link of expected propagation
delay of 1 ms/300 km and the GPS accuracy time is within 0.5 ␮s
with a phase error of 0.01 degrees at 60 Hz [32]). After complet-
ing the sending and receiving process, end 1 relay and end 2 rely
are able to calculate the operating and restraining signals based
on the energy signals sent and received. Eqs. (8) and (9) are used
to calculate the operating and restraining signals. The ratio is also
calculated at each sample for the poles at both ends using (10).
The constant K is found to be 3 based on simulation studies. For
internal faults, the operating signal will be a high value; therefore,
the ratio(s) will be higher than unity for the faulted pole (poles).
As a result, a trip signal will be issued to the DC CB. For exter-
nal faults, the ratios will be less than unity and no trip signal is
needed.
 A.E.B. Abu-Elanien et al. / Electric Power Systems Research 130 (2016) 78–88 85

Fig. 7 shows a negative pole to ground fault 195 km from BB1

with a 50  fault resistance. Fig. 7(a) shows the DWT approxima-
tions of positive pole currents at both ends of section 1. Fig. 7(g)
and (m) shows the same for sections 2 and 3, respectively. It can be
noticed from these figures that the DWT approximations of positive
pole currents at both ends of the three sections coincide with each
other. Fig. 7(b), (h), and (n) depicts the DWT approximations of neg-
ative pole currents at both ends of sections 1, 2, and 3, respectively.
The DWT approximations of the negative pole currents at both ends
coincide for sections 2 and 3; however, the DWT approximations of
the negative pole currents at both ends of section 1 (faulty section)
are distinct. The energy indices based on the DWT approximations
of the positive pole current for both ends of section 1 are shown in
Fig. 7(c). The energy indices for sections 2 and 3 are shown in Fig. 7(i)
and (o). The energy indices of the DWT approximation of the posi-
tive pole currents for both ends of sections 1, 2, and 3 are identical
as shown in Fig. 7(c), (i), (o). The energy indices for the negative pole
currents for sections 2 and 3 are also identical as shown in Fig. 7(j)
and (p). On the other hand, the energy indices for the negative pole
currents for section 1 are dissimilar in shape as can be concluded
from Fig. 7(d). As a result, the discrimination ratios for the posi-
tive pole for all sections are below unity as shown in Figs. 6(e), 7(k)
and (q). Moreover, the discrimination ratios for the negative pole
in sections 2 and 3 are higher than unity for a few samples after
fault inception, after that, they went down below unity as shown
in Fig. 7(l) and (r). The discrimination ratio of the negative pole of
section 1 jumps to a value higher than unity after fault inception,
and the ratio stayed higher than unity for the successive samples
after fault inception. This illustration means that no fault exists in
the positive pole in all sections; no fault exists in the negative pole
of sections 2 and 3; there is a fault in the negative pole of section
1. The energy indices and discrimination ratios are plotted in Fig. 7
for more than 0.15 s following the fault inception to show ratios
robustness.

4.3. Classification of faulted poles

The faulted poles will have a discrimination ratio greater than

unity in the faulted sections. For un-faulted sections, the ratios of
both poles will be below unity. The ratios of faulted poles in the un-
faulted sections may spike to values above unity at the beginning
of the fault, but a few samples after that, the ratios drop to values
below unity. The pole that has a ratio higher than unity for the last
400 samples of the delay is considered to be a faulty pole. For a
single pole to ground fault, the ratio of the faulty pole in the faulty
section will go above unity and stay above unity for the whole 1000
samples. For double pole and double pole to ground faults, the ratios
of the positive and negative poles in the faulty section will go above
unity and stay above unity for the whole 1000 samples. Shapes of
the discrimination ratios of all line sections for different faults will
be given in Section 5.

Fig. 8. Flow chart of the proposed relay algorithm.

4.4. Fault location

Accurate fault location is very important for post fault main-

tenance. The existence of two relays at both ends of each line
facilitates the realization of fault localization using a double ends
traveling wave technique. Faults generate a wide band of high fre-
quency traveling waves that travel in both directions from the fault
location. The traveling waves reach the transmission line ends at
different time instants if the fault is not in the middle of the line,
and at the same instant if the fault occurs exactly at the middle of
the line. By synchronizing the relays at both ends of the line using
Global Positioning System (GPS), the time at which the traveling
waves reach both ends can be recorded and used to locate the fault
point [33];
Assume a fault occurs at some point of the line; the traveling
wave will arrive at both ends at (t1 ) and (t2 ). Assume time t1 is the
arrival time of the traveling wave at the main relay (master relay)
and t2 is the arrival time of the traveling wave at the slave relay.
The master relays in the presented model are R11 , R21 , and R31 .
The distance between the fault point and the master relay point is
calculated using the following equation [33]:
L − v (t2 − t1 )
d= (11)
2
86 A.E.B. Abu-Elanien et al. / Electric Power Systems Research 130 (2016) 78–88

3 3 3 3

Ratios
2 2 2 2
1 1 1 1

0 0 0 0
0 10 20 0 10 20 0 10 20 0 10 20

(a) (d) (g) (j) Rn1

Rn2
Rn3
3 3 Rp1
3 3
Rp2
2
Ratios

2
2 2 Rp3
1 1 1 Unity
1
level
0 0 0 0
0 10 20 0 10 20 0 10 20 0 10 20
(b) (e) (h) (k)

3 3 3
3
Ratios

2 2 2 2
1 1 1 1
0 0 0 0
0 10 20 0 10 20 0 10 20 0 10 20
(c) (f) (i) (l) Time, ms

Fig. 9. Relay ratios at the three sections for different DC faults. (a) p–g fault 3 km from BB1with 30  fault resistance. (b) p–g fault 200 km BB2 with 300  fault impedance.
(c) Bolted p–g fault 30 km BB3. (d) Bolted n–g fault 190 km BB1. (e) Bolted n–g fault 20 km BB2. (f) n–g fault 50 km BB3 with 100  fault impedance. (g) Bolted p–n fault 40 km
BB1. (h) p–n fault 220 km BB2 with 25  fault resistance. (i) Bolted p–n fault 15 km BB3. (j) p–n–g fault 180 km BB1 with 40  fault resistance. (k) Bolted p–n–g fault 10 km
BB2. (l) Bolted p–n–g fault 48 km BB3. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

where d is the fault distance from the master relay point, L is the
transmission line length in km, and v is the propagation speed of
the traveling wave in km/s.
The arrival times of the traveling waves are recorded using d1
of the DWT decomposition of the positive pole or negative pole
signals at both end relays. The detection criterion used to detect
the fault is used here to record the arrival time of the traveling
waves at both ends (t1 and t2 ). Eq. (11) is used to locate the fault.
A flow chart representing the differential relay principle of opera-
tion is shown in Fig. 8. The algorithm starts by loading the moving
window with 1024 (210 ) samples data. The moving window will
move after that sample by sample. At each sample, DWT will be
executed on the data inside the window until a fault is detected. A
fault will be detected if d1 of the DWT of the positive pole or neg-
ative pole exceeds the threshold. If the threshold is exceeded, the
fault inception time is recorded and a trip counter is set to zero.
The counter will be increased by one for each sample. The energy
indices Ep and En are calculated at each sample and recorded. After
1000 samples (i.e. 10 ms), the delay ends and no more energy cal-
culations are done. Relays (R11 , R21 , and R31 ) will wait to receive
the energy data for the 1000 samples for both poles, and also the
fault inception time (traveling wave arrival time) from the relays
at the other line end (R12 , R22 , and R32 ). Relays (R12 , R22 , and R32 )
will receive the same data from (R11 , R21 , and R31 ). Relays (R11 , R21 ,
and R31 ) are considered the master relays for post fault analysis of
fault location. Each relay will calculate the discrimination ratios for
the 1000 sample data for positive and negative poles. If the fault
is internal, the discrimination ratio of the positive pole or negative
pole or both will exceed unity and remain above unity until the
end of the intentional delay (1000 samples). If the fault is external,
the ratio at positive pole or negative pole or both will be below
unity.
The ratio may exceed unity initially after fault inception; how-
ever, it will drop below unity after that toward the end of the 10 ms
delay. The fault is assumed external if the ratio of positive pole or
negative pole or both drops below unity for at least 400 samples
before the end of the intentional delay. The fault is assumed an
internal fault if the ratio of the positive pole or negative pole or
both stays above unity for at least 400 samples before the end of
the intentional delay.
After ratio calculations, fault classification will be done. Pole(s)
involved in the fault will satisfy the internal fault condition, i.e. the
ratio of the faulty pole will be above unity for at least 400 samples
before the end of the intentional delay. After fault classification, a
trip signal is issued to the designated CBs at both ends of the faulty
line. The fault location algorithm can be offline or online before trip
the signal. The algorithm explained in Section 4.4 in this paper is
used for this purpose.
5. Simulation results
Different types of faults with different fault resistances and
different inception times are simulated to verify the proposed
protection algorithm. The three-terminal model is used for the
simulation studies. Numerous faults were simulated to verify
the proposed concepts. For page limitation, samples of these
studies are only presented. 12 cases are presented in Fig. 9 to
show the effectiveness of the presented algorithm. All four fault
types (positive pole to ground, negative pole to ground, positive
pole to negative pole, positive pole to negative pole to ground)
are simulated in each section of the three sections. For each
section, faults are simulated in the first quarter, middle, and last
quarter of the line with different locations. For each fault type,
bolted and high resistance faults are simulated. Fault impedances
are varied from 25  to 300 . A fault impedance of 300  is
tested to show the effectiveness of the proposed technique in
detecting and classifying both low impedance and high impedance
 A.E.B. Abu-Elanien et al. / Electric Power Systems Research 130 (2016) 78–88 87

2000

Current, A
0

-2000
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

(a)
0.2
d1 for positive
pole current 0

-0.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

(b)
d1 for negative

0.2
pole current

-0.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

(c) time (s)

Fig. 10. Wind power jump from 500 MVA to 1000 MVA. (a) Positive and negative poles’ currents at R11 (b) d1 coefficients for positive pole current at R11 (c) d1 coefficients
for negative pole current at R11 .

faults. The detection criterion proposed is very effective in case
of high impedance faults. Although the high-impedance fault
current is low compared with low impedance faults, it contains a
considerable amount of high frequency transients that can be used
effectively in detecting fault and classifying faulted zone.
Fig. 9 shows the discrimination ratios for positive and negative
poles. Actually, the discrimination ratios at both ends are exactly
the same since the absolute value of the energy difference is taken
to calculate the operating signal. Fig. 9(a)–(c) shows the discrim-
ination ratios for positive pole to ground faults in sections 1, 2,
and 3, respectively. The faults are simulated with different fault
inception times and fault resistances. It is clear from the presented
figures that the discrimination ratio for the faulted pole for each
case jumps above unity after fault inception and stays above unity
for the remaining 10 ms delay. The un-faulted poles have two
categories. The ratios of the first category stay below unity from
fault inception until the end of the 10 ms intentional delay, which
means that these poles have no fault. The ratios of the second
category jump above unity after fault inception, then they decay
below unity and stay below unity for more than 4 ms until the
end of the delay, which means that these poles are also un-faulted
poles. Fig. 9(b) represents a high impedance fault (300  fault
impedance. It is clear from the discrimination ratios that this fault
is successfully detected ad classified. Fig. 9(d) to (f) shows the
discrimination ratios for negative pole to ground faults in sections
1, 2, and 3, respectively. Fig. 9(d) to (f) shows the discrimination
ratios for negative pole to ground faults in sections 1, 2, and 3,
respectively. Fig. 9(g) to (i) shows the discrimination ratios for pos-
itive pole to negative pole faults in sections 1, 2, and 3, respectively.
Fig. 9(j) to (l), shows the discrimination ratios for positive pole to
negative pole to ground faults in sections 1, 2, and 3, respectively.
All faults are simulated at different inception times and different
fault resistances. All results show that the ratios of faulted poles
stay above unity after fault inception. Ratios of un-faulted poles
either stay below unity after fault inception or jump above unity
for few samples, then fall below unity for more than 400 samples
before the end of the delay. These results confirm identification of
internal and external faults under different fault conditions.
Wind turbines are subjected to changes to output power. The
proposed protection algorithm is able to differentiate between the
wind generator power variations and fault conditions. Fig. 10(a)
shows the positive pole current and negative pole current of relay
R31 during jump of the wind turbine power from 500 MVA to
1000 MVA at 0.2 s. Fig. 10(b) and (c) shows d1 coefficients of the
DWT decomposition for the positive and negative poles’ currents.
It is clear from the figures, that d1 coefficients for the positive and
negative poles’ current increased slightly over normal case, but they
do not exceed the threshold for operation of protection. This case
study shows that the algorithm will not respond to wind turbine
power variations.

Table 2
Fault location results for the proposed technique.

Fault type Location Fault resistance () t (t2 − t1 ) in samples Calculated location (km) Error (km)

p–g 3 km from BB1 30 64 4.256 1.256

p–g 200 km from BB2 300 −33 199.368 −0.632
p–g 30 km from BB3 0 0 30 0
n–g 190 km from BB1 0 −60 189.76 −0.24
n–g 20 km from BB2 0 88 18.352 −1.648
n–g 50 km from BB3 100 −28 50.824 0.824
p–n 40 km from BB1 0 40 40.16 0.16
p–n 220 km from BB2 25 −48 221.808 1.808
p–n 15 km from BB3 0 20 15.126 0.126
p–n–g 180 km from BB1 40 −54 180.784 0.784
p–n–g 10 km from BB2 0 92 12.368 2.368
p–n–g 48 km from BB3 0 −24 47.849 −0.151
88 A.E.B. Abu-Elanien et al. / Electric Power Systems Research 130 (2016) 78–88
One contribution of this research is to demonstrate the imple-
mentation of a double end fault location technique using GPS
synchronized relays. The arrival times of the first traveling wave
are recorded both sides of the line in order to find the fault location
using (11). Fault location is very important for post fault mainte-
nance operations. The DWT details of the positive and negative
poles at each end are used to record the arrival time of the trav-
eling wave. If the value of the d1 coefficient increases over 0.4, an
abnormal condition is detected and this time is recorded as the
arrival time of the first traveling wave. If the section is identified as
a section with an internal fault, the arrival times of the first trav-
eling waves at both ends of the line are used to find fault location
according to (11). The simulation shows that the propagation speed
in the overhead sections is 299,200 km/s and in the undersea cable
is 148,741.03 km/s. Table 2 shows the calculated fault location and
the error in location for the twelve faults indicated in Fig. 9. Based
on the propagation speed and sampling frequency, the maximum
error is ±2.992 km.
6. Conclusion
A new technique for protection of MTHVDC is presented. This
technique is based on calculating energy contained in the de-noised
signals at both ends of each section of the MTHVDC. Two signals are
calculated in order to identify internal and external faults: operat-
ing and restraining signals. The ratio between the operating and
restraining signals indicates an internal fault if it is larger than
unity; otherwise the fault is external. Double ends traveling waves
technique is used in this research to locate internal faults. Vari-
ous faults with different fault resistances, different fault locations,
and different inception times are simulated to test the performance
of the proposed technique. All simulation results showed that
the proposed technique is reliable in detecting faults, identifying
fault zone, and calculating fault location. The proposed algorithm
attempted to overcome the problems associated with current liter-
ature. The algorithm does not need to shut down the whole system
to perform the protection functions. It works well for bolted and
high resistance faults. Moreover, it is a stable algorithm and does
not respond to power changes of wind generator. Furthermore, the
results of the algorithm for protection of overhead and undersea
cables do not depend on the system parameters.
Acknowledgements
This publication was made possible by NPRP grant NPRP 4-941-
2-356 from the Qatar National Research Fund (a member of Qatar
Foundation). The statements made herein are solely the responsi-
bility of the authors.
References
[1] H. Livani, C.Y. Evrenosoglu, A single-ended fault location method for segmented
HVDC transmission line, EPSR 107 (2014) 190–198.
[2] D. Schmitt, Y. Wang, T. Weyh, R. Marquardt, DC-side fault current management
in extended multiterminal-HVDC-grids, in: Conf. SSD, 2012, pp. 1–5.
[3] M. Hadjikypris, V. Terzija, Transient fault studies in a multi-terminal VSC-
HVDC grid utilizing protection means through DC circuit breakers, in: Conf.
POWERTECH, 2013, pp. 1–6.
[4] G. Song, X. Chu, X. Cai, S. Gao, A novel pilot protection principle for VSC-HVDC
cable lines based on fault component current, Int. J. Electr. Power Energy Syst.
53 (2013) 426–433.
[5] X.-D. Zheng, N.-L. Tai, J.S. Thorp, G.-L. Yang, A transient harmonic current pro-
tection scheme for HVDC transmission line, IEEE Trans. Power Delivery 27
(2012) 2278–2285.
[6] Z. Xiaodong, N. Tai, G. Yang, H. Ding, A transient protection scheme for HVDC
transmission line, IEEE Trans. Power Delivery 27 (2012) 718–724.
[7] J. Suonan, J. Zhang, Z. Jiao, L. Yang, G. Song, Distance protection for HVDC trans-
mission lines considering frequency-dependent parameters, IEEE Trans. Power
Delivery 28 (2013) 723–732.
[8] L. Xiaolei, A.H. Osman, O.P. Malik, Hybrid traveling wave/boundary protection
for monopolar HVDC line, IEEE Trans. Power Delivery 24 (2009) 569–578.
[9] J.I. Marvik, H.G. Svendsen, Analysis of grid faults in offshore wind farm with
HVDC connection, Energy Procedia 35 (2013) 81–90.
[10] T. Lianxiang, O. Boon-Teck, Locating and isolating DC faults in multi-terminal
DC systems, IEEE Trans. Power Delivery 22 (2007) 1877–1884.
[11] Y. Jin, J.E. Fletcher, J. O’Reilly, Multi-terminal DC wind farm collection
and transmission system internal fault analysis, in: Conf. ISIE, 2010,
pp. 2437–2442.
[12] J. Yang, J.E. Fletcher, J. O’Reilly, Multiterminal DC wind farm collection grid
internal fault analysis and protection design, IEEE Trans. Power Delivery 25
(2010) 2308–2318.
[13] T. Lianxiang, O. Boon-Teck, Protection of VSC-multi-terminal HVDC against DC
faults, in: Conf. PESC02, Vol. 2, 2002, pp. 719–724.
[14] K. De Kerf, et al., Wavelet-based protection strategy for DC faults in multi-
terminal VSC HVDC systems, IET Gen. Trans. Distrib. 5 (2011) 496–503.
[15] J. Candelaria, P. Jae-Do, VSC-HVDC system protection: a review of current meth-
ods, in: Conf. PSCE, 2011, pp. 1–7.
[16] W. Yeqi, R. Marquardt, Future HVDC-grids employing modular multilevel con-
verters and hybrid DC-breakers, in: Conf. EPE, 2013, pp. 1–8.
[17] N. Flourentzou, V.G. Agelidis, G.D. Demetriades, VSC-based HVDC power
transmission systems: an overview, IEEE Trans. Power Electron. 24 (2009)
592–602.
[18] S.K. Chaudhary, R. Teodorescu, P. Rodriguez, Wind farm grid integration using
VSC based HVDC transmission—an overview, in: Conf. Energy 2030 Conference,
2008, pp. 1–7.
[19] A. Abdel-Khalik, et al., Optimum power transmission-based droop control
design for multi-terminal HVDC of offshore wind farms, IEEE Trans. Power Syst.
28 (2013) 3401–3409.
[20] M. Prodanovic, T.C. Green, Control and filter design of three-phase inverters
for high power quality grid connection, IEEE Trans. Power Electron. 18 (2003)
373–380.
[21] Y. Jin, J.E. Fletcher, J. O’Reilly, Short-circuit and ground fault analyses and loca-
tion in VSC-based DC network cables, IEEE Trans. Ind. Electron. 59 (2012)
3827–3837.
[22] D. Guanjun, T. Guangfu, H. Zhiyuan, D. Ming, New technologies of voltage source
converter (VSC) for HVDC transmission system based on VSC, in: Conf. Power
General Meeting, 2008, pp. 1–8.
[23] A. Elserougi, A. Abdel-Khalik, A. Massoud, S. Ahmed, A new protection scheme
for HVDC converters against DC-side faults with current suppression capability,
IEEE Trans. Power Delivery 29 (2014) 1569–1577.
[24] L. Xiaoqian, S. Qiang, L. Wenhua, R. Hong, X. Shukai, L. Licheng, Protection of
nonpermanent faults on DC overhead lines in MMC-based HVDC systems, IEEE
Trans. Power Delivery 28 (2013) 483–490.
[25] Andreas Baschnagel, Ulrich Schlapbach, Surge currents for IGBT Diodes, ABB
Switzerland Ltd, Switzerland, 2007 (Application Note 5SYA 2058-01).
[26] A.T. Johns, R.K. Aggarwal, Z.Q. Bo, Non-unit protection technique for EHV trans-
mission systems based on fault-generated noise. Part 1: Signal measurement,
IET Gen. Trans. Distrib. 141 (1994) 133–140.
[27] R. Mardiana, H.A. Motairy, C.Q. Su, Ground fault location on a transmission
line using high-frequency transient voltages, IEEE Trans. Power Del. 26 (2011)
1298–1299.
[28] D.W.P. Thomas, C. Christopoulos, Ultra-high speed protection of series com-
pensated lines, IEEE Trans. Power Delivery 7 (1992) 139–145.
[29] P. Jafarian, M. Sanaye-Pasand, High-frequency transients-based protection of
multiterminal transmission lines using the SVM technique, IEEE Trans. Power
Delivery 28 (2013) 188–196.
[30] A.D. Poularikas, Transforms and Applications Handbook, third ed., CRC Press,
Boca Raton, FL, 2010.
[31] T.B. Littler, D.J. Morrow, Wavelets for the analysis and compression of power
system disturbances, IEEE Trans. Power Delivery 14 (1999) 358–364.
[32] C. Xu, High Accuracy Real-time GPS Synchronized Frequency Measurement
Device for Wide-area Power Grid Monitoring, Electrical and Computer Engi-
neering Department, Virginia Polytechnic Institute, Virginia, 2006 (Ph.D.
Dissertation).
[33] H. Lee, A.M. Mousa, GPS travelling wave fault locator systems: investigation into
the anomalous measurements related to lightning strikes, IEEE Trans. Power
Delivery 11 (1996) 1214–1223.