Digital Object Identifier:
10.1109/TPWRD.2016.2543306
URL:
http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=7435320
© 2016 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained
for all other users, including reprinting/ republishing this material for advertising or promotional
purposes, creating new collective works for resale or redistribution to servers or lists, or reuse of
any copyrighted components of this work in other works.
A Local Backup Protection Algorithm for HVDC
Grids
Willem Leterme, Graduate Student Member, IEEE, Sahar Pirooz Azad, Member, IEEE
and Dirk Van Hertem, Senior Member, IEEE
Abstract—Dc faults in HVDC grids lead to quickly increasing currents which should be interrupted sufficiently fast to prevent damages to power electronics components. Although several primary relaying algorithms for HVDC grids have been proposed, fast backup relaying algorithms are needed to ensure system reliability when primary protection fails. This paper proposes a local backup relaying algorithm for HVDC grids, which leads to a short delay between primary and backup protective actions. The proposed algorithm, consisting of breaker and relay failure subsystems, uses classifiers which detect primary protection mal-functions based on the voltage and current waveforms associated with dc breaker operation. The algorithm initiates detection of uncleared faults during primary protection operation, which results in accelerated actions by the backup protection after primary protection failure. The proposed algorithm is applied to a four-terminal HVDC grid. Study results show that the proposed algorithm accurately detects uncleared faults, identifies the source of primary protection malfunction and expedites backup protective actions by operating during the fault current interruption interval of the primary protection.
Index Terms—backup protection, breaker failure, HVDC grid, power system protection, protective relaying algorithm
I. INTRODUCTION
HVDC grids are expected to be a part of the future power system, either to facilitate the integration of offshore wind power or to upgrade the existing ac system [1], [2]. The pro-tection of an HVDC grid fulfills the same objectives as those of an ac system, e.g., minimizing the impact of faulty elements on grid operation and preventing damages to system equipment. However, protection of an HVDC grid is more difficult than the ac system as prospective dc fault currents rapidly reach high values. These currents should be quickly interrupted to prevent damages to the power electronics components of the HVDC converters [3]. Although recent advances in dc breaker technology have enabled fast fault current interruption, current interruption capabilities of existing prototypes are limited [4]– [7]. Consequently, fast protective relaying algorithms are re-quired to promptly detect the fault and identify its location.
Primary protection provides the fastest fault clearance for elements within the primary protection zone whereas backup The work of W. Leterme is supported by a Ph.D. fellowship from the Re-search Foundation-Flanders (FWO). The reRe-search leading to these results has received funding People Programme (Marie Curie Actions) of the European Union’s Seventh Framework Programme FP7/2007-2013/ under REA grant agreement no. 317221, project title MEDOW.
W. Leterme, S. Pirooz Azad and D. Van Hertem are with KU Leuven, Bel-gium (EnergyVille/Electa research group, Electrical Engineering Department ESAT, Kasteelpark Arenberg 10 (PB2445), 3001 Heverlee).
Contact: [email protected],
[email protected], [email protected]
protection must be provided in case of primary protection fail-ure. The primary protection can fail due to the malfunction of breakers or relays. Fast selective primary relaying algorithms, which detect and discriminate faults within a time-frame of milliseconds, have been investigated in several studies [8]– [11].
The main requirement of HVDC grid backup protection is the high operation speed. A delay in the backup operation will result in dc fault currents with high steady-state values that existing breakers with limited interruption capability will no longer be able to clear [12]. However, a delay between primary and backup protective actions is required to prevent backup prior to primary actions [13].
HVDC grid backup relaying algorithms proposed in liter-ature do not provide a high operation speed. The algorithm proposed in [8] delays backup actions until the primary protection reduces the fault current to a near-zero value. This results in a considerable delay on backup actions as breaker series inductors also limit the rate of change of the current during fault current interruption. In [14], low speed HVDC grid protection based on mechanical dc breakers is proposed. The fault clearing times of the primary and backup protection are 60 and 20 ms, respectively. The backup protection of [14] requires fault current limiting equipment, which results in additional costs.
This paper proposes a fast local backup protection algorithm for HVDC grids, which consists of separate breaker and relay failure subsystems. The backup protection involves in backup actions performed by relays at the local bus rather than those at remote buses [15]. This paper is the first study on HVDC grid backup protection algorithms, which considers both relay and breaker malfunctions, and takes into account the HVDC grid backup protection requirements. The proposed algorithms do not require additional equipment such as fault current limiters. The proposed algorithm is applicable to various topologies of HVDC grids which use a selective primary protection system and fast dc breakers.
The main feature of the proposed algorithms is the resulting short delay between primary and backup protective actions. This is achieved through detecting uncleared faults immedi-ately after initiation of fault current interruption by primary protection, rather than delaying uncleared fault detection until primary protection failure.
The paper is structured as follows: Section II presents local backup protection in ac systems. Section III introduces the proposed local backup protection algorithm for HVDC grids. Furthermore, in this section, the principles of the proposed
Primary protection ∆tpb Backup protection Primary breaker opens
Backup trip signal
t
Fault inception
tdp top tcp tsb tdb tcb
tf
Backup breakers open
tob
Primary trip signal Fault current interruption
Fault current interruption
Fig. 1. Primary and backup protection continua in ac systems [15].
algorithm are compared with those commonly used in ac systems. Section IV describes the four-terminal HVDC grid test system used to evaluate the performance of the proposed algorithm. In Section V, the proposed algorithm is applied to the test system and its performance in case of breaker or relay failure is evaluated. Section VI presents the conclusions.
II. LOCALBACKUPPROTECTION INACSYSTEMS A. Backup Protective Actions in Ac Systems
In ac systems, the fault clearance time in case of primary protection failure is the sum of the primary fault clearance time, a delay between primary and backup actions and the backup protection fault clearance time (Fig. 1) [13]. The primary protection system detects the fault at td
p, which is
the sum of tf, time required for the fault wave to reach the
relay, processing time to detect the fault, and time to send a trip signal to the associated breaker. The associated breaker opens at to
p and interrupts the fault current at tcp. The backup
protection initiates detection of primary protection failure at ts
b, which is the sum of tcp and a delay ∆tpb. If primary
protection failure is detected at td
b, the backup protection trips
the breakers on all lines adjacent to the faulted line. The adjacent breakers open at to
b and interrupt the fault current at
tc
b. The total time required to clear the fault in case of primary
protection failure is tc b− tf.
While in ac systems ∆tpb is used to provide a margin
between primary and backup actions, the delay on backup actions in HVDC grids should be minimized. In ac backup protection, ∆tpb is typically set to multiple cycles of the
fundamental frequency [15] to avoid tripping of multiple circuit breakers due to misoperation of the backup protection. For HVDC grid protection, such a large margin would lead to impractical requirements on ratings of dc breakers and HVDC converters.
B. Local Backup Protection Schemes in Ac Systems
Fig. 2 presents one backup protection scheme for breaker failure and two backup protection schemes in case of primary relay failure. In Fig. 2, the breaker associated with the primary relay is identified as B1 and the adjacent breakers are B2-B4.
The measurements (Mi) for the primary and backup protection
are identified by superscript p and b, respectively.
ure is identified if the current through the breaker associated with the primary relay exceeds a threshold after a certain time following fault detection by the primary relay. Therefore, in an ac system, breaker failure can only be detected after tpc. If
the backup protection system detects a breaker failure, it will trip all adjacent breakers (Fig. 2 (a)).
2) Relay Failure Backup Protection: Two backup protec-tion schemes are proposed in literature to deal with primary relay failures [15]: (i) duplication of the primary protection system and (ii) reverse reach. The primary relay might fail to detect faults due to the failure of transducers, communication system or the relaying algorithm itself [13].
a) Duplication of Primary Protection System: In prac-tice, it is possible to duplicate all protection equipment such as transducers and relays, except for the circuit breakers [15] (Fig. 2 (b)). To avoid common mode failure, the backup relay should use a different protection principle than the primary relay. The main advantage of this backup protection is its fast operation, but this comes at the cost of additional transducers and relaying equipment.
b) Reverse Reach: In this backup scheme, adjacent re-lays provide local backup protection for the faulty relay by setting a reverse reach [16]. As an example, in ac systems, reverse reach can be provided by distance protection if zone 3 is set to detect faults in the backward direction of the relay [13]. The adjacent relays detect the fault independently of the primary relay and, in case of an undetected fault, trip all the breakers connected to the local bus (Fig. 2 (c)). This backup protection scheme does not require extra equipment, but is slower compared to duplication of the primary protection. Furthermore, the entire bus is lost in case of an undetected fault on any of the connected lines.
III. HVDC GRIDLOCALBACKUPPROTECTION
The proposed backup protection algorithm consists of two separate subsystems which utilize classifiers to detect (i) fail-ure of the breaker associated with the primary relay (Fig. 2a) and (ii) failure of the relays on adjacent lines (Fig. 2c). A. Loci of Dc Fault Currents and Voltages with Primary and/or Backup Protection in Service
A short-circuit fault in an HVDC grid, in the initial stage, can be characterized by an increasing current and a decreasing voltage in the faulted pole. In case of pole-to-ground faults in a high impedance grounded system, the fault current in the faulted pole oscillates and returns to zero. In case of a pole-to-ground fault in a low impedance grounded system or pole-to-pole fault, the current increases to a high prospective steady-state value. Fig. 3 provides an overview of the loci of the dc voltage and current for a cleared and an uncleared fault as a function of time (assuming a high prospective steady-state fault current). To simplify Fig. 3, oscillations caused by electromagnetic effects were excluded. In Fig. 3, points 1 and 2 correspond to fault inception and breaker trip instants, tf
and to
p, respectively. The solid line from point 1 to point 2
B1 B2 B3 B4 Primary relay Backup relay M1p Trip Signal Measurement signal
(a) Backup protection for breaker failure
B1 Primary relay Backup relay M1p Mb 1 B2 B3 B4
(b) Backup protection for relay failure-duplication of the primary relay
B1 B2 Backup relay M2b B3M3b B4M4b Primary relay M1p
(c) Backup protection for relay failure -reverse reach
Fig. 2. Backup protection schemes for relay and breaker failures
U I 1 Ifs Ifo,p If0 2 3 5 tf top tcp 4 tob Ifo,b Uf0 Ufo,p Ufo,b Ufs t s f 6 tcb 6’
Fig. 3. Conceptual sketch of loci of the dc fault currents and voltages at the primary and adjacent relay for a cleared and an uncleared fault
primary relay during the [tf top]interval as the fault is detected
by the primary protection system and the primary breaker trip signal is generated. The dashed line between points 2 and 5 shows the loci of dc voltages and currents after the primary breaker trips and the faulty line is removed. As the fault is being cleared, the voltage increases from Uo,p
f to Uf0 and the
current decreases from Io,p f to 0.
In Fig. 3, the solid line from point 2 to point 4 shows the loci of dc voltages and currents measured at the primary relay after primary protection failure. For an uncleared fault, the dc voltage decreases from Uo,p
f to Ufs and the current increases
until a steady-state value, Is
f, is reached at point 4.
If the backup protection clears the fault after primary protection failure, the dc voltages and currents will follow the path from point 2 to point 3 and then to point 6 (in case of relay failure) or 6’ (in case of breaker failure). Based on the backup relaying algorithm, the backup protective actions begin at to
p and continue through the [top, tsb] interval. During
the [to
p, tob] interval (from point 2 to 3) the current increases
from Io,p f to I
o,b
f . If the backup protection does not operate
fast enough, Io,b
f will exceed the current interruption capability
of dc breakers and the fault cannot be cleared. For existing dc breaker prototypes, the maximum interruptible current is typically reached within a few milliseconds [4]–[6]. The dotted line from point 3 to point 6, shows dc currents and voltages at the primary relay as all transmission lines connected to the same bus as the faulty line are disconnected to clear the fault. At point 6 (tc
b), the backup protection clears the fault and the
dc fault current becomes zero.
Primary protection Primary trip signal
Primary breaker opens
Backup trip signal
t
Fault inception
tdp top= tsb tˆdb tcp ˆtcb
tf
Backup breakers open ˆ
tob Fault clearance time by backup protection
Backup protection Fault current
interruption
Fault current interruption
Fig. 4. Primary and proposed local backup protection continua in HVDC grids
B. Backup Protection Principles
The proposed local backup protection algorithm achieves a high operation speed by detecting primary protection failure immediately after the instant at which the primary protection is expected to initiate fault current interruption (Fig. 4). In Fig. 4, similar to Fig. 3, tf, tdpand topcorrespond to the fault inception,
fault detection and expected breaker opening instants. From the instant the primary protection is expected to initiate fault current interruption, to
p, the local backup relaying algorithm
starts tracking uncleared faults. If the primary protection succeeds in interrupting the fault current, the backup relaying algorithm detects a cleared fault and no backup protective actions are carried out. If the primary protection fails, the backup relaying algorithm detects an uncleared fault at ˆtd
b and
trips the adjacent breakers to interrupt the fault current. The adjacent breakers open at ˆto
b and the backup protection clears
the fault at ˆtc b.
Compared to algorithms based on ac backup protection philosophy, the proposed algorithm, poses less stringent re-quirements on ratings of HVDC grid components as the fault clearance time, and hence the system’s exposure to high dc voltages and currents, is shorter (ˆtc
b < tcb). A comparison
between Figs. 1 and 4 shows that the difference between tc b
and ˆtc
b is the sum of the fault current interruption time by
the primary protection (tc
p− top) and the time delay between
primary and backup protective actions ∆tpb, i.e., tcp−top+∆tpb.
C. Fault Classification
The proposed backup relaying algorithm, consisting of separate breaker and relay failure subsystems, is a data-based
to detect the primary protection failure. The breaker and relay failure detection subsystems use classifiers to detect faults at various locations in the grid [17]. The inputs and outputs of the classifiers of each failure detection subsystem are explained in Sections III-C2-III-C3.
1) Classification procedure and algorithm: Various classifi-cation methods can be used for the proposed backup protection algorithm. In all classification methods, two types of data, i.e. training and test, are used. The training data is used to prepare the classifier for the specific application, by assigning a class number to each sample of the training data, and the test data is used to examine the classifier’s performance. To properly train the classifier, the training data should encompass samples from all relevant system operating scenarios. A number of features and a class number are associated with each data sample. The features for both training and test data are extracted from system measurements.
The classifier used in this paper is the K nearest neighbor (KNN), as it provides a simple yet effective classification method [18]. A KNN classifier identifies the class of the unseen sample as the class majority of the k nearest neighbors. A nearest neighbor is defined as the sample from the training set that has the minimal distance to the unseen sample. The distance metric used in this paper is the Euclidean distance. The parameter k is a design parameter of the classifier and can be optimized based on the training samples.
2) Breaker failure detection: The classifier associated with the breaker failure subsystem uses three features from the current and voltage measurements at the primary protection zone; the instantaneous voltage and current and a time tag. To eliminate the impact of the pre-fault operating conditions, the pre-fault currents and voltages are subtracted from the instantaneous quantities. The time tag shows when the sample is taken with respect to the fault detection instant. The time tag can have three values (0, 1 and 2) which correspond to samples taken before td
p (0), during the [tdp top] interval (1),
and after to
p (2). The time tags of the training data are known
and the default time tag of the test samples is 0.
For the test data, the fault inception instant is found using the traveling wave propagation theory, in particular, the wave front concept [19]. Based on this theory, when a fault occurs, current and voltage waves which propagate over the transmission lines, are generated. The fault instant is detected as the wave front is identified. To identify the wave front, the difference between each two subsequent current samples is calculated. If the difference exceeds a threshold, a fault is detected [20]. The accuracy of the determined fault inception instant depends on the sampling frequency of the measurements.
The breaker failure classifier identifies whether the primary breaker operated or the fault remains uncleared although it was detected by the primary relay. This classifier assigns a class number to each unseen data sample (Fig. 5):
• Classbf 0: no fault is detected in the primary protection
zone,
• Classbf 1: fault is detected in the primary protection
zone, but fault current is not interrupted by the primary
tf tdp top
Classbf0: No Fault Classbf1: Detected fault
Classbf3: Breaker fails
t
*
!
Fig. 5. Classifier outputs for breaker failure detection (*: alert state, !: action state)
breaker,
• Classbf 2: the primary breaker operated and fault will
be cleared, and
• Classbf 3: the primary breaker failed to remove the
faulty line.
Classes 1 and 3 indicate the alert and action states, respec-tively. In the alert state, backup actions are delayed until a decision on primary protection failure is made. In the action state, a trip signal is sent to all adjacent breakers. Classes 0 and 2 indicate that no action from the breaker failure backup protection is required.
3) Relay failure detection: The classifier associated with the relay failure subsystem is used to detect uncleared faults in the reverse backup protection zone of the relay. Such an uncleared fault is due to the failure of an adjacent relay for which the fault is in the primary protection zone.
The inputs to the relay failure classifier are the instantaneous current and voltage measurements and a time tag. The time tag determines the relative instant of the data sample with respect to the fault detection instant and takes similar values to that of the breaker failure classifier.
The output of each classifier can obtain six values (Fig. 6):
• Classrf 0: no fault is detected,
• Classrf 1: fault is detected in the primary protection
zone,
• Classrf 2: fault is detected in the reverse backup
pro-tection zone,
• Classrf 3: fault in the primary protection zone is not
cleared,
• Classrf 4: fault is cleared, and
• Classrf 5: fault in the reverse backup protection zone is
not cleared.
Classes 2 and 5 indicate the alert and action states, respec-tively. Classes 0, 1, 3 and 4 do not require any protective action from the relay failure backup protection, but are needed for proper operation of the classification algorithm.
D. Summary
Fig. 7 summarizes the backup relaying algorithm steps and the inputs and outputs of the backup protection system.
These steps are
• feature extraction: Three features, i.e. time tag,
instanta-neous voltage and current, are extracted from the mea-surements and at each sampling instant provided to the subsystems. The time tags for the breaker and relay failure subsystems are generated by timers TBF and TBF
which are activated through fault detection by (i) the primary relay for the breaker failure subsystem or (ii) the
tf tdp top
Classrf0: No fault
t
Classrf4: Cleared fault
Classrf5: Uncleared fault
in backup protection zone
Classrf1: Detected fault in primary protection zone Classrf2: Detected fault in backup protection zone
Classrf3: Uncleared fault
in primary protection zone
*
!
Fig. 6. Classifier outputs for relay failure detection (*: alert state, !: action state) Timer TBF Primary relay Trip Signal Breaker failure classifier Voltage Current Time tag tBF> ∆tBF tBF
Feature extraction Classification Relay logic
tBF
Classbf=3
Start
(a) Breaker failure subsystem
Timer TRF Trip Signal Relay failure classifier Voltage Current Time tag tRF> ∆tRF tRF
Feature extraction Classification Relay logic
Adjacent relays
Classrf=5
Classrf=2
Start tRF
(b) Relay failure subsystem Fig. 7. Block diagram of proposed backup protection system
adjacent relays or the relay failure classifier, respectively. The relay failure classifier detects a fault in the reverse backup zone if the class number becomes 2.
• classification: The classifier associated with each
subsys-tem assigns a class number to the data sample.
• relay logic: The breaker failure and relay failure
subsys-tem generate a trip signal if the class number is 3 or 5 and the timers TBF or TRF exceed a threshold, ∆tBF or
∆tRF, respectively. The breaker failure trip signal is sent
to the breakers on all lines adjacent to the faulted one. The relay failure trip signal is sent to the breakers on the adjacent lines as well as the breaker on the faulted line. The thresholds ∆tBFand ∆tRFon timers TBF and TRF are
used to ensure operation of the backup protection after the primary protection by controlling the time instants, ˆtd
b{BF}
and ˆtd
b{RF}, at which the backup protection starts to interrupt
the fault current. The minimum value for ∆tBF in the breaker
failure subsystem is the primary breaker opening time, i.e. to p−
tdp. In the relay failure subsystem, ∆tRF is equal to the sum of
the primary relay detection time (td
p−tf), breaker opening time,
(to
p− tdp) and a safety margin to account for the uncertainties
on the instant of fault detection by the adjacent relays or the relay failure classifier.
To ensure operation of the breaker failure subsystem prior to the relay failure subsystem, the thresholds ∆tBF and ∆tRF
are set such that ˆtd
b{RF} > ˆtdb{BF}. The reason for this
constraint is that the breaker failure subsystem is more robust than the relay failure subsystem (as the latter detects faults in
R I U R I U R I U B B B
Fig. 8. Single busbar layout with three infeeds, each with an inductor, a primary relay, a breaker and current and voltage sensors
the reverse backup protection zone whereas the former detects faults in the primary protection zone).
IV. TEST SYSTEM
The test system considered in the studies of this paper is the four-bus meshed HVDC grid of [21] (Fig. 9). In all the stations, half-bridge modular multilevel converters with sym-metric configuration are used. The converters are represented by a continuous model which holds IGBT blocking capability. A detailed description of the models for converters, breakers and cables is given in [21]. All ac and dc system parameters are equal to those of [21], except for the series inductors which are considered to be 50 mH. The simulation software used for the studies is PSCAD [22].
Fig. 8 shows the busbar layout and location of measure-ments considered in this study. Hybrid dc breakers, which can interrupt the fault current in 1.7 ms after fault detection, are inserted at the end of each dc transmission line and at the converter’s dc terminals. The dc breakers are assumed to have bidirectional current interruption capability [4], [23]. The pole-to-ground voltage and pole current measurements are assumed to be available at the line end of each inductor. The sign of the current is positive (negative) if the current flows in the direction of the transmission line (bus).
The sample set of each classifier consists of various fault instances generated by applying faults at the two ends of all transmission lines and several points between the two ends (25 km equally-spaced points). To evaluate the classification algorithm, 70% and 30% of the samples are used as training and testing data, respectively. Each classifier was trained for the grid topology as shown in Fig. 9. The classifier does not require retraining for proper operation during converter or line outages. However, the training procedure should be repeated in case of structural changes in the grid, e.g. a change in grid topology or the value of the breaker series inductor.
The proposed backup relaying algorithm is tested for all sys-tem relays and results for R13are provided in this paper. For
the breaker failure backup protection system, measurements at R13 are used. The relay failure backup protection system
uses measurements at R12 and R14. The sampling time for
voltage and current measurements is 0.02 ms, corresponding to a sampling frequency of 50 kHz.
The faults considered in this study are to-pole and pole-to-ground faults on the dc side which occur at 0.025 s. These are detected by the primary protection system after 0.3 ms, which lies within the detection time range of recently proposed algorithms for primary protection [8], [9]. The relaying algo-rithm proposed in [21] is used for primary protection and the
=
Bus 3 Bus 4 F131 F132 F139 25 km L13(200 km) F133 25 km B13 B12=
=
=
L14(200 km) L12(100 km) L24(150 km) L34(100 km) B14 B43 B42 B41 B31 B34 B21 B24 B1c B3c B2c B4cFig. 9. HVDC grid test system
primary protection zone for each relay encompasses the entire transmission line for which the relay provides protection.
To deal with pole-to-ground as well as pole-to-pole faults, separate classifiers are used for positive and negative pole. Although in this paper, only results for pole-to-pole faults are shown, the conclusions obtained from these results can be extended to pole-to-ground faults as well, since in the first milliseconds after fault inception, the voltage and current waveforms in the faulted pole are similar for both faults [21].
V. STUDYRESULTS
A. Breaker Failure
Fig. 10 demonstrates the breaker failure subsystem of R13
for two fault scenarios on line L13 (a solid fault 50 km from
bus 3) where B13 (i) clears the fault (Fig. 10 (a)) or (ii) fails
to clear the fault (Fig. 10 (b)).
The breaker failure classifier distinguishes cleared from uncleared faults within the time frame of primary protection operation and before the current through the primary breaker becomes zero (Fig. 10). Before td
p, the class number is 0 which
shows that no fault is detected. During the interval [td ptop], the
class number is 1 for both scenarios, which indicates that the fault is detected by the primary relay but not yet cleared by the primary breaker. In this stage, the breaker failure subsystem is in the alert state and does not initiate backup actions. At to
p,
i.e. td
p+1.7 ms, the class number becomes 3 and 2 for scenario
(i) and (ii), respectively, indicating that the fault will either be cleared or remains uncleared due to breaker failure. If the class number remains 3, a trip signal is sent to B12 and B14.
For scenario (ii), the breaker failure subsystem ensures a fast response by generating a trip signal for B12 and B14
immediately after B13 fails and the classifier output becomes
3 (Fig. 10 (b)-(c)). This requires a time delay ∆tBF equal to
1.7 ms, i.e. the breaker opening time. Although the classes are separable directly after to
p, a longer delay can be used to
increase the protection system’s reliability [17]. B. Relay Failure
Figs. 11-14 illustrate the backup protection provided by R12
for failure of R13and R14for three scenarios: (i) fault on line
0.024 0.026 0.028 0.03 0.032 0.034 Current I 13 p (kA) 0 5 10 Cleared Uncleared 0.024 0.026 0.028 0.03 0.032 0.034 Voltage U 13 p (kV) 0 300 600 (b) t(s) 0.024 0.026 0.028 0.03 0.032 0.034 Class bf 0 1 2 3 (c) Trip Signal 0 1 Cleared Uncleared
Fig. 10. Breaker failure detection: (a)-(b) positive pole dc current and voltage
at R13 for scenarios (i) and (ii), and (c) classifier output (left y-axis) for
scenarios (i) and (ii) and breaker failure trip signal for scenario (ii) (right y-axis) 0.024 0.026 0.028 0.03 0.032 0.034 Current I 12 p (kA) -5 0 (a) Cleared Uncleared 0.024 0.026 0.028 0.03 0.032 0.034 Voltage U 12 p (kV) 0 150 300 450 (b) t(s) 0.024 0.026 0.028 0.03 0.032 0.034 Class rf 0 1 2 3 4 5 (c) Trip Signal 0 1 Cleared Uncleared
Fig. 11. Relay failure detection for scenario (i): (a)-(b) positive pole dc current
and voltage at R12and (c) classifier output (left y-axis) and relay failure trip
signal (right y-axis)
L13 at bus 3 (ii) fault on line L14 (100 km from bus 1) and
(iii) fault on line L12at bus 1. Each fault scenario studies two
cases where the fault is either detected or not detected by the primary relay. Figs. 11-14 show the output of the classifier associated with R12.
For scenarios (i) and (ii), the relay failure classifier dis-tinguishes cleared from uncleared faults shortly after the instant the primary protection is expected to start fault current interruption (Figs. 11 and 12 (a) and (b)). In both scenarios, the class number is 0 before fault detection and becomes 2 after the fault is detected in the reverse backup zone. During the time interval in which B13 or B14 are expected to open,
i.e. [td
p top], the relay failure subsystem is in the alert state
but does not initiate backup actions. Immediately after to p, the
class number becomes 5 for both uncleared and cleared fault scenarios, indicating that the relay failure subsystem cannot instantly make a correct decision. In less than 0.25 ms from to
p,
the class number for the cleared faults changes to 4, indicating that the primary protection has dealt with the fault. If the class
0.024 0.026 0.028 0.03 0.032 0.034 Current I 12 p (kA) -5 0 (a) Cleared Uncleared 0.024 0.026 0.028 0.03 0.032 0.034 Voltage U 12 p (kV) 0 150 300 450 (b) t(s) 0.024 0.026 0.028 0.03 0.032 0.034 Class rf 0 1 2 3 4 5 (c) Trip Signal 0 1 Cleared Uncleared
Fig. 12. Relay failure detection for scenario (ii): (a)-(b) positive pole dc
current and voltage at R12 and (c) classifier output (left y-axis) and relay
failure trip signal (right y-axis)
0.024 0.026 0.028 0.03 0.032 0.034 Current I 12 p (kA) -5 0 (a) Cleared Uncleared 0.024 0.026 0.028 0.03 0.032 0.034 Voltage U 12 p (kV) 0 150 300 450 (b) t(s) 0.024 0.026 0.028 0.03 0.032 0.034 Class rf 0 1 2 3 4 5 (c) Trip Signal 0 1 Cleared Uncleared
Fig. 13. Relay failure detection, with a sampling frequency of 10 kHz, for
scenario (i): (a) positive pole dc current at R12, (b) positive pole voltage at
R12and (c) classifier output (left y-axis) and relay failure trip signal (right
y-axis)
number remains 5, the relay failure subsystem trips all breakers connected to bus 1.
To avoid tripping of all breakers at bus 1 in case the classifier output becomes 5 while the fault is being cleared by B13, ∆tRF is set to 3 ms (Figs. 11-12 (b) and (c)). This
time delay provides a margin of 1 ms on top of the primary fault detection and breaker opening time (0.3 and 1.7 ms, respectively) to account for the delay in detection of faults in the backup protection zone and the uncertainty on the classifier output shortly after to
p.
The relay failure subsystem also acts correctly for a sam-pling frequency of 10 kHz (Fig. 13). For cleared and uncleared faults, the class numbers follow the same sequence as for a sampling frequency of 50 kHz (Fig. 13 (a)-(b)). The time delay ∆tBF of 3 ms provides a sufficient margin to generate the
correct trip signals for cleared and uncleared faults (Fig. 13 (c)).
For scenario (iii), the relay failure classifier correctly distin-guishes faults in the primary protection zone from those in the
0.024 0.026 0.028 0.03 0.032 0.034 I12 (kA) 0 10 (a) Class rf Number 0 1 2 3 4 5 I12 Class Number t(s) 0.024 0.026 0.028 0.03 0.032 0.034 I12 (kA) 0 5 10 15 (b) Class rf Number 0 1 2 3 4 5 I12 Class Number
Fig. 14. Relay failure detection for scenario (iii): dc fault currents at R12
(left y-axis) and classifier output (right y-axis) for (a) a detected fault and (b) an undetected fault Current (kA) -5 -4 -3 -2 -1 0 1 Voltage (kV) 0 100 200 300 400 Class 4 Class 4 Class 4 Class 5 Class 5 Class 5
Fig. 15. Classes associated with 5 equally spaced samples in time interval
[td
p+3 ms,tdp+4 ms], measured at R12for cleared and uncleared faults on line
L13 (for normal operation (+), with outage of line L14 (4) or outage of
converter 1 (.))
reverse backup protection zone (Fig. 14). After fault detection, the output of the classifier becomes 1, stating that the fault is in the primary zone. About 2 ms after fault detection, the classifier output becomes 2 or remains 1, showing that B12
has either operated or failed to remove the fault. The relay failure subsystem does not initiate backup actions as it does not enter the alert or action state.
The relay failure classifier can unambiguously assign the correct class number to each sample independent of the classification method (Fig. 15). Fig. 15 shows that the samples measured at R12, for cleared and uncleared faults at line L13,
are separable 3 ms after fault detection. Furthermore, the data samples associated with faults on line L13, with and without
the outage of line L14 or converter 1, follow similar patterns.
Therefore, the classifier does not need to be retrained for the aforementioned system changes. Fig. 15 only shows voltage and current samples for faults along line L13as faults on L14
yield similar results.
C. Comparison Between Conventional and Proposed Backup Protection
Fig. 16 (a)-(b) shows the current and voltage signals mea-sured at line L14for fault scenario (ii) as the conventional and
0.025 0.03 0.035 0.04 I14 (kA) 0 5 10 Conventional Proposed Time (s) 0.025 0.03 0.035 0.04 V14 (kV) 0 200 400 600 (b)
Fig. 16. Comparison between conventional and proposed backup protection
subsystems for fault scenario (ii): (a) dc current of L14and (b) dc voltage at
L14
proposed backup protection system initiates backup actions 3 ms after the fault detection instant (Fig. 12 (c)). It is assumed that the conventional backup protection system detects the uncleared fault 1 ms after the expected fault clearance instant by the primary relay (∆tpb+tdb − tsb = 1ms).
With the proposed backup relaying algorithm, the fault is cleared 5.65 ms faster than with the conventional backup protection (Fig. 16 (a)). The dc current of L14 reaches
8.475 kA and 10.46 kA, respectively, before the proposed or conventional backup subsystem removes the faulty line. The 2 kA current difference between the two backup methods affects the ratings of the breakers, current limiting equipment and converters [4]–[6]. With the proposed backup relaying algorithm, the system requires dc breakers with lower fault current interruption and energy dissipation capability as com-pared to conventional algorithms, which results in lower costs. Furthermore, the short response time of the proposed backup relaying algorithm facilitates fast restoration of the dc voltage after fault clearance.
VI. CONCLUSION
The proposed local backup relaying algorithm for HVDC grids reduces the fault clearance time compared to algorithms based on ac backup protection philosophy. The shorter fault clearance time results in reduced exposure of system equip-ment to high dc voltages and currents, and therefore, lower ratings for equipment such as dc breakers and converters.
The proposed backup relaying algorithm consists of breaker and relay failure subsystems and is a data-based scheme which uses the features extracted from local measurements to detect uncleared faults due to primary protection failure. It is applicable to a wide range of HVDC grids which use selective protection and dc breakers, as the principle of the algorithm is based on the voltage and current waveforms associated with dc breaker operation rather than system characteristics. The study results show that both breaker and relay failure subsystems quickly detect primary protection failure and generate the cor-rect trip signals for their associated breakers. The subsystems are insensitive to changes in the grid operation state such as converter and line outages.
In future work, a detailed analysis of measurement unit specifications (e.g., acceptable signal to noise levels and
sam-for the backup relaying algorithm will be considered. Further-more, an in-depth study regarding the use of the algorithm for HVDC grids with overhead lines will be conducted. Although the main principles of the proposed algorithms remain the same, slight adaptations might be needed to deal with various types of faults and fault resistances.
REFERENCES
[1] D. Van Hertem and M. Ghandhari, “Multi-terminal VSC HVDC for the European supergrid: Obstacles,” Renewable and Sustainable Energy Reviews, vol. 14, no. 9, pp. 3156 – 3163, 2010.
[2] J. De Decker and P. Kreutzkamp, “Offshore Electricity Grid Infrastruc-ture in Europe, Final Report,” OffshoreGrid, Tech. Rep., 2011. [3] M. K. Bucher, M. M. Walter, M. Pfeiffer, and C. Franck, “Options for
ground fault clearance in HVDC offshore networks,” in Proc. IEEE ECCE, Raleigh, NC, USA, 15-20 Sep. 2012 2012, pp. 2880–2887. [4] J. Häfner and B. Jacobson, “Proactive Hybrid HVDC Breakers: A
key innovation for reliable HVDC grids,” in CIGRÉ Bologna Symp., Bologna, Italy, 13- 15 Sep. 2011, 8 pages.
[5] C. Davidson, R. Whitehouse, C. Barker, J. Dupraz, and W. Grieshaber, “A new ultra-fast hvdc circuit breaker for meshed dc networks,” in IET ACDC, Birmingham, UK, 10-12 Feb. 2015, 7 pages.
[6] K. Tahata, S. Ka, S. Tokoyoda, K. Kamei, K. Kikuchi, D. Yoshida, Y. Kono, R. Yamamoto, and H. Ito, “HVDC circuit breakers for HVDC grid applications,” in AORC-CIGRE 2014, Tokyo, Japan, 27-29 May. 2014, 9 p.
[7] O. Cwikowski, M. Barnes, R. Shuttleworth, and B. Chang, “Analysis and simulation of the proactive hybrid circuit breaker,” in Proc. IEEE PEDS, Sydney, Australia, 9-12 Jun. 2015, pp. 4–11.
[8] J. Descloux, “Protection contre les courts-circuits des réseaux à courant continu de forte puissance,” Ph.D. dissertation, Université de Grenoble, Grenoble, France, Sep. 2013.
[9] K. De Kerf, K. Srivastava, M. Reza, D. Bekaert, S. Cole, D. Van Hertem, and R. Belmans, “Wavelet-based protection strategy for dc faults in multi-terminal VSC HVDC systems,” IET Generation, Transmission & Distribution, vol. 5, no. 4, pp. 496–503, April 2011.
[10] J. Sneath and A. Rajapakse, “Fault detection and interruption in an earthed hvdc grid using rocov and hybrid dc breakers,” IEEE Trans. Power Del., vol. PP, no. 99, pp. 1–1, 2014.
[11] W. Leterme, J. Beerten, and D. Van Hertem, “Non-unit protection of HVDC grids with inductive dc cable termination,” IEEE Trans. Power Del., vol. PP, no. 99, pp. 1–1, 2015.
[12] CIGRÉ WG B4-52, “HVDC Grid Feasibility Study,” CIGRÉ, Tech. Rep., 2013.
[13] P. Anderson, Power System Protection. Hoboken, NJ, USA: J. Wiley & Sons, 1998.
[14] M. Hajian, L. Zhang, and D. Jovcic, “Dc transmission grid with low speed protection using mechanical dc circuit breakers,” IEEE Trans. Power Del., vol. PP, no. 99, pp. 1–1, 2014.
[15] W. A. Elmore, Protective Relaying, Theory and Appplications. New York, NY: Marcel Dekker, Inc., 2004.
[16] Alstom, Network Protection and Automation Guide. Alstom Grid, 2011. [17] W. Leterme, S. Pirooz Azad, and D. Van Hertem, “Fast breaker failure backup protection for hvdc grids,” in Proc. IPST 2015, Cavtat, Croatia, 15-18 Jun. 2015, 6 pages.
[18] T. M. Cover and P. E. Hart, “Nearest neighbor pattern classification,” IEEE Trans. Inf. Theory, vol. 13, no. 1, pp. 21–27, 1967.
[19] A. Johns and S. Salman, Digital protection for power systems. London, UK: The IET, 1995.
[20] D. Naidoo and N. Ijumba, “HVDC line protection for the proposed future HVDC systems,” in IEEE PowerCon 2004, vol. 2, 21-24 Nov. 2004, pp. 1327–1332 Vol.2.
[21] W. Leterme, N. Ahmed, L. Ängquist, J. Beerten, D. Van Hertem, and S. Norrga, “A new HVDC grid test system for HVDC grid dynamics and protection studies in EMTP,” in Proc. IET ACDC, Birmingham, UK, 10-11 Feb. 2015, 7 pages.
[22] PSCAD. (2010) EMTDC User Guide. [Online]. Available:
https://hvdc.ca/knowledge-library/reference-material
[23] G. Chaffey and T. Green, “Directional current breaking capacity require-ments for hvdc circuit breakers,” in 2015 IEEE ECCE, Montreal, QC, Sep. 2015, pp. 5371–5377.
Willem Leterme (S’12) received the M.Sc. degree in electrical energy engineering from KU Leuven in 2012. Currently, he is pursuing a Ph.D. degree at KU Leuven. His research interests HVDC systems and design of protection algorithms for meshed VSC HVDC grids.
Sahar Pirooz Azadreceived the Ph.D. degree in electrical engineering from
the University of Toronto in 2013. She is currently a post-doctoral fellow at KU Leuven. Her areas of interest include HVDC systems and power systems control.
Dirk Van Hertem(S’02-SM’09) received his Ph.D. degree from KU Leuven
in 2009. In 2010, he was a member of EPS group at the Royal Institute of Technology, in Stockholm, Sweden, where he was the program manager for
controllable power systems for the EKC2competence center. Since spring
2011 he is back at the University of Leuven where he is an Assistant Professor in the Electa group. His special fields of interest are power system operation and control in systems with FACTS and HVDC and building the transmission system of the future, including offshore grids and the supergrid concept. He is an active member of IEEE PES and IAS and Cigré.
