Tutorial on Fault Locating Embedded in

Line Current Differential Relays – Methods,

Implementation, and Application Considerations

Omar Avendano
Portland General Electric

Bogdan Kasztenny, Héctor J. Altuve, Bin Le, and Normann Fischer

Schweitzer Engineering Laboratories, Inc.

Presented at the
41st Annual Western Protective Relay Conference
Spokane, Washington
October 14–16, 2014

Previously published in
Line Current Differential Protection: A Collection of
Technical Papers Representing Modern Solutions, 2014

Originally presented at the

17th Annual Georgia Tech Fault and Disturbance Analysis Conference, April 2014
 1

Tutorial on Fault Locating Embedded in Line

Current Differential Relays – Methods,
Implementation, and Application Considerations
Omar Avendano, Portland General Electric
Bogdan Kasztenny, Héctor J. Altuve, Bin Le, and Normann Fischer, Schweitzer Engineering Laboratories, Inc.

Abstract—Accurate fault locating on transmission lines The following factors make the fault locators embedded in
becomes increasingly beneficial by allowing faster restoration of 87L relays belong to a separate category of fault locating
scarce power system assets back into service. Having access to methods:
synchronized remote current data, line current differential
protection schemes can incorporate multiterminal fault locating  As a rule, the 87L scheme has access to remote
algorithms, allowing for more accurate fault locating compared currents, but not necessarily to remote voltages. As a
with single-ended methods. result, many implementations use local voltages and
This paper describes a new fault locating algorithm for two-, currents but only remote currents.
three-, and four-terminal lines that is suitable for integration in a  The scheme may have permanently or temporarily
line current differential protection scheme. The paper presents
reduced accuracy of data synchronization. Reduced
the algorithm in detail, includes examples of its operation, and
presents test results based on simulations as well as the testing of accuracy may be acceptable for protection functions,
its actual implementation in a particular protective relay. but can be very detrimental for fault locating and
therefore requires special attention.
I. INTRODUCTION  The scheme may lose communications between some
The increasing availability of reliable digital relays and operate in a master-slave mode, with only
communications in electric utilities promotes applications of one relay having access to remote data. When some or
line current differential (87L) schemes. Responding to all all communications are lost, the fault locator may need
currents of the protection zone, the current differential to fall back into a single-ended backup mode. This
principle is sensitive, inherently selective, and secure. Also, requires the fault locator to be adaptive, depending on
differential protection is typically easy to apply because it the availability of remote data.
does not require detailed short-circuit studies and settings  87L relays can be applied to multiterminal lines. Their
calculations. In its application to power lines, the principle is embedded fault locators determine the faulted segment
minimally or not affected by weak terminals, series of the line in addition to the distance to the fault.
compensation, changing short-circuit levels, power swings,  87L schemes must remain fully operational after the
nonstandard short-circuit current sources, and many other fault is cleared and during reclosing. Therefore, any
issues relevant for protection techniques based on additional usage of the channel for fault locating
measurements from a single line end [1]. purposes must be carefully engineered so as not to
Accurate fault locating on transmission lines becomes more negatively impact the line protection.
important as margins in present power systems erode,  Line charging currents on long lines and cables will
requiring fast restoration of transmission lines after faults. affect the fault locating accuracy. Some 87L relays
Embedding multiterminal fault locating algorithms in compensate for the charging current. Their embedded
87L relays is a natural fit. First, fault locators embedded in fault locators can take advantage of this compensation
87L schemes benefit from the data time alignment already in to increase their accuracy.
place for the 87L protection elements and, as such, can often This paper presents a new fault locating algorithm that is
be applied without external time sources for synchronization, optimized specifically for implementation in a typical
such as Global Positioning System (GPS) clocks. 87L scheme and that addresses the following design criteria
Second, the embedded fault locators use the existing and requirements:
87L communications channels, avoiding extra investment and  Minimize the channel bandwidth requirements of the
complexity compared with standalone multiterminal fault fault locator without increasing the amount of data
locators. sent in real time over the existing 87L channel.
 2

 Identify the faulted line section in three- and four- The method used to determine fault location on a two-
terminal applications without the need to exchange terminal line depends on the data available. The types of data
voltage measurements between the 87L relays. available are as follows:
 Report consistent fault location in all relays of the  Local voltage and current only (single-ended method).
87L scheme.  Local voltage and current plus time-aligned remote
 Reduce the impact of fault resistance and its current.
variability on the fault locating accuracy.  Time-aligned local and remote voltages and currents.
 Detect the loss of precise data synchronization in the  Local and remote voltages and currents that are not
scheme and fall back accordingly so that the fault time-aligned.
locating accuracy is not adversely impacted by In this paper, we concentrate on double-ended fault
misalignment between the local and remote currents. locating methods. When double-ended methods are employed,
 Support the master-slave mode of 87L scheme fault locating can be done in real time or offline. Typically,
operation. data generated by a line distance relay or any device that does
 Continue providing fault location information based not have access to remote data will be processed offline. Data
on a single-ended algorithm upon a total loss of data measured by an 87L relay will typically be processed in real
synchronization or communications. time.
In Section II, we review the fundamentals of fault locating A. Two-Ended Method With Synchronized Voltage and
for two-terminal and three-terminal lines. In Section III, we Current Measurements
introduce a new fault locating algorithm, illustrate its
When data are available from both line terminals, we apply
operation with an example, and discuss its accuracy. In
Kirchhoff’s voltage law to the voltage and current phasors in
Section IV, we explain how the new algorithm is applied to
the Fig. 1 circuit.
perform fault locating in three-terminal lines and how the
scheme functions in a master-slave configuration. In VF  VS  m • ZL • IS
(1)
Section V, we discuss how the algorithm is implemented for a VF  VT  1  m  • ZL • IT
four-terminal line. Finally, we provide some key conclusions.
If the measurements at both line terminals are time-aligned
II. REVIEW OF MULTI-ENDED FAULT LOCATING METHODS (synchronized), then we can solve for m as follows:
Various fault locating methods have been introduced over  VS  VT   ZL • IT
m (2)
the last several decades. These methods are based on traveling Z L •  IS  I T 
waves [2] or impedance measurements [3], which each
include single-ended [4] and double-ended [5] methods. This Equation (2) can be applied to any of the symmetrical
paper covers impedance-based fault locating methods only. component networks to solve for m.
As the number of transmission lines with more than two B. Two-Ended Method With Nonsynchronized Voltage and
terminals has increased over the last few years, so has the need Current Measurements
for more accurate fault locating on these multiterminal lines. If the measurements at the line terminals are not time-
This paper shows that one type of double-ended fault locating aligned (nonsynchronized), then it is convenient to use the
method can be adapted for locating faults on multiterminal negative-sequence network to solve for m for unbalanced
lines. faults. Because the negative-sequence network is not
We begin by reviewing the two-ended fault locating influenced by balanced load, load current will not affect the
methods. The problem to solve is to determine the distance to fault location calculation. In addition, the negative-sequence
the fault (m) from the selected line terminal, as shown in network is almost unaffected by charging current and is less
Fig. 1. susceptible to mutual coupling effects than the zero-sequence
network [5]. Furthermore, negative-sequence quantities are
available for all unbalanced faults, while zero-sequence
quantities are available only for ground faults.
Fig. 2 is the negative-sequence network for the fault
condition shown in Fig. 1, valid for any unbalanced fault.

Fig. 1. Equivalent diagram of a two-terminal line with a fault at m per unit

(pu) from Terminal S, with fault resistance RF.

Fig. 2. Negative-sequence network for the fault condition shown in Fig. 1.

 3

The magnitudes of the fault voltages calculated in (1) using Using the sequence network in Fig. 4 and the negative-
the local and remote voltages and currents match even though sequence voltages and currents from the respective terminals,
the local and remote measurements are not synchronized. we can calculate the negative-sequence voltage at
Therefore, we can write (3). Junction Point J and the remote terminal (T or S) twice,
starting from each line terminal, in order to construct two
V2S – m • Z2L • I 2S  V2T – 1 – m  • Z2L • I 2T (3)
negative-sequence voltage profiles [6]. From Fig. 4, we derive
If we square the terms on the left-hand side and right-hand (5), the two negative-sequence voltages calculated at
side of (3) and rearrange terms, we obtain the following Terminal S.
quadratic equation: V2_ SJ  V2S  Z2L_ OH • I 2S
(5)
 
2
Xm  Ym  Z  0 (4) V2_ ST  V2S  Z2L_ OH  Z2L_ UG • I 2S
where:
Similarly, we derive (6), the two negative-sequence
X, Y, and Z are functions of Z2L, V2S, V2T, I2S, and I2T.
voltages calculated at Terminal T.
Solving (4), we obtain two values for m, a value less than
zero and a value greater than zero. The value of m greater than V2_ TJ  V2T  Z2L _ UG • I 2T
(6)
zero is the solution for the distance to the fault. References [1]
and [5] describe this method in detail.
 
V2_ TS  V2T  Z2L _ OH  Z2L _ UG • I 2T
Plotting the magnitudes of the negative-sequence voltages
C. Two-Ended Method for Nonhomogeneous Transmission
given by (5) and (6), we obtain the negative-sequence voltage
Lines
magnitude profiles depicted in Fig. 5. As shown in this figure,
There are nonhomogeneous transmission lines where part the location at which the two voltage magnitude profiles
of the line is an overhead line and part is a cable. For this type intersect is the fault location we seek.
of line configuration, we cannot simply apply the two-ended
fault locating methods described previously and expect
accurate results. One applicable method uses synchronized
voltages and currents from both line terminals and solves the
negative-sequence network for the fault location.
Assume that a nonhomogeneous line composed of an
overhead section and an underground section experiences an
unbalanced fault at a distance m from Terminal S, as shown in
Fig. 3.

Fig. 5. Negative-sequence voltage magnitude profiles as seen from both line

Fig. 3. Equivalent diagram of a nonhomogeneous two-terminal line with a terminals. The intersection of the profiles indicates the fault location.
fault at m pu from Terminal S. J is the junction point of the overhead and
underground sections. D. Three-Ended Method
Fig. 4 is the negative-sequence network for the fault One method of calculating the fault location on a three-
condition shown in Fig. 3. terminal line is to convert the three-terminal line to an
N2 equivalent two-terminal line and apply one of the methods
discussed previously to determine the fault location [6].

Z2S Z2T 1) Synchronized Line Terminal Measurements

V2S V2T When the measurements at all three terminals are
m • Z2L_OH (1 – m) • Z2L_OH Z2L_UG
synchronized, it is easy to obtain the two-terminal equivalent
F2 J
I2T
of the three-terminal line. Assume a fault occurs at a
I2S
distance m from Terminal S of a three-terminal line, as shown
Fig. 4. Negative-sequence network of a nonhomogeneous two-terminal line in Fig. 6.
with a fault at m pu from Terminal S.
 4

Fig. 8. Negative-sequence two-terminal equivalent of the faulted three-

terminal line of the Fig. 7 network for the case when measurements are
synchronized.

2) Nonsynchronized Line Terminal Measurements

When the measurements are not synchronized, the
approach to obtain the two-terminal equivalent of the three-
Fig. 6. Fault on a three-terminal line at m pu from Terminal S. terminal line is similar to the case with synchronized
measurements. We first calculate the negative-sequence tap-
The first step to obtain the two-terminal equivalent of the
point voltage using (7). The magnitudes of the tap-point
three-terminal line is to calculate the voltage at Tap Point P
voltages calculated for the unfaulted line sections will be the
from each line terminal. We can use either the positive-
same, but their angles will be different because of the lack of
sequence network (for three-phase faults) or the negative-
measurement synchronization. Adding the two negative-
sequence network (for unbalanced faults). For example, Fig. 7
sequence currents to form an equivalent terminal current is not
shows the negative-sequence network for the case of an
possible in this case. However, we can calculate the correction
unbalanced fault in the Fig. 6 system.
angle needed to align the currents by using the tap-point
voltages calculated for the two unfaulted line sections, as
shown in (8).
Align  V2_ TP  V2 _ UP (8)
Knowing the alignment angle, we can now calculate the
equivalent terminal current using (9).
I 2 _ Eq  I 2T  I 2U •1Align (9)
Using V2_TP and I2_Eq as the negative-sequence voltage and
Fig. 7. Negative-sequence network for an unbalanced fault at m pu from current of the equivalent terminal, we can create the negative-
Terminal S in the Fig. 6 system. sequence two-terminal equivalent of the faulted three-terminal
Using the negative-sequence voltages and currents line shown in Fig. 9.
measured at the line terminals and the impedances of the line
sections, we calculate the negative-sequence voltages at the
tap point using (7).
V2_ SP  V2S  Z2L _ SP • I 2S
V2_ TP  V2T  Z2L _ TP • I 2T (7)
V2_ UP  V2U  Z2L _ UP • I 2U Fig. 9. Negative-sequence two-terminal equivalent of the faulted three-
terminal line of the Fig. 7 network for the case when measurements are not
The negative-sequence tap-point voltages calculated for the synchronized.
two unfaulted line sections will match (for our case, Once we have the two-terminal equivalent of the three-
V2_TP = V2_UP). The negative-sequence tap-point voltage terminal line for the case with nonsynchronized
magnitude calculated from the faulted line section terminal measurements, we can apply the method described in
will typically be much higher than the tap-point voltage Section II, Subsection B and use (4) to determine the fault
calculated from the unfaulted line section terminals. This location.
allows us to identify the faulted section of the line. If all three
tap-point voltages are equal, then the fault is at the tap point. III. TWO-ENDED FAULT LOCATING IN 87L RELAYS
Because current measurements are synchronized, we can
combine the two unfaulted line sections into one equivalent This section introduces a two-ended fault locating
terminal by adding their currents together and using the tap- algorithm as implemented in an 87L relay. This algorithm is
point voltage as the equivalent terminal voltage. Fig. 8 shows directly applicable to two-terminal lines and is also a part of
the resulting negative-sequence two-terminal equivalent of the the fault locating scheme in three- and four-terminal line
faulted three-terminal line. applications.
Once we have the two-terminal equivalent of the three- A. Fundamentals
terminal line for the case with synchronized measurements, When looking at the faulted line from a single line terminal
we can apply the method described in Section II, Subsection A and trying to find the fault location, we deal with one more
and use (2) to determine the fault location. unknown than the number of equations available.
 5

Single-ended methods solve this problem by making reflect the positive-sequence impedance between the line
reasonable assumptions or approximations. Different terminal and the fault point. For example, for AG faults, we
assumptions yield different fault locating methods. Two-ended apply (13).
methods solve this problem by obtaining at least one
measurement from the other end of the line. m

Im VAS • I 2S*  (13)
Consider the two-terminal line shown in Fig. 1 with a fault
at m pu from Terminal S. The distance to the fault from

Im  I AS  k 0 • I0S  • Z1L • I 2S* 
Terminal S can be calculated using the following fundamental where:
equation [1]: k0 is the zero-sequence current compensating factor.

m

Im VS • I F*  (10)
The use of local negative-sequence current in (13) is
referred to as polarization. The method of (13) is widely used

Im IS • ZL • I F*  and performs well as long as the negative-sequence network is
homogeneous (which is typically the case) [1] [7].
where: This method is further enhanced to create a new method
* is the complex conjugate operator. (described in the next subsection) and becomes a fallback
Neglecting measurement errors in the current (IS) and method in our 87L implementation should the loss of
voltage (VS) phasors, errors in the line impedance (ZL) value communications or loss of data alignment prevent use of the
(magnitude and angle), system nonhomogeneity, and the new method.
impact of charging current, (10) yields accurate results
B. Effect of Polarizing Quantities on Fault Locating
regardless of prefault power flow and fault resistance.
The obvious challenge in implementing (10) is that the Fig. 10 shows that the fault current (normally not available
fault current IF is unknown to any single-ended method. By to any single-ended protection scheme) is the differential
the nature of (10), however, only the angle of the fault current current IDIF naturally available to the 87L scheme.
is required. This angle can be reasonably approximated with I F  IS  IT  I DIF (14)
the angle of the local incremental current (Takagi algorithm)
As a result, (10) is not theoretical anymore but can be
or with the local negative-sequence current or zero-sequence
practically implemented by substituting the fault current with
current (whichever sequence network is more homogeneous).
the 87L differential current IDIF. For example, for AG faults as
This solution leads to a practical and, for most operating
seen from Terminal S, we apply (15).
conditions, accurate single-ended algorithm (Schweitzer
method [4] [7]).
m

Im VAS • I 2DIF*  (15)
Fig. 10 illustrates this particular approach for a single-line-
to-ground fault. 
Im  I AS  k 0 • I0S  • Z1L • I 2DIF* 
We used the two-machine sample system shown in Fig. 11
to illustrate the performance of the new fault locating
Z2S V2S V2T Z2T algorithm shown in (15). The parallel lines in the figure are
m • Z2L (1 – m) • Z2L
mutually coupled. Fig. 11 also shows the impedance values
RF
I2S I2F I2T measured by a distance element for faults at different locations
(different m values from Terminal S) that were calculated in
Positive- and Zero-
Sequence Networks
IF subsequent simulations and are represented by dots in the
figure. We simulated high fault resistance values and a strong
Fig. 10. Negative-sequence voltage and current quantities for a single-line- infeed effect, which caused the apparent impedance to shift
to-ground fault. considerably to the right from the line impedance in the figure.
From Fig. 10, it is clear that:
I F  I 2S  I 2T (11)
Similarly, from the zero-sequence network (not shown in
Fig. 10), we can write (12).
I F  I0S  I0T (12)
Equations (11) and (12) are true as long as the negative-
and zero-sequence networks are homogeneous (i.e., the angles
of the total system impedances at the left and right sides of the
fault point in Fig. 10 are similar).
Practical implementations of this method require fault type
Fig. 11. Impedances measured by a distance element for faults at different
identification and use of the proper loop quantities in order to locations on a two-terminal line mutually coupled with an adjacent parallel
line.
 6

Fig. 12 plots the fault locating results for the fault cases  The algorithm is impedance-based and therefore
shown in Fig. 11. The figure depicts the calculated m values affected by line asymmetry. Even fully transposed
(from Terminal S) when using the Terminal S negative- lines are symmetrical only between their terminals.
sequence current, Terminal S zero-sequence current, and The two line segments created by a randomly located
differential negative-sequence current for polarization. fault are not symmetrical, in general.
 The algorithm is affected by mutual coupling for
ground faults.
 Errors in measuring the currents and voltages affect
the accuracy, as in any impedance-based algorithm.
 Last, but not least, phase errors in the polarizing signal
(the differential current) impact the fault locating
accuracy. This source of error is unique to the new
method and is explained next.
87L relays need to time-align the remote and local currents
before forming the differential current. Two methods are used
in practice for current alignment. When the channel is
symmetrical (equal latencies in the transmitting and receiving
Fig. 12. Comparison of fault locating results for the Fig. 11 fault cases using directions), 87L schemes typically align the data using the
Terminal S zero- and negative-sequence currents and the differential negative- industry standard method known as the ping-pong algorithm
sequence current for polarization. [8]. When the channel is not symmetrical, the ping-pong
Polarizing with the differential current (new method) gives algorithm introduces a time-alignment error proportional to
the best results. The accuracy is only slightly degraded for the amount of asymmetry. The resulting current phase error
faults close to the remote line terminal because of the impact creates a fictitious phase shift in the differential current during
of the zero-sequence mutual coupling with the parallel line. internal faults.
(The fault locating algorithm has no mutual coupling When using asymmetrical channels, 87L relays require a
compensation in this example.) common (external) time reference to align the currents [8].
Using the local negative-sequence current gives good Historically, GPS clocks have been used as the time reference.
results for close-in faults. However, when the fault is farther These clocks may be embedded in the 87L relays (rare) or be
away from the terminal, the results are less accurate. This standalone and connected via an IRIG-B input (more
error is caused by network nonhomogeneity. In this example, common). More recently, terrestrial network-based time-
the angle of the remote source negative-sequence impedance distribution systems have also been used.
is different from the angle of the equivalent negative-sequence In any case, a small phase angle error in the differential
impedance of the line and local source. As the fault moves current (perfectly tolerable by the 87L protection elements)
away from Terminal S, the negative-sequence network seen would cause a considerable fault locating error using (15). To
from the fault point becomes less homogeneous. Using the illustrate this error, consider the system in Fig. 11 and assume
local zero-sequence current gives even worse results because a phase error in the range of ±10 degrees in the differential
we modeled higher nonhomogeneity in the zero-sequence current. Fig. 13 shows the fault locating results using the
network (which is a typical situation) and because the zero- differential current for polarization.
sequence network is more affected by mutual coupling than
the negative-sequence network is.
Of course, for low fault resistance values or with light load
on the line, these three polarizing methods would yield good
fault locating accuracy [7]. In practical situations, however,
the new method using the differential current for polarization
gives much better results.
C. Factors That Affect Fault Locating Accuracy
The accuracy of the described fault locating method can be
analyzed based on (15). In particular:
 The algorithm uses line positive- and zero-sequence Fig. 13. Effect of phase errors in the differential current on the fault locating
accuracy using (15) in the sample system in Fig. 11.
impedances and is affected by errors in their values. In
particular, the zero-sequence impedance (buried in the Consider, for example, an error of 5 degrees. Assuming a
k0 factor) is typically known with less accuracy and perfectly homogeneous network, the local and remote
may change seasonally because of soil resistivity and negative-sequence currents are perfectly in phase for an
conductor sag due to heat or ice. internal fault. If their magnitudes are equal, it would take a
shift in the remote current of 10 degrees in order to shift the
differential current by 5 degrees. A 10-degree shift in a 60 Hz
 7

system can be caused by channel asymmetry equal to Table I shows that the double-ended fault locating
2 • (10 degrees/360 degrees)/60 Hz = 0.93 milliseconds. algorithm is more accurate than the single-ended algorithm for
This level of asymmetry is well within the tolerance of a both faults. However, for the CG fault, the accuracies of both
typical 87L protection scheme, but in the system of Fig. 11, it methods are about the same. The reason for the similar
would cause a fault locating error in the order of 10 to accuracies is that the CG fault resistance was very low, which
20 percent, as shown in Fig. 13. makes both methods behave similarly. The other causes of
Therefore, the precision of current data alignment must be fault locating errors discussed in the previous subsection affect
monitored by the fault locator embedded in the 87L scheme. the single- and double-ended methods similarly. In this
The basic principle of monitoring the precision of data particular case, even though the line is symmetrical when
alignment works as follows: observed from the line terminals, it is not symmetrical when
 When in the time-based mode (GPS), each relay of the viewed from the fault point. Therefore, the error is caused by
87L scheme must be locked to a precise time source. the asymmetry of the transmission line when viewed from the
If the lock is lost or the source of time (clock) reports fault location. In addition, mutual coupling affects both fault
a time error via the IEEE C37.118-compliant time locating methods.
quality bits in the IRIG-B signal, the precision of In the BG fault, the fault resistance had a higher value. The
current data alignment is declared low for use in the accuracy advantage of the double-ended algorithm over the
embedded fault locator. single-ended algorithm is more evident in this case.
 When in the channel-based mode (ping-pong), if the
angle difference between the local negative-sequence IV. THREE-ENDED FAULT LOCATING IN 87L RELAYS
current and the differential negative-sequence current For multiterminal lines, the fault locating process is
is greater than a threshold (a few degrees), the performed in two steps. First, each relay calculates the fault
precision of data alignment is declared low. location and sends the result to the other relays. Second, each
Upon detected or suspected poor precision of current data relay uses the fault location information from all of the
alignment, the algorithm falls back from the version given by terminals to determine the faulted line section and the distance
(15) to the single-ended version given by (13). to the fault.
D. Results Obtained From Field Events A. Faulted Section Identification and Fault Location
We used two actual fault cases, courtesy of Bonneville Determination
Power Administration (BPA), to evaluate the single- and The method presented in Section III, Subsection B requires
double-ended fault locating algorithms. Both faults occurred the 87L relays to exchange their individually calculated fault
on the BPA Goshen-Drummond line. This line has a length of location values (m values) in per unit with each other when
117.11 kilometers and is effectively symmetrical because it applied to three- and four-terminal lines. This exchange is
has a large number of different tower configurations. This line accomplished using a patent-pending method of provisioning
shares towers with another line for a portion of its run. two bits in an 87L data packet and modulating these bits to
Therefore, mutual coupling is a factor for single-line-to- facilitate a virtual serial communication over the 87L channel.
ground faults. Consider the three-terminal line shown in Fig. 14. Three
Table I gives the fault type, the actual fault distance 87L relays comprise the 87L scheme. Each relay has access to
determined by the field crew, the fault distance estimated by the local voltages and currents as well as to the remote
the single-ended fault locating algorithm and its error, and the currents (the method described in this paper does not need
fault distance estimated by the double-ended fault locating access to the remote voltages).
algorithm and its error. Fault distances are given from the
Goshen terminal end. The percentage error was calculated as
defined by IEEE C37.114 [9].
Actual Distance  Calculated Distance
% Error  •100 (16)
Line Length

TABLE I
PERFORMANCE OF THE SINGLE- AND DOUBLE-ENDED FAULT LOCATING
ALGORITHMS FOR TWO ACTUAL FAULTS

Single-Ended Double-Ended Fig. 14. Three-terminal line.

Actual Method Method
Fault For the method described in this paper, each relay assumes
Distance
Type Distance Error Distance Error the fault to be in its local section of the line and uses (15) to
(km)
(km) (%) (km) (%) calculate the fault location.
CG 109.29 105.42 3.30 106.24 2.60
BG 61.41 54.77 5.67 60.69 0.61
 8

The relay at Terminal S calculates:  For faults very close to the tap point, the three

mS 

Im VS • I DIF*  (17a)
m values may be very similar, with more than one
satisfying the m < (1 + margin) condition. This is

Im IS • ZSP • I DIF*  acceptable, because all the m values would be close to
1 pu, indicating even more that the fault is near the tap
The relay at Terminal T calculates: point.

mT 

Im VT • I DIF*  (17b)
 If a given relay cannot use the enhanced algorithm
shown in (15) and falls back to the single-ended

Im IT • ZTP • I DIF*  method shown in (13), the overall scheme still works
in principle (see more discussion in the following
The relay at Terminal U calculates: subsection). The only differences are the method

mU 

Im VU • I DIF*  (17c)
(double-ended versus single-ended) and resulting
accuracy of the m value calculation.

Im I U • Z UP • I DIF*  B. Fault Locating in Master-Slave Mode (Fallback Mode)
If the fault is actually on Section SP of the line (as shown A three-terminal 87L scheme can operate in one of the
in Fig. 14), mS < 1 pu. If the fault is beyond Tap Point P, following two modes:
mS > 1 pu. The method is very unlikely to overreach (i.e.,  Master mode. All relays have access to all the currents
indicate mS < 1 pu for a fault beyond the tap point). of the protection zone; therefore, each relay makes an
For a fault beyond the tap point, extra current flows toward independent trip decision (each relay in the scheme is
the fault on the faulted section (infeed effect), elevating the a master relay).
voltage at the relay location. The relay measures the increased  Master-slave mode. Only one relay has access to all
voltage but not the additional current. Consider the 87L relay the currents of the protection zone; therefore, only that
at Terminal T in Fig. 14. Current IU produces a voltage drop relay is a master and the other two relays (slave
between the tap point and the fault, but this current is not relays) supply data to the master relay and execute trip
measured by the relay at Terminal T. As a result, mT will very decisions received from the master relay.
likely be greater than 1 pu, regardless of the power flow The mode of operation of each relay in the scheme is
through the line. determined by the communications channels available
For example, under one power flow pattern in the system between relays. Consider the three-terminal 87L scheme
of Fig. 14 (power flow from Terminal T to Terminals S shown in Fig. 15.
and U), the following results were obtained for a fault at
m = 0.9 pu from Terminal S: mS = 0.91 pu, mT = 1.30 pu, and
mU = 1.55 pu. Under a different power flow pattern (power
flow from Terminals S and U to Terminal T), the following
results were obtained: mS = 0.89 pu, mT = 1.57 pu, and
mU = 1.34 pu. As expected in both cases, the relay at
Terminal S correctly calculated the location at about 0.9 pu
and the relays at Terminals T and U calculated values
considerably higher than 1 pu.
The faulted line section identification is based on
exchanging the locally calculated values of m and comparing
them with 1 pu. The line section that reports m < (1 + margin)
pu is declared faulted, and the corresponding value of m is Fig. 15. A three-terminal 87L scheme operating in master-slave mode
reported. The value of the margin is in the order of a few because communications are only available between Relays S and U and
hundredths of per unit and accounts for small fault locating Relays U and T.
errors, as discussed in Section III, Subsection C. If the communications channel between Relays S and T
In the previous examples, all three relays would indicate becomes unavailable, then Relay S and Relay T do not have
Section SP as the faulted section and report the fault location access to all the currents that make up the 87L protection zone
as 0.90 pu from Terminal S. and therefore cannot function independently. Relays S and T
The following points are worth observing with respect to will switch from the master mode to the slave mode. Relay U
our faulted line section identification method: has access to all of the currents and therefore will remain in
 The values of m are communicated over the 87L the master mode. The master relay provides differential
channel using our patent-pending method without protection for the three-terminal line. For a fault within the
disturbing the 87L elements. protection zone, the master relay sends tripping commands to
 9

the remote relays via in-band direct transfer trip bits. In C. Simulation Results
addition to providing the line protection function, the master We modeled a three-terminal 525 kV line in a Real Time
relay also provides the overall fault locating function for the Digital Simulator (RTDS®) to illustrate the performance of the
scheme. three-terminal fault locating algorithm. The three-terminal line
The master relay (Relay U) calculates the fault location as modeled in the RTDS is shown in Fig. 14. Sources S and U
seen from its line terminal using (17c). The slave relays have the same strength, and Source T is the weakest source
(Relays S and T) calculate the fault location using the single- with an impedance three times that of Sources S and U. The
ended method given by (13). The slave relays send their line segments and sources are homogeneous, and the lengths
calculated m values to the master relay. Using its own of the line segments are as follows: Segment SP =
calculated m value and those received from the slave relays, 100 kilometers, Segment TP = 25 kilometers, and Segment UP
the master relay identifies the faulted line section by selecting = 50 kilometers. Due to the lengths of the line segments,
the line terminal that calculated an m value less than 1. In our charging current compensation was enabled at all line
implementation, the master relay does not communicate the terminals. The line draws a charging current of approximately
faulted section back to the slave relays. In this case, the true 250 A.
fault location has to be obtained from the master relay. Table II shows the fault type, fault resistance, line segment
Consider Fault F1 on Section UP of the line in Fig. 15. For in which the fault was simulated, actual distance to the fault
this case, Relay S calculates an m value greater than from the line segment terminal, identified faulted line
1 (mS > 1 pu); similarly, Relay T calculates an m value greater segment, distance calculated from the local line terminal using
than 1 (mT > 1 pu). Relays S and T use (13) to calculate their the double-ended fault locating method, and percentage error.
respective fault locations. Relay U (the master relay) From Table II, we can see that for each case, the double-
calculates an m value less than 1 (mU < 1) using (17c). ended fault locating algorithm selected the correct faulted line
Relays S and T transmit their calculated m values (mS and mT) segment. Table II shows a maximum error of 0.75 percent for
to Relay U. Relay U determines the fault location by selecting a fault resistance of 10 ohms and an error of 6.05 percent for a
the m value that is less than 1. For this example, the distance fault resistance of 100 ohms.
to the fault is mU. This is the most accurate result, because mU Factors that affect the accuracy of the fault location are
is calculated using the enhanced method shown in (17c). discussed in the following subsection.
Next, consider Fault F2 on Section SP. In this case,
Relay S calculates mS < 1 pu, Relay T calculates mT > 1 pu, D. Factors That Affect Accuracy
and Relay U calculates mU > 1. As before, Relays S and T The three-ended fault locating algorithm is the same as for
send their calculated m values to Relay U. Relay U then a two-terminal line; therefore, the factors that affect the
selects the faulted section as the one with an m value less accuracy of the algorithm are the same as those discussed in
than 1 and reports the distance to the fault as that value of m, Section III, Subsection C.
in this case mS. This value may have some inaccuracy, For this discussion, we concentrate on two factors, namely
because mS has been calculated in Relay S using the single- the fault resistance and the asymmetry (lack of transposition)
ended method shown in (13). In this particular case, Relay U of the transmission line. The errors for all of the faults in
could theoretically calculate the tap-point voltage and current Table II are directly related to the magnitude of the fault
and execute the method shown in (17c) for the remote line resistance and the asymmetry of the transmission line. The
segment. However, the designers opted against this method used to calculate the distance to the fault is
complication in the actual implementation of the fault locating impedance-based, using the differential current for polarizing.
algorithm. In summary, obtaining data from all relays in an Even though we state that the line is transposed, the fault
87L scheme enables the scheme to correctly identify the current contributions from the line terminals are not perfectly
faulted line section and determine the fault location with in phase with one another due to the asymmetry of the line as
relatively good accuracy, even when the scheme is operating seen from the fault point. This nonhomogeneity of the system,
in master-slave mode. when viewed from the location of the fault, results in the total

TABLE II
PERFORMANCE OF THE DOUBLE-ENDED FAULT LOCATING ALGORITHM FOR FAULTS ON A THREE-TERMINAL LINE

Fault Resistance Simulated Line Actual Distance Identified Line Calculated Distance
Fault Type Error (%)
(ohms) Segment (km) Segment (km)
AG 10 SP 75 SP 74.28 0.41
BC 10 TP 18.75 TP 19.7 0.54
CAG 10 UP 12.5 UP 11.18 0.75
ABC 10 SP 25 SP 24.74 0.15
ABG 100 SP 25 SP 14.41 6.05
 10

fault current not being in phase with the fault current The relay at Terminal U calculates:
contributed by each of the line terminals. The fault resistance I P  I U  IS and VP  VU  Z UP • I U (18b)
magnifies this source of error.
In our case, the fault currents contributed by the strongest The relay at Terminal T calculates:
terminals (S and U) lead the total fault current, and the fault IQ  IT  I W and VQ  VT  ZTQ • IT (18c)
current contributed by Terminal T lags the total fault current.
The relay at Terminal W calculates:
The result of this is that for faults on line segments connected
to the strongest sources, the fault locating algorithm IQ  I W  IT and VQ  VW  Z WQ • I W (18d)
overreaches (i.e., calculates a fault location closer to the In order to calculate the currents in these equations, each
terminal than the actual fault location) because the strong relay is provided with a setting indicating the specific remote
terminal currents lead the fault current. The 100-ohm fault relay that is installed on the line section connected to the same
illustrates how the fault resistance magnifies this tap (Relays S and U monitor line sections that connect to
phenomenon. Tap Point P, and Relays T and W monitor line sections that
For faults located on line segments connected to the connect to Tap Point Q).
weakest terminal, the fault locating algorithm underreaches Having the P (or Q) currents and voltages calculated and
(i.e., locates the fault farther away from the line terminal than having the ZPQ impedance as a setting, each relay executes
the actual fault location) because the weak terminal current (15) and obtains a coherent fault locating result. Relays S and
lags the fault current. U report m pu as the distance to the fault from Tap Point P,
and Relays T and W report (1 – m) pu as the distance to the
V. FOUR-ENDED FAULT LOCATING IN 87L RELAYS fault from Tap Point Q.
A. Faulted Section Identification and Fault Location
B. Fault Locating in Fallback Mode
Determination
In our four-terminal 87L implementation, data are
Consider the four-terminal line shown in Fig. 16.
exchanged between relays using Ethernet as the
communications medium. Because of the use of Ethernet, all
relays are either in the master mode or the 87L scheme is not
operational [5].
In the master mode, all relays have access to all the remote
currents and the distance to the fault is computed as explained
in the previous subsection.
When the 87L scheme is not operational, the fault locating
algorithm switches from the multiterminal mode to single-
terminal mode. As a result, the fault locating accuracy is
typically degraded.
However, the scheme will still be able to correctly
determine the faulted line section. A fault between any line
Fig. 16. Four-terminal line. terminal and a tap point will result in one relay calculating an
m value less than 1 and all other relays calculating m values
In a four-terminal line application, the fault locating
greater than 1. For a fault between the two tap points, all
algorithm is executed in two or three steps, depending on the
relays will calculate m values greater than 1. Because there is
actual fault location.
no communication between the individual relays, it is not
First, each relay assumes the fault is in its local section. If
possible to perform fault locating in real time. Fault locating
the assumption is true for one relay, this relay will calculate a
has to be done offline, either by manually retrieving and
value of m less than 1 pu, and the fault locating process will
processing data from the relays or by using an offline program
continue, as explained in the previous section for the three-
that retrieves and processes the data.
terminal line. For example, for Fault F1 in Fig. 16, Relay U
The following subsection discusses the fault locating
will calculate m < 1 pu, and all four relays will report
accuracy for the four-terminal operating mode and the factors
Section UP as faulted.
that influence this accuracy.
If no relays calculate m < 1 pu, the fault must be in the
middle section (PQ) between the two taps (Fault F2 in C. Factors That Affect Fault Locating Accuracy
Fig. 16). If so, the voltages and currents at both tap points The four-ended fault locating algorithm is an extension of
(P and Q) must be calculated next. Knowing that the local line the two-ended algorithm. Therefore, the accuracy of the four-
sections are free from faults, each relay calculates the ended algorithm is affected by the factors described in
equivalent currents and voltages for the PQ section of the line Section III, Subsection C, except for an important difference:
using the voltage drop equation for the unfaulted section. the effect of the time-alignment method. In the two-ended
The relay at Terminal S calculates: mode, current alignment can be done by using either the ping-
I P  IS  I U and VP  VS  ZSP • IS (18a) pong method (channel-based alignment) or by using a
 11

common time reference. In the four-ended mode, Ethernet is When IEEE C37.238 provides the time reference, all relays
used to exchange data between relays and the ping-pong are synchronized via Ethernet. This means that no relay clock
method cannot be used because channel symmetry cannot be loses synchronism with respect to the other relays while
guaranteed. Therefore, an external time reference is used for communication between the relays exists. Therefore, no time-
current alignment. alignment error exists in this mode as long as the source of the
There are two ways of providing the external time IEEE C37.238 timing signal is accurate.
reference:
 Using local IRIG-B signals (derived from local GPS VI. CONCLUSION
clocks or from a network-based terrestrial time- Embedding multi-ended fault locating in 87L relays brings
distribution system, such as over a synchronous many advantages. A single system serves both protection and
optical network [SONET]). multi-ended fault locating functions, allowing savings in
 Using the Precision Time Protocol (PTP) communications, time synchronization, material costs, and
IEEE C37.238. engineering.
When the time reference is provided by local IRIG-B This paper presents a novel multi-ended fault locating
signals, each relay must monitor the integrity of its clock method designed specifically for ease of integration in 87L
signal. If a clock loses its GPS lock, its time quality will schemes. The method, which uses the differential current,
degrade and the timing accuracy of the data from that relay improves the numerical accuracy of fault locating compared
cannot be guaranteed. The scheme currents cannot be aligned with single-ended methods and indicates the faulted line
with enough accuracy to facilitate multi-ended fault locating, section in three- and four-terminal applications. Practical
and the fault locator needs to fall back to the single-ended implementation aspects are considered to address cases of loss
method. of communications or degraded precision of data alignment.
When the fault locator falls back to the single-ended
method, the accuracy will degrade for the reasons discussed in VII. REFERENCES
Section III, Subsection C. However, if the fault is located [1] H. J. Altuve Ferrer and E. O. Schweitzer, III (eds.), Modern Solutions
between a line terminal and a tap point, the fault locating for Protection, Control, and Monitoring of Electric Power Systems.
accuracy for the relay located at the terminal of the faulted Schweitzer Engineering Laboratories, Inc., Pullman, WA, 2010.
line section is primarily influenced by the fault resistance [2] E. O. Schweitzer, III, A. Guzmán, M. V. Mynam, V. Skendzic,
B. Kasztenny, and S. Marx, “Locating Faults by the Traveling Waves
(assuming that the line positive- and zero-sequence impedance
They Launch,” proceedings of the 40th Annual Western Protective
values are fairly accurate). When the fault resistance is Relay Conference, Spokane, WA, October 2013.
relatively small, the accuracy of the single-ended method for [3] E. O. Schweitzer, III, “Evaluation and Development of Transmission
faults between the line terminal and the tap point for the relay Line Fault-Locating Techniques Which Use Sinusoidal Steady-State
connected to that line terminal will be fairly accurate; fault Information,” proceedings of the 9th Annual Western Protective Relay
Conference, Spokane, WA, October 1982.
location estimation for all other relays will be significantly off
[4] E. O. Schweitzer, III, “A Review of Impedance-Based Fault Locating
because of the infeed effect. The higher the fault resistance, Experience,” proceedings of 15th Annual Western Protective Relay
the more important role the currents from the remote terminals Conference, Spokane, WA, October 1988.
will play and the larger the error will be. Therefore, the relay [5] D. A. Tziouvaras, J. Roberts, and G. Benmouyal, “New Multi-Ended
that calculates an m value less than 1 will have the most Fault Location Design for Two- or Three-Terminal Lines,” proceedings
accurate distance-to-fault value in the scheme. of the 7th International Conference on Developments in Power System
Protection, Amsterdam, Netherlands, April 2001.
For faults between the two tap points, the accuracy of the
[6] Y. Gong, M. Mynam, A. Guzmán, G. Benmouyal, and B. Shulim,
single-ended fault locating algorithm will be determined by “Automated Fault Location System for Nonhomogeneous Transmission
the fault resistance, the distance to the fault from the tap point, Networks,” proceedings of the 65th Annual Conference for Protective
the system nonhomogeneity, and the strength of the sources at Relay Engineers, College Station, TX, April 2012.
the line terminals. As mentioned in Section V, Subsection A, [7] K. Zimmerman and D. Costello, “Impedance-Based Fault Location
Experience,” proceedings of the 58th Annual Conference for Protective
all relays will calculate an m value greater than 1 for a fault
Relay Engineers, College Station, TX, April 2005.
between the two tap points. However, if the fault is located
[8] B. Kasztenny, G. Benmouyal, H. J. Altuve, and N. Fischer, “Tutorial on
closer to one tap point than the other, in general, the relays Operating Characteristics of Microprocessor-Based Multiterminal Line
that are adjacent to that tap point will yield a better result than Current Differential Relays,” proceedings of the 38th Annual Western
those relays adjacent to the tap point farthest from the fault. In Protective Relay Conference, Spokane, WA, October 2011.
this case, the strength of the source behind each relay plays a [9] IEEE Standard C37.114-2004, IEEE Guide for Determining Fault
Location on AC Transmission and Distribution Lines.
significant role.
 12

VIII. BIOGRAPHIES
Omar Avendano received his BS in electrical engineering from Simón
Bolívar University in Caracas, Venezuela, in 1984. He worked for 12 years in
the manufacturing and design of uninterruptible power supply systems applied
to critical industrial loads. During this time, he designed and developed
automated controls for ac and dc distribution centers and performed protective
relay coordination studies. He joined PacifiCorp in November 2001, where he
served as the lead senior protection and control engineer for 10 years,
developing protection designs for transmission and distribution. In 2012, he
joined Portland General Electric in the role of Senior Protection Engineer and
has continued to develop extra-high-voltage transmission and generation
protection.

Bogdan Kasztenny is the research and development director of technology at

Schweitzer Engineering Laboratories, Inc. He has over 23 years of expertise
in power system protection and control, including ten years of academic
career and ten years of industrial experience, developing, promoting, and
supporting many protection and control products. Dr. Kasztenny is an IEEE
Fellow, Senior Fulbright Fellow, Canadian representative of CIGRE Study
Committee B5, registered professional engineer in the province of Ontario,
and an adjunct professor at the University of Western Ontario. Since 2011,
Dr. Kasztenny has served on the Western Protective Relay Conference
Program Committee. He has authored about 200 technical papers and holds
20 patents.

Héctor J. Altuve received his BSEE degree in 1969 from the Central
University of Las Villas in Santa Clara, Cuba, and his Ph.D. in 1981 from
Kiev Polytechnic Institute in Kiev, Ukraine. From 1969 until 1993, Dr. Altuve
served on the faculty of the Electrical Engineering School at the Central
University of Las Villas. From 1993 to 2000, he served as professor of the
Graduate Doctoral Program in the Mechanical and Electrical Engineering
School at the Autonomous University of Nuevo León in Monterrey, Mexico.
In 1999 through 2000, he was the Schweitzer Visiting Professor in the
Department of Electrical Engineering at Washington State University.
Dr. Altuve joined Schweitzer Engineering Laboratories, Inc. (SEL) in
January 2001, where he is currently a distinguished engineer and dean of
SEL University. He has authored and coauthored more than 100 technical
papers and several books and holds four patents. His main research interests
are in power system protection, control, and monitoring. Dr. Altuve is an
IEEE senior member.

Bin Le received his BSEE from Shanghai Jiao Tong University in 2005 and
an MSEE from the University of Texas at Austin in 2008. He has been
employed by Schweitzer Engineering Laboratories, Inc. since 2008. Mr. Le
currently holds the position of lead power engineer in the research and
development division. He is a member of IEEE and a professional engineer
registered in the state of Washington.

Normann Fischer received a Higher Diploma in Technology, with honors,

from Technikon Witwatersrand, Johannesburg, South Africa, in 1988; a
BSEE, with honors, from the University of Cape Town in 1993; and an MSEE
from the University of Idaho in 2005. He joined Eskom as a protection
technician in 1984 and was a senior design engineer in the Eskom protection
design department for three years. He then joined IST Energy as a senior
design engineer in 1996. In 1999, Mr. Fischer joined Schweitzer Engineering
Laboratories, Inc., where he is currently a fellow engineer in the research and
development division. He was a registered professional engineer in South
Africa and a member of the South African Institute of Electrical Engineers.
He is currently a senior member of IEEE and a member of ASEE.

© 2014 by Portland General Electric and

Schweitzer Engineering Laboratories, Inc.
All rights reserved.
20140312 • TP6654-01