IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO.

5, OCTOBER 2018 2207

Automated Failure Diagnosis in Transmission

Network Protection System Using Synchrophasors
B. Cui, Student Member, IEEE, A. K. Srivastava , Senior Member, IEEE, and P. Banerjee, Member, IEEE

Abstract—In this paper, a multihypothesis-based centralized The impact of the protection system and hidden failures on bulk
Failure Diagnosis in Protection System (FDProSys) tool using power system reliability was investigated in [10]. This method
phasor measurement units (PMU) data has been proposed to uses breaker-oriented bulk power system network model which
systematically and automatically detect malfunctions or failures
in protective devices. The proposed diagnostic algorithm and includes substation configuration as well as corresponding pro-
FDProSys tool employs synchrophasor measurements to select a tection schemes. Neural networks [11] is used to model the
section of the network called ProNet, with possible device mal- uncertainties involved in relay and CB operation messages to
function. Multiple hypotheses have been generated automatically estimate faulted section in electric power systems. A decision
based on node incidence matrices and multihypothesis theory. The support system using the CB information for online fault sec-
malfunction of the protective devices in the presence of fault and
relay operation is identified by comparing the five-digit message tion estimation in power systems is presented in [12]. Authors in
of the real event with multiple hypotheses. Trustworthiness of [13] applied pattern recognition techniques and machine learn-
PMU data caused by common failures, such as communication ing in a hidden semi-Markov model for effective online failure
outage and GPS problems, has also been considered by the prediction. Reference [14] proposes a relay supervisory system
proposed algorithm using the PMU status flag. The developed (RSS) as part of relay protection systems to improve condition-
tool FDProSys can also work with supervisory control and data
acquisition data assisted by manual analysis with less accuracy based maintenance and mitigate the impact of hidden failures.
and is robust enough to several PMU/relay failures. Simulation Authors in [15] discussed a hidden failure detection method
results using the IEEE 14-bus, IEEE 57-bus, and IEEE 300-bus through D-S evidence theory by using mutual coordination of
systems demonstrate the accuracy of the developed algorithm. relay protection and fault location result. In [16], a real-time
Index Terms—Failure diagnosis, phasor measurement units, tool was presented to detect, classify, and locate transmission
multi-hypothesis, transmission network, protection system. line faults, and indicate whether the line was tripped due to a
malfunction of protective relays. Temporal Causal Diagram has
I. INTRODUCTION been introduced in [17] for the root cause analysis in the power
HE protection system is critical for the safety personnel system. In power system, a protection system element malfunc-
T and assets of the power grid in the event of faults. The pro-
tection system consists of devices like relays and circuit breaker
tion when it either fails to operate as designed, or operates
unintentionally or outside of its protection zone.
(CB) to isolate the the faulted section of the system from the However, the existing methods or algorithms for power sys-
healthy part. The fault isolation eliminates electrical hazard and tem fault diagnosis do not systematically address the possible
reduces the electrical stress to the equipments. Malfunctions or malfunctions or failures of protective devices [11], [12], [18].
failures in the protection system, increase the risk to life and But when a fault occurs in the system along with failures in
electrical assets. The chances of cascaded outage also increase, protection devices, conflicting information and alarms makes
if the malfunction of protective devices occurs during fault. the problem identification by system engineers/operators very
Restoration of the system following fault requires identifica- difficult. An automated algorithm is required for power system
tion and servicing of malfunction devices as soon as possible, transmission network protection system failure diagnosis to
which requires personnel time and manual effort. A number of precisely identify the malfunctions and failures of protective
fault analysis tools have been reported in the literature. Some of devices. Among these methods mentioned here, authors in [15]
these tools are related to fault detection, classification, or loca- and [16] proposed to use combined fault location techniques
tion [1]–[4], and others are complete fault analysis tools [5]–[9]. to detect if the distance relays behave as they are supposed
to, although the malfunctions of circuit breakers such as stuck
circuit breakers were not discussed. Detection of mal-operation
Manuscript received June 28, 2017; revised December 1, 2017 and February
18, 2018; accepted March 23, 2018. Date of publication April 5, 2018; date of
of circuit breaker is presented in [19] and [20] and requires
current version September 25, 2018. This research was supported by the National additional relay and control logic installed at the substation.
Science Foundation (NSF) under Grant CNS-1329666. Paper no. TPWRD- Such schemes also require protection personnel to physically
00859-2017. (Corresponding author: Anurag Srivastava.)
The authors are with the School of Electrical Engineering and Computer
validate each breaker protection relays for identification of
Science, Washington State University, Pullman, WA 99164 USA (e-mail:,bo.cui mal-operation. Fault identification using current differential
@wsu.edu; [email protected]; [email protected]). protection [21] assumes PMUs installed in both the end of the
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
transmission lines, which may not be a suitable assumption.
Digital Object Identifier 10.1109/TPWRD.2018.2823343 Moreover, current differential protection settings is dependent

0885-8977 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

Authorized licensed use limited to: Rajnish Kumar. Downloaded on August 22,2024 at 09:00:08 UTC from IEEE Xplore. Restrictions apply.
2208 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 5, OCTOBER 2018

on system and line parameters. A simultaneous fault section data for failure diagnosis in a transmission network protection
estimation and protective device failure detection is presented system based on multi-hypothesis theory [25], [26]. The FD-
in [22]. This method uses artificial neural network (ANN) ProSys tool is designed for enough number of PMUs/relays
without two-way correspondence between the neural network installed at the substation for failure diagnosis but can also
inputs and the protection devices alarms. But, it requires a work with some PMUs/relays and SCADA data assisted by
large number of training data to deal with topological changes some manual analysis. Developed tool FDProSys utilizes algo-
and different failure cases which is difficult to obtain from the rithms for selection of zone related to the problem area (ProNet),
real power grid operation under different possible operating data collection, computing 5-digit message, generating multiple
scenarios. Synchrophasor measurement and bus impedance hypothesis, and selecting the hypothesis with maximum credi-
matrix based fault diagnosis is presented in [23], which bility.
matches change in phasor data due to fault at different location.
Generating database of phasor considering all possible fault II. SYNCHROPHASOR BASED FDPROSYS TOOL
location, resistance and fault types is not feasible for large Multiple hypotheses based fault diagnosis method using the
systems specially for different possible operating scenarios. Sequence of Event Recorders (SERs) information was proposed
PMU based protection element failure detection proposed in in [25]. This manual method rank multiple hypothesis by their
[24] assumes transmission of 4-bit binary data from relay and credibility is referred to as Generalized Alarm Analysis Module
breakers, which requires change in computing logic of the relay. (GAAM) and had been tested with data from ENEL, Italy and
In this work, the malfunctions of protective devices during later by many utility companies.
fault which not only comes from distance relays but also from The developed centralized FDProSys tool is designed to run
circuit breakers are considered. None of the methods mentioned in the control center for automated diagnostic of protective de-
here, are related to identifying the malfunctions of protective vice failures and malfunction inspired by multiple hypotheses.
devices during a fault in automated manner, which creates an Confused with conflicting information and alarms, the operator
ambiguous reason for the observed state of the system. Com- can launch the proposed algorithm to identify the malfunctioned
pared with the existing methods, the proposed algorithms use devices. When launching the FDProSys tool, the operator has
synchrophasor measurement, topology information and breaker to provide the starting time tc to start the failure diagnosis. The
status to find the root cause of the observed state of the system time, when the first distance relay tripping is reported will be
evolved through set of events. In the existing industry practice, t0 . tc has to be selected earlier than t0 by at least 1 PMU cycle.
the system operator or engineer may be required to send person- In this work, the maximum operating time for zone 1, zone 2
nel to each substation for identifying the device malfunctions and zone 3 protection are presented as tZ 1 M ax , tZ 2 M ax and
and requiring more effort and a long time to restore the system. tZ 3 M ax . The system has to be at steady state at tc . And t0
In this work, automatic detection of malfunction or failures of has to be more than tZ 1 M ax + tZ 3 M ax after the last reported
distance relays and circuit breakers in transmission system is event. All the synchrophasor data after tc will be used for anal-
presented. The most common causes of malfunction are incor- ysis. The proposed algorithm also needs the base case topology
rect setting/logic/design errors and relay failures/malfunctions. information of transmission network as an input.
The traditional fault location involves demand based down- The proposed method of failure diagnosis provides the most
loading the fault information sent to the control centers by the likely cause of an event or series of events resulting in multiple
relay/CB installed in the field generally triggered by the alarms. line outage. The multiple hypothesis generation, generates all
The relay typically uses local measurements to generate alarms the possible reasons explaining the current set of events and
and fault information or Energy Management System (EMS) filters out the invalid explanations. The initial and final node
can generate alarms based on SCADA data. Hence, any relay incidence matrix used in multiple hypothesis generation use the
operation resulted due to malfunctions of protective devices initial and final state of the network, which evolved through a set
in the neighboring buses presents an apparent reason of fault of events. The node incidence matrix only uses the connectivity
instead of true cause, resulting in ambiguity for the system op- of the buses which can be generated either using the voltage and
erator. In the existing practice, the system operator also do not current phasors from the PMU for high accuracy or the updated
have any means to identify such malfunctions of protective de- state estimation based on SCADA measurements. The accuracy
vices in the neighboring buses using the relay/CB status, which of voltage and current phasors required for generating the initial
are based on local measurements but analyze multiple events and final node incidence matrix is relaxed as high thresholds
files from relays. The FDProSys tool proposed in this work uses can be used. Hence, PMU measurements are not essential for
voltage and current as well as the other parameters related to evaluating node incidence matrix used in multiple hypothesis
relay/CB status to estimate malfunctions in protective devices, generation. The hypothesis selection, use zone digits for gener-
which was not possible using only relay/CB status based on ating the 5-digit message and evaluating their credibility. The
local measurements or EMS alarms. The proposed method also zone digits can either be constructed based on the voltage and
provides the most possible reason for multiple protective device currents phasors from the PMUs or from the distance relay.
malfunctions resulting in multiple lines outage scenario, which However, absence of PMUs and availability of only SCADA
is time consuming following existing manual process by the measurements may render the zone digits unreliable for 5-digit
system operator/ protection engineers. calculation. In such case, more than one hypothesis can be the
The contribution of this work is the development of a central- possible explanation of the state of the system followed by some
ized FDProSys tool using Phasor Measurement Units (PMU) manual analysis.
Authorized licensed use limited to: Rajnish Kumar. Downloaded on August 22,2024 at 09:00:08 UTC from IEEE Xplore. Restrictions apply.
CUI et al.: AUTOMATED FAILURE DIAGNOSIS IN TRANSMISSION NETWORK PROTECTION SYSTEM 2209

A. ProNet Selection and Data Collection

After the occurrence of an observed fault in the system with
inconsistent alarms or information, the first step of FDProSys
tool is selecting the Protection Net (ProNet) to detect malfunc-
tions or failures of protective devices. The ProNet is defined
as the 3-zone area in the transmission network with observed
distance relay tripping. In this work, the zone 1 protection is
set to be 80%, the maximum of zone 2 protection is 120% and
maximum zone 3 protection is 200% of the transmission line.
The ProNet selection will start at t0 from initial ProNet which
virtually select an area which is a union of the first reported
Fig. 1. Generation of 5-digit message.
tripping line and its neighboring lines. If there is an open cir-
cuit breaker event on line p within ProNet within tZ 3 M ax , the
ProNet will be updated by adding line p and its neighboring The failure diagnosis tool FDProSys test the possibility of
lines. All the synchrophasor data from PMUs/relays within the fault at each transmission line within the ProNet to generate hy-
final ProNet will be collected. pothesis that can cover all valid scenarios. The number of trans-
mission lines in the ProNet is i and denoting as a set {1, . . . , i}.
B. 5-Digit Data Calculation The proposed FDProSys tool tests whether a fault at line p ∈ [i]
is a possible explanation of the observed status and ProNet. The
1) Trust Digit: 5-digit data is a message string consists of 5 proposed FDProSys tool will use PMUs/relays data one PMU
digit having a combination of ‘0’ and ‘1’. The 5-digit data is a cycle before t0 to generate prefault node incidence matrix A
combination of one trust digit and 4 protection digit information. and use PMUs/relays data one PMU cycle after t0 + tZ 3 M ax to
The first digit among 5 digits is the trust digit and given a value generate post fault node incidence matrix B.
of ‘1’ when the PMU status flag is indicating a good data. If The prefault node incidence matrix of the ProNet is denoted
the PMU status flag is indicating a bad data, or the FDProSys by A, defined as:
tool can not compute the last 4 digits from corresponding PMU ⎡ ⎤
measurement, the trust digit will be given as ‘0’ denoting invalid A11 A12 · · · A1j
⎢ .. ⎥ (2)
PMU data. A = ((K  × K) ⊕ H) × ⎣ ... ..
.
..
. . ⎦
2) Protection Digit: If the trust digit of certain PMU is ‘1’,
the protection zone setting calculation will be done using PMU Ai1 Ai2 ··· Aij
data. The precise value of V1 , V2 , I1 , I2 can be obtained through where
PMU, where V1 represents the voltage at sending end, V2 rep-
resents the voltage at receiving end, I1 represents the current at K = (Hp• ⊕ H1• ) (3)
sending end, I2 represents the current at receiving end. Then the
line impedance can be obtained as follows: Hp• is the pth row of H, and H is the identity matrix of
dimension i × i, ⊕ denote exclusive OR operator and j is the
(V1 − V2 ) (V1 + V2 ) number of bus within the ProNet. If bus j is connected with
Z= (1)
V1 I2 + V2 I1 line i, the value of Aij is 1, otherwise Aij is 0. Similarly, the
where, Z is the estimated value of transmission line impedance post fault node incidence matrix of the ProNet is denoted by B
based on the PMU measurements. If Z is within any protection which is defined as:
zone settings of the corresponding relay, it will be set to ‘1’, ⎡ ⎤
B11 B12 · · · B1j
which means the certain PMUs/relays ‘see’ the fault. Otherwise, ⎢ .. ⎥ (4)
B = ((K  × K) ⊕ H) × ⎣ ... ..
.
..
. . ⎦
‘0’ is assigned for this digit, which means the PMU did not ‘see’
the fault. The third digit is related to the corresponding breaker Bi1 Bi2 ··· Bij
status. ‘1’ means the breaker is open while ‘0’ will be given if the
where, K  is the transpose of K. when Bus j is connected with
breaker is closed. The second and third digit is together called
Line i, the value of Bij is 1, otherwise Bij it is 0. The tripping
as ‘fault digits’. The fourth and the fifth digits are collectively
of the transmission lines in a ProNet changes the node incidence
called as ‘zone digits’. They represent the zone number: ‘00’
matrix from A to matrix B. The first row of A and B corresponds
for zone 1, ‘01’ for zone 2, ‘10’ for zone 3 and ‘11’ means not
to the status of line p considered for possibility of fault. Each
in any of 3 zones. The composition of a sample 5-digit message
switching of transmission lines which corresponds from rows 2
is shown in Fig. 1.
to i in A and B are tested for possible correct explanation of
fault at line p within ProNet. The difference between A and B is
C. Multiple Hypothesis Generation defined as matrix C = A − B, where 1 in any row from two to
The next step is generating multiple hypothesis to explain all i in C is related to the tripping of circuit breakers. A destination
possible scenarios for the existing status of relays and circuit matrix D is also defined as follows:
breakers. Each hypothesis is a possible explanation of the exist-
ing status of the ProNet. D = [C ⊕ A] × diag (R) (5)
Authorized licensed use limited to: Rajnish Kumar. Downloaded on August 22,2024 at 09:00:08 UTC from IEEE Xplore. Restrictions apply.
2210 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 5, OCTOBER 2018

where R is a matrix denoted by

R = C1• ⊕ A1• (6)
The row k ∈ [i] − {1} (excluding the first row) has value 1,
when the circuit breaker connecting line k at that bus tripped. Fig. 2. Possible zone for fault location on transmission line.
The tripping of the circuit breaker is considered correct, if the
fault is within the protective zone of the corresponding relay. lines which are used to generate multiple hypothesis in the next
To test, if the relay can ‘see’the fault, following operations are step.
performed. Matrix R1 and R2 are defined as follows The multiple hypothesis are generated by considering fault
R1 = Ck • ⊕ Ak • (7) at each suspect transmission lines in the ProNet. Each suspect
transmission lines are divided into three section where sections
and at two ends spans 20% of the total length of the line and the
middle section spans 60% of the whole line. When a fault is
R2 = Ak • (8)
considered in this transmission line, the location of the fault
Using R1 and R2, K1 is derived as follows: varies in each of the three sections at a time as shown in Fig. 2.
When the fault is assumed to occur between 0% to 20% of the
K1 = C × [(R2 × R1) ∨ I] ∧ A (9)
line length from sending end of the line, the ‘zone digits’ mes-
where I is the identity matrix of dimension j × j and ∨ and sage for the PMU at sending end is ‘00’ corresponding to zone
∧ denote logical OR and AND operation, respectively. Now 1, and the ‘zone digits’ message for the PMU at receiving end
K1 is tested with the destination matrix D by evaluating T 1 as of the line is ‘01’ corresponding to zone 2. When the fault is
follows: assumed to occur between 20% to 80% of the line length from
the sending end, the ‘zone digits’ message for the PMU at both
T 1 = [K1 × diag (R1)] ∧ D (10) end of the line is ‘00’ corresponding to zone 1. When the fault
The failure diagnosis FDProSys tool checks the element of T 1 is assumed to occur at 80% to 100% of the line length from the
from second row to ith row. If any such element is equal to 1, sending end, the ‘zone digits’ message for the PMU at receiving
it means the relay at line k can see the fault at line p which is end is ‘00’ corresponding to zone 1, and the ‘zone digits’ mes-
within 120% length of the transmission line and successfully sage for the PMU at sending end is ‘01’ corresponding to zone
tripped the circuit breaker at j to isolate the fault at p. If all 2. Similarly, the ‘zone digits’ message for the PMUs at another
the elements of T 1 from second to row i are 0, the FDProSys suspect lines are also generated according to different assump-
tool checks, if the relay at line k tripped because it can see the tions of the fault location. The failure diagnosis tool checks the
fault at p within 200% of the length of the transmission lines. zone setting of every relays of adjacent lines and generate ‘ zone
The proposed FDProSys tool test the switching of row k by digits’ according to different assumptions of fault location. If
constructing K2 as follows: the FDProSys tool can not decide the fault location should be in
zone 2 or zone 3 of the relays on adjacent lines, two hypotheses
K2 = [[(P 1 × K1) × diag (R1)] ∨ C] (11) will be generated to ensure it covers all the possible conditions.
where In this way, the FDProSys tool can ensure it generates a hypoth-
esis set which covers every single explanation for the existing
P 1 = [A × R1 × Hk • ] ∨ H (12) anomalies in the protection system.
For each row m of K2, where m ∈ [i] − {1}, R3 m, R4 m For the multiple hypothesis generated by the tool, each hy-
and K3m are evaluated as: pothesis set corresponds to different suspect transmission lines.
Each hypothesis can have multiple sub-hypothesis which has
R3 m = K2m • ⊕ Am • (13) different ‘zone digits’ for each relay according to different fault
R4 m = Am • (14) location assumptions. After all the hypothesis has been gener-
ated by the FDProSys tool, the next step is selecting the hypoth-

K3 m = [K2 × [(R4 m × R3 m) ∨ I] ∨ A] × diag (R3 m ) esis which is the correct explanation of the existing status of
(15) relays and circuit breakers.
The second test matrix T 2 is calculated after evaluating K3 m
for m ∈ [i] − {1} in K2 as follows D. Hypothesis Selection
The Hypothesis Selection compares the last 4-digit message
T 2 = [K3 2 + K3 3 + · · · K3 m + · · · K3 i] ∧ D (16)
from each hypothesis with the last 4-digit message constructed
If any element from second row to ith row in T 2 is 1, then from PMU/relay measurements. Different credibility values are
the circuit breaker at line k tripped because the corresponding assigned to each hypothesis. The hypothesis with highest cred-
relay can see the fault at line p. Otherwise, the switching of ibility will be chosen as the correct explanation for the status of
breaker at line k is not correct as the fault at line p is outside the the relays and the circuit breaker as shown in the Fig. 3.
maximum protection zone of 200% of the length of transmission The F ault Digits(k)(H) and Zone Digits(k)(H)
line. After the possibility of fault for each line within ProNet has represent the fault digits and zone digits of PMU k for hypoth-
been tested, all the possible faulty lines are defined as suspect esis H, while F ault Digits(k)(R) and Zone Digits(k)(R)
Authorized licensed use limited to: Rajnish Kumar. Downloaded on August 22,2024 at 09:00:08 UTC from IEEE Xplore. Restrictions apply.
CUI et al.: AUTOMATED FAILURE DIAGNOSIS IN TRANSMISSION NETWORK PROTECTION SYSTEM 2211

Fig. 4. IEEE 14 bus power system in abnormal condition and ProNet Selec-
tion: Case 1.
Fig. 3. Comparing measured data with hypothesis.
the fault. This fault occurred in distance relay’s zone 1 protection
represent the fault digits and zone digits of PMU k obtained settings. The circuit breaker 7 operated by relay 7 is assumed
from the time stamped measurements. Creditability(k)(H) is to be malfunctioned by refusing to open which keeps the fault
the creditability value for PMU k for hypothesis H. connected to Bus 2. Relay 1, 3, 12 and 10 located at transmission
The aggregated creditability value of the hypothesis H which line 1-2, 2-5, 2-4 see the fault in the zone 2 of the protection
consists of n number of PMUs in the ProNet is given as settings. Bus 1 and 2 are connected by a pair of transmission
n lines denoted as line 1-2-1 and line 1-2-2. The relay 3, 12 and 10
Creditability(k) trips the corresponding breaker after a delay of 0.3 sec, which
Creditability(H) = k = 1 × 100%
n is used as zone 2 delay in this work. However, relay 1, which
(17) protects transmission line 1-2-2 fails to send the trip signal to
circuit breaker 3, which still connects the fault with the system.
III. PERFORMANCE EVALUATION Relay 6 which protects transmission line 1-5 detects the fault in
The proposed FDProSys tool is based on the ProNet selection, zone 3 protection settings and trips the circuit breaker after 1s,
which is a section of the network affected by fault and possible which is zone 3 time delay used in this work.
malfunctions of equipment. The sparse nature of the power There are several possible explanations for the existing status
system network and the proximity of malfunctions of equipment of the system after several relays and circuit breaker operated to
with the fault location results in the small size of the ProNet isolate the fault.
even for a large and complex system. The small ProNet results The failure diagnosis tool process the intact system topology
in limited size of the node incidence matrix as well as limited of the IEEE 14 bus system and find the ProNet based on the
number of 5-digit messages for processing. The performance of final system condition and suspect lines. The FDProSys tool
the proposed method is evaluated by creating faults and failure selects the transmission lines with opening breakers as the sus-
of protective devices in the IEEE 14-bus, IEEE 57 and IEEE pect lines and every neighboring lines into the initial ProNet.
300 bus test system. In this scenario, the initial ProNet chosen by the FDProSys tool
Computer-Aided Protection Engineering (CAPE) has been includes Bus 1, 2, 3, 4, 5, 6, 7, 9 and their interconnected trans-
used to calculate the distance relay zone settings for all the mission lines. Within the initial ProNet, the Bus 5 and 6, Bus
transmission lines for both the IEEE 14-bus and IEEE 57 test 4 and 9, Bus 4 and 7 are all connected through transformers.
system. These two test system are modelled in CAPE and the Since the distance relays at the transmission lines are focus of
relay settings obtained from CAPE are used in the relay zone this work, the ProNet excluded corresponding transformers and
settings in the OPAL-RT real time simulation. The zone 1, zone form a new ProNet. So the final ProNet after reduction is shown
2 and Zone 3 settings for all the distance relay is set up to 80%, in Fig. 4. ProNet has been created, including all suspect lines
120% and 200% maximum length of each transmission line, re- and adjacent lines of each suspect line.
spectively. The voltage and current phasors from each substation After selecting the final ProNet, the tool test the possibility
are obtained from the sensors located at each substation. of fault at each transmission line within the ProNet. The intact
system topology before the fault of this ProNet is shown in
Fig. 5. The final topology of the system within the ProNet is
A. Fault Scenario in Abnormal Condition: Case 1 shown in Fig. 6. The proposed FDProSys tool test the possibility
A three-phase line to ground fault is simulated in IEEE 14 of fault at all the transmission lines and verify if the present status
bus system [27] in the transmission line between Bus 2 and Bus of the relays and circuit breaker is a correct explanation for the
3 as shown in Fig. 4. The relay 7 and 8 at transmission line 2-3 fault considered. For testing the possibility of fault at line 3,
detects the fault and trip the corresponding breakers to isolate which is connected between bus 1 and 5, the node incidence
Authorized licensed use limited to: Rajnish Kumar. Downloaded on August 22,2024 at 09:00:08 UTC from IEEE Xplore. Restrictions apply.
2212 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 5, OCTOBER 2018

of relays and circuit breaker. Considering fault at line i7 , node

incidence matrix A and the final node incidence matrix B are
given by:

⎡1 2 3 4 5⎤ ⎡1 2 3 4 5⎤
0 0 1 1 0 0 0 1 1 0 l7
⎢1 1 0 0 0⎥ ⎢1 1 0 0 0⎥
⎢ ⎥ ⎢ ⎥ l2
⎢1 1 0 0 0⎥ ⎢0 1 0 0 0⎥
⎢ ⎥ ⎢ ⎥ l3
A= ⎢
⎢1 0 0 0 1⎥⎥ B= ⎢
⎢1 0 0 0 0⎥⎥ l4
⎢0 1 0 0 1⎥ ⎢0 1 0 0 0⎥
Fig. 5. Pre-fault system within ProNet: Case 1. ⎢ ⎥ ⎢ ⎥ l5
⎢0 1 0 1 0⎥ ⎢0 1 0 0 0⎥
⎢ ⎥ ⎢ ⎥ l6
⎣0 1 1 0 0⎦ ⎣0 1 0 0 0 ⎦ l1
0 0 0 1 1 0 0 0 1 1 l8
(22)

So non-zero elements of C = A − B and D for testing the row

3 of matrix C is given as follows:

Row Col V alue

3 1 1 Row Col V alue
Fig. 6. Post-fault status within ProNet: Case 1.
4 5 1 1 3 1
C= D= (23)
5 5 1 1 4 1
matrix A and the final node incidence matrix B for the intact 6 4 1 8 4 1
system are given by: 7 3 1
⎡1 2 3 4 5⎤ ⎡1 2 3 4 5⎤
1 0 0 0 1 1 0 0 0 0 l3 Evaluating both T1 and T2 using 7 to 15 results in
⎢1 1 0 0 0⎥ ⎢1 1 0 0 0⎥
⎢ ⎥ ⎢ ⎥ l2
⎢1 1 0 0 0⎥ ⎢0 1 0 0 0⎥ T1 = T2 = [0]8×5
⎢ ⎥ ⎢ ⎥ l1 (24)
A= ⎢
⎢0 1 0 0 1⎥⎥ B= ⎢
⎢0 1 0 0 0⎥⎥ l4
⎢0 1 0 1 0⎥ ⎢0 1 0 0 0⎥
⎢ ⎥ ⎢ ⎥ l5 As there is no ‘01’ in both T1 and T2 from row 2 to 7, so
⎢0 1 1 0 0⎥ ⎢0 1 0 0 0⎥
⎢ ⎥ ⎢ ⎥ l6 the switching of row 3 is incorrect response to isloate the fault
⎣0 0 1 1 0⎦ ⎣0 0 1 1 0 ⎦ l7 considered at line l7 , joining Bus 3 and 4.
0 0 0 1 1 0 0 0 1 1 l8 After testing the possibility of fault at each line within ProNet,
(18) the tool collects the voltage and current phasors data from
Then C is a 8 × 5 matrix, obtained by C = A − B and desti- PMUs/relays in the selected ProNet. The voltage and current
nation matrix D is also 8 × 5, obtained using (5) and (6) where phasor from the PMUs are processed to form the 5 digit mes-
the non-zero elements are are given by: sage for the real event which is shown in the Table I. Since
there are 3 different hypotheses for each suspect lines within the
Row Col V alue ProNet. The total number of hypotheses generated for this ex-
1 5 1 ample is 15, out of which credibility values of 3 hypothesis are
Row Col V alue
3 1 1
C= D= 1 1 1 (19) listed in Table I given limited space in this manuscript. Among
4 5 1
2 1 1 the 15 different hypotheses, the Hypothesis (6,1) has the highest
5 4 1 credibility value, hence, selected by tool as the right explana-
6 3 1 tion for the existing scenario. Hypothesis (6,1) assumes that
The T1 matrix is calculated for testing the switching of row the fault occurs at the transmission Line 2-3 (l6 ) at 0-20% and
3 in C, using (7), (8), (9) and (10). circuit breaker at PMU/relay 7 mal-operated. Hypothesis (3,2)
assumes fault at 20-80% section of transmission Line1-2-2 (l2 )
T1 = [0]8×5 (20) with credibility of 0.5625 and Hypothesis (6-2) assumes fault
T1 is tested for the value of one from row 2 to row 7. As there is at 20-80% section of the transmission Line 2-3 (l6 ) with cred-
no ‘01’ in T1 from row 2 to row 7, so T2 matrix is also calculated ibility of 0.84375. A scenario of missing PMU measurements
using 11, 12, 13, 14 and 15, which is given by: at the bus with mal-operated circuit breaker is also considered
in this case. The credibility evaluated for Hypothesis (6,2) with
Row Col V alue
T2 = (21) missing measurements from PMU 7 resulting in value of 0.8333
2 1 1
concludes that the FDProSys tool can identify the correct hy-
The row 2 in T2 has one, which means the consideration of fault pothesis with missing PMU measurement at the mal-operated
at line 3 is a correct explanation of the present system condition circuit breaker bus.

Authorized licensed use limited to: Rajnish Kumar. Downloaded on August 22,2024 at 09:00:08 UTC from IEEE Xplore. Restrictions apply.
CUI et al.: AUTOMATED FAILURE DIAGNOSIS IN TRANSMISSION NETWORK PROTECTION SYSTEM 2213

TABLE I
5 DIGIT MESSAGES COMPARISON BETWEEN REAL EVENT AND TWO SELECTED HYPOTHESIS: CASE 1

Since there are 3 different hypotheses for each suspect lines

within the ProNet. The total number of hypotheses generated
for this case is 9 out of which credibility values of 3 hypothesis
are listed in Table II. Among the 9 different hypotheses, the
Hypothesis (3,2) has the highest credibility value and chosen
by tool as the right explanation for the observed measurements.
Hypothesis (3,2) assumes that the fault occurs at 20-80% section
of the transmission Line 38-48 and circuit breaker at PMU/relay
158 mal-operated. Hypothesis (2,1) assumes fault at 0-20% sec-
tion of transmission Line 47-48 (l2 ) with credibility 0.556 and
Hypothesis (3,3) assumes fault at 80-100% section of the trans-
mission Line 38-48 (l3 ) with credibility 0.917. Missing PMU
measurements at the bus with correct breaker operation is con-
sidered by removing measurements from PMU/relay 157 re-
sulting in credibility of Hypothesis (3,3) as 0.941. The missing
Fig. 7. Anomaly in protection system and ProNet Selection: Case 2. PMU measurements at bus with correct operation of the circuit
breaker do not affect the Hypothesis selection of FDProSys tool.
B. Fault Scenario in Abnormal Condition: Case 2
The three phase to ground fault is simulated at 0.5s at trans-
mission line between Bus 48 and Bus 38 in IEEE 57 bus system
[27] as shown in Fig. 7 (zoom in area of the 57-bus system). The C. Fault Scenario in Abnormal Condition: Case 3
distance relay 157 and 158 located at Bus 38 and Bus 48 detects The 300 bus system [27] is considered in evaluating the per-
this fault and trip the corresponding circuit breaker. However, formance of proposed FDProSys tool. The fault is simulated in
breaker 158 at Bus 48 malfunctioned and do not open. The relay the middle of the line joining Bus 134 and Bus 135 as shown
121 and 124 at Bus 47 and Bus 49 detects the fault in zone 2 pro- in Fig. 10. The relay 1 and 2 detects the fault in line 134-135
tection settings and trips the circuit breaker as shown in Fig. 7. and trips the corresponding circuit breaker. The circuit breaker
Line 38-48, line 47-48 and line 48-49 are chosen as suspect 2 connected to relay 2 malfunctions and fails to open by the
lines which are used to estimate the ProNet. The ProNet includes trip command. The relay 4 sees the fault in the second zone
the 3 possible faulty transmission lines which are line 38-48, of protection and trips the circuit breaker 4 to isolate the fault.
line 47-48 and line 48-49 and their adjacent lines 46-47, 22-38, The network consisting of buses 127, 184, 134, 135, 136, 137
44-38, 37-38, 49-50, 46-47, 49-13 as shown in Fig. 7. and 152 is considered in the ProNet as shown in Fig. 10. The
All of the valid hypotheses are generated using the initial and intact system topology before the fault of this ProNet is shown
final node incidence matrix, which are shown in Figs. 8 and 9, in Fig. 11. The final topology of the system within the ProNet
respectively. The 5 digit message has been calculated by the is shown in Fig. 12. All the valid hypothesis are generated by
FDProSys tool for each relay in the ProNet, which is shown in the algorithm using the initial and final node incidence matrix.
Table II. The FDProSys tool generates multiple hypotheses to For testing the possibility of fault at line l1 , which is connected
cover all possible explanations for existing status of relays and between bus 134 and 135, the node incidence matrix A and the
the circuit breakers. final node incidence matrix B for the intact system are given

Authorized licensed use limited to: Rajnish Kumar. Downloaded on August 22,2024 at 09:00:08 UTC from IEEE Xplore. Restrictions apply.
2214 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 5, OCTOBER 2018

TABLE II
5 DIGIT MESSAGES COMPARISON BETWEEN REAL EVENT AND TWO SELECTED HYPOTHESIS: CASE 2

Fig. 8. Pre-fault system within ProNet: Case 2.

Fig. 10. Anomaly in protection system and ProNet Selection: Case 3.

⎡127 134 135 136 137 152 184⎤

0 0 1 0 0 0 0 l1
⎢0 0 1 0 0 0 0⎥ l2
⎢ ⎥
Fig. 9. Post-fault status within ProNet: Case 2. B= ⎢
⎢0 0 0 1 1 0 0⎥⎥ l3 (26)
⎢0 0 0 1 0 1 0⎥ l4
⎢ ⎥
by: ⎣1 1 0 0 0 0 0⎦ l5
127 134 135 136 137 152 184 0 1 0 0 0 0 1 l6
⎡ ⎤
0 1 1 0 0 0 0 l1 So non-zero elements of C = A − B and D for testing the row
⎢0 0 1 1 0 0 0⎥ l2
⎢ 0⎥
1 of matrix C is given as follows:
A = ⎢0 0 0 1 1 0 ⎥ l3 (25)
⎢0 0 0 1 0 1 0⎥
⎢ ⎥ l4 Row Col V alue Row Col V alue
⎣1 1 0 0 0 0 0⎦ l5 C= 1 2 1 D= 1 3 1 (27)
0 1 0 0 0 0 1 l6 2 4 1 2 3 1
Authorized licensed use limited to: Rajnish Kumar. Downloaded on August 22,2024 at 09:00:08 UTC from IEEE Xplore. Restrictions apply.
CUI et al.: AUTOMATED FAILURE DIAGNOSIS IN TRANSMISSION NETWORK PROTECTION SYSTEM 2215

TABLE III
5 DIGIT MESSAGES COMPARISON BETWEEN REAL EVENT AND TWO SELECTED HYPOTHESIS: CASE 3

Fig. 11. Pre-fault system within ProNet: Case 3. Fig. 12. Post-fault status within ProNet: Case 3.

The T1 matrix is calculated for testing the switching of row 2 in sion Line 135-136 (l2 ) with credibility 0.4167. Scenario of miss-
C, using (7), (8), (9), and (10). ing PMU measurements from the bus connected to un-faulted
Row Col V alue line is considered in this case by removing measurements from
C= (28) PMU/relay 3. The Hypothesis (2,2) with missing measurements
2 3 1
from PMU 3 has credibility of 0.455, resulting in correct se-
The element (2, 3) of matrix T1 is one, which concludes lection of Hypothesis with missing PMU measurement from
that the switching of line 2 is a valid hypothesis of the present un-faulted line.
condition of the system.
The possibility of fault at each line within ProNet is tested
for validity. The 5-digit messages is of the real event and 3 IV. CONCLUSIONS
valid hypothesis is shown in Table III. The Hypothesis (1,1) In this work, a multi-hypothesis based FDProSys tool using
has the highest credibility value, hence, selected by tool as the PMUs/relays data have been developed for the failure diagno-
right explanation for the observed scenario. Hypothesis (1,1) as- sis. The algorithm is designed to run in the control center after
sumes the fault occurs at 0-20% section of the transmission Line fault occurs in the presence of the malfunctioned devices in the
134-135 (l1 ) and circuit breaker at PMU/relay 2 mal-operated. protection system. The proposed FDProSys tool can virtually
The Hypothesis (2,3) assumes fault at 80-100% section of the identify ProNet based on system topology and circuit breaker
transmission Line 135-136 (l2 ) with credibility 0.4167 and Hy- status. The data from PMUs/relays within ProNet is used to cal-
pothesis (2,2) assumes fault at 20-80% section of the transmis- culate the 5-digit data message which includes 1 trust digit and
Authorized licensed use limited to: Rajnish Kumar. Downloaded on August 22,2024 at 09:00:08 UTC from IEEE Xplore. Restrictions apply.
2216 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 5, OCTOBER 2018

4 protection information digits. To generate multiple hypothe- [17] N. Mahadevan, A. Dubey, A. Chhokra, H. Guo, and G. Karsai, “Using
sis, the developed FDProSys tool uses matrix manipulation to temporal causal models to isolate failures in power system protection
devices,” IEEE Instrum. Meas. Mag., vol. 18, no. 4, pp. 28–39, Aug.
filter out non-suspect transmission lines and then create mul- 2015.
tiple hypothesis according to different potential fault location [18] H.-J. Lee, B.-S. Ahn, and Y.-M. Park, “A fault diagnosis expert system for
assumptions on each suspect line. Through 5-digit messages distribution substations,” IEEE Trans. Power Del., vol. 15, no. 1, pp. 92–
97, Jan. 2000.
comparison and credibility score given, developed FDProSys [19] B. Kasztenny, V. Muthukrishnan, and T. S. Sidhu, “Enhanced numeri-
tool selects the hypothesis with highest credibility. The effec- cal breaker failure protection,” IEEE Trans. Power Del., vol. 23, no. 4,
tiveness of the proposed method is demonstrated by considering pp. 1838–1845, Oct. 2008.
[20] V. Muthukrishnan and T. S. Sidhu, “Fast and secure breaker failure detec-
missing PMU measurements from a critical bus in the system, tion algorithms,” IET Gen., Transm. Distrib., vol. 3, no. 2, pp. 198–205,
but still accurately identifying the fault location and breaker Feb. 2009.
failures. By different layered group digits design, the proposed [21] M. M. Eissa, M. E. Masoud, and M. M. M. Elanwar, “A novel back up
wide area protection technique for power transmission grids using phasor
FDProSys tool is robust and can generate the correct explana- measurement unit,” IEEE Trans. Power Del., vol. 25, no. 1, pp. 270–278,
tion even with several PMUs/relays failure. Simulation results Jan. 2010.
for the IEEE 14 bus, IEEE 57 bus and IEEE 300 bus test system [22] W. A. dos Santos Fonseca, U. H. Bezerra, M. V. A. Nunes, F. G. N.
Barros, and J. A. P. Moutinho, “Simultaneous fault section estimation and
demonstrates the accuracy of the proposed algorithm. protective device failure detection using percentage values of the protective
devices alarms,” IEEE Trans. Power Syst., vol. 28, no. 1, pp. 170–180,
REFERENCES Feb. 2013.
[23] A. Salehi-Dobakhshari and A. M. Ranjbar, “Application of synchronised
[1] A. Ghaderi, H. Mohammadpour, H. Ginn, and Y.-J. Shin, “High- phasor measurements to wide-area fault diagnosis and location,” IET Gen.
impedance fault detection in the distribution network using the time- Transm. Distrib., vol. 8, no. 4, pp. 716–729, Apr. 2014.
frequency-based algorithm,” IEEE Trans. Power Del., vol. 30, no. 3, [24] P. Kundu and A. K. Pradhan, “Online identification of protection element
pp. 1260–1268, Jun. 2015. failure using wide area measurements,” IET Gen. Transm. Distrib., vol. 9,
[2] Y. Liao, “Fault location for single-circuit line based on bus-impedance no. 2, pp. 115–123, 2015.
matrix utilizing voltage measurements,” IEEE Trans. Power Del., vol. 23, [25] J. Jung, C.-C. Liu, M. Hong, M. Gallanti, and G. Tornielli, “Multiple
no. 2, pp. 609–617, Apr. 2008. hypotheses and their credibility in on-line fault diagnosis,” IEEE Trans.
[3] N. Kang and Y. Liao, “Double-circuit transmission-line fault location with Power Del., vol. 16, no. 2, pp. 225–230, Apr. 2001.
the availability of limited voltage measurements,” IEEE Trans. Power Del., [26] W.-H. Chen, C.-W. Liu, and M.-S. Tsai, “On-line fault diagnosis of distri-
vol. 27, no. 1, pp. 325–336, Jan. 2012. bution substations using hybrid cause-effect network and fuzzy rule-based
[4] A. Esmaeilian and M. Kezunovic, “An impedance based fault location method,” IEEE Trans. Power Del., vol. 15, no. 2, pp. 710–717, Apr. 2000.
algorithm for tapped lines using local measurements,” in Proc. North [27] IEEE Test Systems. [Online]. Available: www2.ee.washington.edu/
Amer. Power Symp., Sep. 2013, pp. 1–6. research/pstca/pf14/pgtca14bus.htm, Accessed on: Apr. 2018.
[5] M. Korkali, H. Lev-Ari, and A. Abur, “Traveling-wave-based fault-
location technique for transmission grids via wide-area synchronized volt-
age measurements,” IEEE Trans. Power Syst., vol. 27, no. 2, pp. 1003– B. Cui (S’16) received the Master of Science degree
1011, May 2012. in electrical engineering from Carnegie Mellon Uni-
[6] A. Johns, R. Aggarwal, and Z. Bo, “Non-unit protection technique for versity, Pittsburgh, PA, USA, in 2013 and the Ph.D.
EHV transmission systems based on fault-generated noise. Part 1: Signal degree from the Department of Electrical Engineer-
measurement,” Proc. Inst. Elect. Eng., Gen. Transm. Distrib., vol. 141, ing and Computer Science of Washington State Uni-
no. 2, pp. 133–140, Mar. 1994. versity, Pullman, WA, USA. His research interests
[7] R. Aggarwal, A. Johns, and Z. Bo, “Non-unit protection technique for include smart grid, cyber-physical systems, failure
ehv transmission systems based on fault-generated noise. Part 2: Signal diagnosis and prognosis, and cascading outages pre-
processing,” Proc. Inst. Elect. Eng., Gen. Transm. Distrib., vol. 141, no. 2, vention.
pp. 141–147, Mar. 1994.
[8] M. Sanaye-Pasand and P. Jafarian, “An adaptive decision logic to enhance
distance protection of transmission lines,” IEEE Trans. Power Del., vol. 26, A. K. Srivastava (SM’08) received the Ph.D. degree
no. 4, pp. 2134–2144, Oct. 2011. in electrical engineering from the Illinois Institute
[9] C. Pang and M. Kezunovic, “Detection tools for disturbances and protec- of Technology, Chicago, IL, USA, in 2005. He is
tive relay operations leading to cascading events,” in Proc. Power Energy currently an Associate Professor of Electric Power
Soc. Gen. Meeting, Jul. 2009, pp. 1–6. Engineering with Washington State University, Pull-
[10] F. Yang, A. Meliopoulos, G. Cokkinides, and Q. Binh Dam, “Effects of man, WA, USA, and the Director of the Smart Grid
protection system hidden failures on bulk power system reliability,” in Demonstration and Research Investigation Lab, En-
Proc. 38th North Amer. Power Symp., Sep. 2006, pp. 517–523. ergy System Innovation Center (ESIC). His research
[11] G. Cardoso, J. Rolim, and H. Zurn, “Identifying the primary fault section interests include the data-driven algorithm for power
after contingencies in bulk power systems,” IEEE Trans. Power Del., system operation and control. He is an Editor of the
vol. 23, no. 3, pp. 1335–1342, Jul. 2008. IEEE TRANSACTIONS ON SMART GRID; IEEE TRANS-
[12] Y.-C. Huang, “Fault section estimation in power systems using a novel ACTIONS ON POWER SYSTEMS; IET Generation, Transmission and Distribution;
decision support system,” IEEE Trans. Power Syst., vol. 17, no. 2, pp. 439– and Sustainable Computing. He is an IEEE distinguished Lecturer and the coau-
444, May 2002. thor of more than 250 technical publications.
[13] F. Salfner and M. Malek, “Using hidden semi-Markov models for effective
online failure prediction,” in Proc. 26th IEEE Int. Symp. Rel. Distrib. Syst.,
Oct. 2007, pp. 161–174.
[14] X. Gao, J. S. Thorp, and D. Hou, “Case studies: Designing protection P. Banerjee (M’17) received the Ph.D. degree from
systems that minimize potential hidden failures,” in Proc. 66th Annu. the Indian Institute of Technology Kanpur, Kanpur,
Conf. Protect. Relay Eng., Apr. 2013, pp. 384–393. India, in 2015. He is currently an Assistant Research
[15] Z. Jiao, H. Gong, and Y. Wang, “A D-S evidence theory-based relay Professor with the School of Electrical Engineering
protection system hidden failures detection method in smart grid,” IEEE and Computer Science, Washington State University,
Trans. Smart Grid, vol. 9, no. 3, pp. 2118–2126, May 2018. Pullman, WA, USA. His current research mainly fo-
[16] A. Esmaeilian, T. Popovic, and M. Kezunovic, “Transmission line relay cuses on wide-area monitoring and control, real-time
mis-operation detection based on time-synchronized field data,” Elect. simulation, and synchrophasor applications in power
Power Syst. Res., vol. 125, pp. 174–183, 2015. [Online]. Available: systems.
http://www.sciencedirect.com/science/article/pii/S0378779615001145

Authorized licensed use limited to: Rajnish Kumar. Downloaded on August 22,2024 at 09:00:08 UTC from IEEE Xplore. Restrictions apply.