Data-mining for Fault-Zone Detection of Distance

Relay in FACTS-Based Transmission

Nima Salek Gilani Mohammad Tavakoli Bina Fatemeh Rahmani Mahmood Hosseini Imani
Department of Electrical Department of Electrical Department of Electrical Department of Energy
Engineering Engineering Engineering DENERG
[Link] University of [Link] University of Lamar University Politecnico di Torino
Technology Technology
Tehran, Iran Tehran, Iran Beaumont, TX, USA Turin, Italy
[Link]@[Link] Tavakoli@[Link] frahmani@[Link] [Link]@polito.i
t

Abstract—In this study, the problem of fault zone detection of transmission line is presented. In this solution, zone and location
distance relaying in FACTS-based transmission lines is analyzed. of the fault, independent of the fault resistance, have been
Existence of FACTS devices on the transmission line, when they estimated using two subroutines. Some papers developed their
are included in the fault zone, from the distance relay point of methods using measurements from principle components of the
view, causes different problems in determining the exact location main frequency. Saha et al. in [3] developed a method based on
of the fault by changing the impedance seen by the relay. The components of main frequency on one side of the transmission
extent of these changes depends on the parameters that have been line using the Goldsworthy model.
set in FACTS devices. To solve the problem associated with these
compensators, two instruments for separation and analysis of In [4], an algorithm based on forward and backward waves
three line currents, from the relay point of view at fault instance, is presented to protect series compensated transmission lines;
have been utilized. The wavelet transform was used to separate the although only one phase to ground fault has been analyzed in
high-frequency components of the three line currents, and the this paper.
support vector machine (using methods for multi-class usage) was
used for classification of fault location into three protection Data have been presented in the literature [5-7] which
regions of distance relay. Besides, to investigate the effects of suggest methods based on neural networks. It has been
TCSC location on fault zone detection of distance relay, two suggested to use RBFN as a radial neural network to classify
places, one in fifty percent of line length and the other in seventy- fault type and fault location [7]. Reference [6] trains two
five percent of line length, have been considered as two scenarios different types of neural networks to identify fault type and
for confirmation of the proposed method. Simulations indicate location. Also, [5] uses the EDBD algorithm to calculate the
that this method is effective in the protection of FACTS-based fault location. Neural networks have limitations, including their
transmission lines. need for a massive amount of neurons to model network
structure. Therefore, numerous data from the electrical network
Keywords— Data-mining, Distance relay, FACTS, SVM is needed, and also much time to train the network. Nevertheless,
I. INTRODUCTION these papers used a limited number of data to test their methods.

Increasing demand, caused by the development of cities and Girgis et al. in [8] proposed a method based on Kalman filter
growing population, draws more attention on power [9] which estimate the state of a dynamic system from a series
transmission limits. Power Electronics helps to solve the of measurements containing errors. In [10], a method using
problem by introducing flexible ac transmission system wavelet transform is suggested. In both [8] and [10], limited
(FACTS) devices. The general advantages of FACTS devices cases have been used to assess the effectiveness of the proposed
are well known. However, FACTS devices can lead to severe methods. Furthermore, in [10] high sampling frequency of 200
problems in determining the exact location of the fault when KHz is used, which seems to be difficult in practical
they are included in the fault zone. Distance relay’s function is implementation.
based on measured impedance by which they calculate the fault In [11], a data mining method is used for three purposes.
location. Installed FACTS devices on the transmission line Support vector machine (SVM) has become a popular tool in the
change the impedance measured by distance relay, which power grid [12]. In this paper, three separate SVM is trained to
consequently leads to wrong fault zone detection. The amount determine fault location, presence or absence of fault and
of deviation in the impedance depends on the parameters whether the fault includes the ground or not. This reference uses
settings of the FACTS device. samples of firing angle and current, starting from the fault
In recent years, researchers have become increasingly instance till half cycle after it. Considering 400 kV and 50 Hz
interested in solving this issue. For modeling series capacitor in case study and sampling frequency of 1 KHz in this paper, ten
the fault period, most of the papers used the model presented by samples of current from each phase is analyzed. In this paper,
Goldsworthy [1]. In [2], a solution based on time equations of fault location is confined to before and after the compensator,
network with consideration of an extensive model of a and its effectiveness is studied only in 200 cases.

978-1-7281-4436-8/20/$31.00 ©2020 IEEE

Authorized licensed use limited to: University College London. Downloaded on May 25,2020 at [Link] UTC from IEEE Xplore. Restrictions apply.
 In [13], a two stage method is proposed to identify the fault [13], it was reported that from different types of db mother
location. In this paper, samples of three line currents for one wavelets, db2 gives the best results in the detection of fault
cycle after the fault instance is used to train SVM. Considering location, in the presence of series capacitor in a transmission
400 kV and 50 Hz case study and sampling frequency of 1 KHz line. It was also reported that frequencies between 1000 to 2000
in this paper, 80 samples of current from each phase is analyzed. Hz are suitable for best fault zone classification. Therefore,
In the first stage, high frequencies of three line currents are considering a sampling frequency of 4 KHz in PSCAD in this
separated using wavelet transform; then in the second stage, study, single-level discrete wavelet transform with db2 mother
these separated signals are given to SVM to classify the fault wavelet is carried out to obtain high frequency samples. After
location to before and after the compensator. The problem is that that, wavelet transform’s outputs are used as inputs to SVM to
the model used for simulation of the transmission line is not create a hyper plane to separate fault zones.
proper to study non-base frequencies while this paper used high
frequencies to separate the fault locations. This method, III. CASE STUDIES
considering using SVM, can only separate the fault location to A. Simulation of case studies in PSCAD
before and after the compensator. Taking into account 400 kV
and 50 Hz case study and sampling frequency of 1 KHz in this For TCSC simulation in PSCAD, reference model in [13] is
paper, 60 samples are analyzed. used. This model is shown in Fig. 1. The capacitance of SC
capacitor changes in proportion with the firing angle.
Researchers in [14] proposed the random forest method [15-
18] to identify the fault location. This data mining method is The feasibility of the proposed method is evaluated on a 400
founded based on the decision tree (DT) method. In this method, kV 50 Hz power system. Both sides of this network are two
several independent unpruned decision trees are produced using sources that represent two areas which are connected together.
different subsets of the data. Each tree has one vote, and the most The source data at both sides of the transmission line is shown
vote determines the result. Samples of current and voltage in half in Table I. Frequency-dependent (phase) model is used for
cycle after the fault instance are utilized for analysis. Seventy detailed simulation of transmission line. Transmission line data
percent of the samples are used for the training process of is achieved from a real 400 kV single-circuit transmission line
random forest, and the remaining thirty percent is used for the with a length of 250 km. this data is presented in Table II.
testing process. Considering a large amount of training data, Following this line of 250 kilometers, a line with a length of
good results have been achieved. Similarly, in this paper, the 100 km and then a line with length of 50 km with the same
fault location is limited to before and after the compensator. profile was considered. To evaluate the impact of TCSC location
The object of this research is to improve distance protection on the issue of protection, this element was placed in two
performance, using data mining methods while considering locations, first in the middle of the line 1 (125 km of line 1) and
FACTS-based transmission lines. Earlier works using artificial second, in Seventy-five percent of the line 1 (187.5 km of line
neural network suffer from limitations on high number of 1), which in continuation of this paper for the convenience of
neurons. These limitations can be overcome by using more readers will be referred to as scenario 1 and scenario 2,
efficient tools like support vector machines. Furthermore, a respectively. Three protection zones have been defined for
more precise simulation of the transmission line is conducted distance relay, first till 80 percent of the first line, second till 50
here using PSCAD/EMTDC software, as well as separating the percent of the second line and third till 25 percent of the third
fault areas into three zones according to distance relay settings. line.
Besides, to investigate the effects of TCSC1 location on fault
zone detection of distance relay, two places --one in fifty percent G1
SC
G2
of line length and the other in seventy five percent of line length- I
- have been considered as two scenarios for confirmation of the A B

proposed method. The simulations show a 94.2% of success rate MOV Ld

in scenario one and 86.7% in the other, which are high success AG

rates, which indicate that placing TCSC in fifty percent of line CB

length leads to a higher prediction accuracy. Moreover, these
success rates are achieved using only 15.2% of total data as the
training set, which shows the high capability of proposed
method. Fi 1 TCSC i l ti d l [13]
II. PROPOSED METHOD TABLE I. SOURCE DATA AT BOTH ENDS
In this study, PSCAD software is used for simulation. Three
line currents are recorded after creating a variety of fault Parameters Value
conditions on the simulated transmission line. These outputs Source Voltage 400 kV
from PSCAD are used as inputs to wavelet transform in order to Frequency 50 Hz
extract high frequencies (between 1000 to 2000 Hz). It is Positive sequence impedance 1.31+j15 Ÿ
reported in [19] and [20] that the most proper wavelet in
studying faults in power system is Daubechies (db) wavelets. In Zero sequence impedance 2.33+j26.6 Ÿ

1
Thyristor Controlled Series Capacitor

Authorized licensed use limited to: University College London. Downloaded on May 25,2020 at [Link] UTC from IEEE Xplore. Restrictions apply.
 TABLE II . TRANSMISSION LINE DATA

Data type value Data type value

Arrangement of
Horizontal Vertical distance of conductors 15.45 m
wires
Line length 250 km SAG for all conductors 14 m
Number of bundle 2 Number of ground wires 2
Height of ground wires above the
Bundle spacing 45 cm 9.36 m
lowest conductor
Conductor DC
0.0553 ohm/km Ground wire DC resistance 1.463 ohm/km
resistance
Conductor GMR 1.2161 cm Ground wire radius 0.2445 cm
Height of conductors 41.46 m Spacing between ground wires 18.70 m

In scenario one which TCSC is placed in the middle of the B. Classification with One-Against-All method (OAA)
first line, four fault locations are assumed: in 50 km (first zone, There are various ways to use SVM to classify data into more
before TCSC), in 150 km (first zone, after TCSC), in 250 km than two classes. In OAA assuming k classes, k SVM is trained
(second zone), in 325 km (third zone). In scenario 2, three fault for each class with the samples of that class as positive samples
locations are assumed: in 100 km (first zone), in 250 km (second and all other samples as negatives.
zone), in 325 km (third zone).
1) Analysis of scenario 1
In order to evaluate the feasibility of proposed method under Getting a proper output from SVM can be achieved through
various system condition, different kinds of fault type, fault metaheuristic algorithms or trial and error learning [13]. In order
resistance, fault inception angle (FIA), load angle (į), generator to utilize trial and error learning different values for
impedance, compensation percentage and fault location, which regularization parameter (C) and gamma (g) (RBF kernel
include 28800 cases for scenario 1 and 21600 cases for scenario parameter) should be evaluated on the classification of training
2, have been analyzed utilizing multiple run in PSCAD. Table data in order to choose the best ones, then the result is used to
III shows considered cases. 25%, 50%, and 75% are chosen classify the test data. For the range of these parameters, results
TCSC compensation percentages for 250 km transmission line, in reference [13] was a great help. The best result is achieved in
their equivalent value for SC capacitor is 120.45, 60.225 and 21096 and 14.5 for C and g, respectively. With these values,
40.15 μf respectively. 83.7% of success is achieved in fault zone detection.
Randomly chosen cases for the training process of SVM are
2) Analysis of scenario 2
3600 cases for scenario 1 and 2700 cases for scenario 2. These
cases are only 12.5% of the total data set and are shown in Table In result analysis of scenario two, it could be seen that the most
IV. The Gaussian RBF kernel has been used for training and number of correct fault zone detection was reached with
testing process in SVM.

TABLE III. CONSIDERED CASES FOR SYSTEM ANALYSIS

Parameters
Number Number of
Case of fault fault cases
number cases for for Fault Fault
%ZG1 %ZG2 Ψܺܿ ܴ௙ FIA į
scenario 1 scenario 2 types locations

ͳ ͷ͹͸Ͳ Ͷ͵ʹͲ ͳͲͲ ͳͲͲ 4 and 3

location for
ʹ ͷ͹͸Ͳ Ͷ͵ʹͲ ͳͲͲ ͹ͷ ʹͷǡͷͲ Ͳǡͷ ͲǡͶͷ ͳͲǡʹͲ
scenario 1
͵ ͷ͹͸Ͳ Ͷ͵ʹͲ ͳͲͲ ͳʹͷ & & & & 10
and
͹ͷ ʹͷǡͷͲ ͺͳǡͳͳ͹ ͵Ͳ scenario 2
Ͷ ͷ͹͸Ͳ Ͷ͵ʹͲ ͹ͷ ͳͲͲ
ͷ ͷ͹͸Ͳ Ͷ͵ʹͲ ͳʹͷ ͳͲͲ respectively
total number of cases in scenario 1= 5*5760=28800
total number of cases in scenario 2= 5*4320=21600

Authorized licensed use limited to: University College London. Downloaded on May 25,2020 at [Link] UTC from IEEE Xplore. Restrictions apply.
 TABLE IV. CHOSEN CASES FOR TRAINING PROCESS OF SVM

Number Number of Parameters

Case of fault fault cases
number cases for for Fault Fault
%ZG1 %ZG2 %Xc Rf FIA į
scenario 1 scenario 2 types locations

1 720 540 100 100 50 4 and 3

location for
2 720 540 100 75 50 0,5 0,45 10 scenario 1
3 720 540 100 125 50 & & & 10
4 720 540 75 100 50 and
50 117 30
scenario 2
5 720 540 125 100 50 respectively
total number of cases in scenario 1= 5*720=3600
total number of cases in scenario 2= 5*540=2700

C=1080000 and g=38.4. The success rate was 69.7%, which predictions happened in data with load angles (į) equal to 20,
was less than scenario one. In section IV causes of this lower which was not included in the training data set. Therefore,
success rate will be discussed. random data sets with load angles (į) equal to 20 were added to
the training set. Table VII and Table VIII Shows the added data
C. Classification with One-Against-One method (OAO) for scenario 1 and scenario 2, respectively. Consider that now
For this method, a table should be established, and each 15.3% of total data is used as the training data set.
SVM should cast its vote on a class between two classes. The
class that achieves the highest number of votes is chosen by the 1) Analysis of scenario 1
combined classifiers. Possible scenarios are shown in Table V. The maximum number of correct detection of fault zone
using OAA in the modified training set was 22993 cases which
1) Analysis of scenario 1 happened in C=10 and g=8.3. Modified training set increased
The maximum number of correct detection of fault zone the success rate from 83.7% to 94.2%.
using OAO in scenario 1 was 20629 cases which happened in
C=10000 and g=13.1. This number is equivalent to 81.8% of 2) Analysis of scenario 2
success. The maximum number of correct detection of fault zone
using OAA in the modified training set was 15880 cases which
Following up on OAO method, with analysis of results, one happened in C=10 and g=14.5. Modified training set increased
can decide for cases that the table cannot vote for their class. the success rate from 69 .7% to 86.7%.
After checking the results, Table V was replaced with Table VI.
The maximum number of correct detection of fault zone
using Table VI was 21239, which happened in C=10 and
g=10.4. As expected, the success rate increased to 84.2%.
TABLE V. POSSIBLE SCENARIOS IN VOTING
2) Analysis of scenario 2
The maximum number of correct detection of fault zone Groups Possible conditions
using OAO in scenario 2 was 11802 cases which happened in 1,2 1 1 1 1 0 0 0 0
C=10 and g=20. This number is equivalent to %62.2 of success.
1,3 1 1 0 0 1 1 0 0
After replacing Table V by Table VI number of correct cases
increased to 123399 in C=10 and g=21.3, which is equivalent to 2,3 0 1 0 1 0 1 1 0
65.6% of success. result 1 1 3 0 0 2 2 3
D. Classification with OAA method in the modified training
set TABLE VI. MODIFIED SCENARIOS IN VOTING
As can be seen, a substantial percentage of success was not Groups Possible conditions
achieved for scenario 1 and 2 with both methods. One of the
effective parameters to achieve a good result in classification is 1,2 1 1 1 1 0 0 0 0
having a proper and right amount of training data set. One cannot 1,3 1 1 0 0 1 1 0 0
expect an algorithm to predict properly in unseen areas. It is
usual to put 70% of the data set into the training set and 30% to 2,3 0 1 0 1 0 1 1 0
the test set. Until now, only 12.5% of the data set was used in result 1 1 3 3 2 2 2 3
the training set. Cases, which their fault zone was identified
incorrectly, were analyzed. A high proportion of wrong

Authorized licensed use limited to: University College London. Downloaded on May 25,2020 at [Link] UTC from IEEE Xplore. Restrictions apply.
 IV. DISCUSSION ON THE RESULTS
According to the results, the highest success rate is achieved
TABLE VII. ADDED DATA TO TRAINING SET IN SCENARIO 1 in scenario 1 (installing TCSC in the middle of the first line).
Number Parameters This success rate (94.2%) was reached with OAA method on the
Case modified training set. In scenario two, the highest success rate
of fault was 86.7%, which was also reached with OAA method on the
number %ZG1 %ZG2 %XC Rf FIA į
cases modified training set. Thus, the prominent role of selecting a
1 160 100 100 proper training data set had become evident. To investigate why
81 scenario 2 had less success rate than scenario 1, cases with the
2 160 100 75 25 & wrong detection of the fault zone in both scenarios were
3 160 100 125 & 50 117 20 analyzed in Table X and Table XI. They are also visualized in
4 160 75 100 50 Fig. 2 and Fig. 3 for better understanding. By comparing these
5 160 125 100 two tables, it can be concluded that an increase in wrong zone
Total number=5*160=800 detections happened in zone 2 and 3, whereas in zone 1 it almost
remained constant. In other words, SVM made more mistakes
because pushing the TCSC forward increased its area of
influence. Likewise, in both cases, zone 2 presented the highest
TABLE VIII. ADDED DATA TO TRAINING SET IN SCENARIO 2
amount of wrong detection and that’s because TCSC has its
Number Parameters highest effect on this zone as it is placed right before it.
Case
of fault In this research capability and efficiency of the proposed
number %ZG1 %ZG2 %XC Rf FIA į
cases method was evaluated on the sample network. It should be noted
1 120 100 100 that with a change in the network topology, received information
from the distance relay point of view would change. In this case
2 120 100 75 25 0 SVM parameters should be updated.
81
3 120 100 125 & & 20
4 120 75 100 50 50 V. CONCLUSIOIN
5 120 125 100 In this paper, a new method for fault zone detection of
Total number=5*120=600 distance relay in FACTS-based transmission lines has been
proposed. For investigating the effects of TCSC location on fault
zone detection of distance relay, two places, one in fifty percent
of line length and the other in seventy-five percent of line length,
TABLE IX. MODIFIED SCENARIOS IN VOTING FOR SECTION E
have been considered as two scenarios. In total, four cases have
Groups Possible conditions
1,2 1 1 1 1 0 0 0 0 TABLE X. NUMBER OF WRONG PREDICTIONS IN SCENARIO 1
1,3 1 1 0 0 1 1 0 0 Predicted zone Total number
2,3 0 1 0 1 0 1 1 0 of wrong
1 2 3
result 1 1 3 2 3 2 2 3 predictions
Real zone in each zone
1 99 211 310
E. Classification with OAO method in the modified training
set 2 249 512 761
1) Analysis of scenario 1
In this case, the best result is achieved in C=10 and g=12.1. With 3 193 143 336
these values, the success rate is increased to 86.7%. After
analyzing the results, Table V was replaced with Table IX as the
reference voting table. The results showed 91.6% success rate in TABLE XI. NUMBER OF WRONG PREDICTIONS IN SCENARIO 2
C=10 and g=12.1.
Predicted zone Total number
1) Analysis of scenario 2 of wrong
The results showed 76.1% success rate in C=100000 and 1 2 3
predictions in
g=15.3, which is equivalent to 13926 correct fault zone Real zone each zone
detections. After analyzing the results, Table V was replaced
with Table IX as the reference voting table. The results showed 1 91 237 328
78.2% success rate in C=100000 and g=15.4, which is
2 829 613 1442
equivalent to 14318 correct fault zone detections.
3 381 283 644

Authorized licensed use limited to: University College London. Downloaded on May 25,2020 at [Link] UTC from IEEE Xplore. Restrictions apply.
 [2] Sadeh, Javad, A. M. Ranjbar, N. Hadsaid, and R. Feuillet. "Accurate fault
location algorithm for series compensated transmission lines." In 2000
IEEE Power Engineering Society Winter Meeting. Conference
800
Total

700 Proceedings (Cat. No. 00CH37077), vol. 4, pp. 2527-2532. IEEE, 2000.
600
[3] Saha, M. M., J. Izykowski, E. Rosolowski, and B. Kasztenny. "A new
accurate fault locating algorithm for series compensated lines." IEEE
Number of wrong Predictions

Transactions on Power Delivery 14, no. 3 (1999): 789-797.

3
500

400
[4] Thomas, D. W. P., and C. Christopoulos. "Ultra-high speed protection of
Total series compensated lines." IEEE Transactions on Power Delivery 7, no.
300
Total
1 (1992): 139-145.
1

200
3
1 [5] Xuan, Q. Y., Y. H. Song, A. T. Johns, R. Morgan, and D. Williams.
2 "Performance of an adaptive protection scheme for series compensated
100
2
EHV transmission systems using neural networks." Electric power
systems research 36, no. 1 (1996): 57-66.
0
1 2
Fault Zones
3
[6] Song, Y. H., A. T. Johns, and Q. Y. Xuan. "Artificial neural-network-
based protection scheme for controllable series-compensated EHV
Fig. 2. NUMBER OF WRONG PREDICTIONS IN transmission lines." IEE Proceedings-Generation, Transmission and
SCENARIO 1 Distribution 143, no. 6 (1996): 535-540.
[7] Song, Y. H., Q. Y. Xuan, and A. T. Johns. "Protection scheme for EHV
1600
transmission systems with thyristor controlled series compensation using
1400
Total
radial basis function neural networks." Electric machines and power
systems 25, no. 5 (1997): 553-565.
[8] Girgis, Adly A., A. A. Sallam, and A. Karim El-Din. "An adaptive
1200
Number of wrong Predictions

1000 protection scheme for advanced series compensated (ASC) transmission

1 lines." IEEE Transactions on Power Delivery 13, no. 2 (1998): 414-420.
800

Total
[9] Kalman, Rudolph Emil. "A new approach to linear filtering and prediction
600
3
problems." Journal of basic Engineering 82, no. 1 (1960): 35-45.
400 2 [10] Megahed, Ashraf I., A. Monem Moussa, and A. E. Bayoumy. "Usage of
3
Total
3 wavelet transform in the protection of series-compensated transmission
200 lines." IEEE Transactions on Power Delivery 21, no. 3 (2006): 1213-
2

0
1221.
1 2
Fault Zones
3
[11] Dash, P. K., S. R. Samantaray, and Ganapati Panda. "Fault classification
and section identification of an advanced series-compensated
Fig. 3. NUMBER OF WRONG PREDICTIONS IN transmission line using support vector machine." IEEE transactions on
SCENARIO 2 power delivery 22, no. 1 (2006): 67-73.
[12] FattaheianǦDehkordi, Sajjad, Alireza Fereidunian, Hamid GholamiǦ
been considered for each scenario. For scenario 1, with OAA Dehkordi, and Hamid Lesani. "HourǦahead demand forecasting in smart
method, 83.7% of success, and with OAO method, 84.2% of grid using support vector regression (SVR)." International transactions
on electrical energy systems24, no. 12 (2014): 1650-1663.
success have been achieved. After analyzing the result and
[13] Parikh, Urmil B., Biswarup Das, and Rudra Prakash Maheshwari.
increasing the training set, with OAA method, 94.2% of success "Combined wavelet-SVM technique for fault zone detection in a series
and with OAO method, 91.6% of success have been achieved. compensated transmission line." IEEE Transactions on Power
For scenario 2, with OAA method, 69.7% of success, and with Delivery 23, no. 4 (2008): 1789-1794.
OAO method, 65.6% of success have been achieved. After [14] Samantaray, S. R. "A data-mining model for protection of FACTS-based
analyzing the result and increasing the training set, with OAA transmission line." IEEE Transactions on Power Delivery 28, no. 2
method, 86.7% of success and with OAO method, 78.2% of (2013): 612-618.
success have been achieved. Results showed that in solving this [15] Imani, Mahmood Hosseini, et al. "Effect of Changes in Incentives and
Penalties on Interruptible/Curtailable Demand Response Program in
type of problem, ultimately, the OAA method has a comparative Microgrid Operation." 2019 IEEE Texas Power and Energy Conference
advantage over the OAO method. With the OAA method, 94.2% (TPEC). IEEE, 2019.
and 86.7% of success have been achieved in scenario 1 and 2, [16] Lepping, Joachim. "Wiley Interdisciplinary Reviews: Data Mining and
respectively. It was shown that installing TCSC in the middle of Knowledge Discovery." (2018).
the first line leads to better results by decreasing its influence on [17] Imani, Mahmood Hosseini, Payam Niknejad, and M. R. Barzegaran.
zone 2 and 3 of the distance relay. Transmission lines are vital "Implementing Time-of-Use Demand Response Program in microgrid
parts of power system. Improving their protection is necessary considering energy storage unit participation and different capacities of
installed wind power." Electric Power Systems Research 175 (2019):
for reliability of the network. This paper presents a novel 105916.
approach to increase the reliability of the system by improving
[18] Provost, Foster, and Pedro Domingos. "Tree induction for probability-
the protection of transmission lines in fault instance and based ranking." Machine learning 52, no. 3 (2003): 199-215.
minimizing the damage to other network equipment. Rapid [19] Megahed, Ashraf I., A. Monem Moussa, and A. E. Bayoumy. "Usage of
detection and isolation of a faulty transmission line, protect the wavelet transform in the protection of series-compensated transmission
network from its possible harmful effects. lines." IEEE Transactions on Power Delivery 21, no. 3 (2006): 1213-
1221.
REFERENCES [20] Zhang, Nan, and Mladen Kezunovic. "Transmission line boundary
[1] Goldsworthy, Daniel L. "A linearized model for MOV-protected series protection using wavelet transform and neural network." IEEE
capacitors." IEEE Transactions on Power Systems 2, no. 4 (1987): 953- Transactions on Power Delivery 22, no. 2 (2007): 859-869.
957.

Authorized licensed use limited to: University College London. Downloaded on May 25,2020 at [Link] UTC from IEEE Xplore. Restrictions apply.