You are on page 1of 20

This article was downloaded by: [Jamia Millia Islamia]

On: 11 May 2013, At: 22:46


Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Electric Power Components and Systems


Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/uemp20

High-impedance Faults Analysis in


Distribution Networks Using an Adaptive
Neuro Fuzzy Inference System
a b b
M. S. Abdel Aziz , M. A. Moustafa Hassan & E. A. Zahab
a
Shaker Consultancy Group, Mearag City, Cairo, Egypt
b
Electrical Power Department, Faculty of Engineering, Cairo
University, Giza, Egypt
Published online: 01 Aug 2012.

To cite this article: M. S. Abdel Aziz , M. A. Moustafa Hassan & E. A. Zahab (2012): High-impedance
Faults Analysis in Distribution Networks Using an Adaptive Neuro Fuzzy Inference System, Electric
Power Components and Systems, 40:11, 1300-1318

To link to this article: http://dx.doi.org/10.1080/15325008.2012.689418

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions

This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representation
that the contents will be complete or accurate or up to date. The accuracy of any
instructions, formulae, and drug doses should be independently verified with primary
sources. The publisher shall not be liable for any loss, actions, claims, proceedings,
demand, or costs or damages whatsoever or howsoever caused arising directly or
indirectly in connection with or arising out of the use of this material.
Electric Power Components and Systems, 40:1300–1318, 2012
Copyright © Taylor & Francis Group, LLC
ISSN: 1532-5008 print/1532-5016 online
DOI: 10.1080/15325008.2012.689418

High-impedance Faults Analysis in


Distribution Networks Using an Adaptive
Neuro Fuzzy Inference System

M. S. ABDEL AZIZ,1 M. A. MOUSTAFA HASSAN,2 and


E. A. ZAHAB 2
1
Shaker Consultancy Group, Mearag City, Cairo, Egypt
2
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

Electrical Power Department, Faculty of Engineering, Cairo University,


Giza, Egypt

Abstract This article presents a new approach for high-impedance fault analysis
(detection, classification, and location) in distribution networks using the adaptive
neuro fuzzy inference system. The proposed scheme was trained by data from simula-
tion of a distribution system under various faults conditions and tested for different
system conditions. Details of the design process and the results of performance using
the proposed method are discussed in this article. The results show that the proposed
technique has very good performance in detecting, classifying, and locating high-
impedance faults. The third harmonics, magnitude, and angle for the three-phase
currents give superior results for fault detection as well as for fault location in high-
impedance faults. The fundamental components magnitude and angle for the three-
phase currents give superior results for the classification phase of high-impedance
faults over other types of data inputs.

Keywords adaptive neuro fuzzy inference system, discrete Fourier transform, high-
impedance faults, protection of distribution networks

1. Introduction
Detection and discovery of high-impedance faults (HIF) in electrical distribution networks
are a challenge for protection engineers. This is due to the behavior of this kind of fault
and their relatively low fault current levels with respect to feeder load current. HIFs
in power networks represent safety hazards, utility liability problems, and possibility of
equipment damage due to arcing and resistance fires. Different schemes and algorithms
have been proposed by different researchers to cope with the problems associated with
HIFs. These detection techniques were categorized into four classes.

a) Time domain: the time-domain detection algorithms and arc detection algorithms
are based on the arc current waveform [1] and include electromechanical relay
and artificial neural network (ANN) based relaying [2].

Received 24 July 2011; accepted 25 April 2012.


Address correspondence to Prof. Mohamed A. Moustafa Hassan, Electrical Power Department,
Faculty of Engineering, Cairo University, Giza, Egypt, 12613. E-mail: mmustafa_98@hotmail.com

1300
HIF Analysis in Distribution Networks Using ANFIS 1301

Table 1
Typical HIF current values for
<15-kV distribution feeders

Material Current (amps)

Dry asphalt 3
Concrete (non-reinforced) 3
Dry sand 3
Wet sand 15
Dry sod 20
Dry grass 25
Wet sod 40
Wet grass 50
Concrete reinforced 75
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

b) Frequency domain: the detection algorithm of the frequency domain applies


Fourier transform to extract the features of the harmonic components. The identi-
fication approach based on frequency domain involved the third harmonic current
[3], statistical pattern recognition approach [4], energy technique [5], randomness
technique [6], half-cycle asymmetry [7], and amplitude ratio technique.
c) Wavelet transform: wavelet transform can be employed to examine the transient
phenomena of HIFs signals in both the time and frequency domains [8–12].
d) Adaptive neuro fuzzy inference system (ANFIS) mixed with some digital signal
processing techniques.
Also, some of existing detection schemes use conventional over-current, ground relays
that dominate one harmonic detection, and high-frequency-based ripple detection in the
range of 2–10 kHz [13–15]. These schemes utilize the frequency spectra generated by
the non-linearity of the ground path associated with arcing, soil fusing, and temporal
variations in the equivalent fault impedance [16–18]. ANFIS proposed scheme offer one
of the best alternatives to the HIF problem [19]. The design of the (ANFIS) unit-based
approach for an accurate HIF detection algorithm is presented in this article. Neither fault
classification nor fault location problems were addressed in most published results [4, 17].
Most of the published research works depend on heavy mathematics as well as many
steps, while this research work does not need any heavy mathematics and the used steps
are very limited. Typical HIF currents values of this type for distribution feeders <15 kV
are listed in Table 1.

2. Distribution System under Protection


A single-line diagram for the protected distribution feeder is illustrated in Figure 1 [20].
It consists of a 13.8-kV distribution feeder 33 km in length. It is formed by three
sections of equal length and equal segment impedance. Each segment carries one load
behind its appropriate and ideal step-down transformers. The loads are different types
and are behind different kinds of transformer banks. The HIFs are simulated using the
ATP package [21].
1302 M. S. Abdel Aziz et al.

Figure 1. Single-line diagram for distribution feeder. (color figure available online)

3. ANFIS
A fuzzy logic system (FLS) can be viewed as a non-linear mapping from the input space
to the output space. An FLS consists of five main components: fuzzy sets, fuzzifiers,
fuzzy rules, an inference engine, and defuzzifiers. However, a fuzzy inference system is
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

limited in its application to modeling only ill-defined systems.


These systems have a rule structure that is essentially predetermined by the user’s
interpretation of the characteristic of the variables in the model. It has considered only
fixed membership functions (MFs) that were chosen arbitrarily. However, in some mod-
eling situations, what MFs should look like cannot be distinguished simply from looking
at data. Rather than arbitrarily choosing the parameters associated with a given MF,
these parameters could be chosen in order to tailor the MFs to the input/output data
in order to account for these types of variations in the data values. In such a case, the
necessity of the ANFIS becomes obvious. Adaptive neuro fuzzy networks are enhanced
FLSs with learning, generalization, and adaptive capabilities. These networks encode the
fuzzy if–then rules into a neural network-like structure and then use appropriate learning
algorithms to minimize the output error based on the training/validation datasets.
Neuro-adaptive learning techniques provide a method for the fuzzy modeling proce-
dure to learn information about a dataset. It computes the MF parameters that best allow
the associated fuzzy inference to track the given input/output data.
A network-type structure similar to that of an ANN can be used to interpret the
input/output map. Therefore, it maps inputs through input MFs and associated parameters,
and then through output MFs and associated parameters to outputs. These parameters
change through the learning process.
The used ANFIS is assumed to have the following properties [22, 23]:
 it is a zeroth-order Sugeno-type system;
 it has a single output obtained using weighted average defuzzification; all output
MFs are constant;
 it has no rule sharing; different rules do not share the same output MF; the number
of output MFs must be equal to the number of rules; and
 it has unity weight for each rule.
Figure 2 shows the architecture of the ANFIS, comprised by input, fuzzification,
inference, and defuzzification layers. The network can be visualized as consisting of
inputs, with N neurons in the input layer, and F input MFs for each input, with F  N
neurons in the fuzzification layer. There are F N rules with F N neurons in the inference
and defuzzification layers. One neuron is assumed in the output layer.
The ANFIS detector unit consists of six neurons in the input layer, i.e., N D 6, two
trapezoidal MFs for each input, i.e., M D 2, and a constant MF for the output layer. The
HIF Analysis in Distribution Networks Using ANFIS 1303

Figure 2. Architecture of the ANFIS. (color figure available online)


Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

ANFIS classifier unit consist of six neurons in the input layer, i.e., N D 6, two triangular
MFs for each input, i.e., M D 2, and a constant MF for the output layer. The two ANFIS
location units consist of six neurons in the input layer, i.e., N D 6, two triangular MFs
for each input, i.e., M D 2, and constant MFs for the output layers. The details of the
triangular as well as the trapezoidal MFs will be determined by the ANFIS toolbox.

4. Simulation Environment
The simulation environment based on the MATLAB software package (The MathWorks,
Natick, Massachusetts, USA) is selected as the main engineering tool for performing
modeling and simulation of power systems and relays. The ATP program is used for de-
tailed modeling of a power network and simulation of interesting events. Scenario setting
and a relaying algorithm will be implemented in the MATLAB program, while the data
generation for training and testing this algorithm will be executed by the ATP program.
The used training data to train the ANFIS are taken at fault and no-fault conditions.
The fault conditions are carried out at all fault types:
 single line to ground (SLG),
 double line to ground (DLG),
 triple line to ground (TLG), and
 line to line (L-L).
These fault conditions are carried out at different fault distances with inception fault
time (Tf / 3 and 5 msec and fault resistances (Rf ) 100, 150, 175, 200, 225, 250, 500,
750, 1000, 1250, 1500, and 2000 ohms, which give better results than [17].
Three methods were tested for detection, classification, and location of HIFs. These
methods are different according to the inputs to the ANFIS. These inputs in each case
are obtained after Fourier transform and can be classified as
1) fundamental components (magnitudes and angles) of three-phase currents (six
inputs),
2) fundamental and third harmonic magnitudes of three-phase currents (six inputs),
and
3) third harmonic components (magnitudes and angles) of three-phase currents (six
inputs).
1304 M. S. Abdel Aziz et al.

Figure 3 shows the flowchart for the training procedure of the proposed protection
scheme, while Figure 4 presents the flow chart for the testing procedure of the proposed
protection scheme.
Testing data are chosen at different conditions and at different supply voltage values.
The sequence of the proposed technique will first transform the three-phase currents (Iabc )
to the third-order harmonic components (magnitudes and angles) by Fourier transform.
These third-order harmonics are the input to the ANFIS fault detection unit, which
process these inputs, and the HIF index (IHIF ) is calculated according to the input. Then
in the case of HIF presence, the fault classification unit is activated, and the type of
HIF is determined. Then the fault location units are activated and the location of the
fault is calculated, as illustrated in Figure 5.
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

Figure 3. Flowchart for the training procedure.


HIF Analysis in Distribution Networks Using ANFIS 1305
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

Figure 4. Flowchart for the testing procedure.

5. Results and Discussion


The system was simulated using ATP as well as MATLAB, and the results of simulation
are illustrated in this article.

5.1. Detection of HIFs


When inputs to the ANFIS unit are the fundamental components (magnitudes and angles)
1306 M. S. Abdel Aziz et al.

Figure 5. Proposed protection scheme. (color figure available online)

of the three-phase currents, and the testing data are chosen to have data not included in
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

the training process while the supply voltage is kept constant and varied within ˙2.5%,
it is easy to conclude that the outputs (IHIF ) of the ANFIS detection unit were
 IHIF > 0:60 for HIF conditions and
 IHIF < 0:56 for no-fault conditions.
When inputs are the fundamental and third harmonic magnitudes of the three-phase
currents, and the testing data are chosen to have data not included before in the training
process, and supply voltage is kept constant and varied within ˙2.5%, it is easy to
conclude that the outputs (IHIF ) of the ANFIS detection unit were
 IHIF > 0:7 for HIF conditions and
 IHIF < 0:7 for no-fault conditions.
Table 2 presents the testing data when the inputs are the third harmonics components
(magnitudes and angles) of the three-phase currents. The testing data are not included in
the training process. However, the supply voltage is assumed to be constant.
Table 3 illustrates the testing data, considering the supply voltage to be varied within
˙2.5%. Table 4 depicts the testing data at different fault inception time (Tf D 3 msec)
and the supply voltage varied within ˙2.5%. For all these cases (constant generator
voltage, variable generator voltage, and variable fault inception time), it is easy to
conclude that the output of the ANFIS detection unit should be chosen as
 IHIF > 0:90 for HIF conditions and
 IHIF < 0:05 for no-fault conditions.
However, these results are better than the other inputs as illustrated. All obtained HIF
indices match the above given figures. It is thus clear that the third harmonic components
(magnitudes and angles) as inputs for the ANFIS give better results and are very close to
the expected indices. These indices are the output values of the ANFIS unit. Using these
inputs (the third-order harmonics—magnitudes and angles) with this method is near the
expected values for fault and no-fault cases.
The expected value for fault conditions is 1, and the expected value for no-fault
conditions is 0. However, for the matter of reducing any probable errors, these indices
could be fairly chosen as
 IHIF > 0:90 for HIF conditions and
 IHIF < 0:05 for no-fault conditions.
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

Table 2
Testing data using third harmonic components considering constant voltage

Fault
Fault Fault instant Calculated Expected
type distance (msec) Ia3 Øa3 Ib3 Øb3 Ic3 Øc3 IHIF IHIF

SLG 31 5 0.055388 61.994 0.057279 60.819 14.904 76.916 1.023 1

1307
L-L 1 5 0.62118 149.95 1.9501E–03 10.87 0.61934 30.114 1.03 1
DLG 6 5 2.5233E–03 33.585 7878.5 16.764 7875.1 163.23 1.044 1
— — — 7.7304E–04 90.381 2.0379E–03 10.867 2.034E–03 168.98 0.041 0
L-L 24 5 0.61787 149.62 1.9500E–03 10.874 0.61603 30.442 1.03 1
— — — 7.8042E–04 90.381 2.0574E–03 10.866 2.053E–03 168.98 0.04 0
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

Table 3
Testing data using third harmonic components considering variable voltage

Percent
Fault Fault change in Calculated Expected
type distance voltage Ia3 Øa3 Ib3 Øb3 Ic3 Øc3 IHIF IHIF

SLG 19 2.5 0.019429 49.423 0.021478 47.506 6.7223 64.374 1.002 1


L-L 16 1.5 0.60937 149.69 1.9207E–03 10.877 0.60756 30.365 1.03 1
DLG 19 2.5 5.4372E–03 23.722 2600.1 15.558 2596.8 164.42 1.012 1
— — 1.25 7.305E–04 90.204 1.925E–03 10.891 1.923E–03 169 0.04 0

1308
Table 4
Testing data using third harmonic components considering variable voltage and variable inception time

Percent Fault
Fault change in instant Calculated Expected
type voltage (msec) Ia3 Øa3 Ib3 Øb3 Ic3 Øc3 IHIF IHIF

SLG 2.5 3 0.011992 29.057 0.013233 21.008 3.7826 2.5012 0.9138 1


DLG C1 3 5.1681E–04 80.145 5173.1 39.3 5172.9 140.7 1.12 1
SLG C1 3 6.3520E–04 59.628 1.4217E–03 3.852 0.39446 0.65959 1.008 1
L-L 2.5 3 0.77431 172.03 1.9012E–03 10.877 0.77252 8.0181 1.009 1
TLG 2.5 3 930.76 142.54 1122.2 14.123 909.76 112.59 0.989 1
SLG C0.5 3 1.3555E–03 37.474 2.1753E–03 4.8624 0.98246 0.77445 0.9695 1
HIF Analysis in Distribution Networks Using ANFIS 1309

Table 5
Results for different inputs for ANFIS detection unit

Method Calculated IHIF Expected IHIF

Fundamental components  IHIF > 0:60  1


(magnitudes and angles)  IHIF < 0:56  0
Fundamental and third  IHIF > 0:7  1
harmonics magnitudes  IHIF < 0:7  0
Third harmonics components  IHIF > 0:90  1
(magnitudes and angles)  IHIF < 0:05  0

The above choice will narrow any expected errors. Therefore, as illustrated in Table 5,
the proposed technique illustrates promising results in the field of HIF analysis using
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

ANFIS, based on the third-order harmonic components (magnitudes and angles) as inputs
for this ANFIS detection unit. Figure 6 illustrates the error values for the detection
process using the third harmonics technique, which does not exceed 8.62%. This means
91% success in the detection of HIFs, which is better than [4], where success was more
than 89%. Furthermore, none of the published results was tested for high impedance of
more than 2 K ohms.

5.2. Classification of HIFs


Inputs to the ANFIS classification unit are considered to be the third harmonic compo-
nents (magnitudes and angles), and the testing data are chosen to have data not included in
the training process and at constant supply voltage, varied supply voltage within ˙2.5%,
and different fault inception time (Tf D 3 msec).
For all of these cases, it is easy to conclude that the output of the ANFIS classification
unit (CHIF ) should be chosen as
 0:9  CHIF < 1:1 for an SLG fault,
 1:9  CHIF < 2:2 for a DLG fault,
 2:9  CHIF < 3:1 for a TLG fault, and
 3:1  CHIF for an L-L fault

Figure 6. Percent error values for detection process. (color figure available online)
1310 M. S. Abdel Aziz et al.

The inputs to the ANFIS classification unit are considered to be the fundamental and
third harmonic magnitudes, and the testing data are chosen to have data not included in
the training process and at constant supply voltage, varied supply voltage within ˙2.5%,
and different fault inception time (Tf D 3 msec). For all of these cases, it is easy to
conclude that the outputs of the ANFIS classification unit (CHIF ) are not giving the
expected results, so this method is excluded.
Table 6 shows results of when the inputs to the ANFIS classification unit are the
fundamental components (magnitudes and angles) and the testing data are chosen to
have data not included in the training process and chosen at constant supply voltage.
Table 7 illustrates the testing data when the supply voltage is varied within ˙2.5% for
fundamental components (magnitudes and angles) inputs. Table 8 presents the testing
data at different fault inception time (Tf D 3 msec) and supply voltage varied within
˙2.5% with the same inputs. For all of these cases, it is easy to conclude that the output
of the ANFIS classification unit (CHIF ) should be chosen as
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

 0:95  CHIF < 1:05 for an SLG fault,


 1:95  CHIF < 2:05 for a DLG fault,
 2:90  CHIF < 3:05 for a TLG fault, and
 3:30  CHIF < 4:60 for an L-L fault.
The expected values for the SLG fault is 1, for the DLG fault is 2, for the TLG fault
is 3, and for the L-L fault is 4.
However, for the matter of reducing any probable errors, these indices could be fairly
chosen as
 0:95  CHIF < 1:05 for an SLG fault,
 1:95  CHIF < 2:05 for a DLG fault,
 2:90  CHIF < 3:05 for a TLG fault, and
 3:30  CHIF < 4:60 for an L-L fault
This above choice will narrow any expected errors. Therefore, the proposed technique
illustrates promising results in the field of HIF classification using ANFIS based on the
fundamental components (magnitudes and angles) as inputs to the ANFIS classification
unit, as illustrated in Table 9. Figure 7 describes the error values of the proposed method
for HIF classification using the fundamental components technique, which does not ex-
ceed 15% [24]. Furthermore, the ANFIS is used here in this research for HIF analysis (de-
tection, classification, and location). The obtained results are completely different from the
published results in [22, 23, 25]. These articles deal with the distance protection system;
however, fault location was not considered in [25], while it was considered in [22, 23].

5.3. Location of HIFs


Using one ANFIS unit for the location process will increase the location percentage
error; therefore, two ANFIS location units were proposed. These proposed two ANFIS
location units minimized the location percentage error. The first ANFIS location unit is
used to determine the section on which the fault occurred, which is called the primary
ANFIS unit, while the second ANFIS location unit is used to determine the distance
of the fault from the beginning of the section on which the fault occurred, which is
called the secondary ANFIS unit. The inputs to the primary ANFIS location unit are
considered to be the fundamental components (magnitudes and angles); the testing data
were obtained as illustrated in Table 10. The inputs to the primary ANFIS location unit
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

Table 6
Testing data using fundamental components for HIF classification considering constant voltage

Fault
Expected Calculated instant Fault
CHIF CHIF Øc1 Ic1 Øb1 Ib1 Øa1 Ia1 (msec) type

1 1 152.78 89.464 29.899 58.873 90.119 58.86 5 SLG


2 1.999 127.38 38,942 52.682 39,006 90.108 58.865 5 DLG

1311
3 3 155.98 42,581 79.346 45,413 41.829 40,929 5 TLG
4 3.4 148.5 60.803 29.863 58.868 89.256 61.083 5 L-L
2 1.998 125.7 13,821 54.45 13,873 90.107 58.867 5 DLG
4 3.4 148.54 60.808 29.897 58.874 89.219 61.083 5 L-L
2 1.995 125.42 9363.2 54.798 9413.9 90.134 58.863 5 DLG
4 3.33 148.51 60.805 29.879 58.871 89.248 61.097 5 L-L
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

Table 7
Testing data using fundamental components for HIF classification considering variable voltage

Percent Fault
Expected Calculated change in instant Fault
CHIF CHIF Øc1 Ic1 Øb1 Ib1 Øa1 Ia1 voltage (msec) type

1 1 153.05 84.018 29.923 57.403 90.117 57.376 2.5 5 SLG


2 1.997 125.68 12,749 54.498 12,810 90.092 57.386 2.5 5 DLG
4 3.4 148.53 59.893 29.883 57.988 89.233 60.164 1.5 5 L-L
1 1 152.78 91.701 29.899 60.345 90.119 60.332 C2.5 5 SLG
4 3.34 148.53 59.895 29.889 57.99 89.23 60.17 1.5 5 L-L

1312
Table 8
Testing data using fundamental components for HIF classification considering variable voltage and variable inception time

Percent Fault
Expected Calculated change in instant Fault
CHIF CHIF Øc1 Ic1 Øb1 Ib1 Øa1 Ia1 voltage (msec) type

1 1 152.19 74.841 29.913 57.403 90.112 57.384 2.5 3 SLG


2 1.975 126.86 40,708 53.168 40,737 89.722 33.888 C0:5 3 DLG
4 4.5 148.64 59.82 29.883 57.399 88.74 59.744 2.5 3 L-L
2 2.007 127.6 55,338 52.427 55,366 89.734 34.055 C1 3 DLG
3 2.905 148.77 10,844 80.18 10,775 33.892 8956.5 2.5 3 TLG
4 4.5 148.6 59.818 29.863 57.396 88.769 59.758 2.5 3 L-L
HIF Analysis in Distribution Networks Using ANFIS 1313

Table 9
Results for different inputs for ANFIS classification unit

Expected CHIF Calculated CHIF Method


 1  0:95  CHIF < 1:05 Fundamental components
 2  1:95  CHIF < 2:05 (magnitudes and angles)
 3  2:90  CHIF < 3:05
 4  3:30  CHIF < 4:60
This method is excluded Fundamental and third
harmonics magnitudes
 1  0:90  CHIF < 1:1 Third harmonics components
 2  1:90  CHIF < 2:2 (magnitudes and angles)
 3  2:90  CHIF < 3:1
 4  3:10  CHIF
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

are the fundamental and third harmonic magnitudes, and the testing data were obtained as
illustrated in Table 11. On the other hand, Table 12 shows the testing data when the inputs
to the primary ANFIS location unit are the third harmonic components (magnitudes and
angles). From the previous results, knowing that the expected values for fault occurrences
in the first section is 1, in the second section is 2, and in the last section is 3, it will be
evident that the third harmonics method is the best method, as it gives the best results
with the least error (6.67%).
Also for the secondary ANFIS location unit, when the inputs are the fundamental
components (magnitudes and angles), the results will be as illustrated in Table 13.
Table 14 illustrates the testing data when the inputs to the secondary ANFIS location
unit are the fundamental and third harmonic magnitudes.
On the other hand, Table 15 presents the testing data when the inputs to the secondary
ANFIS location unit are the third harmonic components (magnitudes and angles).

Figure 7. Percent error values for classification process. (color figure available online)
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

Table 10
Testing data using fundamental components for HIF primary ANFIS unit

Fault Fault
Percent Expected Calculated instant Fault distance Fault
error LP HIF LP HIF Øc1 Ic1 Øb1 Ib1 Øa1 Ia1 (msec) type (km) section

10 1 0.9 154.12 128.03 29.885 58.87 90.124 58.864 5 SLG 1 1


70 1 1.7 155.27 127.29 29.901 58.869 90.154 58.831 5 SLG 8 1
5 2 1.9 153.79 97.95 29.91 58.872 90.133 58.84 5 SLG 12 2
15 2 2.3 154.64 15,412 82.113 16,062 38.455 14,970 5 TLG 17 2
10 3 2.7 125.4 8655.6 54.874 8718.9 90.096 58.853 5 DLG 27 3
8 3 2.76 125.43 8389.5 54.848 8452.7 90.096 58.853 5 DLG 28 3

1314
Table 11
Testing data using fundamental and third harmonic magnitudes for HIF primary ANFIS unit

Fault Fault
Percent Expected Calculated instant Fault distance Fault
error LP HIF LP HIF Ic3 Ic1 Ib3 Ib1 Ia3 Ia1 (msec) type (km) section

80 1 1.8 17.837 128.03 5.7948E–03 58.87 3.6051E–03 58.864 5 SLG 1 1


70 1 1.7 17.174 127.29 0.02965 58.869 0.027551 58.831 5 SLG 8 1
12 2 1.76 3712.3 21,398 4299.2 22,372 1782.5 20,750 5 TLG 12 2
12 2 2.24 2813 13,812 2817.8 13,882 7.5387E–03 58.851 5 DLG 17 2
21 3 2.37 1764.6 8652.6 1769.4 8721.9 9.3093E–03 58.845 5 DLG 27 3
7.67 3 2.77 1644.4 9412.9 1856.1 9760.5 751.27 9169.5 5 TLG 28 3
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

Table 12
Testing data using third harmonic components for HIF primary ANFIS unit

Fault Fault
Percent Expected Calculated instant Fault distance Fault
error LP HIF LP HIF Øc3 Ic3 Øb3 Ib3 Øa3 Ia3 (msec) type (km) section

5 1 0.95 60.576 17.837 36.956 5.7948E–03 41.236 3.6051E–03 5 SLG 1 1


1 1 0.99 150.39 5292.4 4.9911 6202.2 127.24 2607 5 TLG 8 1
0 2 2 66.906 16.789 50.417 0.039902 51.589 0.037833 5 SLG 12 2
6.67 3 2.8 164.63 1763 15.266 1771 11.137 0.015737 5 DLG 27 3
2.33 3 2.93 152.04 1644.4 4.1472 1856.1 122.06 751.27 5 TLG 28 3

1315
Table 13
Testing data using fundamental components for HIF secondary ANFIS unit

Fault
Percent Calculated Expected Calculated distance
error distance LSHIF LSHIF Øc1 Ic1 Øb1 Ib1 Øa1 Ia1 (km)

275 3.75 1 3.75 154.12 128.03 29.885 58.87 90.124 58.864 1


43.75 4.5 8 4.5 155.27 127.29 29.901 58.869 90.154 58.831 8
30.8 15.7 1 4.7 153.79 97.95 29.91 58.872 90.133 58.84 12
10 16.4 6 5.4 154.64 15,412 82.113 16,062 38.455 14,970 17
0.6 26.97 5 4.97 125.4 8655.6 54.874 8718.9 90.096 58.853 27
18 26.9 6 4.9 125.43 8389.5 54.848 8452.7 90.096 58.853 28
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

Table 14
Testing data using fundamental and third harmonic magnitudes for HIF secondary ANFIS unit

Fault
Percent Calculated Expected Calculated distance
error distance LSHIF LSHIF Ic3 Ic1 Ib3 Ib1 Ia3 Ia1 (km)

360 4.6 1 4.6 17.837 128.03 5.7948E–03 58.87 3.6051E 03 58.864 1


19 3.58 3 3.58 13,221 79,579 17,400 87,286 8345.7 75,895 3
0.3125 16.05 5 5.05 2825.7 16,236 3237.8 16,915 1328.7 15,773 16
17.8 15.6 8 4.6 16.086 126.05 0.051841 58.867 0.049835 58.805 19
11 26.64 2 4.64 15.595 125.47 0.056317 58.867 0.054357 58.802 24
14 26.65 9 4.65 14.904 124.61 0.057279 58.864 0.055388 58.801 31

1316
Table 15
Testing data using third harmonic components for HIF secondary ANFIS unit

Fault
Percent Calculated Expected Calculated distance
error distance LSHIF LSHIF Øc3 Ic3 Øb3 Ib3 Øa3 Ia3 (km)

1 0.99 1 0.99 155.96 35,088 24.03 35,096 43.957 1.5156E–03 1


0 3 3 3 145.95 13,221 6.4984 17,400 139.38 8345.7 3
4.5 14.32 4 3.32 164.32 3172.1 15.652 3176.9 21.747 6.9906E–03 15
1.55 17.72 7 6.72 164.4 2661 15.538 2669.2 14.983 0.013313 18
4.5 20.9 9 9.9 151.77 2278.8 4.2661 2592.3 123.12 1056.7 20
4 28.1 5 6.1 152.06 1697.4 4.1432 1913.9 121.92 773.98 27
2.1 31.65 9 9.65 164.71 1539.9 15.227 1544.6 16.44 9.5345E–03 31
HIF Analysis in Distribution Networks Using ANFIS 1317

From the previous results and from the error values, it is clear that the third harmonics
is the best method to determine the location of the HIFs, as it gives the best results
compared with other input techniques.

6. Conclusion
This research presents a novel study of HIF detection, classification, and location schemes
using an ANFIS. The proposed artificial intelligent approach demonstrates successful
performance for fault detection, classification, and location. Three different groups of
inputs were tested. These groups of inputs are output from discrete Fourier transform
(DFT). It was found that the third harmonics component plays the essential role in
HIF detection and location, while the fundamental component (magnitudes and angles)
plays the essential rule in HIF classification. For the fault detection task, all testing data
for the ANFIS detector in fault and no-fault conditions give the correct output. For fault
classification and location tasks, all testing data for the ANFIS classifier unit and ANFIS
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

locator units give the correct output. Data for testing include both kinds of data, either
used in training or not used in training. Suggested indices for occurrence of the HIFs
were introduced. The practical implementation will be easy after settling the theoretical
research.

References
1. Sultan, A. F., Swift, G. W., and Fedirchuk, D. J., “Detecting arcing downed-wires using fault
current flicker and half-cycle asymmetry,” IEEE Trans. Power Delivery, Vol. 9, No. 1, pp. 461–
470, 1994.
2. Calhoun, H., Bishop, M. T., Eichler, C. H., and Lee, T. E., “Development and testing of and
electromechanical relay to detect fallen distribution conductors,” IEEE Trans. Power Apparat.
Syst., Vol. PAS-101, No. 6, pp. 1643–1650, June 1982.
3. Sharaf, A. M., and Wang, G., “High impedance fault detection using feature-pattern based
relaying,” IEEE PES Transm. Distribut. Conf. Expo., Vol. 1, pp. 222–226, September 2003.
4. Sedighi, A. R., Haghifam, M. R., and Malik, O. P., “High impedance fault detection based on
wavelet transform and statistical pattern recognition,” IEEE Trans. Delivery, Vol. 20, pp. 241–
2421, October 2005.
5. Aucoin, B. M., and Russell, B. D., “Distribution high impedance fault detection using high
frequency current components,” IEEE Trans. Power Apparat. Syst., Vol. PAS-101, No. 6,
pp. 1596–1606, June 1982.
6. Russell, B. D., and Chinchali, R. P., “A digital signal processing algorithm for detecting arcing
faults on power distribution feeders,” IEEE Trans. Power Delivery, Vol. 4, No. 1, pp. 132–140,
January 1989.
7. Kwon, W. H., Lee, G. W., Park, Y. M., Yoon, M. C., and Yoo, M. H., “High impedance fault
detection utilizing incremental variance of normalized even order harmonic power,” IEEE
Trans. Delivery, Vol. 6, pp. 557–564, April 1991.
8. Aucoin, B. M., and Jones, R. H., “High impedance fault detection implementation issues,”
IEEE Trans. Power Delivery, Vol. 11, No. 1, pp. 139–148, January 1996.
9. Eldin, E. S. T., Ibrahim, D. K., Aboul-Zahab, E. M., and Saleh, S. M., “High impedance faults
detection in EHV transmission lines using the wavelet transforms,” Proceedings of the IEEE
Power Engineering Society General Meeting, Tampa, FL, 24–28 June 2007.
10. Shaaban, S. A., “Transmission line faults classification using wavelet transform,” 14th Inter-
national Middle East Power Systems Conference (MEPCON’10), Paper ID 225, Cairo, Egypt,
19–21 December 2010.
11. Aucoin, B. M., “Detection of distribution high impedance faults using burst noise signals near
60 Hz,” IEEE Trans. Power Delivery, Vol. PWRD-2, No. 2, pp. 342–348, April 1987.
1318 M. S. Abdel Aziz et al.

12. Reddy, M. J., and Mohanta, D. K., “A wavelet-neuro-fuzzy combined approach for digital
relaying of transmission line faults,” Elect. Power Compon. Syst., Vol. 35, pp. 1385–1407,
2007.
13. Snider, L. A., and Shan, Y. Y., “The artificial neural networks based relay algorithm for
distribution system high impedance fault detection,” Proc. APSCom-97, Vol. 1, pp. 100–106,
November 1997.
14. Wester, C. G., “High impedance fault detection on distribution systems,” 42nd Annual Rural
Electric Power Conference, St. Louis, MO, 26–28 April 1998.
15. Elkalashy, N. I., Lehtonen, M., Darwish, H. A., Taalab, A.M. I., and Izzularab, M. A., “DWT
based detection and transient power direction based location of high impedance faults due
to leaning trees in unearthed MV networks,” IEEE Trans. Power Delivery, Vol. 23, No. 1,
pp. 94–101, January 2008.
16. Jeerings, D. J., “Ground resistance revisited,” IEEE Trans. Power Delivery, Vol. 4, No. 2,
pp. 949–956, April 1989.
17. Benner, C. L., and Russell, B. D., “Practical high-impedance fault detection on distribution
feeders,” IEEE Trans. Industry Appl., Vol. 33, pp. 635–640, 1997.
Downloaded by [Jamia Millia Islamia] at 22:46 11 May 2013

18. Shebl, K. M., Badran, E. A., and Abdalla, E., “A combined MODELS-TACS ATPdraw general
model of the high impedance faults in distribution networks,” 14th International Middle East
Power Systems Conference (MEPCON’10), Paper ID 220, Cairo, Egypt, 19–21 December
2010.
19. Chan, D., and Yibin, X., “A novel technique for high impedance faults identification,” IEEE
Trans. Power Delivery, Vol. 13, No. 3, pp. 738–744, July 1998.
20. Uriarte, F. M., Modeling, Detection, and Localization of High-impedance Faults in Low-voltage
Distribution Feeders, M.Sc. Thesis, Virginia Tech Polytechnic Institute and State University,
Blacksburg, VA, 2003.
21. Alternative Transient Program ATPDraw Version 3.5, 2002.
22. Kamel, T. S., Moustafa Hassan, M. A., and El-Morshedy, A., “An ANFIS based distance relay
protection for transmission lines in EPS,” Int. J. Innovat. Elect. Power Syst., Vol. 3, No. 1,
pp. 49–65, 2011.
23. Kamel, T. S., Moustafa Hassan, M. A., and El-Morshedy, A., “Advanced distance protection
technique based on multiple classified ANFIS considering different loading conditions for long
transmission lines in EPS,” Int. J. Model., Ident. Control, Vol. 16, No. 2, pp. 108–121, 2012.
24. Abdel Aziz, M. S., Hassan, M. A., and Zahab, E. A., “Applications of ANFIS in high impedance
faults detection and classification in distribution networks,” 8th IEEE International Symposium
on Diagnostics for Electrical Machines, Power Electronics & Drives, (SDEMPED 2011),
Bologna, Italy, 5–8 September 2011.
25. Elbaset, A. A., and Hiyama, T., “Fault detection and classification in transmission lines using
ANFIS,” IEEJ Trans. Industry Appl., Vol. 129, No. 7, pp. 705–713, 2009.

You might also like