Available online at www.sciencedirect.

com
Available online at www.sciencedirect.com
ScienceDirect
ScienceDirect
Energy
Available
Available Procedia
online
online 00 (2017) 000–000
atatwww.sciencedirect.com
www.sciencedirect.com
Energy Procedia 00 (2017) 000–000 www.elsevier.com/locate/procedia
ScienceDirect
ScienceDirect
www.elsevier.com/locate/procedia

Energy
EnergyProcedia
Procedia117 (2017) 000–000
00 (2017) 417–423
www.elsevier.com/locate/procedia
1st International Conference on Power Engineering, Computing and CONtrol, PECCON-2017, 2-
1st International Conference on Power
4 March Engineering,
2017, VIT Computing
University, Chennai and CONtrol, PECCON-2017, 2-
Campus
4 March 2017, VIT University, Chennai Campus
Combined Mathematical Morphology and Data Mining Based High
Combined Mathematical Morphology
The 15th International Symposium on and Data
District Mining
Heating Based High
and Cooling
Impedance Fault Detection
Impedance Fault Detection
Assessing the feasibility a,of* using the heat demand-outdoor
b
Kavaskar
temperature function Sekar , Nalin Kant Mohanty
Kavaskar Sekar , Nalin Kant Mohanty b demand forecast
for a long-term
a, * district heat
a
Research scholar, Department
a,b,c of Electricala and Electronics aEngineering, Panimalar
b Engineering College, cChennai, Tamil Nadu, cIndia.
a I. Andrić *, A. Pina , P. Ferrão , J. Fournier ., B. Lacarrière , O. Le Corre
Research scholar, Department of Electrical and Electronics Engineering, Panimalar Engineering College, Chennai, Tamil Nadu, India.
b
Professor, Department of Electrical & Electronics Engineering, Sri Venkateswara College of Engineering, Tamil Nadu, India.
a
IN+
b Center for Innovation, Technology and Policy Research - Instituto Superior Técnico, Av. Rovisco Pais 1, 1049-001 Lisbon, Portugal
Professor, Department of Electrical & Electronics Engineering, Sri Venkateswara College of Engineering, Tamil Nadu, India.
b
Veolia Recherche & Innovation, 291 Avenue Dreyfous Daniel, 78520 Limay, France
c
Département Systèmes Énergétiques et Environnement - IMT Atlantique, 4 rue Alfred Kastler, 44300 Nantes, France
Abstract
Abstract
This paper presents an intelligent scheme for high impedance fault detection using mathematical morphology and decision tree.
Abstract
The
This current signals are
paper presents pre-processed
an intelligent usingformathematical
scheme high impedance morphology and estimation
fault detection of the signalmorphology
using mathematical features is used to generate
and decision tree.a
decision
The treesignals
current model.areThe final relayingusing
pre-processed operation based onmorphology
mathematical generated dataand mining decision
estimation of thetree model.
signal Theisproposed
features method isa
used to generate
District
tested
decision heating
ontree
a standardnetworks
model. The arerelaying
test final
system commonly addressed
with aoperation
wide range
based inon
of thegenerated
powerliterature
system asoperating
data one
miningof the most tree
effective
conditions.
decision Simulation
model. solutions
Theresultsforshow
proposed decreasing
that the
method the
is
proposed
tested on method
greenhouse can be
a standard testhighly
gas emissions system
fromreliable
with
the in detecting
a wide
building range
sector. high
of impedance
Thesepower fault
system
systems for high
harmless
operating
require and secured
conditions.
investments operations.
Simulation
which results through
are returned show thatthe the
heat
proposed method
sales. Due canchanged
to the be highlyclimate
reliableconditions
in detectingand high impedance
building fault forpolicies,
renovation harmlessheat
and secured
demandoperations.
in the future could decrease,
©prolonging
2017 The Authors. Published
the investment by period.
return Elsevier Ltd.
© 2017 The Authors. Published by Elsevier Ltd.
© 2017
The Thescope
Peer-review
main Authors.
underof Published
this paper isby
responsibility toElsevier
of the
assess Ltd.
scientific
the committee
feasibility of thethe
of using 1st heat
International
demand –Conference on Power Engineering,
outdoor temperature function for heat demand
Peer-review under responsibility of the scientific committee of the 1st International Conference on Power Engineering,
Peer-review
Computing
Computing under
and
forecast. Theand responsibility
CONtrol.
district
CONtrol. of Alvalade,of thelocated
scientific
in committee of the 1stwas
Lisbon (Portugal), International
used as aConference
case study.onThe
Power Engineering,
district is consisted of 665
Computing
buildings and
that CONtrol.
vary in both construction period and typology. Three weather scenarios (low, medium, high) and three district
Keywords:
renovationMathematical
scenarios were Morphology,
developedDecision
(shallow,tree, High impedance
intermediate, deep).fault, Data mining.
To estimate the error, obtained heat demand values were
Keywords:
comparedMathematical
with results from Morphology,
a dynamicDecision tree, High
heat demand model, impedance
previouslyfault, Data mining.
developed and validated by the authors.
The results showed that when only weather change is considered, the margin of error could be acceptable for some applications
(the error in annual demand was lower than 20% for all weather scenarios considered). However, after introducing renovation
1.scenarios,
Introductionthe error value increased up to 59.5% (depending on the weather and renovation scenarios combination considered).
1.The
Introduction
value of slope coefficient increased on average within the range of 3.8% up to 8% per decade, that corresponds to the
HighinImpedance
decrease the numberFault (HIF)hours
of heating occurs when theduring
of 22-139h conductors break
the heating and (depending
season touch the poor
on theconducting
combinationsurface. As and
of weather a
result, faultImpedance
High
renovation current isconsidered).
scenarios less
Faultthan
(HIF)normal
Onoccurs load
when
the other current
hand, level [1],
thefunction
conductors making
break
intercept conventional
and
increased touch theprotection
poor per
for 7.8-12.7% schemes
conducting
decade not suitable
surface.
(depending As a
on the
for thisfault
result,
coupled category
scenarios).of fault.
current is less
The The
than
values fallen
normalenergized
suggested load conductor
couldcurrent to may
be usedlevel [1],result
modify the in
making fire and endanger
conventional
function parameters fortothe
protectionhuman life. considered,
schemes
scenarios not suitableand
for this category
improve of fault.
the accuracy Thedemand
of heat fallen estimations.
energized conductor may result in fire and endanger to human life.

© 2017 The Authors. Published by Elsevier Ltd.

* Corresponding author. Tel.:+91-9894909616;
Peer-review
E-mail
*
under
address:
Corresponding
responsibility of the Scientific Committee of The 15th International Symposium on District Heating and
thisiskavas@gmail.com
author. Tel.:+91-9894909616;
Cooling.
E-mail address: thisiskavas@gmail.com
1876-6102
Keywords:©Heat
2017demand;
The Authors. Published
Forecast; bychange
Climate Elsevier Ltd.
Peer-review©under
1876-6102 2017responsibility
The Authors. of the scientific
Published committee
by Elsevier Ltd. of the 1st International Conference on Power Engineering, Computing and CONtrol.
Peer-review under responsibility of the scientific committee of the 1st International Conference on Power Engineering, Computing and CONtrol.

1876-6102 © 2017 The Authors. Published by Elsevier Ltd.

Peer-review under responsibility of the Scientific Committee of The 15th International Symposium on District Heating and Cooling.
1876-6102 © 2017 The Authors. Published by Elsevier Ltd.
Peer-review under responsibility of the scientific committee of the 1st International Conference on Power Engineering, Computing
and CONtrol.
10.1016/j.egypro.2017.05.161
418 Kavaskar Sekar et al. / Energy Procedia 117 (2017) 417–423
2 Author name / Energy Procedia 00 (2017) 000–000

An extensive literature review of HIF detection methods in [2]. Several articles reported for HIF detection so far.
These are low order current harmonic ratio [3], Extended Kalman filter approach [4], Wavelet transforms [5, 6],
Mathematical Morphology (MM) based filters [7], and knowledge discovery in databases and the application of data
mining [8]. Significant developments have been made in the earlier decades for HIF detection. But the solution
methodology still to be improved. The main goal of this work is to develop an intelligent protection method which
can accurately identify and differentiate HIF from other normal power system events with the knowledge of HIF
characteristics and data mining Decision Tree (DT) model.

2. Methodology of the proposed work

In the proposed method, current at targeted places are received and processed through a tool based on MM to
find features such as change in energy, mean and standard deviations. These features are used to train DTs for HIF
detection. The data mining based decision tree takes a final decision on trip signal for HIF detection. The proposed
method provides a suitable solution for HIF. Fig.1 shows the detailed flow chart of the proposed method.

Data collection
(Three phase current signals)

Pre-processing by Mathematical
Morphology

Feature extraction
(Energy and Standard deviation)

Data mining based

HIF detection

Fig.1. The proposed method.

3. System studied

3.1 Distribution system studied

The effectiveness of the proposed method is demonstrated on the standard test systems. The standard IEEE 13
bus system shown in Fig. 2 and IEEE 34 bus system shown in Fig. 3 was modelled by MATLAB / SIMULINK. The
system data for both the system is given in [9]. The modelled substation bus has an equivalent voltage and a series
of impedance of 0.01+ j 0.08 p.u. Distribution lines are modelled with lumped parameters. The voltage regulators
are modelled by three phase two winding transformer with on-load tap changing transformer.

3.2 HIF model studied

HIFs are highly nonlinear and complex phenomenon and this HIF current have five significant attributes, namely
asymmetry, nonlinearity, build up, shoulder and intermittence. The current with different peak values in both
positive and negative cycles produces an asymmetry featured signal. Non-linearity arises from odd harmonics. The
 Kavaskar Sekar et al. / Energy Procedia 117 (2017) 417–423 419
Author name / Energy Procedia 00 (2017) 000–000 3

magnitude of HIF current gradually increases until it reaches the maximum value, which is termed as build up. The
HIF current stops in certain cycles are called shoulder. The energized conductor break off the contact of the surface
in a few of the cycles termed as intermittence.
The HIF model used in this work is shown in Fig. 4 and it is based on several Emmanuel arc model. The HIF
Current at fault location is shown in Fig. 5. The v-i relationship of HIF model is shown in Fig. 6. It is observed that
the chosen HIF model has all aforementioned attributes of HIFs.

650

646 645 632 633 634

611 684 692 675

671

652 680

Fig. 2. IEEE 13 node test feeder

848

822 846

820 844

864
818 842

802 806 808 812 814 850 824 826 834 860 836
858 840
816
832 862
800 888 890
810
838
852

828 830 854 856

Fig. 3. IEEE 34 node test feeder

420 Kavaskar Sekar et al. / Energy Procedia 117 (2017) 417–423
4 Author name / Energy Procedia 00 (2017) 000–000

Fig. 4. HIF model

Fig. 5. HIF Current.

Fig. 6. v-i characteristics of HIF

 Kavaskar Sekar et al. / Energy Procedia 117 (2017) 417–423 421
Author name / Energy Procedia 00 (2017) 000–000 5

4. Pre-processing

The current signal is pre-processed by MM filter which is non-linear and time domain processing of the signal.
The basic functions in MM are defined as dilation and erosion used to design a filter is termed as Dilation and
Erosion Morphological Filter (DEMF). Structure Element (SE) playing an important role in pre-processing of the
given signal [10]. Since the choice of SE depends on the anticipated signal of the chosen applications, based on trial
and error method SE with the length of three [0.2, 0.8, 0.2] is chosen in this work.
Let f (n) be a signal to be transformed and defined in the domain Df = { x0, x1,……xn } and g(m) be SE defined in
the domain Dg = {z0, z1,…….Zm} where n > m and are integers. Dilation of the signal f (n) by g (m) is given in
equation (1):

 – (1)

Erosion of the signal f(n) by g(m) is given in equation (2):

 , (2)

Based on the equations (1) and (2), DEMF termed as

  (3)

It can be observed that equations (1)-(3) require simple arithmetic operations, that reduces computational work,
which in turn attracts real time applications.

5. Feature extraction:

The current signal for the HIF and other disturbances such as load switching, capacitance switching and
transformer energizing are observed using the MATLAB / SIMULINK model of standard test system. The current
signal (one cycle data window length) is retrieved for each phase and pre-processed through MM. The following
features such as standard deviation, the change in energy, and mean are estimated for each phase of the system.

 A1= Standard deviation of one cycle window length of phase A

 A2= Change in energy of one cycle window length of phase A
 A3= Mean of one cycle window length of phase A
 B1= Standard deviation of one cycle window length of phase B
 B2= Change in energy of one cycle window length of phase B
 B3= Mean of one cycle window length of phase B
 C1= Standard deviation of one cycle window length of phase C
 C2= Change in energy of one cycle window length of phase C
 C3= Mean of one cycle window length of phase C

6. Building data mining model for HIF detection:

The DT is more transparent, rule based and easy for implementation compare to black box solutions [11]. So it
attracts power system protection applications for decision making. The proposed method uses open source data
mining software package 'R' for generating the DT [12]. Building the data mining decision tree model for HIF
detection commences at collecting single cycle post-disturbance data window length of current signals at the target
feeder processed through the MM filter and assess the features of chosen window length. To detect the HIF, features
(change in energy, mean and standard deviation of the current signals) are used to train DT. These features are
considered as input with 0 or 1 as output (0 for non-HIF and 1 for HIF). The DT built for HIF detection is given in
Fig. 7.
6422 Kavaskar
Author nameSekar et al.
/ Energy / Energy00Procedia
Procedia 117 (2017) 417–423
(2017) 000–000

Fig. 7. DT for HIF detection

7. Simulation results and discussion

To test validity of the proposed HIF detection strategy a standard IEEE 13 bus distribution system has been
undertaken with the different simulation condition in this work. A total of 250 (5x5x10) HIF cases was simulated at
five different locations of the IEEE 13 bus system in combination with five different fault inception angles and ten
different fault resistances. Twenty (2x2x5) cases of capacitance switching were simulated by connecting and
disconnecting (on & off) of capacitor bank at nodes 611 & 671 in combination with five different inception angles.
Similarly, 20 (2x2x5) cases of load switching were simulated by connecting and disconnecting (on & off) of load at
nodes 692 & 671 in combination with five different inception angles. Ten (2 x 5) cases of transformer inrush
current at five different instances in a cycle and two remnant flux were also simulated. From the above operating
conditions 300 test cases (including 200 HIF and 100 non-HIF) are simulated, out of which 70% cases are used for
training and 30% cases are used for testing purpose. Testing is carried out for data set of 90 test cases (including 60
HIF and 30 non-HIF cases), out of 60 HIF cases, 59 cases are classified as HIF and one HIF case is misclassified as
non-HIF. All 30 non-HIF cases are classified correctly.

Evaluation of the proposed method is judged through the performance indices such as dependability, accuracy,
and reliability. Dependability of the system is a predicted number of HIF cases against the actual number of HIF
cases. Security of the system is a predicted number of non-HIF cases against the actual number of non-HIF cases.
Accuracy of the system is a total number of both non-HIF and HIF cases predicted against the total number of actual
non-HIF and HIF cases. Table 1 depicts the evaluation of the proposed method. It is also observed that the time
taken for the complete process is sum of data window length, pre-processing delay, and data mining processing
delay. Thus, the total time delay for HIF detection is 30 ms (20 ms + 5 ms + 5 ms) which is less compared to
previously published work and it is presented in Table 2.

To enhance the validity of the proposed HIF detection strategy a standard IEEE 34 bus system has been
undertaken with the simulation conditions similar to 13 bus system. Though IEEE 34 system is lightly loaded and
 Kavaskar Sekar et al. / Energy Procedia 117 (2017) 417–423 423
Author name / Energy Procedia 00 (2017) 000–000 7

long, the result shows that the exceptional classification rate for HIF as like 13 bus system. Thus, it is to be observed
that loading conditions not affecting the proposed method.

Table 1 Evaluation of the proposed method.

Performance indices Predicted

Accuracy 98.33%
Dependability 98.88%
Security 100%

Table 2 Comparison of detection time.

Method Time
MM in [7] 1 sec
The proposed method 30 ms

8. Conclusions:

A comprehensive data mining model based HIF protection has been proposed and it is extensively tested on
IEEE 13 and 34 test systems with the different system operating conditions. The proposed method is found suitable
for HIF detection with performance indices about 99% and detection time delay of 30ms. Hence, the proposed
method can reliably detect high impedance fault for secured and stable operations in electrical power system.

References

[1]High Impedance Fault Detection Technology, Mar. 1996, Report of PSRC Working Group D15.
[Online]. Available:http://www.pessrc.org/Reports/High_Impedance_Fault_Detection_ Technology.pdf.
[2] Sedighizadeh. M, Rezazadeh. A, Elkalashy. NI, Approaches in high impedance fault
detection-A chronological review, Adv. Elect. Comput. Eng., Vol. 10, No. 3, 2010,pp.114 –128.
[3] Emanuel.A.E, Cyganski .D, Orr. J.A, Shiller. S, and Gulachenski. E.M, High impedance fault arcing on sandy soil in 15 kV distribution
feeders: Contribution to the evaluation of the low frequency spectrum, IEEE Trans. Power Del., vol. 5, no. 2, Apr. 1990, pp. 676–686.
[4] Samantaray. S.R, Ensemble decision trees for high impedance fault detection in power distribution network, Int J Electr Power Energy
Syst.,vol.43, 2012, pp.1048-1055.
[5] Mahari. A, Seyed. H, High impedance fault protection in transmission lines using a WPT-based algorithm, Int. J. Electr. Power Energy Syst.,
vol.67, 2015, pp.537–545.
[6] Baqui. I, Zamora. I, Mazón. J, Buigues. G, High impedance fault detection methodology using wavelet transform and artificial neural
networks, Electr. Power Syst. Res. Vol.81, 2011, pp.1325-1333.
[7] Gautam. S, Brahma. S.M, Detection of High Impedance Fault in Power Distribution Systems Using Mathematical Morphology, IEEE Trans.
on Power Sys., vol.28, 2013, pp.1226-1234.
[8] Valero Masa. A, Werben. S, and Maun. J.C, Incorporation of Data mining in protection technology for high impedance fault detection, In.
Proc. IEEE Power & Energy Society General Meeting 2012, pp. 1–8.
[9] Distribution Test Feeders IEEE Distribution System Analysis Subcommittee [online]. Available http//ewh.ieee.org./soc/pes/dsacom/
testfeeders/index.html.
[10] Gautam. S, Brahma. S.M, Overview of mathematical morphology in power systems - A tutorial approach, In: Proc. IEEE Power & Energy
Society General Meeting 2009, pp. 1–7.
[11] Hastie. T, Tishirani. R, Friedman. J, "The Elements of Statistical Learning: Data Mining, Inference, and Prediction", 2 nd Edition, Springer -
Verlag: Newyork, 2009, pp.745.
[12] Rattle (The R Analytical tool to learn easily) by D.Williams ver. 2.6. Dec 2010 [online]: http://rattle togaware.com/