Decision Tree Supported Distance relay for Fault
Detection and Classification in a series compensated line
Subodh Kumar Mohanty, Member IEEE, Anshudip Karn, Shobhan Banerjee, Student Member IEEE
School of Electrical Engineering, KIIT Deemed to be University, Bhubaneswar, India
Abstract—Fault detection and classification in a series
compensating power transmission line is a big challenge for
protective relays in power system. The proposed method is
based on machine learning algorithm application in distance
protection for fault detection and classification in power
transmission network. In this proposed method one cycle post
fault current data is taken into consideration for fault
classification and detection using decision tree method. This
technique is tested on 400 kV, 50 Hz two area system which is
simulated in PSCAD platform. The post fault current data
extracted from relay point fed in to decision tree algorithm in
machine learning application. The data extracted from
relaying point in a different condition like different distance
with several fault resistance.
Keywords—Fault detection, Transmission Network, Machine
Learning, Decision Tree
I. INTRODUCTION
Power transmission line is the tie line between the
generation end and consumer end. Now-a-days the increased
demand of power and restriction of new building
transmission line creates headache for power system
engineer due to heavy transmission line loading and
increased power losses. To fulfill these requirements series
capacitor is mostly preferred in long transmission system
day by day. By the use of series compensating line, it further
supports the stability of power system, increase of power
transfer capability and voltage profile improvement. But the
inclusion of series capacitor in long transmission line creates
huge protection problem due to several reasons.
a) The current in steady state increases remarkably
compare to the fault current of line to ground at boundary of
the line due to series compensation.
b) The used MOV (metal oxide varistor) is a non linear
resistor for over voltage protection of capacitor during fault
increases complexity problem for protection point of view in
the series compensation arrangement.
c) Current and Voltage Inversion signals
d) The transmission line produced current and voltage
signals containing various type of frequency components.
The various frequency elements consisting non sinusoidal
fundamental deteriorate as well as DC components decaying
due to series capacitor and line inductance, subsynchronous
frequencies having frequency elements varies about the half
of the fundamental component frequency value, MOV
conduction during faults for odd harmonics, elements of
high frequency caused by line inductance and capacitance
resonance and fundamental elements of the fault current
during steady state. Therefore, traditional technique of
digital distance protection, which depends on the calculation
of fundamental current and voltage phasors for series
compensated transmission lines are of little effectiveness.
Fig 1: Current and Voltage waveform during L-G fault
Symmetrical faults, even though occur less, but when
occur are more dangerous than asymmetrical faults and need
to be analyzed under certain circumstances. Again the
current magnitude may be the same for different location of
the transmission line (after and before the series
compensation by TCSC/UPFC) for similar type of fault.
Therefore, fault classification is a big challenging task for
TCSC/UPFC connected series compensated transmission
line. The impedance assessment [1] without the FACTs
device is just like an normal transmission line for the fault
but the impedance assessment must include the impedance
for the FACTs device when the fault relates to FACTs. The
relay gives a trip signal to circuit breaker when the line
impedance compared with the relay setting impedance and if
it is less than a predefined value in the protective region.
In reference [2] for large number of features and
classes in multivariable analysis generally one wants to
determine defined parameters of class distributions like high
dimensional distribution. This problem is avoided by using
decision tree classifier for a smaller number of attributes at
each internal node without degrading the presentation.
References [3]-[4] explain the significance of
Decision Trees in machine learning and application in
Power systems. In [1], the author has compared Support
Vector Machine with Decision Tree to identify fault zone in
TCSC and UPFC lines, and type of fault in which Decision
trees have been concluded superior to Support Vector
Machine.
2020 IEEE International Conference on Power Electronics, Smart Grid and Renewable Energy (PESGRE2020)
978-1-7281-4251-7/20/$31.00 ©2020 IEEE 1