06-02-2013, 04:52 PM
Fault Classification in EHV Transmission Lines Using Artificial Neural Networks
Fault Classification in EHV.pdf (Size: 531.75 KB / Downloads: 53)
ABSTRACT
This paper investigates a new approach based on Artificial
Neural Networks (ANNs) for real-time fault classification in
power transmission lines which can be used in digital power
system protection. The technique uses sampled current and
voltage data of each phase at one terminal as inputs to the
corresponding ANN. The ANN outputs indicate the type of
the fault within a time less than 5 ms. The ANN-based
classifier is tested under different fault types, fault location,
fault resistance and fault inception angle. All the test results
show that the proposed fault classifier can be used for
supporting a new generation of very high speed protective
relaying systems.
I. INTRODUCTION
Reliable detection and isolation of transmission lines fault
is very important for maintaining safe, continued and
economic operation of power systems. The design of high
performance protective techniques is a subject to
significant development within the academic community
and in industry. Various approaches of fault detection and
classification have been proposed in the literature. In
almost all these protective techniques, sampled voltage
and current data at the relaying point are used for fault
recognition.
INPUTS AND OUTPUTS
To evaluate the performance of the proposed neural
network-based FC, a 400 kV, 150 km transmission line
extending between two sources is considered in this
study. The transmission line is represented by distributed
parameters and the frequency dependence of the line
parameters is taken into account. The physical
arrangement of the conductors is resumed in Fig. 2 and
the line characteristics can be found in [14].
NEURAL NETWORK STRUCTURES
The FC tasks can be formulated as a pattern classification
problem. A fully-connected multi layer (input, hidden and
output) feed-forward neural network (FFNN) has been
used to classify faulty/non-faulty data sets. The number of
inputs to the network and the number of neurons in the
input and hidden layers are decided empirically through
extensive simulations. Various network configurations are
trained and tested in order to establish an appropriate
network with satisfactory performance. Performance
criteria are fault tolerance, time response and
generalization capabilities. The three layer FFNN is
selected to implement the algorithm for single ended fault
detection using current and voltage data. Data strings of
four consecutive samples of the current and voltage
signals taken every 1 kHz are found to be appropriate
inputs to the neural network. This represents a moving
window with a length of 3 ms.